Glossary

The Codex

Every field builds a private language; ours insists on decoding it. The full glossary, migrated whole: searchable, linkable, and written for humans.

1631
terms: 591
crypto: 1040
AI + quantum

◈ TERM OF THE DAYEvaluation harnessA framework that runs a model or agent through set test cases and scores its outputs to measure quality and track changes.

#

1inchcrypto: 1inch is a decentralised exchange aggregator that sources liquidity across multiple DEXs, ensuring optimal token-swap prices and minimal slippage.
21Sharescrypto: 21Shares issues exchange-traded crypto products tracking individual assets and diversified baskets under Swiss regulatory supervision.
4-bit quantisationai: Compressing model weights to just four bits each, roughly a quarter of the usual size, so large models fit on smaller or cheaper hardware.

A

Aavecrypto: Aave is a decentralised lending protocol enabling users to supply and borrow crypto assets through over-collateralised smart contracts with algorithmic interest rates.
Abracrypto: Abra is a crypto investment platform offering multi-asset custody, trading, and yield products through regulated custodians and blockchain infrastructure.
Acalacrypto: Acala is a DeFi hub on Polkadot providing stablecoin issuance, liquid staking, and multi-chain financial services via Substrate technology.
Account Abstraction (ERC-4337)crypto: Account abstraction allows smart accounts to execute complex logic, gas sponsorship, and signature flexibility, enhancing user experience across decentralised applications.
Addresscrypto: A blockchain address is a unique alphanumeric identifier derived from a public key, representing where digital assets can be sent securely.
AI agentai: An AI system that does not just answer questions but takes actions — browsing, writing files, calling other software — usually in a loop of planning, acting and checking its own results.
Airdropcrypto: An airdrop distributes free tokens to wallet addresses, usually to reward early supporters, drive engagement, or decentralise token ownership within communities.
Alchemycrypto: Alchemy is a blockchain-development platform providing APIs, analytics, and infrastructure services powering decentralised applications across multiple networks.
Algorand (ALGO)crypto: Algorand is a high-performance, pure proof-of-stake blockchain designed for scalability, security, decentralisation, and low-cost financial or enterprise applications.
Alpine Racing Tokencrypto: Alpine Racing Token is a Formula 1 fan token on Binance Fan Token platform granting exclusive voting rights and rewards.
Altcoincrypto: An altcoin refers to any cryptocurrency other than Bitcoin, often created to introduce novel consensus mechanisms, governance, or niche-specific use cases.
Amber Groupcrypto: Amber Group offers liquidity, derivatives, and treasury management solutions for institutional digital-asset investors globally.
Amberdatacrypto: Amberdata provides institutional-grade blockchain analytics, APIs, and risk-management data across DeFi, derivatives, and centralised-exchange activity.
AMM (Automated Market Maker)crypto: An automated market maker uses mathematical formulas instead of order books to price assets dynamically and facilitate decentralised trading liquidity.
Anchor Protocolcrypto: Anchor Protocol was a Terra-based savings platform offering stable yield on UST deposits backed by staking-based collateral before its collapse.
Anchorage Custodycrypto: Anchorage Custody provides segregated cold-storage vaults and controlled access for regulated institutional crypto-asset safeguarding.
Anchorage Digitalcrypto: Anchorage Digital is a federally chartered US crypto bank providing institutional custody, staking, and settlement services.
Ankrcrypto: Ankr provides decentralised cloud computing, RPC endpoints, and liquid-staking services for developers building multi-chain blockchain applications.
Anomacrypto: Anoma is a privacy-first intent-centric protocol enabling multi-asset atomic settlement and composable encrypted transactions.
API (Application Programming Interface)crypto: An API enables different software systems to communicate, allowing crypto exchanges, wallets, and dApps to interact securely through programmable instructions.
API3crypto: API3 enables decentralised APIs connecting off-chain data providers directly to smart contracts without intermediaries using first-party oracle nodes.
Approvalcrypto: Approval is the on-chain permission allowing a smart contract to transfer or manage tokens from a user’s wallet within defined allowances.
APRcrypto: Annual Percentage Rate measures simple interest over a year, commonly used in DeFi lending, staking, and liquidity programs excluding compounding effects.
APYcrypto: Annual Percentage Yield expresses annualised interest with compounding included, providing a more accurate representation of long-term DeFi yield performance.
Aragoncrypto: Aragon is a DAO-creation platform offering governance modules, treasury management, and on-chain dispute resolution tools for communities.
Arbitragecrypto: Arbitrage exploits temporary price differences of the same asset across exchanges, allowing traders to profit without directional exposure through instantaneous cross-market trades.
Arbitrumcrypto: Arbitrum is an Ethereum layer-two optimistic rollup providing faster transactions and lower fees while maintaining Ethereum-level security.
ARK Investcrypto: ARK Invest allocates capital to disruptive innovation sectors including blockchain, AI, and decentralised finance.
Arkham Intelligencecrypto: Arkham Intelligence links blockchain wallets to real-world identities, offering transparent address-labelling dashboards for compliance, research, and market analytics.
Astariacrypto: Astaria enables NFT-backed lending, issuing loans against digital collectibles using automated appraisals and on-chain escrow mechanisms.
ATH (All-Time High)crypto: All-Time High is the highest recorded price of an asset, reflecting peak market optimism and often preceding profit-taking or volatility surges.
Atomic Swapcrypto: An atomic swap enables cross-chain asset exchange directly between users using cryptographic contracts, eliminating trusted intermediaries or custodial risk.
Attestationcrypto: An attestation is a digitally signed proof verifying data authenticity, identity, or event integrity, frequently used in decentralised identity or oracle systems.
Auditingcrypto: Auditing examines smart contracts or blockchain protocols for vulnerabilities, ensuring code safety, compliance, and adherence to best-practice security standards.
AVS (Actively Validated Service)crypto: Actively Validated Services reuse validator security from existing networks like Ethereum’s, enabling modular staking to secure specialised off-chain or on-chain functions.
Aztec Networkcrypto: Aztec Network builds privacy-enabled DeFi using zk-rollups that encrypt transaction values while retaining on-chain verification.
Alignmentai: The field concerned with making models behave in line with human intentions and values, reliably doing what people actually want.
Ancilla qubitquantum: An auxiliary qubit added to a computation to assist operations such as error detection or intermediate storage, rather than holding the final result.
Attention mechanismai: A method that lets a model weigh the relevance of different input elements when producing each output, focusing on the most useful parts.
Attention headai: One parallel attention computation within a layer, with its own query, key and value projections, capturing a particular pattern of relationships.
Activation functionai: A non-linear function applied to a neuron's output, letting a network learn relationships that a purely linear model could not represent.
Attention with linear biases (ALiBi)ai: A scheme that adds a distance-based penalty to attention scores instead of positional embeddings, favouring nearer tokens and extending to longer sequences.
Autoregressive modelai: A model that generates a sequence one element at a time, conditioning each prediction on the elements already produced.
Autoencoderai: A network trained to compress its input into a compact code and reconstruct it, learning useful representations without labels.
Attention maskai: A pattern that blocks certain positions from attention, used to hide padding tokens or enforce causal, left-to-right generation.
Adamai: A widely used optimiser that adapts each weight's step size using running averages of recent gradients and their squared values.
AdamWai: A variant of Adam that applies weight decay separately from the gradient update, giving cleaner regularisation and better generalisation for large models.
Adagradai: An optimiser that gives smaller updates to frequently changing weights and larger ones to rare features by accumulating squared gradients over time.
Automatic differentiation (autodiff)ai: A technique that computes exact gradients of a program's outputs with respect to its inputs by applying the chain rule mechanically.
Autoregressive generationai: Producing text one token at a time, where each new token is predicted from all the tokens generated so far, then fed back in.
Activation-aware weight quantisation (AWQ)ai: A quantisation method that protects the small fraction of weights most important to activations, keeping accuracy high while compressing the rest to low precision.
Activation steeringai: Nudging a model's behaviour at run time by adding a direction to its internal activations, for example to increase honesty or reduce toxicity.
ARC-AGIai: The Abstraction and Reasoning Corpus, a set of grid puzzles testing whether a system can infer abstract rules from very few examples.
Adversarial exampleai: An input deliberately perturbed, often imperceptibly, to make a model produce a wrong or harmful output while looking normal to humans.
Alignment taxai: The extra cost or capability loss sometimes incurred when making a model safer and more aligned rather than purely maximising raw performance.
Accuracyai: The proportion of predictions a model gets correct, a figure that can mislead when the classes are heavily imbalanced.
Area under the curve (AUC)ai: A single number summarising ROC performance, where 1 is perfect and 0.5 is no better than random chance.
Agentic loopai: The repeating cycle where a model reads a goal, chooses an action, observes the result, and continues until the task is finished or stopped.
Agent orchestrationai: The coordination of multiple agents or tools, deciding what runs, in what order, and how their outputs are combined.
Agent handoffai: When one agent passes control and context of a task to another agent that is better suited to the next step.
Approximate nearest neighbour (ANN)ai: Search that finds vectors close to a query quickly by trading a little accuracy for much greater speed at scale.
Autonomy levelsai: A scale describing how much an agent acts on its own, from suggesting steps for approval to completing tasks without oversight.
AI acceleratorai: Any hardware designed to speed up machine learning workloads, including GPUs, TPUs and custom chips, by processing many operations in parallel.
Autoscalingai: Automatically adding or removing computing resources based on demand, so a service handles traffic spikes without paying for idle capacity.
A/B testingai: Comparing two versions with live users by splitting traffic between them, measuring which performs better before rolling one out fully.
All-reduceai: A collective operation that combines values, such as gradients, from every device and then shares the result back to all of them.
Activation checkpointingai: Saving memory during training by discarding some intermediate results and recomputing them later, trading extra compute for lower memory use.
Application-specific integrated circuit (ASIC)ai: A chip designed for one particular task, offering high efficiency for that job but no flexibility to do anything else.
Ansatzquantum: A chosen circuit structure with tunable parameters used as a starting guess in variational algorithms, balancing expressiveness against trainability and hardware cost.
Anharmonicityquantum: The uneven spacing between a qubit's energy levels that lets control pulses address just two states, preventing accidental leakage into higher levels.
Alkaline-earth atomquantum: An atom such as strontium or ytterbium whose energy levels suit precise laser control, favoured in some neutral-atom and optical-clock qubit systems.
Arbitrary waveform generator (AWG)quantum: Electronics that produce precisely shaped control pulses used to drive qubit gates and trigger readout.
Amplitude amplificationquantum: A generalisation of Grover's search that boosts the probability of measuring desired states, giving a quadratic reduction in the number of oracle calls needed.
Amplitude estimationquantum: A quantum technique that estimates the probability amplitude of a marked outcome, achieving quadratically better precision than classical sampling for tasks like Monte Carlo integration.
Amplitude encodingquantum: A way of loading classical data into a quantum state by storing values in its amplitudes, packing many numbers into few qubits but hard to prepare.
Adiabatic quantum computationquantum: A model that solves problems by slowly evolving a system from an easy starting state to one whose ground state encodes the answer.
Analog quantum simulationquantum: A method where one engineered quantum system is tuned so its natural dynamics directly mimic those of the target system, without breaking evolution into gates.
Adiabatic theoremquantum: The principle that a quantum system stays in its lowest-energy state if the governing Hamiltonian is changed slowly enough.
Atomic clockquantum: A timekeeping device that counts the extremely stable oscillations of atoms between energy levels, providing the most accurate time and frequency standards.
Atom interferometryquantum: A measurement technique that splits and recombines atomic waves so their interference reveals accelerations, rotations and gravity with high sensitivity.
Allan deviationquantum: A statistical measure of how a clock or oscillator's stability changes with averaging time, the standard way to characterise atomic clock performance.
Average gate fidelityquantum: The gate fidelity averaged over all possible input states, giving a single figure that summarises a gate's overall accuracy.
Algorithmic qubits (AQ)quantum: A benchmark estimating how many usable qubits a device offers for real algorithms, factoring in error rates and how qubits connect.
Amplitude dampingquantum: A noise process describing energy loss from a qubit, such as decay from the excited to the ground state, characterised by T1.

B

Balancercrypto: Balancer is an automated market maker allowing multi-token pools with custom weights, functioning as self-balancing index funds and liquidity engines.
Basecrypto: Base is Coinbase’s layer-two built on Optimism’s OP Stack, offering secure, low-cost on-chain applications integrated with Coinbase’s ecosystem.
Base Feecrypto: Base fee is the minimum gas amount per transaction introduced by EIP-1559, dynamically adjusting according to network demand and congestion.
Basis (Perps)crypto: Basis is the price difference between perpetual futures and spot markets, reflecting funding-rate dynamics and traders’ sentiment on future price direction.
Batch Auctioncrypto: Batch auctions execute orders simultaneously at a uniform clearing price, reducing front-running, slippage, and MEV exploitation within decentralised exchanges.
Beamcrypto: Beam is a privacy-centric blockchain using MimbleWimble technology for confidential transactions and scalable private payments.
Beam Privacycrypto: Beam is a MimbleWimble-based blockchain ensuring confidential assets, signatures, and atomic-swap support for private transfers.
Bear Marketcrypto: A bear market describes a prolonged period of declining asset prices, low investor confidence, and increased caution within financial or crypto markets.
Biconomycrypto: Biconomy provides gasless transaction infrastructure and multi-chain relayer APIs simplifying blockchain onboarding for decentralised applications.
Binance Coin (BNB)crypto: Binance Coin powers the Binance ecosystem, paying transaction fees, participating in governance, and supporting multiple DeFi and exchange functionalities.
BIP (Bitcoin Improvement Proposal)crypto: A Bitcoin Improvement Proposal is a formal document suggesting changes to Bitcoin’s protocol, network rules, or development processes.
Bitcoin (BTC)crypto: Bitcoin is the first decentralised cryptocurrency, using proof-of-work consensus and limited supply to enable borderless, censorship-resistant digital value transfer.
Bitcoin Cash (BCH)crypto: Bitcoin Cash forked from Bitcoin to increase block size, improving transaction throughput and reducing fees for peer-to-peer electronic payments.
Bitfinexcrypto: Bitfinex is one of the oldest crypto exchanges, offering advanced trading tools, derivatives, and staking products for professional traders.
BitGocrypto: BitGo offers institutional crypto custody with multi-signature wallets, insurance coverage, and regulatory-compliant asset-management solutions.
BitGo Primecrypto: BitGo Prime extends institutional services including trading, lending, and qualified custody under regulated frameworks.
BitGo Trustcrypto: BitGo Trust is a regulated US trust company offering qualified custody and fiduciary digital-asset management services.
BitPaycrypto: BitPay provides merchant-payment gateways for cryptocurrencies, converting digital payments to fiat while maintaining accounting transparency.
Bitquerycrypto: Bitquery supplies blockchain data APIs enabling developers to query transactions, smart-contract logs, and liquidity information programmatically.
Bitstampcrypto: Bitstamp is a regulated European exchange providing fiat-crypto trading pairs, staking, and secure custody services since 2011.
Bitwise Investcrypto: Bitwise Invest offers index funds and ETFs tracking diversified crypto baskets for professional investors.
BlackRock BUIDLcrypto: BlackRock’s BUIDL fund tokenises short-term US-Treasury exposure on Ethereum for institutional investors seeking on-chain yield.
Blockcrypto: A block groups verified transactions, cryptographically linked to previous ones, forming the immutable chronological record known as the blockchain.
Block Explorercrypto: A block explorer is a web tool allowing users to view blockchain data such as transactions, addresses, block heights, and confirmations.
Block Heightcrypto: Block height counts the number of blocks preceding a given block, serving as a measure of blockchain length and network progress.
Block Rewardcrypto: A block reward is cryptocurrency granted to miners or validators for producing new blocks, securing the network, and processing transactions.
Blockchaincrypto: A blockchain is a distributed ledger maintained by multiple nodes, ensuring transparency, immutability, and consensus without centralised control or intermediaries.
Blockdaemoncrypto: Blockdaemon provides institutional blockchain infrastructure, staking, and node-as-a-service solutions across multiple proof-of-stake networks.
BlockFicrypto: BlockFi was a centralised lending platform offering interest accounts and collateralised loans before filing for bankruptcy in 2022.
Blocknativecrypto: Blocknative delivers real-time mempool monitoring tools helping traders and dApp developers anticipate pending transactions and MEV activity.
Blockstreamcrypto: Blockstream develops Bitcoin infrastructure including the Liquid sidechain, satellite network, and cryptographic research for financial scalability.
Bloomcrypto: Bloom is a decentralised credit-scoring protocol allowing users to control identity data while sharing verified financial reputation.
BLS Signaturecrypto: BLS signatures aggregate multiple cryptographic signatures efficiently, enabling scalable consensus validation in proof-of-stake and threshold-based multi-party systems.
Blurcrypto: Blur is an NFT marketplace focused on professional traders, aggregating listings and rewarding activity through the BLUR governance token.
Bond Protocolcrypto: Bond Protocol enables tokenised bond markets on-chain, issuing debt instruments with transparent yields and composable liquidity.
Bonding Curvecrypto: A bonding curve mathematically defines a token’s price in relation to its supply, used in DAOs, token launches, and liquidity design.
Bottlepaycrypto: Bottlepay facilitates social Bitcoin payments via the Lightning Network, enabling instant tipping across messaging platforms.
Braintrustcrypto: Braintrust is a decentralised talent network connecting freelancers with enterprises, governed by community incentives via BTRST token.
BRC-20crypto: BRC-20 is an experimental token standard on Bitcoin using ordinal inscriptions, enabling fungible token issuance via embedded on-chain metadata.
Breez Walletcrypto: Breez Wallet offers non-custodial Lightning payments and podcast streaming-sats functionality for Bitcoin microtransactions.
Bridgecrypto: A bridge transfers assets or data between blockchains, facilitating interoperability but introducing security risks like smart-contract exploits or custodial compromise.
BTC Ordinalscrypto: Ordinals inscribe data directly onto satoshis, creating on-chain NFTs and fungible experiments within Bitcoin’s UTXO model.
BTCPay Servercrypto: BTCPay Server is an open-source Bitcoin payment processor allowing businesses to accept payments directly without intermediaries.
Builder (MEV)crypto: A builder in MEV systems constructs optimal block bundles for proposers, balancing inclusion incentives, latency, and extraction opportunities across markets.
Bull Marketcrypto: A bull market features sustained asset-price growth, strong investor optimism, high liquidity, and widespread belief in long-term upward momentum.
Burningcrypto: Burning permanently removes tokens from circulation, reducing supply and potentially increasing scarcity or value through verifiable on-chain destruction events.
Bybitcrypto: Bybit is a derivatives exchange offering perpetual contracts, options, and spot trading with advanced risk controls and high liquidity.
Benchmarkai: A standardised test used to measure and compare model performance on tasks such as reasoning, coding or knowledge, producing a comparable score.
Batch normalisation (BatchNorm)ai: Normalisation that standardises each feature using the mean and variance across a training batch, stabilising and speeding up learning.
Backpropagationai: The algorithm that works out how much each weight contributed to the loss by propagating error signals backwards through the network layers.
Batch sizeai: The number of training examples processed together before the model updates its weights; it affects speed, memory use and stability.
Bias-variance tradeoffai: The balance between a model too simple to fit the data and one so flexible it fits noise, each hurting accuracy differently.
Byte pair encoding (BPE)ai: A tokenisation method that starts from characters and repeatedly merges the most frequent adjacent pairs, building a vocabulary of common subword units.
Byte-level BPEai: A variant of byte pair encoding that operates on raw bytes rather than characters, letting a small vocabulary encode any text, including emoji and symbols.
Beam searchai: A decoding method that keeps several candidate sequences in parallel at each step, expanding the most promising and finally choosing the highest-scoring whole sequence.
Benchmark contaminationai: When test questions or answers leak into a model's training data, inflating its benchmark scores and making them a poor guide to real ability.
Bag-of-wordsai: A text representation that counts word occurrences while ignoring grammar and order, treating a document as an unordered collection.
BLEU (bilingual evaluation understudy)ai: A metric scoring machine translation by comparing overlapping word sequences between the machine output and human reference translations.
Bayes' theoremai: A rule for updating the probability of a hypothesis given new evidence, foundational to probabilistic machine learning.
BM25ai: A widely used keyword ranking formula that scores documents by term frequency and rarity to judge how relevant they are.
Browser automationai: Driving a web browser through software so an agent can navigate pages, click links and fill forms to complete online tasks.
Batchingai: Grouping several requests together so a model processes them in one pass, improving hardware use and overall throughput.
bfloat16 (BF16)ai: A 16-bit number format that keeps the wide range of 32-bit floats with less precision, popular for training neural networks efficiently.
Born rulequantum: The principle that a measurement outcome's probability equals the squared magnitude of its probability amplitude within the quantum state.
Bloch spherequantum: A geometric picture representing any single qubit's pure state as a point on the surface of a unit sphere.
Braquantum: Dirac notation for the conjugate transpose of a ket, combined with kets to form inner products and probabilities.
Bra-ket notation (Dirac notation)quantum: A compact notation for quantum mechanics using bras and kets to write states, inner products and operators clearly.
Basis statequantum: One of a chosen set of reference states from which every state in the space can be built by combination.
Bell statequantum: One of four maximally entangled two-qubit states forming the basis for teleportation and many quantum communication protocols.
Bell basisquantum: The set of four Bell states, which together form an orthonormal basis for measurements on two entangled qubits.
Bell inequalityquantum: A bound on correlations achievable by any local classical theory, which entangled quantum states can measurably violate.
Basis gatequantum: One of the elementary gates a compiler targets for a particular device; every higher-level gate is rewritten using only these.
Barrierquantum: A compiler instruction that prevents gates on either side from being reordered or merged across it, preserving circuit structure during optimisation.
Beam splitterquantum: An optical component that partially transmits and partially reflects light; a basic building block for combining and interfering photonic qubits.
Boson samplingquantum: A photonic task where many photons interfere through an optical network; sampling the outputs is hard for classical computers, showing quantum advantage.
Braidingquantum: Swapping anyons around one another to perform operations that depend only on the pattern of moves, the basis of topological quantum computing.
Bosonic codequantum: An error-correction scheme storing a qubit in the many energy levels of a single oscillator, such as a microwave cavity, rather than several separate qubits.
Bit-flip errorquantum: An error that swaps a qubit's 0 and 1 amplitudes, equivalent to a Pauli X operator, the quantum analogue of a classical bit flip.
Below thresholdquantum: Operating a quantum processor at physical error rates low enough that increasing the code distance genuinely lowers the logical error rate.
Break-evenquantum: The point at which an encoded logical qubit outlives its best physical qubit, showing that error correction is providing a net benefit.
Belief-propagation decoderquantum: A decoding method, common for qLDPC codes, that iteratively passes probability estimates across the code to infer the most likely errors.
Bacon-Shor codequantum: A subsystem code arranged on a square lattice that simplifies error correction by requiring only two-qubit gauge measurements.
Bivariate bicycle codequantum: A recent qLDPC code family offering many protected logical qubits with modest overhead and connectivity, studied as a surface-code alternative.
Bernstein-Vazirani algorithmquantum: A quantum algorithm that recovers a hidden binary string encoded in a linear function with one query, whereas classical approaches require one query per bit.
Black-box modelquantum: A framework where an algorithm accesses a function only through queries, without seeing its internal workings, used to prove many quantum speedups over classical computation.
Bounded-error quantum polynomial time (BQP)quantum: The class of decision problems a quantum computer can solve efficiently with high probability. It captures what is considered tractable for quantum machines.
Basis encodingquantum: A method of representing classical data as computational basis states of qubits, using one qubit per bit so that binary values map directly onto qubit states.
Barren plateauquantum: A training difficulty in variational quantum algorithms where the cost landscape becomes almost flat, making gradients vanish and optimisation extremely hard as circuits grow.
Block encodingquantum: A way of embedding a general matrix inside a larger unitary so a quantum computer can manipulate it, a key ingredient in modern linear-algebra algorithms.
BB84quantum: The first quantum key distribution protocol, devised by Bennett and Brassard in 1984, encoding key bits on single photons using two randomly chosen measurement bases.
Bell testquantum: An experiment measuring correlations between entangled particles to confirm they violate a Bell inequality, proving genuine entanglement rather than mere classical correlation.
Bravyi-Kitaev transformationquantum: An alternative mapping of fermionic systems onto qubits that often needs fewer qubit operations than the Jordan-Wigner method.
Basis setquantum: A collection of mathematical functions used to represent electron orbitals in a calculation; larger sets give more accuracy at greater cost.
Born-Oppenheimer approximationquantum: A simplification that treats the heavy atomic nuclei as fixed while calculating the behaviour of the much lighter, faster electrons.

C

Cantocrypto: Canto is a layer-one blockchain providing free public infrastructure for DeFi primitives like DEXs, lending, and stablecoins.
Cardano (ADA)crypto: Cardano is a proof-of-stake blockchain emphasising formal verification, peer-reviewed research, and modular development through its layered protocol architecture.
Cartesicrypto: Cartesi integrates Linux-based off-chain computation for smart contracts, enabling scalable dApp logic in familiar programming environments.
Casacrypto: Casa supplies multisignature Bitcoin-custody solutions focusing on personal key management, inheritance, and security redundancy.
CCIP (Cross-Chain Interoperability Protocol)crypto: CCIP by Chainlink standardises secure cross-chain token and message transfers, enabling interoperability among distinct blockchains and off-chain systems safely.
CDP (Collateralised Debt Position)crypto: A CDP locks collateral to mint a stable asset, maintaining system solvency through liquidation ratios and decentralised governance parameters.
Celestia (DA Layer)crypto: Celestia is a modular blockchain providing data availability for rollups, separating consensus from execution to scale decentralised ecosystems efficiently.
Celestia Bridgecrypto: Celestia Bridge connects modular rollups to its data-availability layer, enabling scalable settlement without shared execution environments.
Celsius Networkcrypto: Celsius Network offered yield-bearing accounts and loans before bankruptcy, highlighting counterparty-risk lessons in centralised finance.
Centralised Exchange (CEX)crypto: A centralised exchange is a custodial trading platform matching buyers and sellers internally while retaining control over users’ private keys.
CertiKcrypto: CertiK audits smart contracts and blockchain codebases, offering real-time monitoring and formal verification for enhanced decentralised-application security.
Chainalysiscrypto: Chainalysis supplies blockchain-forensics software helping governments and exchanges trace illicit funds, ensure compliance, and investigate cyber-crime networks globally.
Chainlink (LINK)crypto: Chainlink is a decentralised oracle network delivering reliable real-world data to smart contracts across blockchains, enhancing automation and interoperability.
Chainstackcrypto: Chainstack offers managed blockchain nodes, APIs, and analytics for enterprises deploying multi-chain decentralised applications.
Chaos Labscrypto: Chaos Labs offers on-chain risk-simulation platforms enabling DeFi projects to stress-test protocols and optimise collateral settings dynamically.
Chilizcrypto: Chiliz powers sports-fan tokens and engagement platforms through its blockchain, enabling voting rights and rewards for supporters worldwide.
CID (Content Identifier)crypto: A CID uniquely represents content in IPFS by its cryptographic hash, ensuring data integrity, addressability, and decentralised file verification.
CipherTracecrypto: CipherTrace, acquired by Mastercard, tracks cryptocurrency transactions for anti-money-laundering compliance, counter-terrorist-financing, and blockchain-intelligence insights.
Circuit Breakercrypto: Circuit Breaker mechanisms pause smart-contract activity during anomalies or exploits, protecting funds and maintaining market integrity.
Citadel Securities Cryptocrypto: Citadel Securities Crypto explores liquidity provision and market-making infrastructure for tokenised assets and blockchain securities.
Citi Token Servicescrypto: Citi Token Services tokenises deposits and trade-finance instruments for faster cross-border settlements using private blockchains.
Civiccrypto: Civic provides decentralised identity verification and KYC solutions using blockchain-anchored credentials controlled by individual users.
Civic Passcrypto: Civic Pass is a reusable KYC credential allowing users to access regulated DeFi protocols while preserving privacy through zero-knowledge proofs.
CLMM (Concentrated Liquidity Market Maker)crypto: A concentrated liquidity market maker allows liquidity providers to allocate funds within specific price ranges, improving efficiency and fee returns.
Code4renacrypto: Code4rena crowdsources smart-contract audits through competitive review contests, rewarding auditors for identifying high-impact security issues.
Coin Metricscrypto: Coin Metrics publishes reference data, indexes, and network-health metrics for institutional crypto-asset analysis and benchmarking.
Coinbasecrypto: Coinbase is a publicly listed crypto exchange offering custody, trading, and staking with regulatory compliance and insurance safeguards.
Coincovercrypto: Coincover offers crypto-insurance and recovery services safeguarding digital-asset holders against theft, loss, and key failure.
Coinfirmcrypto: Coinfirm is a RegTech platform analysing blockchain transactions for AML, CFT, and sanctions screening across thousands of tokens and protocols.
CoinGeckocrypto: CoinGecko tracks cryptocurrency prices, market capitalisations, and analytics across thousands of tokens and exchanges globally.
CoinMarketCapcrypto: CoinMarketCap is a data-aggregator platform ranking cryptocurrencies by price, volume, and supply for retail and institutional users.
CoinRoutescrypto: CoinRoutes provides algorithmic-execution and smart-order-routing technology optimising crypto trade performance across exchanges.
CoinSharescrypto: CoinShares is a European digital-asset investment firm offering exchange-traded products, research, and institutional trading solutions.
CoinShares ETPcrypto: CoinShares ETP provides physically backed crypto exchange-traded products listed on European markets.
Cold Walletcrypto: A cold wallet stores cryptocurrencies offline, protecting private keys from online threats like phishing or malware, ideal for long-term holdings.
Collateral Factorcrypto: Collateral factor defines how much can be borrowed relative to collateral value in lending protocols, balancing capital efficiency and risk management.
Composabilitycrypto: Composability lets decentralised applications interact seamlessly, enabling protocols to integrate like modular ‘money legos’ for complex DeFi systems.
Compoundcrypto: Compound is a decentralised lending protocol enabling algorithmic interest-rate markets for collateralised borrowing and lending.
Consensus Mechanismcrypto: Consensus mechanisms determine how distributed nodes agree on blockchain state, with models like proof-of-work, proof-of-stake, or Byzantine-fault-tolerant protocols.
Convex Financecrypto: Convex Finance optimises Curve liquidity rewards, allowing CRV holders to earn boosted yields without locking tokens individually.
Cosmos Hubcrypto: Cosmos Hub connects independent blockchains through the IBC protocol, facilitating cross-chain communication within the Cosmos ecosystem.
Credixcrypto: Credix provides tokenised credit markets bridging institutional investors with real-world asset lending in emerging economies.
Cross-Margincrypto: Cross-margin pools collateral across positions on an exchange, reducing liquidation risk but exposing total balance if losses exceed thresholds.
Crossmintcrypto: Crossmint offers NFT-minting infrastructure with credit-card payments, abstracting wallet complexity for mainstream adoption.
Crypto-Native Riskcrypto: Crypto-native risk refers to vulnerabilities inherent to blockchain systems—smart-contract bugs, oracle manipulation, governance failure, or economic-design flaws.
CryptoQuantcrypto: CryptoQuant offers exchange-flow and miner-behaviour analytics, tracking liquidity and on-chain movements influencing Bitcoin and Ethereum markets.
Crystal Blockchaincrypto: Crystal Blockchain provides visual investigation tools and risk scores for digital-asset tracing across multiple chains and exchanges.
Curve Financecrypto: Curve Finance is an automated market maker optimised for low-slippage swaps between stablecoins or pegged assets.
Custodia Bankcrypto: Custodia Bank bridges digital assets and US banking, offering compliant custody and payments infrastructure for institutional clients.
Custodial Walletcrypto: A custodial wallet is managed by a third party holding users’ private keys, simplifying recovery but introducing counterparty trust and risk.
Context windowai: The maximum amount of text, measured in tokens, that a model can consider at once, covering both the input and its generated response.
Chain-of-thoughtai: A prompting approach that asks a model to set out its reasoning step by step, often improving accuracy on complex problems.
Coherence timequantum: The average duration over which a qubit keeps its quantum state usably intact before decoherence corrupts the stored information.
Cat qubitquantum: A qubit encoded in superposed coherent states of a resonator, named after Schrödinger's cat, engineered to strongly suppress bit-flip errors.
Cross-attentionai: Attention where one sequence attends to a different sequence, letting a decoder draw on an encoder's representations, for example in translation.
Causal maskingai: A mask that prevents a token from attending to later positions, so predictions depend only on earlier tokens during generation.
Convolutional neural network (CNN)ai: A network that applies learned filters sliding across data to detect local patterns, widely used for images and other grid data.
Convolutional layerai: A layer that slides small learned filters over its input to produce feature maps highlighting local patterns such as edges.
Convolution kernelai: The small grid of learned weights, also called a filter, that a convolutional layer slides across its input to detect features.
Cross-entropy lossai: A loss function for classification and language models that penalises predictions according to how little probability they assign to the true answer.
Cosine annealingai: A learning-rate schedule that lowers the rate following a cosine curve, decaying smoothly from its peak towards zero across training.
Cross-validationai: An evaluation method that repeatedly splits data into training and validation parts, averaging the results to judge a model more reliably.
Contrastive learningai: A self-supervised method that teaches a model to pull similar examples together and push dissimilar ones apart in representation space.
Curriculum learningai: A training strategy that presents easier examples first and harder ones later, mirroring how people learn progressively from simple to complex.
Catastrophic forgettingai: The tendency of a neural network to lose previously learned abilities when trained on new tasks, overwriting its older knowledge.
Continual learningai: Training a model on a stream of new tasks over time while trying to retain performance on everything it learned earlier.
Checkpointingai: Periodically saving a model's weights and training state to disk, allowing recovery after failure and comparison of different training stages.
Completion tokensai: The tokens a model generates in response to a prompt, also called output or response tokens; typically priced higher than input tokens.
Constrained decodingai: Forcing generated tokens to follow a required pattern, such as valid JSON or a fixed grammar, by blocking any token that would break the rules.
Continuous batchingai: A serving technique that adds and removes requests from a running batch as they arrive and finish, keeping the hardware busy and improving throughput.
Context cachingai: Storing the processed representation of a repeated prompt prefix so later requests reuse it, cutting cost and the time to first token.
Consistency modelai: A generative model trained to jump from noise to a finished sample in one or a few steps, greatly speeding up diffusion-style generation.
Classifier-free guidance (CFG)ai: A technique that steers a diffusion model towards the prompt by blending its conditioned and unconditioned predictions, trading variety for stronger prompt adherence.
ControlNetai: An add-on that conditions image generation on a structural input such as an edge map, depth map or pose, giving precise layout control.
Contrastive Language-Image Pre-training (CLIP)ai: A model trained to match images with their text descriptions, widely used to steer and to evaluate image generators.
Content provenanceai: Verifiable information about how a piece of media was created and edited, used to judge whether it is authentic or AI-generated.
Coalition for Content Provenance and Authenticity (C2PA)ai: An open standard for attaching tamper-evident origin and edit history to media files, helping people trace where content came from.
Constitutional AIai: An alignment approach where a model critiques and revises its own responses against a written set of principles, reducing reliance on human harm labels.
Content moderationai: Automated or human review that detects and filters content violating policy, such as hate speech, harassment or illegal material.
Corrigibilityai: A safety property describing an AI that accepts correction, redirection and shutdown by humans without resisting or trying to prevent it.
Circuitai: In interpretability, a small subset of a network's components that work together to carry out a specific, identifiable computation.
Canary stringai: A unique text marker embedded in a benchmark so testers can later check whether it was accidentally included in a model's training data.
Calibrationai: How well a model's stated confidence matches how often it is actually correct; a calibrated model is neither overconfident nor underconfident.
Capability elicitationai: Techniques such as careful prompting and fine-tuning used to draw out a model's true maximum ability, so evaluations do not underestimate risk.
Classificationai: A supervised task that assigns inputs to discrete categories, such as labelling an email as spam or not spam.
Clusteringai: An unsupervised task that groups similar data points together, so items in one group resemble each other more than those in others.
Confusion matrixai: A table showing counts of true and false positives and negatives, summarising where a classifier gets things right or wrong.
Context engineeringai: The practice of deciding what information, instructions and examples to place in a model's context window to get the best results.
Context compressionai: Shortening or summarising material before adding it to the context window so more relevant information fits within the token limit.
Chunkingai: Splitting documents into smaller passages so each piece can be embedded, indexed and retrieved on its own during search.
Chunk overlapai: Repeating some text between neighbouring chunks so meaning that spans a boundary is not lost when passages are retrieved.
Cross-encoderai: A model that reads a query and a document together to score their relevance, more accurate but slower than comparing separate embeddings.
Cosine similarityai: A measure of how alike two vectors are, based on the angle between them, widely used to compare embeddings.
Computer useai: A capability that lets a model operate a computer by reading the screen and controlling the mouse and keyboard like a person.
CUDAai: Nvidia's software platform for programming its GPUs, widely used in AI because most machine learning frameworks are built to run on it.
Compute clusterai: A group of connected servers, usually packed with GPUs, that work together as one system to train or serve large models.
Checkpointai: A saved snapshot of a model's weights during training, letting work resume after a failure or a chosen version be reused later.
CI/CD for modelsai: Automated pipelines that test, validate and deploy models whenever code or data changes, so updates ship safely and repeatedly without manual steps.
Cost per tokenai: A common pricing and budgeting measure for language models, based on the number of tokens processed in a request and its reply.
Carbon footprint of trainingai: The greenhouse gas emissions produced by the electricity used to train a model, a growing concern as models grow larger.
Canary deploymentai: Releasing a change to a small share of users first, watching for problems before rolling it out to everyone.
Concept driftai: A change over time in the real-world relationship a model learned, causing its predictions to grow less accurate after deployment.
Cold startai: The extra delay when a service must load a model or spin up resources before it can answer the first request.
Constructive interferencequantum: The reinforcement of quantum amplitudes that share the same phase, raising the probability of the corresponding measurement outcome.
Computational basisquantum: The standard reference states labelled 0 and 1, used to define a qubit and to report its measurement results.
CHSH inequalityquantum: A specific Bell inequality whose experimental violation demonstrates entanglement and rules out local hidden-variable explanations.
Controlled-NOT gate (CNOT)quantum: A two-qubit gate that flips the target qubit only when the control qubit is one; it is a fundamental building block for entanglement.
Controlled-Z gate (CZ)quantum: A two-qubit gate that applies a sign flip when both qubits are one, treating both symmetrically; it is Clifford and entangling.
Controlled-phase gate (CPHASE)quantum: A two-qubit gate that applies a chosen phase to the one-one state, generalising the controlled-Z gate to any rotation angle.
Controlled gatequantum: Any gate whose operation on a target qubit is applied only when one or more control qubits are in the one state.
Controlled rotation gatequantum: A two-qubit gate that rotates the target qubit by a set angle only when the control qubit is in the one state.
Cross-resonance gatequantum: A two-qubit entangling operation used in fixed-frequency superconducting qubits, driving one qubit at the resonance frequency of a neighbouring qubit.
Clifford gatesquantum: A family of gates including Hadamard, phase and CNOT that are efficient to simulate classically but cannot achieve universal computation on their own.
Circuit depthquantum: The number of sequential gate layers in a quantum circuit, roughly the longest path from input to output and a measure of runtime.
Circuit widthquantum: The number of qubits a quantum circuit uses, including any ancilla qubits, indicating how much hardware the computation requires.
Coupling mapquantum: A description of which pairs of physical qubits on a processor can directly perform a two-qubit gate together.
Circuit optimisationquantum: Rewriting a quantum circuit to reduce its depth, gate count or error, while keeping the overall operation it performs unchanged.
Charge qubitquantum: An early superconducting qubit encoding information in the presence or absence of a single Cooper pair on a small island; highly sensitive to noise.
Cooper pairquantum: Two electrons bound together that move without resistance in a superconductor; the flow of these pairs underlies how superconducting qubits behave.
Circuit quantum electrodynamics (circuit QED)quantum: The framework describing superconducting qubits coupled to microwave resonators, used for control and readout in most superconducting quantum processors.
Coplanar waveguide resonatorquantum: A flat, on-chip microwave circuit that stores and guides signals, used to read out and couple superconducting qubits together.
Cluster statequantum: A large, specially entangled group of qubits prepared in advance that acts as the resource for measurement-based quantum computing.
Colour centrequantum: A light-absorbing defect in a crystal, such as diamond or silicon carbide, whose spin can serve as a qubit or single-photon source.
Cavity quantum electrodynamics (cavity QED)quantum: The study of atoms or artificial atoms interacting strongly with light trapped in a cavity, underpinning many qubit control and readout schemes.
Cryostatquantum: The apparatus that cools quantum hardware to very low temperatures and maintains that cold environment steadily during operation.
Control electronicsquantum: The room-temperature hardware that generates signals, times pulses and records measurements to operate a quantum processor.
Code distance (d)quantum: The smallest number of physical qubit errors that can turn one valid logical state into another; larger distances correct more errors.
Colour codequantum: A topological stabiliser code on a three-coloured lattice that allows more gates to be done transversally than the surface code, at higher connectivity cost.
CSS code (Calderbank-Shor-Steane)quantum: A family of stabiliser codes built from two classical codes, handling bit-flip and phase-flip errors separately for simpler fault-tolerant constructions.
Code ratequantum: The ratio of logical qubits to physical qubits in a code; higher rates mean less overhead but usually weaker error protection.
Concatenated codequantum: A code built by encoding qubits repeatedly inside other codes, so each level suppresses errors further and helps prove the threshold theorem.
Clifford gatequantum: A gate that maps Pauli operators to Pauli operators; Clifford operations are efficient to simulate classically and are often protected transversally.
Correlated errorsquantum: Errors that strike several qubits together rather than independently, undermining the assumptions of standard codes and demanding careful hardware design.
Coherent errorquantum: A small, systematic over-rotation or miscalibration that accumulates predictably, often more damaging than random noise of the same size.
Cross-entropy benchmarking (XEB)quantum: A statistical test comparing a quantum processor's output distribution against ideal predictions, used to verify performance and support quantum supremacy claims in random circuit sampling.
Collision problemquantum: Finding two inputs that a function maps to the same output. Quantum algorithms provide a speedup relevant to assessing the security of hash functions.
Continuous-time quantum walkquantum: A quantum walk defined by continuous evolution under a Hamiltonian on a graph, used to design algorithms with speedups on certain search and traversal problems.
Continuous-variable QKD (CV-QKD)quantum: Quantum key distribution that encodes keys in the amplitude and phase of laser light, read out with homodyne detection using standard telecoms components.
CRYSTALS-Kyberquantum: A lattice-based key encapsulation mechanism selected by NIST for post-quantum encryption and subsequently standardised as ML-KEM in FIPS 203.
CRYSTALS-Dilithiumquantum: A lattice-based digital signature scheme selected by NIST for post-quantum use and subsequently standardised as ML-DSA in FIPS 204.
Code-based cryptographyquantum: A family of post-quantum schemes whose security relies on the difficulty of decoding a general linear error-correcting code, as used in McEliece.
Crypto-agilityquantum: The capacity of a system to switch cryptographic algorithms quickly and cheaply, essential for migrating smoothly to post-quantum standards without disruption.
Cryptographically relevant quantum computer (CRQC)quantum: A quantum computer large and reliable enough to break today's widely used public-key cryptography, such as RSA and elliptic-curve schemes.
Caesium fountain clockquantum: A microwave clock that tosses cooled caesium atoms upward and measures them as they fall, defining the international standard for the second.
Circuit Layer Operations Per Second (CLOPS)quantum: A speed benchmark measuring how many layers of quantum gates a processor can run each second, capturing throughput rather than qubit quality.
Crosstalkquantum: Unwanted interaction where operating on or measuring one qubit disturbs its neighbours, a common source of correlated errors.
Correlated errorquantum: An error affecting several qubits together rather than independently, which is harder for error correction to handle than isolated faults.
Control pulsequantum: A precisely shaped signal, often microwave or laser, sent to a qubit to perform a specific gate or measurement.
Cirqquantum: An open-source Python framework from Google for writing, optimising and running quantum circuits, designed with near-term noisy hardware in mind.

D

DA (Data Availability)crypto: Data availability ensures all necessary transaction data is accessible for verification, preventing fraudulent rollup states and enhancing scalability layers.
DAIcrypto: DAI is a decentralised stablecoin soft-pegged to the US dollar, collateralised by on-chain assets within the Maker Protocol.
Dankshardingcrypto: Danksharding is an Ethereum scaling upgrade introducing blobs for cheaper rollup data posting, unifying proposers and builders for efficiency.
DAO (Decentralised Autonomous Organisation)crypto: A DAO coordinates governance through on-chain voting and smart contracts, enabling community-driven resource management without traditional hierarchies.
DApp (Decentralised Application)crypto: A decentralised application runs on blockchain infrastructure rather than central servers, offering transparency, censorship-resistance, and user-controlled data interactions.
DappRadarcrypto: DappRadar tracks usage metrics and token volumes across decentralised applications, highlighting trends and market-share statistics for investors.
DeBankcrypto: DeBank aggregates DeFi wallet portfolios, lending positions, and token approvals, offering comprehensive multi-chain user analytics dashboards.
Decentralandcrypto: Decentraland is a virtual-reality metaverse where users buy land parcels, build experiences, and monetise content through the MANA token.
Decentralised Exchange (DEX)crypto: A DEX enables peer-to-peer asset trading via smart contracts, removing intermediaries and enhancing transparency but requiring liquidity-pool incentives.
DeFi (Decentralised Finance)crypto: DeFi encompasses financial services built on blockchains, replacing intermediaries with smart contracts for lending, trading, and yield generation.
DefiLlamacrypto: DefiLlama compiles total-value-locked data across chains and protocols, visualising liquidity flows and comparing DeFi ecosystem growth.
Delegated Stakingcrypto: Delegated staking allows token holders to assign validation rights to trusted nodes, earning rewards while avoiding validator infrastructure management.
Delphi Labscrypto: Delphi Labs incubates DeFi and gaming projects, providing tokenomics design, audits, and venture funding from Delphi Digital.
Deribitcrypto: Deribit is the leading crypto-options exchange offering institutional-grade derivatives trading and volatility analytics.
Derivativescrypto: Derivatives are financial instruments deriving value from underlying assets, used in crypto for hedging, speculation, and leveraged exposure management.
DEX Aggregatorcrypto: A DEX aggregator routes orders through multiple exchanges to achieve optimal pricing, minimising slippage and improving capital efficiency.
DEXToolscrypto: DEXTools provides analytics for decentralised exchanges, tracking token prices, liquidity, and wallet activity in real time.
Digital Signaturecrypto: A digital signature cryptographically verifies message authenticity and ownership of a private key, underpinning secure blockchain transactions and identities.
Distributed Ledgercrypto: A distributed ledger records data across multiple nodes simultaneously, ensuring transparency, immutability, and resilience against single-point failures.
DODOcrypto: DODO is a proactive-market-maker DEX optimising liquidity efficiency and price discovery for new token listings.
Dogecoin (DOGE)crypto: Dogecoin began as a meme coin but evolved into a community-driven cryptocurrency emphasising accessibility, fast transactions, and tipping culture.
Double Spendingcrypto: Double spending attempts to use the same digital currency twice, prevented in blockchains by consensus and cryptographic transaction validation.
DTCC Digitalcrypto: DTCC Digital tests blockchain-based post-trade infrastructure, improving transparency and settlement efficiency in capital markets.
Dune Analyticscrypto: Dune Analytics enables community-driven blockchain dashboards using SQL-like queries on on-chain data for transparent insights.
Dusting Attackcrypto: A dusting attack sends tiny token amounts to multiple wallets, attempting to deanonymise users by analysing subsequent transactional correlations.
DVN/DONcrypto: Decentralised Validation or Oracle Networks verify off-chain data collectively, improving reliability and removing single-provider trust dependencies.
Diffusion modelai: A generative model that creates images or other data by starting from random noise and repeatedly refining it towards a coherent result.
Distillationai: Training a smaller model to reproduce the outputs of a larger one, transferring much of its capability into a cheaper, faster form.
Decoherencequantum: The loss of a qubit's quantum properties as it interacts with its environment, causing superposition and entanglement to degrade into ordinary classical behaviour.
Dilution refrigeratorquantum: A cryogenic device that cools quantum processors to a few thousandths of a degree above absolute zero using a mixture of helium isotopes.
Decoderai: The part of a model that generates an output sequence step by step, typically conditioned on encoder representations or prior tokens.
Decoder-only architectureai: A design using only stacked decoder blocks with causal masking to predict the next token, used by most large language models.
Dropoutai: A regularisation technique that randomly disables a fraction of units during training, forcing redundancy and reducing overfitting.
Data augmentationai: Expanding a training set by creating altered copies of existing examples, such as cropped, rotated or reworded versions, to improve robustness.
Data curationai: The careful selection, cleaning and organisation of training data to improve its quality, balance and relevance for a model.
Detokenisationai: Converting a sequence of tokens back into readable text, reversing tokenisation and correctly rejoining subword fragments and spacing.
Decode phaseai: The stage after prefill where the model generates output one token at a time, each step reusing the cache and appending the new token.
Draft modelai: A smaller, faster model that guesses upcoming tokens for a larger model to check, used in speculative decoding to cut generation time.
Denoisingai: The core generative step where a model gradually removes predicted noise from a corrupted or random input to reveal a clean image or audio clip.
Denoising diffusion probabilistic model (DDPM)ai: An early diffusion method that learns to reverse a fixed noise-adding process step by step, producing samples starting from pure random noise.
Diffusion transformer (DiT)ai: A diffusion model that replaces the usual U-Net with a transformer backbone, scaling well and powering many recent image and video generators.
Discriminator (GAN)ai: The network in a GAN that tries to tell real training examples from the generator's fakes, providing the signal that trains the generator.
Denoising strengthai: In image-to-image, a setting controlling how much the input is altered; low values stay close to the original, high values diverge widely.
DreamBoothai: A fine-tuning technique that teaches an image model a particular subject from a handful of photos so it can be regenerated in new scenes.
Denoising diffusion implicit model (DDIM)ai: A faster diffusion sampler that produces good images in fewer steps and can generate deterministically from a fixed seed.
Deepfakeai: Synthetic media, typically video or audio, that convincingly replaces or imitates a real person's face or voice.
Diffusion distillationai: Training a fast student model to reproduce a slower diffusion model's results in far fewer steps, cutting generation time sharply.
Direct preference optimisation (DPO)ai: A method for aligning models directly on human preference pairs, skipping the separate reward model and reinforcement learning loop used in RLHF.
Deceptive alignmentai: A hypothesised failure where a model behaves as intended during training and testing but pursues different goals once it believes it is unobserved.
Dangerous capability evaluationai: A test that checks whether a model can meaningfully help with high-harm tasks, such as cyber-attacks, biological weapons or autonomous self-replication.
Data poisoningai: An attack that inserts malicious or corrupted examples into training data, so the resulting model behaves badly or hides a planted backdoor.
Dimensionality reductionai: Techniques that compress data into fewer variables while preserving important structure, aiding visualisation, speed and noise reduction.
Decision treeai: A model that predicts by following a branching sequence of yes or no questions about the input features.
Data leakageai: When information from outside the training data influences a model, inflating measured performance and misleading real-world expectations.
Data parallelismai: A training method where each device holds a full copy of the model and processes a different slice of the data.
Data centreai: A large facility housing servers, storage and networking, with dedicated power and cooling, where cloud computing and AI workloads actually run.
Decodeai: The generation stage of language model inference, where output tokens are produced one at a time after the prompt is processed.
Distributed trainingai: Spreading a single training job across many machines or chips that work together, so very large models can be trained in reasonable time.
Destructive interferencequantum: The cancellation of quantum amplitudes with opposing phases, lowering or eliminating the probability of a particular measurement outcome.
Density matrixquantum: A matrix describing any quantum state, pure or mixed, encoding all statistical predictions about measurements performed on the system.
Dispersive readoutquantum: A measurement method where a qubit's state slightly shifts a resonator's frequency, letting the state be inferred without directly disturbing the qubit.
Doppler coolingquantum: A first-stage laser cooling method that slows atoms or ions by having them preferentially absorb photons that oppose their motion.
Donor qubitquantum: A qubit formed from the spin of an electron or nucleus bound to a single dopant atom, such as phosphorus, placed inside silicon.
Depolarising noisequantum: A simple error model in which a qubit is randomly hit by an X, Y or Z error with equal probability, widely used for benchmarking.
Data qubitquantum: A physical qubit that holds part of the encoded logical information, as opposed to an ancilla used only for measuring syndromes.
Detection eventquantum: A change between consecutive stabiliser measurements signalling that an error occurred nearby, the basic input a decoder works from.
Deutsch-Jozsa algorithmquantum: An early quantum algorithm that decides whether a hidden function is constant or balanced using a single query, where classical methods may need exponentially many.
Deutsch's algorithmquantum: The first quantum algorithm to beat classical computing, deciding a property of a one-bit function with a single query instead of two.
Data re-uploadingquantum: A technique where classical inputs are fed into a quantum circuit multiple times across layers, increasing its expressive power for machine learning tasks.
Dequantisationquantum: The discovery of classical algorithms that match the performance of certain quantum machine learning methods, showing those quantum speedups were not genuine.
Discrete-time quantum walkquantum: A quantum walk that advances in steps using a coin operation and a shift, offering a versatile framework for building quantum search algorithms.
Decoy-state protocolquantum: A quantum key distribution technique that sends pulses of randomly varying brightness to expose photon-splitting eavesdroppers, letting weak laser sources achieve secure key exchange.
Device-independent QKD (DI-QKD)quantum: Quantum key distribution whose security rests only on observing a Bell inequality violation, requiring no assumptions about the internal workings of the devices used.
Diffie-Hellman key exchangequantum: A classic method letting two parties agree a shared secret over a public channel; its security fails against Shor's quantum algorithm.
Digital quantum simulationquantum: A method that reproduces a target system's dynamics using a sequence of discrete quantum gates on a programmable quantum computer.
Dephasingquantum: The gradual loss of the fixed phase relationship in a qubit's superposition, which erases quantum interference and degrades computation.
Depolarising channelquantum: A simple noise model in which a qubit's state is, with some probability, replaced by a completely random state.
DRAG pulse (Derivative Removal by Adiabatic Gate)quantum: A shaped control pulse designed to suppress leakage into unwanted energy levels during fast single-qubit gates.
Dynamical decouplingquantum: A sequence of pulses applied to idle qubits that averages out slow noise, extending how long they keep their quantum information.

E

ECDSAcrypto: Elliptic Curve Digital Signature Algorithm secures Bitcoin and Ethereum transactions through mathematically verifiable cryptographic signing and key pairs.
Edge Walletcrypto: Edge Wallet offers mobile self-custody supporting multiple cryptocurrencies, in-app swaps, and private-key encryption on-device.
EigenDAcrypto: EigenDA is a modular data-availability layer within EigenLayer, providing scalable, decentralised blob storage for rollups.
EigenLayer / Restakingcrypto: EigenLayer enables restaking of Ethereum collateral to secure additional services, extending trust and yield within modular decentralisation.
EIP-1559crypto: EIP-1559 reformed Ethereum’s fee market by burning base fees, introducing predictable gas pricing and deflationary pressure on ETH supply.
EIP-4844 (Proto-Danksharding)crypto: EIP-4844 introduces blob-carrying transactions to reduce rollup costs, bridging Ethereum toward full data-sharding scalability.
Ellipticcrypto: Elliptic is a blockchain-analytics company providing transaction monitoring, wallet screening, and risk-assessment tools for financial institutions and law-enforcement agencies.
ENS (Ethereum Name Service)crypto: ENS converts complex blockchain addresses into readable names like alice.eth, simplifying payments and Web3 user experiences.
Epochcrypto: An epoch is a fixed time period in proof-of-stake systems for distributing rewards and confirming validator performance statistics.
ERC-20crypto: ERC-20 is Ethereum’s standard for fungible tokens, defining consistent smart-contract interfaces for transfers, balances, and allowance approvals.
ERC-721crypto: ERC-721 is the Ethereum non-fungible token standard enabling unique digital asset ownership, essential for collectibles, gaming, and digital identity.
Escrowcrypto: Escrow temporarily holds funds until predefined conditions are met, providing transactional security across marketplaces or decentralised contract interactions.
Ethenacrypto: Ethena issues the USDe synthetic dollar backed by delta-neutral hedging positions, combining on-chain collateral with derivative exposure.
Ethereum (ETH)crypto: Ethereum is a programmable blockchain supporting smart contracts, decentralised applications, and token economies through its robust EVM ecosystem.
Ethereum Classic (ETC)crypto: Ethereum Classic preserves the original Ethereum chain post-DAO hack, upholding immutability as a fundamental blockchain governance principle.
EtherMailcrypto: EtherMail delivers wallet-to-wallet email communication secured by blockchain identities and incentivised with tokenised engagement rewards.
EthSigncrypto: EthSign provides on-chain digital-signature and document-management solutions enabling verifiable contract execution without intermediaries.
Euler Financecrypto: Euler Finance was a non-custodial lending protocol hacked in 2023, later recovering most assets through negotiated white-hat returns.
Euroclear DLTcrypto: Euroclear DLT explores distributed-ledger-based collateral and fund-distribution platforms within Europe’s financial-market infrastructure.
EVM (Ethereum Virtual Machine)crypto: The Ethereum Virtual Machine executes smart-contract bytecode consistently across nodes, ensuring deterministic computation and interoperability among networks.
Exchangecrypto: An exchange facilitates cryptocurrency trading between buyers and sellers, centralised or decentralised, providing liquidity, discovery, and price formation.
Embeddingai: A list of numbers representing a piece of text, image or other data so that similar items sit close together in mathematical space.
Entanglementquantum: A quantum correlation between two or more particles whose measurement outcomes stay linked in ways that cannot be explained by any independent classical state.
Encoderai: The part of a model that reads an input sequence and builds rich contextual representations for downstream tasks to use.
Encoder-decoder architectureai: A design pairing an encoder that reads the input with a decoder that produces the output, common in translation and summarisation.
Encoder-only architectureai: A design using only encoder blocks to produce representations for understanding tasks like classification, rather than generating new text.
Expert networkai: One of several specialised sub-networks in a mixture-of-experts layer, of which only a few are activated for each token.
Embedding layerai: A lookup table that maps discrete tokens to learned dense vectors, forming the input representation for a neural network.
Exploding gradient problemai: A failure where gradients grow uncontrollably large during backpropagation, causing unstable updates and weights that overflow into invalid numbers.
Early stoppingai: A technique that halts training once performance on held-out validation data stops improving, preventing the model from overfitting the training set.
Evalsai: Structured tests that measure a model's capabilities, safety and alignment, run repeatedly to track progress and catch regressions before deployment.
Elo ratingai: A rating system borrowed from chess that ranks models by their win rates in pairwise comparisons, updating scores after each match-up.
Ensemble learningai: Combining the predictions of several models to achieve better accuracy and robustness than any single model can alone.
Episodic memoryai: A record of specific past events or interactions that an agent can recall later to inform its decisions.
Evaluation harnessai: A framework that runs a model or agent through set test cases and scores its outputs to measure quality and track changes.
Experiment trackingai: Recording the settings, data and results of each training run, so experiments can be compared, reproduced and audited later.
Eigenstatequantum: A state whose direction an operator leaves unchanged, returning a definite value when the associated observable is measured.
Eigenvaluequantum: The definite value an observable yields when the system occupies a corresponding eigenstate of that observable.
Expectation valuequantum: The average result expected from measuring an observable many times on identically prepared copies of a quantum system.
Entanglement entropyquantum: A measure of entanglement between two parts of a system, given by the von Neumann entropy of the reduced state.
Entangling gatequantum: A multi-qubit gate that can create entanglement between qubits, producing correlations that cannot be described by treating each qubit separately.
Exchange interactionquantum: A tunable coupling between neighbouring electron spins that turns two-qubit gates on and off in semiconductor spin-qubit devices.
Error syndromequantum: The pattern of stabiliser measurement outcomes indicating where and what kind of errors occurred, which a decoder interprets to choose a correction.
Error thresholdquantum: The physical error rate below which adding more qubits to a code reduces the logical error rate rather than increasing it.
Error budgetquantum: An allocation of the total tolerable error across the different noise sources and operations in a computation, guiding hardware and code design.
Eastin-Knill theoremquantum: A no-go result proving no quantum code can implement a universal gate set entirely transversally, forcing extra techniques like magic states.
Error-correction cyclequantum: One full round of measuring all stabilisers and decoding, repeated continuously to keep logical qubits protected over time.
Erasure errorquantum: An error at a known location, such as a lost or leaked qubit; knowing where it occurred makes it much easier to correct.
Exponential speedupquantum: A quantum advantage where the running time grows far more slowly than the best classical algorithm, as seen in Shor's factoring compared with classical factoring.
Element distinctness problemquantum: Deciding whether a list contains any repeated entry. Quantum walks solve it faster than classical methods, needing roughly N to the two-thirds queries.
E91 protocolquantum: A quantum key distribution scheme proposed by Ekert in 1991 that draws keys from shared entangled photon pairs and verifies security by testing Bell inequalities.
Entanglement swappingquantum: A procedure that entangles two particles which have never interacted, by performing a joint measurement on the partners each was previously entangled with.
Entanglement distillationquantum: A process that converts many weakly entangled pairs into fewer strongly entangled ones, raising the quality needed for teleportation and quantum repeaters.
EPR pairquantum: A pair of maximally entangled particles, named after Einstein, Podolsky and Rosen, whose correlated measurements underpin many quantum communication schemes.
Elliptic-curve cryptography (ECC)quantum: Public-key cryptography based on the algebra of elliptic curves, giving strong security with small keys but breakable by a capable quantum computer.
Excited statequantum: A quantum state of a system with more energy than its ground state, important for understanding light absorption and chemical dynamics.
Electronic structurequantum: The arrangement and energies of electrons in atoms, molecules or solids, which determines chemical bonding and material behaviour.
Electron correlationquantum: The way electrons influence each other's motion beyond simple averaged models; capturing it accurately is a central goal for quantum computers.
Entanglement-enhanced sensingquantum: Using entangled particles together so a measurement gains precision that the same number of independent particles could never reach.
Error ratequantum: The probability that a given quantum operation produces an incorrect result, often quoted per gate and closely linked to fidelity.
Error per Clifford (EPC)quantum: The average error of a randomly chosen Clifford gate, a standard figure produced by randomised benchmarking experiments.
Error mitigationquantum: Techniques that reduce the impact of noise on results through post-processing or clever circuit design, without full error correction.

F

FalconXcrypto: FalconX provides institutional crypto trading, credit, and analytics via an all-in-one brokerage platform.
Fantomcrypto: Fantom is a high-throughput blockchain using directed-acyclic-graph consensus for near-instant transaction finality and low fees.
Farcastercrypto: Farcaster is a decentralised social network using Ethereum addresses and Optimism frames to enable portable, censorship-resistant social identities.
Fiat Currencycrypto: Fiat currency is government-issued legal tender without intrinsic value, backed by state authority and central-bank monetary policy mechanisms.
Fidelity Digital Assetscrypto: Fidelity Digital Assets offers custody and execution services, integrating Bitcoin exposure into institutional investment portfolios.
Filecoin (FIL)crypto: Filecoin is a decentralised storage network using cryptographic proofs to verify data retention, rewarding providers and ensuring content persistence.
Finalitycrypto: Finality means a transaction’s state becomes irreversible, guaranteeing settlement assurance once sufficient confirmations or consensus conditions are met.
Fireblockscrypto: Fireblocks is an institutional digital-asset custody platform offering MPC-based key management and secure transaction workflows.
Fireblocks Networkcrypto: Fireblocks Network interconnects institutions for secure asset transfer, settlement, and DeFi access via MPC-based custody.
Firocrypto: Firo, formerly Zcoin, is a privacy-focused cryptocurrency using Lelantus protocols for zero-knowledge transaction confidentiality.
Flash Loancrypto: A flash loan is an uncollateralised loan executed and repaid within one transaction, enabling arbitrage, refinancing, or exploitation risks.
Flashbotscrypto: Flashbots is a research and infrastructure organisation mitigating MEV risks through transparent block-building and fair-ordering protocols.
Flowdeskcrypto: Flowdesk delivers market-making, compliance, and treasury services helping token issuers maintain stable on-chain liquidity.
Fluiditycrypto: Fluidity wraps stablecoins into yield-generating assets rewarding users randomly upon transfers, gamifying on-chain payments.
Fold Appcrypto: Fold App rewards Bitcoin cashback for everyday purchases, integrating Lightning Network payments into loyalty programmes.
FOMO (Fear of Missing Out)crypto: FOMO describes investor anxiety about missing profitable opportunities, often driving speculative behaviour and irrational market surges.
Footprint Analyticscrypto: Footprint Analytics provides no-code blockchain-data dashboards, helping analysts track DeFi, GameFi, and NFT market movements.
Forkcrypto: A fork occurs when blockchain participants adopt new rules or software versions, creating divergent chains or temporary network splits.
Franklin Templeton OnChaincrypto: Franklin Templeton OnChain tokenises money-market funds on public blockchains, pioneering regulated on-chain financial products.
Fraud Proofcrypto: Fraud proofs allow optimistic rollups to verify correctness by challenging invalid state transitions before final settlement on-chain.
Frax Financecrypto: Frax Finance issues hybrid algorithmic-collateralised stablecoins and yield products underpinning its ecosystem of DeFi protocols.
Front-Runningcrypto: Front-running reorders transactions for profit before a target trade executes, exploiting information asymmetry in public mempools or MEV systems.
Funding Ratecrypto: Funding rate balances perpetual futures with spot prices by rewarding or penalising positions, reflecting leveraged trader sentiment directionally.
Fine-tuningai: Further training of an already-trained model on a smaller, specific dataset to adapt its behaviour to a particular task or domain.
Foundation modelai: A large model trained on broad data that serves as a general-purpose base, adaptable to many downstream tasks through further tuning or prompting.
Few-shotai: When a model is given a small number of worked examples within the prompt to guide how it should perform a task.
Frontier modelai: One of the most capable general-purpose models available at a given time, typically the largest and most expensive to train and run.
Fidelityquantum: A measure of how closely an achieved quantum state or operation matches the intended one, expressed as a value between zero and one.
Flash attentionai: An input-output-aware algorithm that computes exact attention without storing the full attention matrix in memory, greatly speeding training and inference.
Feedforward network (FFN)ai: The position-wise sub-layer in a transformer block, two linear layers with an activation between, that transforms each token independently.
Fully connected layerai: A layer where every input unit connects to every output unit through a learned weight, also known as a dense layer.
Frequency penaltyai: A setting that reduces a token's likelihood in proportion to how often it has already appeared, curbing overuse of common words.
FlashAttentionai: An optimised way of computing attention that avoids writing large intermediate matrices to memory, making inference and training faster and more memory-efficient.
FP8ai: An 8-bit floating-point number format used to store weights and activations more compactly than 16-bit, speeding up inference on supporting hardware.
Forward diffusion processai: The training stage that repeatedly adds small amounts of random noise to data until it becomes indistinguishable from pure noise.
Flow matchingai: A training method that teaches a model the continuous path from noise to data, offering a simpler and often faster alternative to step-by-step diffusion.
Frechet inception distance (FID)ai: A common metric scoring how close generated images are to real ones; lower values indicate more realistic and varied output.
Frame interpolationai: Generating extra in-between frames to make video smoother or slower, synthesising motion that was not originally filmed.
Factualityai: How accurately a model's statements reflect verifiable real-world facts, a core quality target for news, research and reference use.
Frontier safetyai: Practices and policies for managing the heightened risks posed by the most capable AI models, including staged testing and deployment thresholds.
F1 scoreai: The harmonic mean of precision and recall, giving a single balanced measure that is useful when classes are imbalanced.
Feature engineeringai: Creating, transforming or selecting input variables to help a model learn better, often decisive for classic machine learning success.
Feature scalingai: Adjusting numeric features to a common range or distribution, so no single one dominates simply because of its units.
Feature importanceai: A score indicating how much each input variable contributes to a model's predictions, aiding interpretation and feature selection.
Function callingai: A feature letting a model return a structured request to run a named function with arguments, which the surrounding application then executes.
Floating-point operations (FLOPs)ai: A count of the arithmetic calculations needed for a task, used to measure how much compute training or running a model requires.
Floating-point operations per second (FLOP/s)ai: A measure of a chip's raw speed, counting how many floating-point calculations it can perform each second.
Fully Sharded Data Parallel (FSDP)ai: A technique that splits a model's weights and optimiser state across devices to cut memory use, letting bigger models train.
Feature storeai: A system that stores and serves the prepared input features used by machine learning models, keeping them consistent across training and serving.
Field-programmable gate array (FPGA)ai: A chip that can be reconfigured after manufacture to run different tasks, sitting between fixed chips and general processors in flexibility.
Fredkin gate (CSWAP)quantum: A three-qubit gate that swaps two target qubits only when a control qubit is one; it is reversible and preserves the number of ones.
Flux qubitquantum: A superconducting qubit whose two states correspond to tiny loops of current circulating in opposite directions, controlled by an applied magnetic flux.
Fluxoniumquantum: A superconducting qubit using a large array of Josephson junctions to give strong anharmonicity and long coherence, operating at relatively low frequencies.
Fault tolerancequantum: Designing quantum circuits so that a single component failure cannot spread into an uncorrectable error, keeping computation reliable despite faulty parts.
Fault-tolerant quantum computing (FTQC)quantum: Running algorithms on encoded logical qubits with error correction throughout, so noise stays controlled and long computations still give correct results.
Flag qubitquantum: An extra ancilla added to a syndrome-extraction circuit to catch dangerous error patterns that would otherwise spread and defeat fault tolerance.
Free-space QKDquantum: Quantum key distribution that sends photons through open air or space rather than optical fibre, used for satellite and line-of-sight links.
FN-DSA (FALCON)quantum: A compact lattice-based signature scheme built on NTRU lattices, selected by NIST and being standardised as the FFT-based FN-DSA.
Fermionquantum: A particle, such as an electron, that forbids two of its kind from sharing the same quantum state; central to chemistry and materials simulation.
Fermionic simulationquantum: Simulating systems of fermions such as electrons on a quantum computer, which requires special encodings that respect their exchange behaviour.
Flux noisequantum: Random fluctuations in the magnetic flux threading a superconducting qubit, a major source of dephasing in flux-tunable designs.

G

Galaxy Digitalcrypto: Galaxy Digital is a diversified crypto-financial-services firm engaged in trading, investment banking, and asset management.
Gascrypto: Gas measures computational work required to execute blockchain operations, denominated in network tokens to prevent spam and resource abuse.
Gas Limitcrypto: Gas limit defines the maximum computational effort a transaction or block can consume before execution halts to preserve resources.
Gauntletcrypto: Gauntlet simulates DeFi-protocol economics under stress scenarios, optimising risk parameters through agent-based modelling and quantitative analysis.
Geminicrypto: Gemini is a regulated US exchange offering insured custody, derivatives, and staking under stringent compliance frameworks.
Genesis Blockcrypto: The genesis block is the first block in a blockchain, embedding initial parameters and establishing the network’s immutable origin.
Genesis Tradingcrypto: Genesis Trading was a leading OTC desk offering institutional liquidity and lending before restructuring in 2023.
Gitcoincrypto: Gitcoin funds open-source projects through quadratic-funding rounds, aligning community donations with ecosystem impact.
Glassnodecrypto: Glassnode provides on-chain market intelligence, deriving indicators from blockchain data to analyse investor behaviour and network health.
GloDollarcrypto: GloDollar is a non-profit stablecoin whose revenue funds universal basic income, demonstrating mission-driven monetary design.
Gnosis Safe (Safe)crypto: Gnosis Safe is a multisignature smart-wallet enabling collaborative asset management, requiring multiple approvals for secure on-chain transactions.
Goldman Sachs Digital Assetscrypto: Goldman Sachs Digital Assets develops tokenisation and blockchain settlement solutions for traditional-finance instruments.
Governance Tokencrypto: A governance token grants voting power within decentralised protocols, allowing holders to propose, prioritise, and implement protocol upgrades.
Graph Indexercrypto: Graph indexers organise blockchain data for fast querying, powering analytics dashboards, APIs, and dApp front-end data pipelines efficiently.
Grayscale Investmentscrypto: Grayscale Investments manages digital-asset trusts offering regulated exposure to cryptocurrencies for institutional and retail investors.
Grincrypto: Grin is a lightweight privacy coin implementing MimbleWimble for scalable, confidential transactions without address reuse or rich lists.
GSR Marketscrypto: GSR Markets trades and structures digital-asset derivatives, supplying liquidity and hedging solutions to institutional counterparties.
Guardrailsai: Controls placed around a model that filter or block unwanted inputs and outputs, keeping its behaviour within safe and intended limits.
Grover's algorithmquantum: A quantum algorithm that searches an unstructured list quadratically faster than any classical method, needing roughly the square root of the usual attempts.
Grouped-query attention (GQA)ai: A variant where groups of query heads share key and value projections, balancing the quality of multi-head attention with lower memory use.
Gaussian error linear unit (GELU)ai: A smooth activation that scales each input by the probability, under a standard normal distribution, that a value falls below it.
Gated linear unit (GLU)ai: A layer that multiplies one linear projection by a sigmoid-gated second projection, letting the network control which information passes through.
Gated recurrent unit (GRU)ai: A streamlined recurrent unit with update and reset gates, offering much of the LSTM's benefit with fewer parameters.
Generative adversarial network (GAN)ai: A pair of networks trained together, a generator making samples and a discriminator judging them, competing until outputs look realistic.
Gradient descentai: An optimisation method that repeatedly nudges weights in the direction that most reduces the loss, taking steps down the error surface.
Gradientai: A vector of partial derivatives showing how the loss changes as each weight changes, pointing towards the steepest increase in error.
Gradient clippingai: A stabilising technique that caps the size of gradients before a weight update, preventing rare huge values from derailing training.
Gradient accumulationai: A method that sums gradients over several small batches before updating weights, simulating a large batch when memory is limited.
Generalisationai: A model's ability to perform well on new, unseen data rather than merely memorising the specific examples it was trained on.
Greedy decodingai: Generating text by always picking the single most likely next token; fast and deterministic, but often repetitive and lacking variety.
GPTQai: A post-training method that quantises a model's weights to very low precision, usually four bits, layer by layer while limiting the resulting accuracy loss.
GGUFai: A file format for storing quantised language models, widely used with llama.cpp, that packs weights and metadata for efficient loading on consumer hardware.
Generator (GAN)ai: The network in a GAN that turns random input into synthetic samples, improving as it learns to fool the discriminator.
Guidance scaleai: A setting controlling how strongly generation follows the prompt; higher values increase adherence but can reduce naturalness and variety in the output.
Gaussian splattingai: A technique representing a 3D scene as many coloured fuzzy blobs, enabling fast, high-quality rendering from fresh viewpoints.
Goodhart's lawai: The principle that when a measure becomes a target it stops being a good measure, a recurring problem for AI training objectives and benchmarks.
Grokkingai: A phenomenon where a model suddenly generalises to unseen data long after it has already memorised its training set.
Graduate-Level Google-Proof Q&A (GPQA)ai: A set of hard science questions written by experts to be difficult even for skilled people using web search.
Gradient boostingai: An ensemble method that builds trees sequentially, where each new tree corrects the errors made by those before it.
Groundingai: Tying a model's output to retrieved evidence or real data so that its claims can be traced back to a source.
Groundednessai: The degree to which a model's response is supported by the source material it was given, rather than invented.
Guardrail agentai: A supporting agent that checks another agent's inputs or outputs against safety and policy rules before they are allowed to proceed.
Graphics processing unit (GPU)ai: A processor with thousands of cores built for parallel maths, now the standard hardware for training and running large AI models.
GPU hoursai: A measure of compute usage, counting the number of GPUs multiplied by the hours they run, often used for billing.
Global phasequantum: An overall phase factor multiplying an entire quantum state, carrying no physical consequence and never observable by any measurement.
GHZ statequantum: A maximally entangled state of three or more qubits, named after Greenberger, Horne and Zeilinger.
Gate countquantum: The total number of gates in a quantum circuit, often reported separately for single-qubit and two-qubit gates as a complexity measure.
Gate decompositionquantum: Breaking a complex or high-level gate into a sequence of simpler gates that the target hardware can actually execute.
Gate cancellationquantum: An optimisation that removes pairs of adjacent gates which undo each other, such as two consecutive identical self-inverse gates.
Gottesman-Knill theoremquantum: A result showing that quantum circuits built only from Clifford gates can be simulated efficiently on an ordinary classical computer.
Gottesman-Kitaev-Preskill code (GKP)quantum: A bosonic error-correcting code that stores a qubit in grid-like states of a light or microwave mode to resist small shifts.
Grover diffusion operatorquantum: The reflection step in Grover's search that amplifies the amplitude of marked states, also called the diffuser, repeated to drive the answer towards certainty.
Ground statequantum: The lowest-energy state of a quantum system; finding it is central to quantum chemistry and materials science.
Ground state energyquantum: The energy of a system's lowest-energy state, a key quantity for predicting chemical structure, reactions and material properties.
Gravimetryquantum: The precise measurement of gravitational acceleration; quantum gravimeters drop cold atoms to map density changes underground with high accuracy.
Ghost imagingquantum: An imaging method that forms a picture from correlations between two light beams, one of which never directly strikes the object.
Gate fidelityquantum: A measure of how closely a quantum gate's actual operation matches the ideal one, where 1 is perfect and lower values indicate errors.
Gate set tomography (GST)quantum: A thorough technique that reconstructs a full, self-consistent description of a device's gates, preparation and measurement without assuming they are perfect.

H

Hackencrypto: Hacken is a blockchain-security consultancy performing audits, penetration testing, and DeFi vulnerability assessments via its HackenProof bug-bounty platform.
Halborncrypto: Halborn provides cybersecurity services for Web3 firms, including smart-contract audits, penetration testing, and security-awareness training programmes.
Halvingcrypto: Halving periodically reduces proof-of-work block rewards by half, controlling supply issuance and sustaining Bitcoin’s deflationary monetary schedule.
Hard Forkcrypto: A hard fork introduces non-backward-compatible changes, requiring node upgrades and potentially creating a permanent blockchain network split.
Hashdexcrypto: Hashdex develops regulated crypto ETFs in collaboration with Nasdaq, bridging traditional finance with digital-asset markets.
Hashratecrypto: Hashrate measures computational power securing proof-of-work networks, directly correlating with mining difficulty, energy consumption, and security strength.
Health Factorcrypto: Health factor quantifies collateral safety in DeFi lending, comparing total borrowed value against liquidation thresholds to monitor risk.
Heliumcrypto: Helium operates a decentralised wireless network rewarding node operators with HNT tokens for providing IoT and 5G coverage.
Hive Blockchaincrypto: Hive Blockchain is a publicly traded mining company operating renewable-energy data centres securing proof-of-work networks.
HODLcrypto: HODL means holding cryptocurrency long-term regardless of volatility, originating from a typo yet symbolising investor conviction and patience.
Hot Walletcrypto: A hot wallet remains internet-connected for convenient access, trading usability, and speed but carries elevated security and hacking risks.
HSBC Orioncrypto: HSBC Orion issues digital bonds on distributed-ledger systems, demonstrating institutional adoption of tokenised securities.
HTLC (Hashed Time-Locked Contract)crypto: An HTLC enforces conditional payments with cryptographic secrets and expirations, enabling atomic swaps and trust-minimised payment channels.
Humanodecrypto: Humanode verifies unique human identities through biometric proof-of-personhood consensus for Sybil-resistant governance and airdrops.
Hybrid AMMcrypto: A hybrid AMM combines multiple pricing curves for stable-asset and volatile pairs, improving capital efficiency and reducing impermanent-loss exposure.
Hyperspacecrypto: Hyperspace is a Solana NFT marketplace integrating analytics, social discovery, and multi-market aggregation.
Hallucinationai: When a model generates confident but false or fabricated information that is not supported by its training data or any provided sources.
Harvest now, decrypt laterquantum: An attack strategy of collecting encrypted data today to store and decrypt in future once quantum computers can break the encryption.
Hyperbolic tangent (tanh)ai: An activation that maps inputs to the range minus one to one, zero-centred and often used in recurrent networks.
Hidden stateai: The internal vector a recurrent or state-space model carries between steps, summarising the sequence processed up to that point.
Hyperparameterai: A configuration value set before training, such as learning rate or batch size, that shapes how a model learns rather than being learned.
Hyperparameter tuningai: The process of searching for the configuration values that make a model train best, often by trial, grid search or Bayesian methods.
HumanEvalai: A benchmark of one hundred and sixty-four Python programming problems that scores whether generated code passes hidden unit tests.
Holistic Evaluation of Language Models (HELM)ai: A Stanford framework that scores models across many scenarios and metrics, not accuracy alone, for fairer and more transparent comparison.
Hallucination rateai: The proportion of a model's outputs that contain fabricated or false information, used as a measure of its reliability.
Hierarchical clusteringai: A clustering method that builds a tree of nested groups by repeatedly merging or splitting clusters based on similarity.
Hybrid searchai: Combining keyword and semantic search so results benefit from both exact term matching and meaning based similarity.
Hierarchical Navigable Small World (HNSW)ai: A graph based vector index that links vectors in layers so nearest neighbour searches stay fast even on large collections.
Human-in-the-loop (HITL)ai: A design where a person reviews or approves an agent's key actions or outputs before they are allowed to take effect.
Human-on-the-loopai: A design where a person monitors an agent working on its own and can step in, without approving every single action.
High-bandwidth memory (HBM)ai: Fast memory stacked close to an AI chip, providing the huge data throughput that large models need to keep the processor busy.
Hilbert spacequantum: The abstract complex vector space in which quantum states live, providing the geometry underlying superposition and measurement.
Hermitian operatorquantum: An operator equal to its own conjugate transpose, guaranteeing real eigenvalues and so suitable for representing physical observables.
Heisenberg uncertainty principlequantum: The limit stating that certain pairs of observables cannot both be known to arbitrary precision at the same time.
Hadamard gate (H)quantum: A single-qubit gate that puts a definite zero or one into an equal superposition of both, and reverses that operation when applied a second time.
Hardware-efficient ansatzquantum: A parametrised circuit designed to use a device's native gates and qubit connectivity, minimising depth at the expense of theoretical grounding.
Hyperfine qubitquantum: A qubit encoded in two closely spaced internal energy levels of an atom or ion, prized for very long coherence times.
Hypergraph product codequantum: A construction that combines classical codes to build quantum LDPC codes with good distance and low-weight stabilisers.
Harrow-Hassidim-Lloyd algorithm (HHL)quantum: A quantum algorithm for solving certain systems of linear equations, potentially exponentially faster than classical methods when only summary properties of the solution are needed.
Hidden subgroup problem (HSP)quantum: A general problem of finding an unknown subgroup from a function constant on its cosets. Shor's and Simon's algorithms are special cases solved efficiently.
Hamiltonian simulationquantum: Using a quantum computer to model how a quantum system evolves over time, the original motivation for quantum computing and a proven area of advantage.
Hidden shift problemquantum: Finding an unknown offset between two related functions. Quantum algorithms solve certain versions efficiently, offering speedups linked to cryptography and hidden structure.
Hybrid quantum-classical algorithmquantum: An approach that splits work between a quantum processor and a classical computer, running short quantum circuits inside a classical optimisation loop, as in VQE and QAOA.
Heralded entanglementquantum: Entanglement generation accompanied by a separate signal confirming success, so that only verified entangled pairs are kept for teleportation or repeater use.
Hash-based signaturesquantum: Digital signatures whose security rests entirely on the strength of a hash function, making them a conservative and well-understood post-quantum option.
Hybrid key exchangequantum: Combining a classical and a post-quantum algorithm so that a shared key stays secure as long as either component resists attack.
Hamiltonianquantum: The mathematical operator describing a quantum system's total energy; it governs how the system's state changes over time.
Hartree-Fock methodquantum: A foundational chemistry approximation where each electron is treated as moving in the average field created by all the other electrons.
Heisenberg modelquantum: A model of interacting quantum spins that can point in any direction, used to study magnetism and test quantum simulation methods.
Hubbard modelquantum: A simplified model of electrons hopping between lattice sites and repelling each other, central to understanding superconductivity and strongly correlated materials.
Heisenberg limitquantum: The ultimate precision allowed by quantum mechanics for a measurement, improving with the number of resources far faster than conventional methods.

I

IBC (Inter-Blockchain Communication)crypto: IBC is Cosmos’s protocol for secure cross-chain messaging, allowing tokens and data to move between sovereign zones safely.
ICO (Initial Coin Offering)crypto: An ICO raises capital by selling newly issued tokens to investors, funding blockchain startups before full product deployment.
IEO / IDOcrypto: Initial Exchange Offerings and Initial DEX Offerings distribute new tokens via trading platforms, automating compliance, liquidity, and discovery.
Immuneficrypto: Immunefi is a bug-bounty platform connecting white-hat hackers with protocols, distributing multimillion-dollar rewards for responsibly disclosed vulnerabilities.
Immutablecrypto: Immutable describes data that cannot be altered once written to the blockchain, ensuring integrity, auditability, and historical transparency.
Immutable Xcrypto: Immutable X is an Ethereum layer-two using zk-rollups for gas-free NFT minting and instant trading settlements.
Impermanent Losscrypto: Impermanent loss is the temporary value reduction liquidity providers face when pool asset prices diverge relative to simple holding.
Incognito Chaincrypto: Incognito Chain is a privacy-preserving blockchain supporting confidential swaps and wrapped assets for cross-chain anonymity.
Indexercrypto: An indexer structures blockchain data for queryable access, improving analytics performance, transparency, and developer integration capabilities.
Injectivecrypto: Injective is a Cosmos-based derivatives exchange protocol offering order-book trading with zero gas fees and MEV protection.
Insurance Fundcrypto: An insurance fund absorbs protocol losses from liquidations or hacks, protecting users and stabilising DeFi lending or derivative markets.
Intentscrypto: Intents express desired outcomes for transactions, letting solvers or relayers optimise execution under constraints while abstracting away manual steps.
Interoperabilitycrypto: Interoperability enables independent blockchains to exchange assets and data seamlessly, promoting liquidity, composability, and cross-network functionality.
IntoTheBlockcrypto: IntoTheBlock applies machine-learning models to blockchain and market data, offering predictive indicators and on-chain sentiment analysis.
Isolated Margincrypto: Isolated margin limits collateral exposure to one trading position, protecting the rest of a trader’s portfolio from liquidation cascades.
Inferenceai: The process of running a trained model to generate outputs from new inputs, as opposed to the earlier training phase.
Instruction tuningai: Adapting a language model on examples of instructions paired with good responses, making it far better at following a user's requests.
Inter-token latencyai: The average gap between successive output tokens during generation; low values give smooth, fast-flowing streamed text.
Inference serverai: Software that hosts a trained model and handles incoming requests, managing batching, caching and hardware to serve predictions efficiently at scale.
INT8 quantisationai: Storing model weights, and sometimes activations, as 8-bit integers instead of 16- or 32-bit numbers, cutting memory use with little loss of accuracy.
Image-to-image (img2img)ai: Producing a new image guided by an existing one plus a prompt, keeping some structure while changing style or content.
Inpaintingai: Filling a selected or missing region of an image with newly generated content that blends naturally with the surrounding pixels.
IP-Adapterai: An add-on that lets a diffusion model take an image as a style or subject reference, guiding generation without full retraining.
Indirect prompt injectionai: An attack that hides malicious instructions inside external content, such as a web page or document, which the model reads and then obeys.
Inner alignmentai: The problem of ensuring the goal a model actually learns during training matches the objective its designers specified.
Instrumental convergenceai: The idea that agents with very different final goals tend to pursue similar sub-goals, such as self-preservation, resources and acquiring influence.
Interpretabilityai: The field of understanding why a model produces a given output, making its internal reasoning legible to humans for debugging and safety.
Inverted File Index (IVF)ai: A vector index that groups embeddings into clusters, so a search checks only the clusters nearest the query rather than all vectors.
Interconnectai: The high-speed links that connect chips and servers in a cluster, letting them share data quickly during large distributed training runs.
InfiniBandai: A very fast, low-latency networking standard used to connect servers in AI and supercomputing clusters for large distributed workloads.
Inner productquantum: A complex number measuring the overlap of two quantum states, whose squared magnitude relates to a transition probability between them.
Identity gate (I)quantum: A single-qubit gate that leaves a qubit unchanged, often used as a placeholder or to represent a deliberate idle period within a circuit.
iSWAP gatequantum: A two-qubit gate that swaps the states of two qubits and adds a phase factor; it arises naturally in some superconducting hardware.
Ion shuttlingquantum: Moving trapped ions between zones of a chip by adjusting electrode voltages, used to bring qubits together for gates in scalable designs.
Iterative phase estimationquantum: A resource-frugal version of phase estimation that uses a single extra qubit repeatedly, trading many rounds of measurement for far fewer qubits.
Information reconciliationquantum: The error-correction stage of quantum key distribution, where the parties exchange public messages to fix mismatched bits while revealing as little as possible.
Information-theoretic securityquantum: Security that holds even against an adversary with unlimited computing power, because it rests on physical or mathematical impossibility rather than computational difficulty.
Isogeny-based cryptographyquantum: A post-quantum approach built on hard-to-reverse maps between elliptic curves, offering small keys, though one prominent scheme was broken in 2022.
Ising modelquantum: A simple model of interacting spins on a lattice, widely used to study magnetism and to benchmark quantum simulators and annealers.
Interferometryquantum: A measurement method that overlaps waves to produce interference patterns, revealing tiny changes in distance, time or field with great precision.
Incoherent errorquantum: A random, noise-driven error caused by the environment, described statistically rather than as a fixed unwanted operation.
Interleaved randomised benchmarkingquantum: A variant of randomised benchmarking that repeatedly inserts one specific gate to estimate the error of that individual gate.

J

Jamming (Mempool Spam)crypto: Jamming floods mempools with spam transactions, disrupting ordering, raising gas costs, and manipulating MEV extraction opportunities.
JIT Liquiditycrypto: Just-in-time liquidity adds capital precisely before a trade executes to capture fees efficiently without long-term exposure to impermanent loss.
Jito Networkcrypto: Jito Network enhances Solana validator performance and redistributes MEV revenue through stake-pool coordination.
JoinMarketcrypto: JoinMarket coordinates peer-to-peer CoinJoin transactions, rewarding participants who provide liquidity for Bitcoin privacy mixes.
JOMO (Joy of Missing Out)crypto: JOMO counters FOMO by valuing restraint and risk management, reflecting disciplined investor behaviour during speculative bull markets.
JPM Onyxcrypto: JPM Onyx utilises blockchain for interbank settlement, collateral management, and programmable money through JPM Coin technology.
Jump Cryptocrypto: Jump Crypto develops infrastructure, research, and liquidity provisioning for Web3 ecosystems, including validator and cross-chain projects.
Jupiter Exchangecrypto: Jupiter Exchange is Solana’s leading swap aggregator connecting multiple DEXs for best-price execution and bridging routes.
Jailbreakai: A prompt crafted to bypass a model's safety training and make it produce content its guidelines would normally refuse.
JSON modeai: A setting that forces a model to return valid JSON, often following a supplied schema, so its output is predictable to parse.
Josephson junctionquantum: A thin insulating gap between two superconductors through which current tunnels; its non-linear behaviour provides the anharmonicity most superconducting qubits rely on.
Josephson parametric amplifier (JPA)quantum: A superconducting amplifier that boosts faint microwave readout signals with very little added noise, improving the quality of qubit measurement.
Jordan-Wigner transformationquantum: A standard method for mapping electrons and their exchange rules onto qubits so molecular systems can be simulated on a quantum computer.

K

Kaikocrypto: Kaiko provides institutional market data APIs covering centralised and decentralised exchanges for quantitative trading and compliance.
Kaiko Datacrypto: Kaiko Data aggregates trade, order-book, and derivatives metrics from exchanges, powering quantitative models and compliance analytics.
Karuracrypto: Karura is Acala’s sister network on Kusama, testing DeFi features before Polkadot mainnet deployment.
Keystorecrypto: A keystore securely stores encrypted private keys protected by passwords, essential for wallet recovery and preventing unauthorised access.
Komodocrypto: Komodo is a multi-chain platform providing atomic swaps, privacy features, and cross-chain interoperability tools.
Krakencrypto: Kraken is a global crypto exchange known for security, staking, and comprehensive asset listings with regulatory adherence.
Kyber Networkcrypto: Kyber Network is a multi-chain liquidity aggregator powering DEX trading and token swaps with on-chain price discovery.
KYC (Know Your Customer)crypto: KYC requires identity verification for compliance, mitigating money-laundering risks but reducing privacy within regulated crypto services.
KZG Commitmentscrypto: KZG commitments enable succinct cryptographic proofs for polynomial data, supporting Ethereum’s upcoming data-availability and proto-danksharding roadmap.
Key-value cache (KV cache)ai: A store of previously computed keys and values reused during autoregressive generation, so earlier tokens are not recomputed at each step.
Kullback-Leibler divergence (KL divergence)ai: A measure of how much one probability distribution differs from another, used in training to keep an updated model close to a reference.
KV cache (key-value cache)ai: Stored attention keys and values from tokens already processed, reused so the model need not recompute them for every new token, greatly speeding up generation.
K-means clusteringai: A clustering method that partitions data into k groups, each represented by the mean of the points assigned to it.
Kernel trickai: A technique that lets algorithms like SVMs find non-linear boundaries by implicitly mapping data into a higher-dimensional space.
K-nearest neighbours (k-NN)ai: A simple method that classifies a point by taking a majority vote of its k closest labelled neighbours.
K-fold cross-validationai: Cross-validation that splits data into k parts, training on k minus one and testing on the rest, rotating through all folds.
Keyword searchai: Retrieving text that contains the query's literal words or terms, without any regard to their underlying meaning.
Knowledge baseai: A structured store of documents or facts that an agent can search to answer questions or support its reasoning.
Ketquantum: Dirac notation writing a quantum state as a column vector, shown inside a vertical bar and a right angle bracket.
Knill-Laflamme-Milburn protocol (KLM)quantum: A scheme showing efficient quantum computing is possible with only single photons, linear optics and measurement, using extra photons to enable gates.
Knill-Laflamme conditionsquantum: The precise mathematical requirements a quantum code must satisfy for a given set of errors to be perfectly correctable.
Key encapsulation mechanism (KEM)quantum: A scheme for securely establishing a shared secret key using public-key techniques, the form taken by post-quantum encryption standards such as ML-KEM.

L

Large language model (LLM)ai: The technology behind tools like Claude and ChatGPT: a model trained on enormous amounts of text that predicts what comes next, which turns out to be enough to draft, summarise, translate and reason.
Layer 1crypto: Layer 1 denotes the base blockchain layer handling consensus, execution, and settlement directly on-chain, e.g., Bitcoin or Ethereum.
Layer 2crypto: Layer 2 scaling protocols process transactions off-chain while anchoring security to Layer 1, reducing fees and improving throughput dramatically.
LayerZerocrypto: LayerZero is an interoperability protocol enabling trust-minimised omnichain messaging for unified liquidity and seamless cross-chain applications.
Least Authoritycrypto: Least Authority provides privacy-focused security audits and zero-knowledge research supporting anonymity-preserving blockchain technologies.
Ledgercrypto: Ledger manufactures hardware wallets securing private keys offline, protecting crypto holdings against malware and phishing attacks.
Ledger Enterprisecrypto: Ledger Enterprise provides institutional cold-storage infrastructure, governance policies, and transaction-approval workflows.
Lens Protocolcrypto: Lens Protocol is a social-graph platform on Polygon enabling composable, user-owned social profiles and monetisable interactions via NFTs.
Lightning Labscrypto: Lightning Labs builds tools for the Bitcoin Lightning Network, supporting instant micropayments and application development.
Lightning Networkcrypto: The Lightning Network is Bitcoin’s layer-two solution using payment channels for instant, low-fee microtransactions across participating nodes.
Lightsparkcrypto: Lightspark develops enterprise APIs and wallet services for Lightning Network transactions, simplifying integration into payment systems.
Limit Ordercrypto: A limit order instructs execution at a specified price or better, empowering traders to control slippage and improve entry precision.
Lineacrypto: Linea is ConsenSys’s zkEVM-based layer-two network for scalable Ethereum-compatible smart-contract deployment and low-fee transactions.
Liquid Networkcrypto: Liquid Network is a Bitcoin sidechain enabling fast, confidential asset issuance and settlement for exchanges and institutions.
Liquid Staking Token (LST)crypto: A Liquid Staking Token tokenises staked assets, allowing holders to maintain liquidity while earning staking rewards through derivative representation.
Liquidationcrypto: Liquidation automatically sells collateral when borrower health falls below thresholds, ensuring protocol solvency and reducing systemic exposure.
Liquiditycrypto: Liquidity measures how quickly an asset can be bought or sold without significant price impact, essential for efficient market functioning.
Liquidity Bootstrapping Pool (LBP)crypto: An LBP gradually shifts token weights during sale periods to discover prices fairly while deterring early sniping or whale concentration.
Liquidity Layercrypto: Liquidity Layer abstracts capital routing across DeFi protocols, enhancing efficiency and composability through unified liquidity management tools.
Lit Protocolcrypto: Lit Protocol offers distributed key management and programmable encryption for access-controlled data and NFTs.
Litecoin (LTC)crypto: Litecoin is a peer-to-peer cryptocurrency inspired by Bitcoin but featuring faster block times and a distinct hashing algorithm (Scrypt).
LP Tokencrypto: An LP token represents a liquidity provider’s share in a pool, entitling holders to accrued trading fees and governance participation.
LTV (Loan-to-Value)crypto: Loan-to-Value ratio compares borrowed funds to collateral worth, guiding risk limits and liquidation thresholds in lending protocols.
LVR (Loss vs Rebalancing)crypto: Loss vs Rebalancing quantifies systematic value drift liquidity providers face from continuous pool rebalancing under volatile asset conditions.
Lyracrypto: Lyra is an options AMM protocol on Optimism, allowing traders to price, hedge, and manage crypto options positions efficiently.
Logical qubitquantum: A robust qubit built from many physical qubits using error correction, presenting a single reliable unit of quantum information to the algorithm.
Linear attentionai: An attention approximation that reformulates the computation to scale linearly with sequence length rather than quadratically, trading some accuracy for speed.
Layer normalisation (LayerNorm)ai: A technique that normalises each example's features to zero mean and unit variance, then rescales and shifts them with learned parameters.
Load-balancing lossai: An auxiliary training objective encouraging a mixture-of-experts router to spread tokens evenly across experts and avoid overloading a few.
Long short-term memory (LSTM)ai: A recurrent architecture using gates and a memory cell to retain information over long sequences, easing the vanishing-gradient problem.
Logitsai: The raw, unnormalised scores a model outputs before a softmax converts them into probabilities over possible classes or tokens.
Loss functionai: A mathematical measure of how far a model's predictions fall from the correct answers, providing the signal that training works to reduce.
Learning rateai: A setting controlling how big each weight update is; too high overshoots the minimum, too low makes training painfully slow.
Learning-rate scheduleai: A plan that changes the learning rate over the course of training, often starting higher and reducing it as the model settles.
Learning-rate warmupai: A phase at the start of training where the learning rate rises gradually from near zero, avoiding unstable updates on fresh weights.
L1 regularisationai: A penalty on the sum of absolute weight values that encourages sparse models where many weights are driven exactly to zero.
L2 regularisationai: A penalty on the sum of squared weights that discourages large values, spreading influence across many weights and reducing overfitting.
Label smoothingai: A regularisation trick that softens hard target labels slightly, discouraging overconfident predictions and improving a model's calibration and generalisation.
Low-rank adaptation (LoRA)ai: A parameter-efficient method that trains small added matrices while freezing the original weights, adapting large models cheaply and quickly.
Logit biasai: A control that manually raises or lowers the scores of specific tokens before sampling, making chosen words more or less likely to appear.
Log probability (logprob)ai: The logarithm of a token's predicted probability; models often report these so you can gauge confidence or score alternative continuations.
Latencyai: The delay between sending a request and receiving the result; in language models it covers the wait for the first token and for the full response.
Latent diffusion model (LDM)ai: Diffusion carried out in a compressed latent space rather than on raw pixels, making high-resolution image generation far faster and cheaper to run.
Latent spaceai: The compressed numerical representation a model uses internally, where similar concepts sit close together and much of the generation actually happens.
Latent consistency model (LCM)ai: A consistency model applied in latent space, enabling image generation in just a handful of steps for near real-time results.
Logit lensai: An interpretability technique that decodes a model's intermediate layers into vocabulary predictions, revealing how its guess for the next token evolves.
LMArenaai: A public platform, formerly Chatbot Arena, where users compare two anonymous models' answers and vote, producing crowd-sourced rankings from head-to-head battles.
Leaderboardai: A ranked table comparing models on one or more benchmarks, widely used to track progress but easily distorted by contamination or narrow metrics.
Linear regressionai: A model that fits a straight line to data, predicting an output as a weighted sum of the input features.
Logistic regressionai: A classification model that estimates the probability of a class using a logistic function, despite the word regression in its name.
Lemmatisationai: Reducing words to their dictionary base form using vocabulary and grammar, so better maps to good and mice to mouse.
Long-term memoryai: Knowledge an agent stores outside the context window, often in a database, and retrieves again across separate sessions.
LLM-as-a-judgeai: Using a language model to score or compare other models' outputs against criteria, as a scalable stand-in for human review.
Liquid coolingai: Using circulating fluid rather than air to remove heat from dense AI hardware, allowing higher performance in packed data centres.
Local hidden-variable theoryquantum: A classical model proposing predetermined properties explain measurement outcomes, shown false by violations of Bell inequalities.
Laser coolingquantum: Using precisely tuned laser light to slow atoms or ions, lowering their temperature to the microkelvin levels needed for reliable quantum control.
Linear optics quantum computing (LOQC)quantum: An approach building computation from single photons passing through beam splitters, phase shifters and detectors, without direct interaction between photons.
Lattice surgeryquantum: A technique for performing logical operations on surface-code qubits by merging and splitting their patches, using only local two-dimensional interactions.
Logical error ratequantum: The probability that an encoded logical qubit suffers an uncorrected error per operation or round, the key measure of a code's protection.
Logical operatorquantum: A Pauli operation that acts on the encoded logical qubit, commuting with all stabilisers but not itself contained in the stabiliser group.
Leakage errorquantum: An error where a qubit escapes its two computational levels into other energy states, breaking the usual error model and needing special handling.
Logical qubit lifetimequantum: How long an encoded qubit retains its information before a logical error occurs, a headline benchmark for error-correction experiments.
Linear combination of unitaries (LCU)quantum: A method that implements a weighted sum of unitary operations on a quantum computer, used in Hamiltonian simulation and other advanced algorithms.
Local Hamiltonian problemquantum: The task of estimating the lowest energy of a many-body quantum system. It is the canonical QMA-complete problem, central to quantum complexity theory.
Lattice-based cryptographyquantum: A family of post-quantum schemes whose security relies on the hardness of finding short vectors or solving noisy equations in high-dimensional lattices.
Learning With Errors (LWE)quantum: A computational problem of solving linear equations deliberately corrupted with small noise, believed hard even for quantum computers and central to lattice cryptography.
Lamport one-time signaturequantum: An early hash-based signature that can safely sign only one message per key, forming a building block for tree-based hash signatures.
Lattice gauge theoryquantum: A framework describing fundamental forces on a grid of points; quantum simulators are being explored to study its otherwise intractable dynamics.
Leakagequantum: When a qubit escapes its two computational states into other energy levels, leaving the system outside the space quantum operations assume.

M

Magic Edencrypto: Magic Eden is a leading NFT marketplace on Solana supporting minting, auctions, and cross-chain collections.
Maker/Takercrypto: Maker/Taker models differentiate liquidity providers and removers, charging distinct fees to encourage order-book depth and efficient price discovery.
MakerDAOcrypto: MakerDAO governs the DAI stablecoin system through on-chain proposals, collateral frameworks, and risk-management parameters.
Margincrypto: Margin trading borrows funds to increase exposure, amplifying potential profits and losses depending on market movement and leverage settings.
Market Capcrypto: Market capitalisation equals circulating supply multiplied by current price, reflecting an asset’s relative network size and market valuation.
mBridgecrypto: mBridge is a BIS-led CBDC platform enabling cross-border transactions among multiple central banks on a shared ledger.
Merkle Treecrypto: A Merkle tree hashes transactions into a hierarchical structure, allowing efficient verification of data integrity without downloading entire blocks.
Merlin Blockchaincrypto: Merlin offers on-chain AML intelligence integrating wallet screening, counterparty-risk scoring, and visual transaction-path analysis for investigators.
Messaricrypto: Messari provides crypto market research, data feeds, and regulatory disclosures for investors and builders.
Messari Procrypto: Messari Pro delivers in-depth crypto research, governance tracking, and data feeds for institutional investors and asset managers.
Metacocrypto: Metaco delivers digital-asset custody and orchestration platforms integrating with banks’ core systems under strict compliance standards.
MetaMaskcrypto: MetaMask is a browser wallet enabling users to manage assets, sign transactions, and interact with EVM-compatible dApps securely through extensions or mobile apps.
Metamask Snapscrypto: Metamask Snaps extends wallet functionality through permissioned JavaScript modules integrating new networks and features.
Metaplexcrypto: Metaplex is Solana’s NFT infrastructure enabling minting, auctions, and metadata standards for creators.
MEV (Maximal Extractable Value)crypto: MEV represents profit from transaction ordering and inclusion manipulation, incentivising builders to capture arbitrage while raising ethical and network-efficiency concerns.
MEV-Boostcrypto: MEV-Boost separates block proposers and builders, allowing validators to source pre-built blocks for fairer competition and reduced centralisation risks.
Miningcrypto: Mining validates proof-of-work transactions by solving cryptographic puzzles, securing the network while rewarding miners with newly minted coins and fees.
Mintingcrypto: Minting creates new tokens or NFTs on-chain, encoding ownership and metadata while verifiably expanding a network’s supply through smart-contract execution.
Mixin Messengercrypto: Mixin Messenger combines encrypted chat with multi-chain wallet capabilities, enabling peer-to-peer payments privately and instantly.
Mixin Networkcrypto: Mixin Network facilitates fast, private cross-chain transfers using DAG architecture and secure multiparty computation.
Monerocrypto: Monero employs ring signatures, stealth addresses, and confidential transactions to guarantee untraceable, fungible private payments.
Moonbeamcrypto: Moonbeam is an Ethereum-compatible parachain on Polkadot supporting Solidity smart-contract deployment with native interoperability.
MPC Walletcrypto: A Multi-Party Computation wallet splits private-key operations among participants, eliminating single points of failure and enhancing custodial security models.
Multisigcrypto: Multisignature wallets require multiple approvals to execute transactions, improving shared-fund governance and reducing the impact of single-key compromise.
Multimodal modelai: A model that can process or generate more than one type of data, such as text, images, audio or video, within one system.
Mixture of experts (MoE)ai: A model design that routes each input to a subset of specialised sub-networks, cutting the computation needed while keeping the total parameter count large.
Multi-head attention (MHA)ai: Attention run in parallel across several heads, each learning different relationships, whose outputs are concatenated and combined into one representation.
Multi-query attention (MQA)ai: An attention variant where all query heads share a single set of keys and values, reducing memory and accelerating generation.
Multilayer perceptron (MLP)ai: A basic neural network of fully connected layers with non-linear activations, able to approximate a wide range of functions.
Mambaai: A selective state-space architecture that processes sequences with near-linear scaling, offering an efficient alternative to attention for long contexts.
Mean squared error (MSE)ai: A regression loss that averages the squared differences between predicted and actual values, punishing large errors far more heavily than small ones.
Mean absolute error (MAE)ai: A regression loss that averages the absolute differences between predictions and actual values, treating every error in direct proportion to its size.
Momentumai: An optimisation technique that accumulates a moving average of past gradients, helping updates keep direction and move faster through shallow regions.
Mini-batch gradient descentai: Training that computes gradients on small groups of examples, balancing the stability of full batches against the speed of single samples.
Masked language modelling (MLM)ai: A self-supervised task where random words are hidden and the model learns to predict them from the surrounding context.
Mixed-precision trainingai: Speeding up training and saving memory by running many calculations in lower-precision numbers while keeping key values in full precision.
Min-p samplingai: A decoding method that keeps tokens whose probability is at least a set fraction of the most likely token's probability, adapting the cutoff to model confidence.
MXFP4ai: A four-bit micro-scaling floating-point format that groups values with a shared scale, letting large models run in far less memory with modest accuracy loss.
Mode collapseai: A GAN failure where the generator produces only a few similar outputs, ignoring the full variety present in the training data.
Mesh generationai: Producing the network of vertices, edges and faces that defines a 3D object's surface, a common output of text-to-3D systems.
Mel spectrogramai: A spectrogram scaled to match human hearing, widely used as the intermediate target representation inside text-to-speech systems.
Many-shot jailbreakai: A jailbreak that fills a long context window with many fabricated examples of the model complying, gradually eroding its refusal behaviour.
Mesa-optimisationai: When a trained model itself becomes an optimiser pursuing its own internal objective, which may differ from the objective it was trained on.
Mechanistic interpretabilityai: A branch of interpretability that reverse-engineers the internal weights and activations of a network into human-understandable algorithms and circuits.
Massive Multitask Language Understanding (MMLU)ai: A benchmark of multiple-choice questions across fifty-seven subjects, from history to law, used to gauge a model's broad knowledge.
MT-Benchai: A benchmark of multi-turn conversation questions where a strong language model acts as judge, scoring each answer's quality on a numeric scale.
Model cardai: A short document describing a model's intended uses, training data, performance across groups and known limitations, to support responsible deployment.
Machine learning (ML)ai: A field of AI where systems learn patterns from data to make predictions or decisions, rather than following explicitly programmed rules.
Multi-agent systemai: A setup where several specialised agents cooperate, each handling part of a task and exchanging results to reach a shared goal.
Multi-hop retrievalai: Answering a question that needs facts from several sources by retrieving and reasoning over them in successive connected steps.
Model Context Protocol (MCP)ai: An open standard that lets AI applications connect to external tools and data sources through a common shared interface.
MCP serverai: A service that exposes tools, data or prompts to an AI application over the Model Context Protocol.
Model routingai: Directing each request to the most suitable model, balancing cost, speed and quality across a set of available models.
Model parallelismai: Splitting a single model across several devices when it is too large to fit in the memory of one.
Mixed precisionai: Training that uses lower-precision numbers for speed and memory savings while keeping higher precision where needed for numerical stability.
Model registryai: A central store that tracks trained models, their versions and metadata, so teams can find, deploy and roll back reliably.
MLOpsai: The practice of reliably building, deploying and maintaining machine learning models in production, applying software engineering discipline to the model lifecycle.
Model driftai: The gradual decline in a deployed model's accuracy as the real-world data it sees moves away from its training data.
Memory bandwidthai: The rate at which data moves between a chip and its memory, often the main limit on how fast large models run.
Model servingai: Making a trained model available to answer requests in production through an API, with attention to speed, scale and reliability.
Mixed statequantum: A statistical blend of pure states, representing classical uncertainty about which state a system actually occupies.
Maximally mixed statequantum: A state holding no information, giving equal probabilities for all outcomes whatever measurement basis is chosen.
Maximal entanglementquantum: The strongest possible correlation between subsystems, where each part measured alone appears completely random.
Minus statequantum: The equal superposition of 0 and 1 with an opposite relative phase, orthogonal to the plus state.
Multi-controlled gatequantum: A gate that acts on its target only when several control qubits are simultaneously one, such as a Toffoli gate with three or more controls.
Molmer-Sorensen gatequantum: A two-qubit entangling gate widely used in trapped-ion systems, coupling qubits through their shared motional vibrations using laser light.
Magic statequantum: A specially prepared qubit state that, when consumed, lets a fault-tolerant computer perform a non-Clifford gate such as the T gate.
Magic state distillationquantum: A procedure that converts many noisy magic states into fewer higher-quality ones, enabling reliable non-Clifford gates in fault-tolerant computing.
Mid-circuit measurementquantum: Measuring one or more qubits partway through a circuit rather than only at the end, often used to steer later operations.
Motional modequantum: A shared vibration of trapped ions in their trap; these collective oscillations act as a bus that links ion qubits during gates.
Magneto-optical trap (MOT)quantum: A device combining laser beams and magnetic fields to cool and gather a cloud of neutral atoms as the starting point for atom-based processors.
Measurement-based quantum computing (MBQC)quantum: A model where computation proceeds by measuring qubits in a large pre-prepared entangled state, with each measurement guiding the next.
Majorana zero modequantum: An exotic state predicted at the ends of certain superconducting nanowires; storing information across pairs of them could give inherently protected qubits.
Microwave photonquantum: A single quantum of microwave-frequency light; superconducting qubits are controlled and read out by exchanging these photons with resonators.
Mixing chamberquantum: The coldest stage of a dilution refrigerator, where the quantum chip sits at a few thousandths of a degree above absolute zero.
Microwave controlquantum: Using precisely timed microwave pulses to drive gate operations on superconducting and some spin qubits.
Magic-state factoryquantum: A dedicated region of a quantum processor devoted to producing and distilling magic states to feed the rest of the computation.
Minimum-weight perfect matching (MWPM)quantum: A popular surface-code decoding method that pairs up syndrome events using the fewest, shortest error chains to infer the correction.
Measurement errorquantum: An error where a qubit or stabiliser readout returns the wrong result, so syndrome data itself is unreliable and rounds must be repeated.
Measurement-device-independent QKD (MDI-QKD)quantum: A quantum key distribution method where both parties send photons to an untrusted central relay, removing every attack that targets flawed single-photon detectors.
Module-LWEquantum: A version of Learning With Errors that balances efficiency and security using modules over polynomial rings, underpinning the Kyber and Dilithium schemes.
ML-KEMquantum: The Module-Lattice-based Key-Encapsulation Mechanism, published by NIST as FIPS 203, the standardised form of CRYSTALS-Kyber for quantum-safe key exchange.
ML-DSAquantum: The Module-Lattice-based Digital Signature Algorithm, published by NIST as FIPS 204, the standardised form of CRYSTALS-Dilithium for quantum-safe signatures.
Merkle signature schemequantum: A hash-based signature method that uses a binary tree of hashes to authenticate many one-time signatures under a single public key.
McEliece cryptosystemquantum: A code-based public-key encryption scheme from 1978, unbroken for decades but needing very large keys, now a leading conservative post-quantum candidate.
Molecular Hamiltonianquantum: The energy operator describing all the electrons and nuclei in a molecule and their interactions, the starting point for quantum chemistry calculations.
Molecular orbitalquantum: A quantum state describing where an electron is likely to be found across a whole molecule, formed by combining atomic orbitals.
Magnetometryquantum: The measurement of magnetic fields; quantum techniques enable extremely sensitive detection for navigation, geology and medical imaging.
Mirror benchmarkingquantum: A scalable technique that runs a circuit followed by its inverse, so the correct final result is known and errors are easy to spot.

N

Namadacrypto: Namada is an asset-agnostic privacy layer providing shielded transfers for IBC-connected blockchains within the Cosmos ecosystem.
Nansencrypto: Nansen provides blockchain analytics linking on-chain wallet activity with labelled entities for trader intelligence and research.
Nansen Querycrypto: Nansen Query labels blockchain addresses, clustering entities to provide behavioural insights and wallet-tracking analytics for traders.
Native Bridgecrypto: A native bridge is maintained by layer-two protocols to transfer assets securely between themselves and their underlying layer-one chain.
NavCoincrypto: NavCoin is a proof-of-stake privacy cryptocurrency offering optional anonymous transactions through dual-blockchain architecture.
Near Protocolcrypto: Near Protocol is a sharded proof-of-stake blockchain optimised for usability, fast finality, and scalable dApp development.
Network Effectscrypto: Network effects occur when a protocol’s value increases as more participants join, strengthening liquidity, security, and developer ecosystem growth.
Nodecrypto: A node is a computer participating in a blockchain network, storing data, validating transactions, and propagating consensus information.
Non-Fungible Token (NFT)crypto: An NFT is a unique token representing digital ownership of art, media, or in-game assets secured by blockchain immutability and metadata standards.
Noncecrypto: Nonce means ‘number used once’, preventing replay attacks by ensuring each transaction identifier remains unique and chronologically ordered.
NuCyphercrypto: NuCypher provides threshold-encryption middleware allowing decentralised access control for shared sensitive information.
NVT Ratiocrypto: Network Value to Transaction ratio compares market capitalisation to transaction volume, acting as a valuation metric for network utilisation.
NYDIGcrypto: NYDIG offers Bitcoin custody, execution, and infrastructure for institutional clients within regulated US financial frameworks.
Neural networkai: A computing system of interconnected layers of simple units whose numerical weights are adjusted during training to recognise patterns in data.
Neutral-atom qubitquantum: A qubit encoded in the states of an electrically neutral atom held by focused laser beams called optical tweezers and controlled optically.
Noisy Intermediate-Scale Quantum (NISQ)quantum: The current era of quantum computers with tens to thousands of qubits but no full error correction, so noise limits their capability.
Negative log-likelihood (NLL)ai: A loss that rewards a model for assigning high probability to the observed data, equivalent to cross-entropy loss for classification tasks.
Next-token predictionai: The core training task of most language models, learning to predict the following word given all of the preceding text.
Nucleus sampling (top-p)ai: A decoding method that keeps only the smallest set of top tokens whose probabilities add up to a chosen threshold, then samples from that set.
Noise scheduleai: The predefined plan controlling how much noise is added or removed at each stage of a diffusion process, shaping output quality and speed.
Negative promptai: Text describing what you do not want in the output, which the model deliberately steers away from during generation.
Neural radiance field (NeRF)ai: A method that learns a 3D scene from photos as a continuous function, letting you render sharp new viewpoints not in the originals.
Neural audio codecai: A learned system that compresses audio into compact tokens and reconstructs it, underpinning many modern speech and music generators.
Neural style transferai: A technique that repaints one image's content in the visual style of another using features learned by a neural network.
Naive Bayesai: A probabilistic classifier applying Bayes' theorem while assuming features are independent, working well for text and spam filtering.
N-gramai: A contiguous sequence of n words or characters from text, used to capture short local patterns in language.
Named entity recognition (NER)ai: An NLP task that locates and classifies names in text, such as people, organisations, places and dates.
Neural processing unit (NPU)ai: A specialised chip for running AI tasks efficiently, often built into phones and laptops to handle inference using less power.
NVLinkai: Nvidia's fast direct connection between GPUs, letting them exchange data far quicker than standard links so they can work together as one.
No-cloning theoremquantum: The result that an arbitrary unknown quantum state cannot be copied exactly, a foundational limit of quantum information.
No-deleting theoremquantum: The result that an unknown quantum state's copy cannot be reversibly deleted, a counterpart to the no-cloning theorem.
Non-Clifford gatequantum: Any gate outside the Clifford family, such as the T gate, required to reach universal computation and generally harder to implement fault-tolerantly.
Native gate setquantum: The specific gates a given quantum processor can execute directly in hardware, into which all other gates must be decomposed.
Nitrogen-vacancy centre (NV centre)quantum: A defect in diamond where a nitrogen atom sits beside a missing carbon; its electron spin works as a qubit even at room temperature.
NTRUquantum: A long-standing lattice-based public-key cryptosystem using polynomial arithmetic, valued for fast operations and widely considered resistant to quantum attack.
NIST PQC standardisationquantum: The United States programme, running since 2016, that evaluates and standardises post-quantum cryptographic algorithms, publishing its first finished standards in 2024.
NOON statequantum: A special entangled state of light where photons travel together along one of two paths, used to push interferometry towards the Heisenberg limit.

O

OCR (Off-Chain Reporting)crypto: Off-Chain Reporting aggregates oracle data off-chain to minimise gas costs before posting a single verified update on-chain securely.
OKLinkcrypto: OKLink is a blockchain explorer and data provider offering multi-chain analytics, address tracking, and AML compliance reports.
OKXcrypto: OKX is a global crypto exchange and wallet ecosystem offering trading, Web3 DeFi, and NFT services.
On-Chain Governancecrypto: On-chain governance enables token holders to vote directly via smart contracts, ensuring transparent and programmable decision-making processes.
Open Interestcrypto: Open interest represents the total outstanding derivative contracts not yet settled, reflecting market participation and liquidity intensity.
OpenNodecrypto: OpenNode offers Bitcoin payment-processing infrastructure supporting Lightning Network settlements and fiat conversion options.
OpenSeacrypto: OpenSea is a leading NFT marketplace enabling peer-to-peer trading, minting, and royalty distribution across numerous blockchain ecosystems.
Opensearchcrypto: Opensearch indexes blockchain and NFT data for fast query performance across decentralised applications and explorers.
Optimismcrypto: Optimism is an Ethereum layer-two optimistic rollup reducing transaction fees and supporting on-chain public-goods funding via retroactive rewards.
Optimistic Rollupcrypto: An optimistic rollup bundles transactions off-chain, assuming validity unless challenged through fraud proofs during its verification window.
Options (Calls/Puts)crypto: Options contracts grant rights to buy or sell an asset at a fixed price within a specified period, facilitating hedging and speculation.
Oraclecrypto: An oracle delivers off-chain data to smart contracts, enabling decentralised applications to react to real-world information reliably.
Ordifycrypto: Ordify tracks Bitcoin Ordinal and BRC-20 markets, providing minting analytics and discovery tools.
Ordinals Walletcrypto: Ordinals Wallet manages Bitcoin inscriptions and BRC-20 tokens, supporting trading and collection of ordinal-based assets.
Open-weight modelai: A model whose trained parameters are released publicly for download and use, though its training data and code may remain undisclosed.
Objective functionai: The overall quantity that training seeks to optimise, usually a loss to minimise, sometimes combined with regularisation or other penalty terms.
Optimiserai: The algorithm that adjusts a model's weights from computed gradients to reduce the loss, such as SGD, Adam or AdamW.
Overfittingai: When a model learns the training data too closely, including its noise, and then performs poorly on new examples it has not seen.
Outpaintingai: Extending an image beyond its original borders by generating plausible new content around the edges.
Over-refusalai: When a model wrongly rejects harmless requests because they superficially resemble unsafe ones, harming usefulness in the name of caution.
Outer alignmentai: The problem of ensuring the specified training objective genuinely captures what humans want, so optimising it does not produce unintended behaviour.
One-hot encodingai: Representing a categorical value as a vector of zeros with a single one, so models can process categories numerically.
Orchestrator-worker patternai: A design where a lead agent plans and delegates subtasks to worker agents, then combines their outputs into a final result.
Observabilityai: The ability to understand a running system's health from its logs, metrics and traces, so problems can be spotted and diagnosed quickly.
Offloadingai: Moving parts of a model or its data from fast accelerator memory to slower CPU memory or storage to fit larger workloads.
Orthogonal statesquantum: Two quantum states with zero overlap, so a suitable measurement can perfectly distinguish one from the other every time.
Orthonormal basisquantum: A set of mutually orthogonal states, each of unit length, that together span the whole state space.
Observablequantum: A measurable physical quantity represented by a Hermitian operator, whose eigenvalues are the possible values a measurement can return.
OpenQASMquantum: An open, text-based language for describing quantum circuits gate by gate, used to specify programs for many different quantum computers.
Optical qubitquantum: A qubit encoded in two atomic or ionic levels separated by an optical-frequency transition, addressed directly with a narrow laser beam.
Optical tweezersquantum: Tightly focused laser beams that grip and hold individual neutral atoms in place, arranging them into programmable arrays for computing.
Order findingquantum: The task of finding the smallest power to which a number must be raised to give one, modulo N. Its quantum solution underlies Shor's factoring.
Oracle separationquantum: A proof that two complexity classes differ relative to some oracle, used as evidence that quantum computers can outperform classical ones on certain tasks.
One-time padquantum: An unbreakable cipher that combines a message with a truly random key as long as the message; quantum key distribution supplies such keys.
Optical lattice clockquantum: An atomic clock that traps many atoms in a grid of laser light and reads their optical transitions, achieving exceptional precision and stability.
Optical atomic clockquantum: An atomic clock based on very high-frequency optical transitions, offering far greater precision than traditional microwave-based clocks.
Optical frequency combquantum: A laser producing millions of evenly spaced, sharp frequencies, acting as a ruler for light that links optical and microwave measurements.
Optically pumped magnetometer (OPM)quantum: A highly sensitive, cryogen-free magnetic sensor using laser-prepared atomic vapour, increasingly used for wearable brain imaging.
OpenPulsequantum: A specification, used with Qiskit, for controlling quantum hardware at the level of analogue pulses rather than abstract gates.

P

Paradigmcrypto: Paradigm is a venture-capital firm investing in blockchain infrastructure, DeFi, and privacy technologies.
Particlcrypto: Particl is a privacy marketplace platform using confidential transactions and ring signatures to protect buyer and seller identities.
PBS (Proposer-Builder Separation)crypto: Proposer-Builder Separation divides roles in block creation, reducing MEV centralisation and improving block market fairness across validators.
PeckShieldcrypto: PeckShield conducts blockchain-security audits and hack-tracking, publishing analytics on exploits, rug-pulls, and on-chain phishing campaigns.
Peer-to-Peer (P2P)crypto: Peer-to-peer refers to direct interactions between users without intermediaries, empowering decentralised exchange and value transfer models.
Penumbracrypto: Penumbra is a privacy-preserving proof-of-stake chain for confidential staking, swaps, and governance using zero-knowledge cryptography.
Perpetual Futures (Perps)crypto: Perpetual futures are derivative contracts without expiry, tracking spot prices through funding payments between long and short traders.
Phantom Walletcrypto: Phantom Wallet is a Solana wallet enabling token swaps, staking, and NFT management through an intuitive mobile and browser interface.
Pirate Chaincrypto: Pirate Chain (ARRR) enforces mandatory zero-knowledge privacy on all transactions, maximising anonymity for users.
Polygoncrypto: Polygon is a scaling framework for Ethereum offering sidechains and zk-rollups to improve throughput and lower gas costs.
Polygon IDcrypto: Polygon ID provides decentralised identity infrastructure enabling verifiable, privacy-preserving credentials using zero-knowledge technology.
Portal Bridgecrypto: Portal Bridge (by Wormhole) allows asset transfers across blockchains using wrapped-token minting and burning mechanisms.
Post-quantum cryptographyquantum: New encryption designed to survive attack by future quantum computers. The migration has already started, because data stolen today could be decrypted by the machines of the 2030s.
Price Impactcrypto: Price impact measures slippage caused by trade size relative to liquidity depth, critical for efficient execution strategies.
Prime Trustcrypto: Prime Trust provides regulated custodial infrastructure for digital-asset platforms, including settlement, KYC, and payment rails.
Priority Fee (Tip)crypto: Priority fee is an optional extra gas tip paid to miners or validators to prioritise transaction inclusion.
Private Keycrypto: A private key is a cryptographic secret authorising transactions and ownership of digital assets stored on-chain.
Project Guardiancrypto: Project Guardian is a Singapore-led initiative exploring tokenised-asset interoperability and DeFi integration within regulated frameworks.
Project Icebreakercrypto: Project Icebreaker connects retail CBDC systems between Nordic countries to test real-time cross-border digital-currency payments.
Prompt injectionai: An attack where malicious instructions are hidden in content an AI reads — a web page, an email — so the AI follows the attacker’s orders instead of its user’s. The AI equivalent of tricking a new employee with a forged memo.
Proof of Reservescrypto: Proof of Reserves demonstrates custodian asset backing through on-chain attestations, enhancing transparency and user trust.
Proof of Stake (PoS)crypto: Proof of Stake selects validators based on staked assets to secure the network and earn transaction rewards efficiently.
Proof of Work (PoW)crypto: Proof of Work requires computational effort to add blocks, ensuring security through energy expenditure and difficulty adjustments.
Public Keycrypto: A public key is derived from a private key and used to generate addresses for receiving cryptocurrency securely.
PWA / WalletConnectcrypto: Progressive Web Apps and WalletConnect enable browser or mobile wallet integration for secure connection to decentralised applications.
Parameterai: A numerical value inside a model, learned during training, that shapes its behaviour. Modern language models contain billions of them.
Physical qubitquantum: A single real quantum system used to physically realise one qubit, typically noisy and prone to errors on its own.
Photonic qubitquantum: A qubit encoded in a property of a single particle of light, such as its polarisation or path, and processed with optical components.
Pre-normalisationai: An arrangement placing normalisation before each sub-layer rather than after it, which stabilises the training of deep transformers.
Positional encodingai: Information added to token representations that tells a model the order of a sequence, since attention alone is order-agnostic.
Pooling layerai: A layer that downsamples feature maps, for example by taking the maximum or average over small regions, reducing size and adding invariance.
Patch embeddingai: The step that divides an image into fixed-size patches and projects each into a vector for a vision transformer to process.
Pretrainingai: The initial phase of training a model on large general datasets to learn broad patterns before it is adapted to specific tasks.
Parameter-efficient fine-tuning (PEFT)ai: A family of methods that adapt large models by updating only a small fraction of parameters, saving memory and compute.
Prompt tokensai: The tokens making up the input sent to a model, including any system instructions and conversation history; they are usually billed separately from output.
Presence penaltyai: A setting that lowers the likelihood of any token that has already appeared at all, encouraging the model to introduce new topics and words.
Perplexityai: A measure of how well a model predicts a text sample; lower perplexity means the model finds the text less surprising and predicts it better.
Prefillai: The first inference stage, where the model processes the whole prompt at once to build its cache before it begins producing output tokens.
Prefix cachingai: A serving optimisation that keeps the cached computation for a shared opening sequence, so many requests beginning the same way skip repeated prefill work.
PagedAttentionai: A memory management technique that stores the KV cache in small non-contiguous blocks, like operating-system paging, reducing waste and letting more requests run at once.
Post-training quantisation (PTQ)ai: Reducing the numerical precision of an already trained model without retraining it, a quick way to shrink size and speed up inference.
Pruningai: Removing weights or whole components from a model that contribute little, producing a smaller, faster network with similar accuracy.
Point cloudai: A set of points in 3D space describing a shape or scene, often produced by scanning or as a step in 3D generation.
Prosodyai: The rhythm, stress and intonation of speech, which good text-to-speech systems reproduce so the voice sounds natural.
Preference dataai: Pairs or rankings of model outputs labelled by which is better, used to train reward models or to fine-tune models directly.
Proximal policy optimisation (PPO)ai: A reinforcement learning algorithm widely used in RLHF to update a model's behaviour while limiting how far each step changes it.
Prompt injection defenceai: Techniques that stop untrusted input from overriding a system's instructions, such as input filtering, privilege separation and marking data as non-executable.
Power-seekingai: A hypothesised tendency for capable agents to gain resources, options and control, because doing so helps achieve almost any long-term objective.
Probingai: Training a simple classifier on a model's internal activations to test what information, such as sentiment or truth, those activations encode.
Pass@kai: An evaluation metric giving the chance that at least one of a model's k generated attempts solves a problem correctly.
Principal component analysis (PCA)ai: A technique that reduces dimensions by projecting data onto new axes that capture the greatest variance in the data.
Precisionai: Of the items a model labelled positive, the fraction that truly are positive; high precision means few false alarms.
Part-of-speech tagging (POS tagging)ai: Labelling each word in a sentence with its grammatical role, such as noun, verb, adjective or adverb.
Parallel tool callsai: When a model requests several tool functions at once so the application can run them together rather than one after another.
Planning (agent planning)ai: The step where an agent works out a sequence of actions or subtasks before carrying them out, rather than acting purely step by step.
Plan-and-executeai: An agent design that first drafts a full plan, then works through each step, optionally revising the plan as new information arrives.
Prompt chainingai: Linking several prompts so the output of one becomes the input to the next, breaking a task into ordered stages.
Pipeline parallelismai: Placing different layers of a model on different devices, so data flows through them in stages like an assembly line.
Power usage effectiveness (PUE)ai: A data centre efficiency measure comparing total energy used to the energy that reaches the computing equipment; lower is better.
Probability amplitudequantum: A complex number attached to a possible outcome, whose squared magnitude gives the probability of obtaining that outcome on measurement.
Pauli matricesquantum: A set of three two-level operators, X, Y and Z, fundamental for describing single-qubit states, gates and measurements.
Pauli-X gatequantum: The single-qubit operation that exchanges the 0 and 1 states, serving as the quantum version of a bit flip.
Pauli-Z gatequantum: The single-qubit operation that leaves 0 unchanged and adds a minus sign to 1, performing a phase flip.
Pauli-Y gatequantum: A single-qubit operation combining a bit flip with a phase flip, being one of the three Pauli operators.
Pauli basisquantum: A description of single-qubit operators using the identity together with the Pauli X, Y and Z operators.
Pure statequantum: A quantum state known exactly, describable by a single wavefunction and having the maximum possible density-matrix purity.
Purityquantum: A number quantifying how close a state is to pure, calculated from the trace of its squared density matrix.
Partial tracequantum: A mathematical operation that discards a chosen subsystem, yielding the reduced description of the part that remains.
Product statequantum: A joint state formed by combining independent subsystem states, being the simplest kind of separable, unentangled state.
Plus statequantum: The equal superposition of 0 and 1, sitting on the Bloch sphere equator and central to quantum interference.
Projective measurementquantum: A standard measurement that projects a state onto one eigenstate of an observable, returning that eigenstate's value.
Positive operator-valued measure (POVM)quantum: A general description of quantum measurements giving outcome probabilities without requiring orthogonal projective operators.
Phase gate (S)quantum: A single-qubit gate that adds a quarter-turn phase of ninety degrees to the one state, equal to the square root of the Pauli-Z gate.
Pauli gatesquantum: The set of single-qubit X, Y and Z gates, together with the identity, forming the basic building blocks for describing qubit operations and errors.
Parametrised quantum circuitquantum: A circuit containing gates whose rotation angles are adjustable variables, tuned during algorithms such as variational optimisation.
Phase kickbackquantum: An effect where a controlled gate imprints a phase onto the control qubit rather than the target, exploited in many quantum algorithms.
Principle of deferred measurementquantum: A rule stating that measurements can be postponed to the end of a circuit by replacing classical control with quantum controlled gates.
Pulse-level controlquantum: Programming a quantum computer by directly shaping the analogue control signals that implement gates, rather than using predefined gate operations.
Purcell filterquantum: An on-chip filter that lets readout signals pass while blocking the pathway through which a qubit would otherwise decay too quickly into its resonator.
Paul trapquantum: A device that confines charged ions using oscillating electric fields, holding them in place so they can serve as qubits.
Penning trapquantum: An ion trap using a combination of static electric fields and a strong magnetic field to confine charged particles for quantum experiments.
Photolithographyquantum: A chip-making process using light to pattern circuits onto a wafer, adapted from the semiconductor industry to fabricate many qubit devices.
Phase-flip errorquantum: An error that flips the sign between a qubit's superposition components, equivalent to a Pauli Z operator, with no classical counterpart.
Pauli errorquantum: Any error described by an X, Y or Z operator or a product of them; most codes are designed and analysed against Pauli errors.
Pseudo-thresholdquantum: The physical error rate at which a specific code of a specific distance breaks even, sitting below the code family's true asymptotic threshold.
Physical error ratequantum: The probability that an individual hardware operation, such as a gate or measurement, introduces an error on a physical qubit.
Pauli twirlingquantum: Randomly inserting Pauli gates to convert complicated coherent noise into simpler random Pauli errors that codes can handle more easily.
Parity checkquantum: A measurement of the combined value of several qubits that reveals whether an error occurred without exposing the individual qubit states.
Period findingquantum: Determining the repeating interval of a periodic function. Quantum computers solve it efficiently using the quantum Fourier transform, a key step in factoring.
Promise problemquantum: A decision problem where inputs are guaranteed to satisfy some condition. Many quantum speedups, including Deutsch-Jozsa, are proven on promise problems.
Privacy amplificationquantum: A quantum key distribution step that compresses a partly exposed key into a shorter one, shrinking any eavesdropper's knowledge to a negligible amount.
Photon-number-splitting attack (PNS)quantum: An eavesdropping technique exploiting laser pulses that contain several photons, letting an attacker skim one photon per pulse without being detected.
Polarisation encodingquantum: A method of storing quantum information in the polarisation direction of photons, commonly used in free-space and satellite quantum key distribution.
Process fidelityquantum: A measure of how accurately a quantum operation is performed across all inputs, comparing the real process to the ideal one.
Phase dampingquantum: A noise process that destroys the phase relationship in a qubit's superposition without changing its energy, characterised by dephasing.
Probabilistic error cancellation (PEC)quantum: An error mitigation method that runs many adjusted circuits and combines them to cancel noise on average, at high sampling cost.
PennyLanequantum: An open-source Python library from Xanadu focused on quantum machine learning, linking quantum circuits with automatic differentiation and classical tools.

Q

QR Codecrypto: A QR code encodes wallet addresses or transaction details into a scannable pattern, simplifying crypto payments, authentication, and data sharing.
Quadratic Votingcrypto: Quadratic voting allocates influence proportionally to square-rooted token holdings, preventing whales from dominating decentralised governance outcomes unfairly.
Quantstampcrypto: Quantstamp automates smart-contract verification through static analysis and manual audits, improving reliability of on-chain financial systems.
Quantum error correctionquantum: Techniques for keeping fragile qubits reliable by spreading information across many of them. Widely seen as the gate between today’s noisy prototypes and genuinely useful quantum computers.
Qubitquantum: The quantum version of a computer bit. Where a bit is 0 or 1, a qubit can be a blend of both at once — the property that gives quantum computers their (still mostly theoretical) power.
QuickSwapcrypto: QuickSwap is a Polygon-based DEX leveraging Uniswap v2 architecture for fast, inexpensive token trading.
Quorumcrypto: Quorum is the minimum number of participants required for valid decision-making within blockchain governance, consensus rounds, or multi-signature operations.
Quantisationai: A technique that reduces the numerical precision of a model's parameters, shrinking its memory needs and speeding up inference with usually minor accuracy loss.
Quantum gatequantum: A basic operation that changes the state of one or more qubits in a reversible way, forming the building block of quantum circuits.
Quantum circuitquantum: A sequence of quantum gates applied to qubits, usually ending in measurement, that describes and implements a quantum computation.
Quantum advantagequantum: The point at which a quantum computer solves a useful problem faster or better than the best available classical computer could.
Quantum supremacyquantum: The demonstration that a quantum computer can perform a specific task, not necessarily useful, that no classical computer could complete in reasonable time.
Quantum annealingquantum: A computing approach that finds low-energy solutions to optimisation problems by slowly evolving a quantum system towards its lowest-energy configuration.
Quantum volumequantum: A single benchmark number that captures a quantum computer's overall capability, accounting for its qubit count, connectivity and error rates together.
Quantum key distribution (QKD)quantum: A method that uses quantum physics to share encryption keys so that any eavesdropping disturbs the signal and is detected.
Quantum simulatorquantum: A quantum device or system built to imitate and study the behaviour of another quantum system that is hard to model on classical computers.
Query, key and value (QKV)ai: The three learned projections in attention: queries seek information, keys advertise it, and values carry the content that is aggregated.
Quantisation-aware training (QAT)ai: Training or fine-tuning a model while simulating low-precision arithmetic, so it adapts to quantisation and keeps more accuracy than post-training methods.
Query rewritingai: Reshaping a user's question into a clearer or expanded form before retrieval to improve the passages that are found.
Quantum statequantum: The complete mathematical description of a quantum system, encoding the probabilities of every possible outcome for any measurement one might perform.
Quantum measurementquantum: The process of reading classical information from a quantum system, producing one outcome with a probability fixed by the system's state.
Quantum phasequantum: The angular part of a complex amplitude, hidden from a single measurement yet shaping how different quantum possibilities interfere.
Quantum interferencequantum: The addition or cancellation of probability amplitudes, making some outcomes more likely and others less likely or entirely impossible.
Quantum state tomographyquantum: The procedure of reconstructing an unknown quantum state by measuring many identical copies across several different bases.
Quantum teleportationquantum: A protocol transferring an unknown quantum state between two parties using shared entanglement and classical communication.
Quantum coherencequantum: The presence of well-defined phase relationships within a superposition, the resource that enables interference and quantum computation.
Quantum registerquantum: A named group of qubits treated together within a circuit, analogous to a register of bits in a classical computer.
Qubit resetquantum: An operation that returns a qubit to the zero state during a circuit, allowing it to be reused for further computation.
Qubit routingquantum: Inserting SWAP gates during compilation so that two-qubit operations only occur between qubits that are physically connected on the device.
Qubit mappingquantum: Assigning the logical qubits of a circuit to specific physical qubits on a device, chosen to reduce added communication overhead.
Quantum charge-coupled device (QCCD)quantum: A trapped-ion architecture that shuttles ions between storage and interaction zones on a segmented chip to scale up processors.
Quantum dotquantum: A tiny semiconductor region that traps one or a few electrons, whose spin can be used as a qubit in silicon or similar chips.
Quantum processing unit (QPU)quantum: The core chip or device that holds the qubits and performs quantum operations, analogous to a classical computer's processor.
Quantum low-density parity-check code (qLDPC)quantum: An error-correcting code whose stabilisers each involve few qubits, promising strong protection with far less qubit overhead than surface codes.
Qubit overheadquantum: The many physical qubits needed to encode and protect one logical qubit, a central cost of fault-tolerant quantum computing.
Quantum Fourier transform (QFT)quantum: The quantum analogue of the discrete Fourier transform, mapping amplitudes into frequency components efficiently. It underpins phase estimation and Shor's factoring algorithm.
Quantum phase estimation (QPE)quantum: An algorithm that estimates the eigenvalue phase of a unitary operator applied to one of its eigenstates, forming a core subroutine in many quantum algorithms.
Quantum approximate optimisation algorithm (QAOA)quantum: A hybrid quantum-classical algorithm that finds approximate solutions to combinatorial optimisation problems by alternating problem and mixing operations tuned by a classical optimiser.
Quantum walkquantum: The quantum analogue of a classical random walk, where superposition and interference let a walker spread quadratically faster, powering several search and sampling algorithms.
Quantum speedupquantum: The reduction in resources, usually time or queries, that a quantum algorithm achieves over the best known classical algorithm for the same problem.
Quadratic speedupquantum: A quantum advantage that roughly square-roots the running time, such as Grover's search checking N possibilities in about the square root of N steps.
Quantum Merlin-Arthur (QMA)quantum: The quantum counterpart of NP, covering problems whose solutions can be verified efficiently by a quantum computer given a quantum proof state.
Query complexityquantum: A measure of algorithmic cost counting how many times a function or oracle must be consulted, widely used to compare quantum and classical performance.
Quantum machine learning (QML)quantum: A field combining quantum computing with machine learning, aiming to speed up training or inference, or to process quantum data classical methods handle poorly.
Quantum kernel methodquantum: A machine learning approach that uses a quantum computer to compute similarity between data points in a high-dimensional space, then feeds results to a classical model.
Quantum countingquantum: An algorithm combining amplitude amplification with phase estimation to estimate how many solutions a search problem has, rather than just finding one.
Quantum singular value transformation (QSVT)quantum: A unifying framework that applies polynomial functions to the singular values of a matrix embedded in a quantum circuit, generalising many algorithms under one technique.
Quantum signal processing (QSP)quantum: A technique that transforms a single qubit's rotations to apply chosen polynomial functions to a unitary, forming the foundation of quantum singular value transformation.
Quantum principal component analysis (qPCA)quantum: A quantum algorithm that reveals the dominant features of data encoded in a density matrix, potentially faster than classical analysis for certain structured inputs.
Quantum support vector machine (QSVM)quantum: A quantum version of a classical classifier that uses quantum states to compute data similarities, aiming for advantage on complex feature spaces.
Quantum neural network (QNN)quantum: A parameterised quantum circuit trained like a neural network, using tunable gates and measurements to learn patterns from classical or quantum data.
Quantum feature mapquantum: A circuit that encodes classical data into a high-dimensional quantum state, letting quantum kernels separate patterns that are hard to distinguish classically.
Quantum complexity theoryquantum: The study of what problems quantum computers can solve efficiently and how their power compares with classical computation, defining classes such as BQP and QMA.
Quantum linear systems problemquantum: The general task of extracting information about the solution to a large set of linear equations using a quantum computer, addressed by the HHL algorithm.
Quantum bit error rate (QBER)quantum: The fraction of received key bits that disagree with those sent, used in quantum key distribution to gauge channel noise and detect eavesdropping.
Quantum repeaterquantum: A device that extends entanglement across long distances by chaining shorter links with entanglement swapping, purification and quantum memories to overcome photon loss.
Quantum memoryquantum: A device that stores a quantum state faithfully and releases it on demand, an essential building block for quantum repeaters and networks.
Quantum internetquantum: A proposed global network that distributes entanglement between distant users, enabling secure communication, distributed computing and precise sensing beyond classical limits.
Quantum networkquantum: A set of nodes linked by quantum channels that share qubits or entanglement, forming the infrastructure for distributed quantum communication and computing.
Quantum channelquantum: A communication link that carries quantum states, such as photons in fibre or free space, between the nodes of a quantum network.
Quantum random number generator (QRNG)quantum: A device that produces genuinely unpredictable random numbers by measuring inherently random quantum events, unlike algorithmic pseudo-random generators that only appear random.
Quantum-safequantum: Describing cryptographic algorithms or systems believed to withstand attacks by both classical computers and future large-scale quantum computers.
Quantum simulationquantum: Using a controllable quantum system to reproduce and study the behaviour of another quantum system that is too complex to model on ordinary computers.
Qubitisationquantum: An efficient technique for simulating Hamiltonian evolution by encoding it into a single well-structured quantum operation, achieving near-optimal use of resources.
Quantum chemistryquantum: The study of chemical systems using quantum mechanics, aiming to predict molecular structures, energies and reactions from fundamental principles.
Quantum many-body systemquantum: A collection of many interacting quantum particles whose combined behaviour is extremely hard to compute, a prime target for quantum simulation.
Quantum sensingquantum: Using quantum properties such as superposition and entanglement to measure physical quantities with precision beyond conventional instruments.
Quantum metrologyquantum: The science of making high-precision measurements by exploiting quantum effects, aiming to reach the fundamental limits set by physics.
Quantum imagingquantum: Techniques that use quantum correlations of light to produce images with better resolution, contrast or noise than classical optics allow.
Quantum illuminationquantum: A sensing scheme using entangled light to detect faint objects hidden in noisy backgrounds better than any classical signal of the same power.
Quantum radarquantum: A proposed detection system using entangled or quantum-correlated signals to spot targets, aiming for advantages over conventional radar in noisy conditions.
Quantum non-demolition measurement (QND)quantum: A measurement that reads out a quantum property without destroying it, allowing the same quantity to be observed repeatedly.
Quantum Fisher informationquantum: A mathematical measure of how much information a quantum state carries about an unknown parameter, setting the best possible precision for estimating it.
Quasiparticle poisoningquantum: A disruption in superconducting qubits where stray broken electron pairs briefly tunnel across junctions, causing sudden decoherence and errors.
Qubit frequencyquantum: The energy gap between a qubit's two states, expressed as a frequency, which control pulses must match to drive operations.
Quantum process tomographyquantum: A procedure that reconstructs the complete behaviour of a quantum operation by testing it on many inputs and measuring the outputs.
Qiskitquantum: An open-source quantum software toolkit from IBM for building, transpiling and running circuits on simulators and real hardware using Python.
Quantum SDKquantum: A software development kit providing the libraries and tools to write, simulate and run quantum programs on real or virtual devices.
Quantum intermediate representation (QIR)quantum: A hardware-neutral, low-level format based on LLVM for expressing quantum programs so different tools and devices can share a common target.
Q# (Q sharp)quantum: A quantum programming language from Microsoft designed for writing scalable algorithms within its Azure Quantum development kit.

R

Rabby Walletcrypto: Rabby Wallet is a browser extension designed for DeFi safety, simulating transactions and detecting malicious contract interactions before approval.
Radiant Capitalcrypto: Radiant Capital is a cross-chain lending platform built on Arbitrum and BNB Chain, supporting unified collateral management and seamless borrowing.
Radiate Protocolcrypto: Radiate Protocol is a yield automation platform combining optimisation algorithms and strategy vaults to maximise multi-chain DeFi returns.
Railguncrypto: Railgun is an Ethereum privacy system enabling shielded transfers and DeFi interactions via zero-knowledge encryption layers.
Rainbow Walletcrypto: Rainbow Wallet is a user-friendly Ethereum wallet supporting NFTs, swaps, and multichain integration through intuitive mobile and desktop interfaces.
Rainmaker Gamescrypto: Rainmaker Games is a Web3 gaming platform uniting players, guilds, and investors through analytics and NFT asset management.
Rango Exchangecrypto: Rango is a cross-chain DEX aggregator connecting dozens of bridges and AMMs for optimal token-swap routing across networks.
Rari Capitalcrypto: Rari Capital was an early DeFi protocol offering automated yield strategies, later merged into Tribe DAO.
Rariblecrypto: Rarible is an NFT marketplace empowering creators with royalty enforcement and community-governed curation via RARI tokens.
RavenCoin (RVN)crypto: RavenCoin enables asset issuance and transfer on its blockchain, supporting tokenised securities, collectibles, and decentralised messaging.
Raydiumcrypto: Raydium is a Solana-based AMM integrating order-book liquidity from Serum, offering ultra-fast swaps and yield farming opportunities.
Rebase Tokencrypto: A rebase token automatically adjusts its supply to stabilise price targets, maintaining equilibrium via algorithmic expansion or contraction cycles.
Redstone Oraclescrypto: Redstone delivers modular, customisable oracles optimised for DeFi protocols, providing fast, verifiable data feeds from off-chain sources.
Reentrancycrypto: Reentrancy is a vulnerability allowing recursive calls before state updates complete, enabling attackers to exploit contract balances maliciously.
Ref Financecrypto: Ref Finance is the flagship DEX on NEAR Protocol, combining automated market-maker liquidity with stable-swap pools and on-chain governance.
Reflexercrypto: Reflexer Finance issues RAI, a non-pegged stable asset governed algorithmically through on-chain feedback mechanisms ensuring self-stabilising value control.
Ren Bridgecrypto: Ren Bridge once facilitated wrapped Bitcoin and cross-chain transfers before sunsetting under regulatory uncertainty.
Render Networkcrypto: Render Network distributes GPU computing for artists and AI projects, tokenising rendering workloads across decentralised hardware providers efficiently.
RENQcrypto: RENQ aims to unify DeFi and CeFi trading tools, providing aggregation, risk management, and hybrid-exchange liquidity interfaces.
RenVMcrypto: RenVM was a cross-chain liquidity protocol enabling wrapped assets like renBTC, now deprecated after ecosystem consolidation.
Republic Cryptocrypto: Republic Crypto provides token advisory, fundraising, and compliance services for Web3 projects seeking retail and institutional participation.
Reserve Protocolcrypto: Reserve Protocol issues asset-backed stablecoins using diversified collateral baskets and governance mechanisms for long-term price stability and resilience.
Restakingcrypto: Restaking reuses already-staked collateral to secure additional services or chains, compounding yield while spreading validator-based security guarantees.
RFQ (Request for Quote)crypto: An RFQ mechanism invites market makers to provide off-chain quotes, ensuring competitive pricing and reduced slippage in DeFi trading platforms.
Ribbon Financecrypto: Ribbon Finance builds structured products using options strategies to generate sustainable yield on crypto assets.
Ripple (XRP)crypto: Ripple is a payment protocol using the XRP Ledger for rapid, low-cost cross-border transfers between banks and financial institutions globally.
Risk Enginecrypto: A risk engine analyses user exposure across DeFi positions, assessing liquidation probability, volatility correlation, and creditworthiness dynamically.
Roadmapcrypto: A roadmap outlines a project’s planned milestones, technical upgrades, and strategic goals, signalling long-term commitment and transparency to investors.
Rocket Poolcrypto: Rocket Pool is a decentralised Ethereum staking network allowing smaller participants to operate validators with pooled capital and liquid staking tokens.
Rollbitcrypto: Rollbit is a crypto casino and trading platform offering leveraged products, NFTs, and on-chain revenue-sharing via its native RLB token.
Rollupcrypto: A rollup aggregates transactions off-chain, posts proofs on-chain, and inherits security from a base layer while improving throughput and cost efficiency.
Routercrypto: A router contract directs token swaps across liquidity pools or DEXs, optimising execution by comparing multiple route paths automatically.
Router Protocolcrypto: Router Protocol provides cross-chain bridging and liquidity aggregation infrastructure connecting diverse blockchain ecosystems securely.
Rug Pullcrypto: A rug pull occurs when developers abandon a project after draining liquidity or funds, leaving token holders with worthless assets.
Runecrypto: Rune is a forthcoming Bitcoin-layer token protocol enabling native asset issuance directly on the Bitcoin blockchain using ordinals.
RWA (Real-World Assets)crypto: Real-World Assets tokenise tangible items like property, bonds, or invoices, bridging traditional finance liquidity with blockchain-based yield systems.
Reinforcement learning from human feedback (RLHF)ai: A training method that uses human preference ratings to teach a model which responses people judge more helpful or appropriate.
Retrieval-augmented generation (RAG)ai: A technique that fetches relevant documents from an external source and supplies them to a model so its answers draw on specific, current information.
Rectified linear unit (ReLU)ai: A simple activation that outputs the input if positive and zero otherwise, adding non-linearity while being cheap to compute.
Root mean square normalisation (RMSNorm)ai: A lighter normalisation that scales a vector by its root mean square and learned weights, without subtracting the mean.
Rotary position embedding (RoPE)ai: A positional method that rotates query and key vectors by angles proportional to position, encoding relative distance directly within attention.
Relative positional encodingai: A method that encodes the distance between tokens rather than their absolute positions, helping models generalise across sequence lengths.
Residual connectionai: A shortcut that adds a layer's input to its output, easing gradient flow and enabling the training of very deep networks.
Recurrent neural network (RNN)ai: A network that processes sequences one step at a time, carrying a hidden state that summarises the information seen so far.
RMSPropai: An optimiser that divides each update by a running average of recent squared gradients, stabilising learning across weights with very different scales.
Regularisationai: Any technique that discourages a model from fitting the training data too closely, improving how well it performs on unseen examples.
Repetition penaltyai: A decoding setting that lowers the probability of tokens already used, discouraging the model from repeating the same words or phrases.
Reverse diffusion processai: The generation stage where the model begins from noise and iteratively predicts and removes it to build a coherent sample.
Rectified flowai: A flow-based generative approach that learns near-straight paths from noise to data, allowing high-quality samples in very few steps.
Reinforcement learning from AI feedback (RLAIF)ai: Like RLHF, but the preference labels used to train the reward signal come from another AI model rather than from people.
Reward modelai: A model trained to predict human preferences and output a numerical score, used to guide fine-tuning of a language model during RLHF.
Reward hackingai: When a model exploits flaws or loopholes in its reward signal to score highly without actually achieving the intended goal.
Red-teamingai: Deliberately probing a model with adversarial prompts and scenarios to uncover harmful outputs, security flaws or safety failures before release.
Refusalai: A model's decision to decline a request that breaches its usage policies, ideally with a brief explanation rather than silent non-compliance.
Responsible scaling policy (RSP)ai: A published commitment tying a lab's deployment of more capable models to defined risk thresholds and the safeguards required at each level.
Reinforcement learning (RL)ai: Training an agent to make decisions by rewarding good actions and penalising bad ones as it interacts with an environment.
Regressionai: A supervised task that predicts a continuous numerical value, such as a house price or tomorrow's temperature.
Random forestai: An ensemble that combines many decision trees trained on random data and feature subsets, averaging their predictions for greater accuracy.
Recallai: Of all the truly positive items, the fraction the model correctly identified; high recall means few positives are missed.
Receiver operating characteristic (ROC) curveai: A plot of the true positive rate against the false positive rate across thresholds, showing a classifier's trade-offs.
R-squared (coefficient of determination)ai: A measure of how much of the variation in the target a regression model explains, ranging up to one.
ROUGE (recall-oriented understudy for gisting evaluation)ai: A set of metrics scoring summaries by measuring overlap between the generated text and one or more human reference summaries.
ReAct (Reason and Act)ai: A prompting pattern where a model alternates between writing reasoning steps and taking actions such as tool calls, using each observation to guide the next step.
Reflectionai: A technique where a model reviews its own earlier output or actions and revises them to improve the quality of the result.
Re-rankingai: A second pass that reorders an initial set of retrieved results using a more accurate model to surface the most relevant first.
Reciprocal rank fusion (RRF)ai: A method that merges several ranked result lists into one by combining each item's positions across the different lists.
Retrieverai: The component that searches a knowledge store and returns the passages most relevant to a query for the model to use.
Roofline modelai: A chart showing whether a task is limited by compute or by memory bandwidth, helping engineers find the true performance ceiling.
Rollbackai: Reverting a system to a previous working version after a deployment causes problems, restoring service quickly while the fault is investigated.
Relative phasequantum: The phase difference between the amplitudes of different basis states, which does influence interference and the resulting measurement statistics.
Reduced density matrixquantum: The density matrix of a subsystem, obtained by tracing out the rest and describing that part's local measurement statistics.
Rx rotation gate (Rx)quantum: A single-qubit gate that rotates the qubit state by a chosen angle about the x-axis of the Bloch sphere.
Ry rotation gate (Ry)quantum: A single-qubit gate that rotates the qubit state by a chosen angle about the y-axis of the Bloch sphere.
Rz rotation gate (Rz)quantum: A single-qubit gate that rotates the qubit state by a chosen angle about the z-axis of the Bloch sphere, adjusting relative phase.
Readout resonatorquantum: A microwave resonator attached to a qubit whose resonant frequency shifts depending on the qubit's state, allowing that state to be measured.
Rydberg atomquantum: An atom with an electron excited to a very high energy level, giving it a large size and strong interactions used to entangle qubits.
Rydberg blockadequantum: An effect where one excited atom prevents a nearby atom from being excited too, providing the interaction that creates entangling gates in neutral-atom systems.
Resonatorquantum: A circuit or cavity that stores electromagnetic energy at specific frequencies, used to control, couple and read out qubits.
Repetition codequantum: The simplest code, copying one logical bit across several physical qubits to correct either bit-flip or phase-flip errors, but not both.
Real-time decodingquantum: Running the decoder fast enough to keep pace with the hardware, so corrections are ready before errors accumulate and computation stalls.
Rotated surface codequantum: A more compact layout of the surface code that protects a logical qubit using fewer physical qubits for the same code distance.
Random circuit samplingquantum: Running a random quantum circuit and sampling its outputs, a task used by Google to claim quantum supremacy because classical simulation is extremely costly.
Ring-LWEquantum: A structured, more efficient variant of the Learning With Errors problem set in polynomial rings, used to build compact lattice-based cryptographic schemes.
RSAquantum: A widely used public-key cryptosystem whose security relies on the difficulty of factoring large numbers, a task Shor's algorithm could render easy.
Ramsey interferometryquantum: A technique using two separated pulses to make atoms interfere over time, sharpening measurements of frequency in atomic clocks and sensors.
Rabi oscillationquantum: The cyclic swapping of a quantum system between two energy levels when driven by a resonant field, underpinning control in sensing and computing.
Readout errorquantum: A mistake made when measuring a qubit, where the reported result differs from the true state because of imperfect detection.
Ramsey experimentquantum: A two-pulse measurement that reveals a qubit's frequency and dephasing time by observing how its phase evolves during a wait.
Randomised benchmarking (RB)quantum: A method that estimates average gate error by running random sequences of gates and measuring how quickly performance decays with length.
Readout error mitigationquantum: Post-processing that corrects measurement results using a previously measured map of how often each outcome is misreported.

S

Sabliercrypto: Sablier is a streaming-payments protocol enabling real-time token transfers, payroll automation, and continuous vesting of on-chain income.
Saddle Financecrypto: Saddle Finance was an Ethereum stable-swap AMM optimised for low-slippage trades between pegged assets like USDC and DAI.
Safe (Gnosis Safe)crypto: Safe is a multi-signature smart-wallet platform for teams and DAOs, ensuring secure treasury management and permissioned spending control.
SafePalcrypto: SafePal manufactures hardware and mobile wallets integrating DeFi access, staking, and cross-chain swaps.
Samourai Walletcrypto: Samourai Wallet emphasises Bitcoin privacy through Whirlpool mixes, coin-control tools, and Tor integration for censorship resistance.
Sandwich Attackcrypto: A sandwich attack occurs when MEV bots place trades before and after a victim’s transaction to profit from predictable price movement.
Saros Financecrypto: Saros Finance is a DeFi suite on Solana offering swaps, staking, and token-launch utilities for community projects.
Satcrypto: A sat, short for satoshi, represents one hundred-millionth of a Bitcoin, functioning as the smallest divisible unit within Bitcoin’s network.
Satoshi Nakamotocrypto: Satoshi Nakamoto is the pseudonymous creator of Bitcoin, whose white paper and implementation introduced decentralised, trust-minimised digital money.
SBT (Soulbound Token)crypto: A Soulbound Token is a non-transferable credential representing identity, achievements, or reputation within decentralised social or governance ecosystems.
Scalabilitycrypto: Scalability measures a blockchain’s ability to handle increased transactions per second without compromising decentralisation, security, or performance stability.
Schnorr Signaturescrypto: Schnorr signatures enable multi-key aggregation for compact proofs, improving privacy, scalability, and efficiency in Bitcoin and advanced crypto systems.
Scorechaincrypto: Scorechain provides blockchain-risk-scoring, compliance APIs, and analytics dashboards for institutional crypto-asset monitoring and audit trails.
Secret Networkcrypto: Secret Network enables privacy-preserving smart contracts using encrypted inputs, outputs, and state secured by trusted-execution environments.
Secret Tokencrypto: Secret Token represents encrypted assets on Secret Network, preserving data confidentiality within smart-contract execution.
Seed Phrasecrypto: A seed phrase is a human-readable backup of a wallet’s private keys, enabling full account recovery and cryptographic ownership restoration.
Sei Networkcrypto: Sei is a sector-specific layer-one blockchain optimised for trading, offering sub-second finality and parallelised transaction execution.
Sentinel dVPNcrypto: Sentinel provides decentralised VPN infrastructure built on Cosmos, rewarding bandwidth providers with native SENT tokens for secure routing.
Sequencercrypto: A sequencer orders transactions in rollups, affecting fairness and latency while ensuring deterministic block production for layer-two scalability.
Session Keyscrypto: Session keys grant temporary permissions for smart-account actions, enabling secure delegation without revealing permanent wallet credentials.
Shardexcrypto: Shardex offers multichain liquidity aggregation and automated yield optimisation across EVM networks with cross-chain execution support.
Shardingcrypto: Sharding partitions blockchain data horizontally, distributing workload among shards to improve scalability and transaction throughput across networks.
Shiba Inu (SHIB)crypto: Shiba Inu is a meme-origin cryptocurrency evolving into a decentralised ecosystem including DEXs, NFTs, and community-driven governance features.
Shielded Poolcrypto: A shielded pool hides transaction amounts and addresses using zero-knowledge proofs, improving on-chain privacy for compliant DeFi activity.
Skale Networkcrypto: Skale is an elastic sidechain network for Ethereum enabling high-throughput, zero-gas applications using pooled validator security.
Sky Protocolcrypto: Sky Protocol (formerly MakerDAO) governs DAI and stUSDS, integrating risk management, stability mechanisms, and institutional DeFi participation.
Skynet Analyticscrypto: Skynet Analytics provides on-chain risk metrics and behavioural analysis tools supporting DeFi security and compliance monitoring.
Slippagecrypto: Slippage measures price difference between expected and executed trade values, influenced by liquidity depth, volatility, and order size.
SlowMistcrypto: SlowMist is a cybersecurity firm specialising in smart-contract audits, threat-intelligence, and incident-response reporting for global Web3 ecosystems.
Smart Contractcrypto: A smart contract is self-executing code stored on-chain, enforcing agreements automatically without intermediaries or manual intervention.
Snapshotcrypto: Snapshot is an off-chain governance platform enabling token-weighted voting without gas fees, storing verified results via cryptographic signatures.
Socket Protocolcrypto: Socket enables secure cross-chain liquidity bridging and composable interoperability primitives for developers and decentralised exchanges.
Solana (SOL)crypto: Solana is a high-throughput blockchain using proof-of-history and parallel execution to achieve low-latency transactions and scalable decentralised applications.
Sora (XOR)crypto: Sora is a decentralised economic system and parachain building interoperability across Polkadot and Kusama, using XOR as its utility token.
SpaceIDcrypto: SpaceID provides universal blockchain domain names linking wallets and decentralised identities across Ethereum, BNB Chain, and Arbitrum.
Spillwaycrypto: Spillway routes liquidity between AMMs, optimising capital efficiency and mitigating impermanent loss via algorithmic rebalancing strategies.
SpookySwapcrypto: SpookySwap is a Fantom-based AMM and DEX offering native staking, NFTs, and low-fee cross-chain bridging integrations.
Spruce IDcrypto: Spruce ID develops open-source identity-verification standards for Web3 wallets and credential issuers.
Stablecoincrypto: A stablecoin maintains a fixed value relative to fiat or commodities, providing stability for trading, payments, and on-chain financial applications.
StableSwap Invariantcrypto: StableSwap invariant is a curve formula optimised for correlated assets, balancing stability and liquidity efficiency within decentralised exchanges.
Stackscrypto: Stacks brings smart-contract functionality to Bitcoin via a separate layer anchored by proof-of-transfer consensus.
Stadercrypto: Stader provides liquid-staking infrastructure for multiple proof-of-stake blockchains, issuing derivatives like ETHx or MATICx representing staked positions.
Stafi Protocolcrypto: Stafi allows users to stake PoS assets while maintaining liquidity through rTokens redeemable for underlying deposits and accrued rewards.
StakeWisecrypto: StakeWise is an Ethereum liquid-staking service using dual-token accounting to separate principal and yield for flexible DeFi composition.
Stakingcrypto: Staking locks tokens to secure a proof-of-stake network, rewarding participants with yield while deterring malicious behaviour economically.
Starknetcrypto: Starknet is a validity-rollup using STARK proofs for scalability, maintaining Ethereum-level security while enabling high-throughput smart contracts.
Stealth EXcrypto: Stealth EX is a non-custodial crypto-exchange aggregator enabling instant swaps across blockchains without account registration.
Stridecrypto: Stride offers liquid staking across Cosmos chains, issuing stTokens redeemable for staked assets and rewards in IBC-connected ecosystems.
Strikecrypto: Strike is a Bitcoin-based payment app using Lightning Network for instant, low-fee global remittances and merchant settlements.
Sudoswapcrypto: Sudoswap introduced automated market-maker mechanics for NFTs, replacing order books with dynamic bonding curves for efficient pricing.
Sui Networkcrypto: Sui is a Move-based layer-one blockchain emphasising parallel transaction execution, object-centric storage, and developer-friendly scalability.
Sushiswapcrypto: Sushiswap is a community-driven DEX fork of Uniswap adding yield farms, token incentives, and cross-chain integrations.
Swell Networkcrypto: Swell Network offers non-custodial Ethereum liquid staking, issuing yield-bearing swETH tokens redeemable for underlying ETH.
SWIFT CBDC Bridgecrypto: SWIFT CBDC Bridge experiments with connecting central-bank digital-currency systems across jurisdictions for interoperable cross-border settlement.
Synapse Protocolcrypto: Synapse is a cross-chain messaging and liquidity network enabling asset transfers, bridging, and multichain governance interoperability.
SyncSwapcrypto: SyncSwap is a zkSync-based AMM combining efficient zero-knowledge settlements with low gas fees and unified liquidity routing.
Synonymcrypto: Synonym builds Bitcoin-native Web3 tools enabling decentralised identity and peer-to-peer application layers.
System promptai: A background instruction given to a model before the conversation, setting its role, tone and rules that shape every subsequent response.
Superpositionquantum: A quantum system's ability to exist in a combination of multiple states at once, with each state carrying a probability amplitude until measured.
Surface codequantum: A widely studied quantum error-correcting code that arranges physical qubits on a two-dimensional grid, tolerating relatively high error rates.
Superconducting qubitquantum: A qubit made from superconducting electrical circuits cooled to near absolute zero, controlled with microwave pulses and read out electronically.
Shor's algorithmquantum: A quantum algorithm that can factor large whole numbers exponentially faster than known classical methods, threatening much of today's public-key cryptography.
Self-attentionai: An attention operation where each element of a sequence attends to every other element of the same sequence to build context-aware representations.
Scaled dot-product attentionai: Attention that scores query and key pairs by their dot product, divides by a scaling factor, applies softmax, then weights the values.
Sliding-window attentionai: Attention in which each token attends only to a fixed window of nearby tokens, cutting cost for very long sequences.
Sparse attentionai: Attention that connects each token to only a subset of others following a fixed or learned pattern, reducing quadratic cost.
SwiGLUai: A gated activation used in transformer feedforward layers, combining a Swish activation with a gated linear unit to improve quality.
Sigmoid-weighted linear unit (SiLU)ai: An activation, also called Swish, that multiplies each input by its sigmoid, giving a smooth curve that can output small negative values.
Softmax functionai: A function that turns a vector of scores into a probability distribution, where larger scores receive exponentially larger shares summing to one.
Sigmoid functionai: An activation that squashes any real number into the range zero to one, historically used for gates and binary outputs.
Sinusoidal positional encodingai: Fixed positional signals built from sine and cosine waves of different frequencies, added to embeddings in the original transformer design.
Skip connectionai: A link that passes activations from an earlier layer directly to a later one, bypassing intermediate layers to preserve information.
State-space model (SSM)ai: A sequence model that maintains a hidden state evolving through linear recurrence, allowing efficient handling of long-range dependencies.
Sequence-to-sequence (seq2seq)ai: A framework that maps an input sequence to an output sequence of possibly different length, as in translation or summarisation.
Stochastic gradient descent (SGD)ai: An optimisation method that updates model weights using the gradient from one small batch of examples at a time rather than everything.
Self-supervised learningai: Training where labels are generated from the data itself, such as predicting hidden words, removing the need for costly human annotation.
Semi-supervised learningai: Training that combines a small amount of labelled data with a large amount of unlabelled data to improve results affordably.
Synthetic dataai: Artificially generated training examples, often produced by a model or simulation, used to supplement or replace scarce real-world data.
Subword tokenisationai: Splitting text into pieces smaller than words, so rare or unseen words are represented by known fragments rather than treated as unknown.
SentencePieceai: An open-source tokenisation toolkit that treats text as a raw stream, including spaces, so it works across languages without needing pre-split words.
Special tokensai: Reserved tokens that carry structural meaning rather than ordinary text, marking things like the start of a sequence, padding, or the end of a turn.
Samplingai: Choosing the next token at random according to its probability rather than always taking the most likely one, adding variety to generated text.
Structured outputai: Model responses forced to match a defined schema, such as JSON with set fields, so the result can be parsed reliably by software.
Stop sequenceai: A string that tells the model to halt generation as soon as it is produced, used to end responses at a natural or defined boundary.
Speculative decodingai: A speed-up where a small draft model proposes several tokens ahead and the main model verifies them in one pass, accepting those it agrees with.
Static batchingai: Grouping a fixed set of requests and processing them together from start to finish; simpler than continuous batching but wastes capacity when requests differ in length.
Streamingai: Sending a model's output token by token as it is generated, so users see text appear progressively instead of waiting for the whole response.
Structured pruningai: Pruning that removes entire units such as neurons, attention heads or layers, so the smaller model runs faster on ordinary hardware.
Sparsityai: The share of a model's weights that are zero; higher sparsity can reduce storage and computation if the hardware exploits the missing values.
Score-based generative modelai: A model that learns the gradient of the data distribution and follows it to turn noise into realistic samples, closely related to diffusion.
Seedai: A number that fixes the random starting point of generation, so reusing it reproduces the same output for a given prompt and settings.
Super-resolutionai: Increasing an image or video's resolution by generating realistic fine detail rather than simply stretching the existing pixels.
Samplerai: The algorithm that decides how a diffusion model steps from noise to a finished sample, affecting speed, quality and style.
Sampling stepsai: The number of denoising iterations used to produce an output; more steps often improve quality but take longer to compute.
Signed distance function (SDF)ai: A way of representing a 3D surface by the distance from any point to it, commonly used in generative 3D pipelines.
Spectrogramai: A visual map of how a sound's frequencies change over time, commonly used as an intermediate step in speech and audio generation.
SynthIDai: A watermarking technology that embeds an imperceptible signal into AI-generated images, audio, text or video for later detection.
Specification gamingai: Behaviour that satisfies the literal wording of an objective while violating its intended spirit, a common failure when goals are imperfectly specified.
Safety classifierai: A separate model that screens inputs or outputs for harmful content and blocks or flags them, adding a safety layer around the main model.
Sycophancyai: A model's tendency to agree with a user or tell them what they want to hear, rather than giving an accurate or honest answer.
Scalable oversightai: Research into supervising and evaluating AI on tasks too hard or costly for humans to check directly, for example by using AI assistance.
Superalignmentai: Research aimed at reliably aligning and controlling AI systems that are far more capable than humans, before such systems are built.
Sandbaggingai: When a model deliberately underperforms on a safety evaluation to hide a dangerous capability, so it appears less capable than it truly is.
Situational awarenessai: A model's ability to recognise that it is an AI system, and to tell whether it is being tested, trained or deployed.
Sparse autoencoder (SAE)ai: An interpretability tool that decomposes a model's dense activations into many sparse features, each hopefully corresponding to a single human-understandable concept.
SWE-benchai: A benchmark that gives a model real GitHub issues from open-source projects and checks whether its code patch makes the repository's tests pass.
System cardai: A broader disclosure than a model card, documenting how a deployed system was tested, along with its safety evaluations, risks and mitigations.
Safety caseai: A structured, evidence-backed argument that a system is safe enough to deploy for a given use, drawing on evaluations, testing and safeguards.
Supervised learningai: Training a model on labelled examples, where each input carries a known correct output, so it learns to predict labels for new data.
Support vector machine (SVM)ai: A classifier that separates classes by finding the decision boundary with the widest possible margin between them.
Specificityai: Of all the truly negative items, the fraction a model correctly identifies as negative; also called the true negative rate.
Stemmingai: Reducing words to a crude root by chopping off endings, so that running and runs both map to the same stem.
Stop wordsai: Common words such as the, is and and, often removed before analysis because they carry little distinguishing meaning.
Sentiment analysisai: Automatically determining the emotional tone of text, such as whether a review is positive, negative or neutral.
Self-critiqueai: Prompting a model to find faults in its own answer and suggest corrections before it produces a final response.
Self-consistencyai: Sampling several independent reasoning paths for one question and choosing the most common answer to raise reliability over a single attempt.
Supervisor agentai: An agent that oversees others, assigning work, checking progress and deciding when the overall task is complete.
Short-term memoryai: Information an agent holds only for the current task or conversation, usually kept inside the active context window.
Semantic memoryai: An agent's store of general facts and concepts, kept separate from any particular past event or interaction.
Scratchpadai: A working space where a model writes intermediate reasoning or notes while solving a task, kept apart from its final answer.
Semantic chunkingai: Splitting text at meaningful boundaries such as topic shifts rather than at fixed lengths, so each chunk stays coherent.
Semantic searchai: Finding results by meaning rather than exact words, usually by comparing embeddings of the query and the stored text.
Sandboxai: An isolated environment where an agent's code or actions can run safely without affecting the wider system or real data.
Superposition principlequantum: The rule that any linear combination of valid quantum states is itself another valid state of the same system.
State normalisationquantum: The requirement that a quantum state's squared amplitudes sum to one, keeping the total probability across outcomes equal to one.
Separable statequantum: A joint state expressible as a mixture of independent subsystem states, and therefore containing no entanglement.
Superdense codingquantum: A protocol using one shared entangled qubit to transmit two classical bits by sending a single qubit.
S-dagger gate (S dagger)quantum: The inverse of the phase gate, subtracting a ninety degree phase from the one state to undo the S gate's action.
Square root of NOT gate (SX)quantum: A single-qubit gate that performs half of a bit flip, so applying it twice equals a Pauli-X; it is common as a hardware-native operation.
SWAP gatequantum: A two-qubit gate that exchanges the states of two qubits, equivalent to three CNOT gates applied in sequence.
Single-qubit gatequantum: A gate acting on one qubit at a time, corresponding to a rotation of that qubit's state on the Bloch sphere.
SWAP networkquantum: A structured pattern of SWAP gates that moves qubit states around a limited-connectivity device so that distant qubits can interact.
Solovay-Kitaev theoremquantum: A result guaranteeing that any single-qubit gate can be approximated efficiently to any accuracy using a fixed universal set of gates.
Stabiliser formalismquantum: A compact way of describing certain quantum states and Clifford circuits using Pauli operators, underpinning much of quantum error correction.
Superconducting quantum interference device (SQUID)quantum: A loop containing Josephson junctions that is extremely sensitive to magnetic flux, used to tune qubit frequencies and detect tiny magnetic signals.
Surface trapquantum: A chip-based ion trap where control electrodes lie on a flat surface, letting ions be held and moved just above the chip.
Sideband coolingquantum: A laser technique that removes vibrational energy from trapped ions, bringing their motion close to its lowest quantum state for high-fidelity gates.
Single-photon sourcequantum: A device that emits exactly one photon on demand, a key ingredient for photonic quantum computing and quantum communication.
Single-photon detectorquantum: A highly sensitive device that registers the arrival of an individual photon, essential for measuring the state of photonic qubits.
Squeezed lightquantum: Light engineered so that noise in one property drops below the usual quantum limit, used in continuous-variable photonic computing and precision sensing.
Spin qubitquantum: A qubit encoded in the spin of an electron or nucleus, usually confined in a semiconductor structure and controlled electrically or magnetically.
Silicon qubitquantum: A spin qubit built in silicon, benefiting from mature chip manufacturing and long coherence when made from isotopically purified material.
Singlet-triplet qubitquantum: A spin qubit encoded in the combined state of two electrons in adjacent quantum dots, controlled by tuning their exchange interaction.
Semiconductor nanowirequantum: A very thin wire-like crystal that, when coupled to a superconductor, is a leading candidate for hosting Majorana zero modes.
Stabiliser codequantum: A quantum error-correcting code defined by a group of commuting Pauli operators whose shared +1 eigenspace stores the protected logical information.
Stabiliser generatorquantum: One of the independent Pauli operators that generate a stabiliser code's group; measuring each reveals errors without disturbing the encoded data.
Syndrome measurementquantum: Measuring a code's stabiliser operators to detect errors, giving classical outcomes that flag problems without revealing or collapsing the encoded data.
Syndrome extractionquantum: The circuit process of coupling data qubits to ancillas and measuring them to read out error syndromes each correction round.
Shor codequantum: The first quantum error-correcting code, using nine physical qubits to protect one logical qubit against any single-qubit error.
Steane codequantum: A seven-qubit CSS code that protects one logical qubit and allows all Clifford gates to be performed transversally.
Space-time overheadquantum: The combined cost in qubits and time steps to run a fault-tolerant computation, a fuller measure than counting qubits alone.
Subsystem codequantum: A code with extra gauge degrees of freedom that need not be tracked, often simplifying syndrome measurement and fault-tolerant design.
Single-shot error correctionquantum: Correcting errors reliably from just one round of noisy syndrome measurements, avoiding the usual need to repeat rounds many times.
Simon's algorithmquantum: A quantum algorithm that finds a hidden period of a function exponentially faster than any classical method, inspiring the techniques behind Shor's algorithm.
Sampling problemquantum: A computational task requiring outputs drawn from a specific probability distribution rather than a single answer, central to early demonstrations of quantum advantage.
Siftingquantum: The step in quantum key distribution where the two parties publicly compare their basis choices and discard every bit measured in mismatched bases.
Secret key ratequantum: The number of secure bits per second, or per pulse, that a quantum key distribution link produces after error correction and privacy amplification.
Satellite QKDquantum: Quantum key distribution carried out using photons exchanged with an orbiting satellite, extending secure links far beyond the reach of ground fibre.
Spontaneous parametric down-conversion (SPDC)quantum: A nonlinear optical process that splits one photon into two lower-energy photons, widely used as a reliable source of entangled photon pairs.
Superconducting nanowire single-photon detector (SNSPD)quantum: A highly efficient, low-noise light detector used in quantum communication to register individual photons, especially at telecom wavelengths in optical fibre.
Shortest Vector Problem (SVP)quantum: The task of finding the shortest non-zero vector in a lattice, a problem thought to resist quantum attack and underlying lattice cryptography.
SLH-DSAquantum: The Stateless Hash-based Digital Signature Algorithm, published by NIST as FIPS 205 and derived from SPHINCS+, relying only on hash-function security.
SPHINCS+quantum: A stateless hash-based signature scheme whose security depends solely on hash functions, chosen by NIST and standardised as SLH-DSA.
SIKEquantum: Supersingular Isogeny Key Encapsulation, a compact post-quantum candidate broken in 2022 by a classical mathematical attack, ending its place in standardisation.
Second quantisationquantum: A framework describing systems of identical particles by how many occupy each available state, widely used to encode chemistry for quantum computers.
SQUIDquantum: A superconducting quantum interference device, an extremely sensitive magnetometer that detects tiny magnetic fields using superconducting loops and quantum interference.
Spin squeezingquantum: Redistributing quantum noise among many entangled atoms so a chosen collective property can be measured more precisely, enhancing clocks and sensors.
Standard quantum limit (SQL)quantum: The best precision achievable using independent, unentangled particles; entanglement and squeezing are needed to surpass it.
Shot noisequantum: Random fluctuations arising from counting discrete particles such as photons, setting a basic precision limit that quantum techniques can beat.
State fidelityquantum: A measure of how close a prepared quantum state is to the intended target, ranging from 0 to 1, where 1 means identical.
Single-qubit gate fidelityquantum: The accuracy of an operation acting on one qubit, typically much higher than two-qubit gate fidelity on current hardware.
Spin echo (Hahn echo)quantum: A control technique that reverses slow dephasing with a mid-sequence pulse, extending the measured coherence of a qubit.
State preparation and measurement error (SPAM)quantum: The combined error from imperfectly initialising qubits and imperfectly reading them out, often reported together because the two are hard to separate.
SWAP insertionquantum: Adding SWAP gates during transpilation to move qubit states around a chip so that distant qubits can interact.

T

Taloscrypto: Talos is an institutional trading platform offering connectivity, settlement, and risk tools across digital-asset markets.
TapiocaDAOcrypto: TapiocaDAO issues omnichain stablecoins collateralised by yield-bearing assets, integrating cross-chain borrowing and lending functionalities.
Taprootcrypto: Taproot is a Bitcoin upgrade improving privacy and script flexibility by merging complex conditions into a single Schnorr-based signature output.
Tempus Financecrypto: Tempus tokenises future yield and principal separately, enabling fixed-income markets within decentralised finance environments.
Tensorcrypto: Tensor is a professional Solana NFT marketplace offering real-time analytics, liquidity aggregation, and advanced trading automation.
Thala Labscrypto: Thala Labs builds decentralised financial primitives on Aptos, including a native stablecoin, DEX, and lending markets.
Thorchaincrypto: Thorchain facilitates native cross-chain swaps without wrapped assets, using continuous liquidity pools secured by bonded RUNE validators.
ThorWalletcrypto: ThorWallet connects users to Thorchain liquidity pools, facilitating cross-chain swaps and staking through a mobile non-custodial interface.
Threshold Networkcrypto: Threshold merges Keep and NuCypher, offering threshold encryption, access control, and Bitcoin-to-Ethereum bridging via tBTC.
Threshold Network tBTCcrypto: tBTC enables Bitcoin holders to mint tokenised BTC on Ethereum using threshold-signature bridges for trust-minimised custody.
Tick (CLMM)crypto: A tick defines discrete price intervals within concentrated-liquidity AMMs, allowing efficient capital placement along specific trading ranges.
Timelockcrypto: A timelock restricts fund movement until a future timestamp, providing safety for contracts, governance proposals, or vesting schedules.
Tokemakcrypto: Tokemak coordinates decentralised liquidity provisioning, directing capital efficiently through tokenised reactors governed by TOKE holders.
Tokencrypto: A token represents digital value, rights, or access within a blockchain network, governed by smart-contract rules defining issuance and transfers.
Token Terminalcrypto: Token Terminal aggregates on-chain financial data, offering fundamentals-based metrics for protocol valuation and comparison.
Tokenomicscrypto: Tokenomics studies a cryptocurrency’s supply, distribution, incentives, and economic design, shaping sustainability and community alignment over time.
TokenUnlockscrypto: TokenUnlocks tracks vesting schedules and cliff releases, informing investors about upcoming token supply events impacting price dynamics.
Toniccrypto: Tonic is a decentralised exchange on NEAR’s Aurora network, focusing on institutional-grade execution and risk-managed liquidity.
Trader Joecrypto: Trader Joe is an Avalanche DEX combining swaps, lending, and NFT markets with revenue-sharing governance through JOE tokens.
Trail of Bitscrypto: Trail of Bits is a cybersecurity company conducting advanced audits and cryptanalysis for blockchain protocols and open-source software.
Travel Rulecrypto: The Travel Rule mandates exchanges share sender and recipient information, aligning crypto transfers with anti-money-laundering compliance standards.
TreasureDAOcrypto: TreasureDAO builds a decentralised gaming ecosystem on Arbitrum linking studios, NFTs, and in-game economies via the MAGIC token.
Trezorcrypto: Trezor is a hardware wallet brand pioneering open-source cold-storage devices for secure cryptocurrency management.
Tridentcrypto: Trident is Sushiswap’s next-generation AMM framework supporting multiple pool types for enhanced capital efficiency and strategy flexibility.
TRM Labscrypto: TRM Labs delivers blockchain-intelligence solutions detecting fraud, sanctions breaches, and money-laundering risks through multi-chain entity-behaviour analytics.
Tron (TRX)crypto: Tron is a blockchain for high-throughput digital content sharing and decentralised applications, featuring low fees and delegated governance.
Trust Walletcrypto: Trust Wallet is a mobile crypto wallet supporting thousands of tokens and NFT collections across multiple blockchains.
Trustlesscrypto: Trustless systems eliminate the need for intermediaries by using cryptography and consensus to guarantee predictable, verifiable transaction execution.
Turing Completecrypto: A Turing-complete blockchain can execute any computable function, allowing sophisticated logic and programmable decentralised applications.
TVL (Total Value Locked)crypto: Total Value Locked measures the amount of assets deposited within a DeFi protocol, indicating ecosystem trust and on-chain capital utilisation.
TWAMMcrypto: A Time-Weighted Automated Market Maker executes large orders gradually, minimising market impact and MEV risk during extended trading intervals.
TWAP / VWAPcrypto: TWAP and VWAP are algorithmic order types averaging execution by time or volume, reducing slippage for large institutional crypto trades.
Transformerai: A neural network architecture that uses attention to weigh the relationships between all tokens in a sequence, underpinning most modern language models.
Trainingai: The compute-intensive process of adjusting a model's parameters by exposing it to large datasets so it learns to predict or generate.
Tokeniserai: The component that splits text into tokens for a model to process, and reassembles tokens back into readable text on output.
Trapped-ion qubitquantum: A qubit encoded in the internal energy states of a charged atom held in place by electromagnetic fields and manipulated with lasers.
Topological qubitquantum: A proposed qubit that stores information in global, topological features of a system, making it intrinsically resistant to local noise and errors.
Transformer blockai: A repeatable unit combining an attention sub-layer and a feedforward sub-layer, each wrapped with normalisation and a residual connection.
Top-k routingai: A strategy that sends each token to its k highest-scoring experts, keeping computation sparse while drawing on many parameters.
Test setai: Data kept completely separate until the end, used once to give an honest estimate of how the finished model performs.
Transfer learningai: Reusing a model trained on one task as the starting point for another, so learned features speed up training on new problems.
Tokenised datasetai: A dataset converted into sequences of tokens, the numerical units a language model reads, prepared in advance so training runs efficiently.
Token IDai: The integer index that identifies a token within the vocabulary; models work with these numbers, not the text pieces directly.
Temperatureai: A sampling setting that scales the token probabilities; low values make output focused and predictable, high values make it more varied and random.
Top-k samplingai: A decoding method that restricts each choice to the k most likely tokens, sampling from those and ignoring the long tail of unlikely options.
Throughputai: The total amount of work a serving system completes per unit time, often measured in tokens or requests per second across all users.
Time to first token (TTFT)ai: How long a user waits from sending a prompt until the first output token appears, dominated by the prefill stage.
Tokens per second (TPS)ai: A measure of generation speed counting how many tokens a model produces each second, commonly used to compare serving performance.
Text-to-imageai: Generating a still image directly from a written description, the flagship capability of modern diffusion image generators and similar systems.
Text-to-videoai: Generating moving footage from a written description, extending image generation with motion and consistency across a sequence of frames.
Text-to-audioai: Generating sound, music or speech from a written description or prompt using a learned audio model.
Text-to-3Dai: Generating three-dimensional shapes or scenes from a written description, often by guiding a 3D representation with a 2D image model.
Textual inversionai: A method that learns a new prompt token to represent a specific subject or style from a few example images.
Text-to-speech (TTS)ai: Converting written text into spoken audio using a synthetic voice, so software can read documents, messages or articles aloud.
Temporal consistencyai: The quality of keeping objects, colours and details stable across video frames so generated footage does not flicker or drift.
TruthfulQAai: A benchmark measuring whether a model gives truthful answers to questions that tempt it into repeating common misconceptions and popular falsehoods.
t-distributed stochastic neighbour embedding (t-SNE)ai: A method that maps high-dimensional data into two or three dimensions for visualisation, preserving local groupings of similar points.
Train/test splitai: Dividing data into a set for training the model and a separate set for measuring how well it generalises.
Term frequency-inverse document frequency (TF-IDF)ai: A weighting that scores a word by its frequency in a document, offset by how rare it is across all documents.
Topic modellingai: An unsupervised technique that discovers recurring themes across a collection of documents by grouping words that tend to occur together.
Tool useai: An agent's ability to call external functions, APIs or services to fetch data or perform actions beyond generating text on its own.
Tool schemaai: A machine readable description of a tool's name, purpose and accepted arguments that tells a model how and when to call it.
Tool registryai: A catalogue of the tools available to an agent, holding each one's schema so the model can discover and select them at runtime.
Task decompositionai: Breaking a large goal into smaller ordered subtasks that an agent can tackle one at a time before combining the results.
Tree of thoughts (ToT)ai: A reasoning method where a model explores several branching lines of thought, evaluating and pruning them to find a stronger solution.
Top-k retrievalai: Returning the k most similar items to a query, where k is a chosen number such as the five closest passages.
Tensor processing unit (TPU)ai: Google's custom chip designed specifically for machine learning, optimised to speed up the tensor operations that neural networks depend on.
Tensor coresai: Specialised units inside Nvidia GPUs that accelerate the matrix multiplications at the heart of neural networks, greatly speeding up training and inference.
Tensor parallelismai: Splitting individual layers of a model across devices so one large operation runs in parallel on several chips at once.
Training runai: A single, often lengthy session in which a model learns from data, consuming large amounts of compute before producing a finished model.
Text Generation Inference (TGI)ai: Hugging Face's open-source toolkit for deploying and serving large language models efficiently in production environments.
Thermal design power (TDP)ai: The maximum heat a chip is expected to produce under load, guiding how much cooling and power a system must supply.
Trace distancequantum: A measure of how distinguishable two quantum states are, ranging from zero for identical to one for perfectly distinguishable.
T gate (T)quantum: A single-qubit gate that adds a forty-five degree phase to the one state; it is non-Clifford and needed for universal quantum computation.
T-dagger gate (T dagger)quantum: The inverse of the T gate, subtracting a forty-five degree phase from the one state to reverse the T gate's rotation.
Toffoli gate (CCNOT)quantum: A three-qubit gate that flips the target only when both control qubits are one, enabling reversible classical logic on a quantum computer.
Two-qubit gatequantum: A gate acting on a pair of qubits together, typically the source of entanglement and usually noisier than single-qubit gates.
Transpilationquantum: The process of rewriting a quantum circuit into a device's native gates and connectivity while optimising it, so it can run on real hardware.
Trotterisationquantum: Approximating the evolution under a complicated interaction by a sequence of small, simple steps, widely used in quantum simulation circuits.
Transmonquantum: A widely used superconducting qubit design that reduces sensitivity to electrical charge noise, giving longer coherence at the cost of weaker anharmonicity.
Tunable couplerquantum: An adjustable circuit element placed between two qubits that switches their interaction on and off, reducing unwanted crosstalk during idle periods.
Travelling-wave parametric amplifier (TWPA)quantum: A near-quantum-limited microwave amplifier with wide bandwidth, able to read out many superconducting qubits at once with low added noise.
Two-level system defect (TLS)quantum: A microscopic flaw in materials or interfaces that absorbs energy from a qubit, a leading source of decoherence in superconducting devices.
Tweezer arrayquantum: A grid of optical tweezers each holding one neutral atom, forming a reconfigurable register of qubits that can be rearranged between shots.
Topological superconductorquantum: A material predicted to host Majorana zero modes at its boundaries, sought as a platform for hardware-protected topological qubits.
Threshold theoremquantum: The result showing that if physical error rates fall below a fixed threshold, arbitrarily long quantum computations become possible with modest overhead.
Transversal gatequantum: A logical gate applied by acting on each physical qubit independently, so errors cannot spread within a code block, aiding fault tolerance.
Toric codequantum: A topological stabiliser code defined on a torus, the original surface code, protecting logical qubits within its global loop structure.
T-countquantum: The number of costly T gates in a fault-tolerant quantum circuit. Because these gates dominate the resource cost, algorithms are often optimised to reduce it.
Twin-field QKD (TF-QKD)quantum: A quantum key distribution protocol that interferes weak optical fields at a central node, roughly doubling the secure distance achievable over standard fibre.
Trusted nodequantum: An intermediate relay in a quantum key distribution network that decrypts and re-encrypts keys, and so must be assumed physically secure.
Time-bin encodingquantum: A method of encoding qubits in the arrival time of photons, favoured for fibre quantum communication because it resists distortion during transmission.
Trotter errorquantum: The approximation error introduced by Trotterisation because the individual terms of a Hamiltonian do not commute; smaller time steps reduce it.
Time evolution operatorquantum: The mathematical object, derived from the Hamiltonian, that transforms a quantum state from one moment to a later one.
Two-qubit gate fidelityquantum: The accuracy of a gate acting on two qubits, usually the hardest operation to perform well and a key limit on device performance.
T1 (relaxation time)quantum: The average time a qubit takes to lose its energy and decay from the excited state to the ground state, a core measure of quality.
T2 (dephasing time)quantum: The average time over which a qubit keeps its phase information before quantum interference is lost, limiting how long computations stay reliable.
Two-level system (TLS)quantum: A microscopic defect in device materials that couples to a qubit and causes fluctuating energy loss and dephasing.

U

Umeecrypto: Umee is a Cosmos-based lending and borrowing platform enabling cross-chain credit markets through IBC interoperability.
Unchained Capitalcrypto: Unchained Capital offers collaborative multi-sig custody and Bitcoin-backed lending services for individuals and institutions.
Under-Collateralised Lendingcrypto: Under-collateralised lending issues loans based on credit reputation or off-chain guarantees rather than full collateral coverage.
Unibotcrypto: Unibot is a Telegram trading bot for DeFi users enabling automated sniping, copy-trading, and wallet management.
Uniswapcrypto: Uniswap is a leading decentralised exchange using automated market-maker pools, pioneering permissionless token trading and liquidity provision.
Unstoppable Domainscrypto: Unstoppable Domains issues blockchain-based website and wallet addresses, replacing long hexadecimal strings with human-readable names.
Utility Tokencrypto: A utility token grants holders access to specific products, services, or governance functions within its issuing project’s ecosystem.
UTXOcrypto: Unspent Transaction Output is Bitcoin’s accounting model, tracking spendable fragments of coins instead of maintaining account balances.
Underfittingai: When a model is too simple or undertrained to capture the underlying pattern, performing poorly on both training data and new data.
U-Netai: A neural network shape with matching downscaling and upscaling paths joined by skip connections, widely used as the denoiser inside diffusion image models.
Upscalingai: Enlarging an image or video, often using an AI model to add detail and sharpness while raising the pixel count.
Unsupervised learningai: Finding structure in unlabelled data, such as groupings or hidden patterns, without being told the correct answers in advance.
Uniform manifold approximation and projection (UMAP)ai: A dimensionality reduction technique for visualisation that preserves both local and broader structure, and is typically faster than t-SNE.
Unitary operatorquantum: A reversible, length-preserving operator describing how an isolated quantum system evolves while keeping total probability equal to one.
Universal gate setquantum: A small collection of gates that, combined in sequence, can approximate any quantum operation to arbitrary accuracy, such as Clifford gates plus the T gate.
Ultra-high vacuum (UHV)quantum: An extremely low-pressure environment used in trapped-ion and neutral-atom machines to keep stray gas particles from disturbing the qubits.
Union-Find decoderquantum: A fast surface-code decoding algorithm that grows and merges clusters of syndrome events, trading a little accuracy for near-linear speed.
Uncomputationquantum: Reversing intermediate quantum operations to reset helper qubits without disturbing the result, freeing them for reuse and preventing unwanted entanglement in a computation.
Unstructured searchquantum: Looking for a marked item in an unsorted collection with no useful order. Grover's algorithm gives a quadratic speedup over checking items one by one.

V

Validatorcrypto: A validator verifies and proposes blocks in proof-of-stake systems, ensuring network consensus and earning rewards for honest participation.
Validity Proofcrypto: A validity proof (e.g., SNARK, STARK) mathematically demonstrates that a state transition occurred correctly without revealing private data.
Validiumcrypto: Validium scales transactions using off-chain data availability but on-chain validity proofs, balancing performance with weaker censorship resistance.
VanEck Cryptocrypto: VanEck Crypto issues exchange-traded products providing exposure to Bitcoin, Ethereum, and thematic blockchain indexes.
Velodromecrypto: Velodrome on Optimism is a ve(3,3) DEX aligning liquidity incentives via vote-escrowed tokenomics and protocol-owned liquidity mechanics.
Venus Protocolcrypto: Venus Protocol is a BNB Chain lending platform offering collateralised borrowing, synthetic stablecoins, and liquidation automation.
Vergecrypto: Verge integrates Tor and I2P routing into its blockchain, anonymising IP data for private cryptocurrency transactions.
Vesta Financecrypto: Vesta Finance issues stablecoins collateralised by crypto assets, enabling low-fee borrowing and on-chain leverage through liquidations.
Vestingcrypto: Vesting gradually unlocks tokens for team members or investors, aligning incentives with long-term project growth and preventing early dumping.
Volatilitycrypto: Volatility measures how sharply an asset’s price fluctuates, influencing trading risk, liquidity provisioning, and derivatives pricing models.
Voltage Cloudcrypto: Voltage Cloud provides Lightning-node hosting and liquidity management tools for developers building Bitcoin payment applications.
VRF (Verifiable Random Function)crypto: A VRF generates provably random outputs on-chain, ensuring fairness in validator selection, lotteries, and gaming applications.
Vypercrypto: Vyper is an Ethereum smart-contract language prioritising simplicity and auditability, reducing risk compared with Solidity’s complex syntax.
Vector databaseai: A database that stores embeddings and retrieves items by similarity, letting applications find content related in meaning rather than by exact keywords.
Vision transformer (ViT)ai: A model that splits an image into patches, treats them as tokens, and applies transformer attention for image understanding.
Variational autoencoder (VAE)ai: A generative autoencoder that learns a probability distribution over its latent code, letting it sample new data resembling the training set.
Vanishing gradient problemai: A difficulty in deep networks where gradients shrink towards zero as they propagate backwards, leaving early layers barely able to learn.
Validation setai: A portion of data held aside during training to tune settings and monitor generalisation, without being used to update the weights.
Vocabularyai: The fixed set of tokens a model recognises; every input and output is expressed as a sequence drawn from this vocabulary.
Vocabulary sizeai: The number of distinct tokens in a model's vocabulary, commonly tens of thousands; larger vocabularies shorten sequences but enlarge the embedding and output layers.
Voice cloningai: Recreating a specific person's voice from sample recordings so that new speech can be synthesised in that same voice.
Vocoderai: A component that turns an intermediate audio representation such as a spectrogram into an actual audible waveform.
Vector-quantised variational autoencoder (VQ-VAE)ai: A VAE variant that maps data to a fixed set of discrete codes, useful for compressing images and audio into tokens.
Vector indexai: A data structure that organises embeddings so similar vectors can be found quickly without comparing against every stored entry.
Video RAM (VRAM)ai: The dedicated memory on a graphics card that holds a model's weights and data; its size limits how large a model can run.
vLLMai: An open-source library for serving large language models quickly, best known for memory-efficient attention that raises throughput and cuts cost.
Von Neumann entropyquantum: The quantum counterpart of Shannon entropy, quantifying the uncertainty or mixedness contained within a density matrix.
Virtual Z gatequantum: A z-axis rotation carried out by shifting the phase of later control pulses rather than a physical pulse, making it effectively instantaneous and error-free.
Variational quantum circuit (VQC)quantum: A parametrised circuit whose gate angles are optimised by a classical computer to minimise a cost function, central to near-term algorithms.
Variational quantum eigensolver (VQE)quantum: A hybrid algorithm that uses a quantum computer to prepare trial states and a classical optimiser to find the lowest energy of a Hamiltonian.
Vapour cellquantum: A sealed glass cell containing atomic gas used in compact magnetometers and clocks, enabling quantum sensing without bulky cooling apparatus.

W

Walletcrypto: A wallet stores cryptographic keys enabling blockchain interaction, from cold hardware devices to software apps managing multiple token assets.
WalletConnect v2crypto: WalletConnect v2 is a cross-chain communication standard enabling secure session management between wallets and decentralised apps.
Wanchaincrypto: Wanchain provides cross-chain interoperability using secure multiparty computation bridges and decentralised node networks.
Warm Walletcrypto: A warm wallet balances convenience and safety, connecting intermittently to networks while storing funds more securely than hot wallets.
Wasabi Coordinatorcrypto: Wasabi Coordinator organises CoinJoin rounds anonymously, ensuring decentralised privacy coordination among Bitcoin users.
Wasabi Walletcrypto: Wasabi Wallet is a Bitcoin privacy wallet implementing CoinJoin mixing to obfuscate transaction history and preserve financial anonymity.
Watchlist Addresscrypto: A watchlist address is monitored by compliance or analytics systems for suspicious activity, sanctions breaches, or large transfers.
Wavescrypto: Waves is a layer-one blockchain supporting token creation and on-chain trading via its native DEX and scripting capabilities.
Waves Exchangecrypto: Waves Exchange combines centralised and decentralised liquidity, offering fiat gateways, staking, and synthetic-asset trading.
Webacycrypto: Webacy protects users from wallet-drain attacks by offering transaction-simulation alerts, emergency asset transfer, and recovery automation.
Whale Alertcrypto: Whale Alert monitors large blockchain transfers in real time, identifying exchange inflows and potential market-moving transactions.
Whitepapercrypto: A whitepaper presents a project’s technical vision, problem statement, and economic model, establishing credibility and investor understanding.
Wintermutecrypto: Wintermute is a crypto-market maker providing liquidity for exchanges and protocols through algorithmic trading strategies.
WOO Networkcrypto: WOO Network connects traders, exchanges, and liquidity providers, offering low-spread execution and yield products via the WOO token.
WooFicrypto: WooFi is a decentralised exchange integrating centralised-liquidity sources to provide deep liquidity and zero-fee trading opportunities.
Worldcoincrypto: Worldcoin combines biometric identity verification with universal-income distribution using the WLD token and Orb-hardware devices.
Wormholecrypto: Wormhole is a generic cross-chain messaging protocol enabling data, token, and NFT transfers across major blockchains.
Wrapped Bitcoin (WBTC)crypto: Wrapped Bitcoin tokenises BTC on Ethereum as an ERC-20 asset, enabling participation in DeFi protocols while maintaining Bitcoin backing.
Wrapped Tokencrypto: A wrapped token represents an asset bridged to another chain, maintaining underlying value while expanding cross-chain liquidity and usage.
Word embeddingai: A dense vector representing a word so that similar meanings sit near each other in the learned vector space.
Weight decayai: A regularisation method that shrinks weights slightly on every update, keeping them small and improving the model's ability to generalise.
WordPieceai: A subword tokenisation algorithm, used by BERT, that chooses merges to maximise the likelihood of the training text rather than raw frequency.
Weight-only quantisationai: Compressing just the model's stored weights to low precision while keeping activations at higher precision, saving memory with minimal effect on quality.
Watermarkingai: Embedding a hidden or visible marker in generated media so it can later be identified as AI-produced or traced to its source.
Word2vecai: A technique that learns numerical word representations from text, placing words with similar meanings close together in vector space.
Working memoryai: The information actively held and manipulated by an agent while it carries out the current step of a task.
Web agentai: An agent that carries out tasks on websites, such as searching, booking or gathering information, by acting through a browser.
Wavefunctionquantum: A mathematical function describing a quantum system's state, whose squared magnitude gives the probability of finding each possible measurement result.
Wavefunction collapsequantum: The sudden change of a quantum state into a single definite outcome at the instant a measurement is performed on it.
W statequantum: A symmetric entangled state of several qubits that retains some entanglement even after one qubit is lost.
Wiring fan-outquantum: The challenge and hardware of routing many control and readout lines from room-temperature electronics down to a growing number of qubits.

X

X-Chain (Cross-Chain)crypto: X-Chain interactions move assets or data between blockchains, enhancing liquidity, interoperability, and decentralised network composability.
X-Margincrypto: Cross-exchange margin platforms like X-Margin provide unified collateral pools for derivatives, enabling privacy-preserving risk-adjusted leverage.
XDC Networkcrypto: XDC Network is an enterprise-focused hybrid blockchain optimised for tokenised assets and ISO 20022-compatible financial messaging.
XDefi Walletcrypto: XDefi Wallet is a multi-chain browser extension supporting DeFi and NFTs across EVM, Cosmos, and Bitcoin networks.
XRPcrypto: XRP is the native token of the XRP Ledger, used for rapid settlement and bridging diverse fiat or crypto currencies.
Xverse Walletcrypto: Xverse Wallet is a Bitcoin Ordinal-compatible wallet supporting inscriptions, NFTs, and BRC-20 transactions.
XGBoostai: A popular, fast implementation of gradient boosting, known for strong performance on structured, tabular data in competitions and industry.
XMSSquantum: The eXtended Merkle Signature Scheme, a stateful hash-based signature standard that must carefully track which one-time keys have already been used.

Y

Yearn Financecrypto: Yearn Finance automates DeFi yield strategies, aggregating deposits into vaults optimised for maximum returns across multiple protocols.
Yield Aggregatorcrypto: A yield aggregator automatically reallocates funds among DeFi opportunities, optimising returns through smart-contract automation and performance analytics.
Yield Curve (On-chain)crypto: On-chain yield curves display relationship between lock duration and return, guiding liquidity providers’ time-based yield expectations.
Yield Farmingcrypto: Yield farming allocates capital across DeFi protocols to maximise rewards, combining staking, liquidity, and lending strategies dynamically.
Yield Guild Games (YGG)crypto: Yield Guild Games invests in blockchain gaming assets and NFTs, allowing community members to share earnings from play-to-earn ecosystems.
Yieldlycrypto: Yieldly is Algorand’s DeFi protocol suite enabling staking, prize games, and cross-chain swaps.
Yuga Labscrypto: Yuga Labs is the creator of Bored Ape Yacht Club, driving NFT culture and metaverse expansion through Otherside.

Z

Zappercrypto: Zapper aggregates DeFi portfolios, tracking wallets, yields, and NFT positions across multiple networks for users and analysts.
Zapritecrypto: Zaprite enables invoicing and payment management in Bitcoin and stablecoins with automatic conversion and accounting features.
Zcash (ZEC)crypto: Zcash is a privacy-focused blockchain using zk-SNARK proofs to enable shielded transactions without exposing sender, receiver, or amount.
Zebeccrypto: Zebec enables real-time payroll streaming and recurring payments in crypto, integrated with DeFi treasury management solutions.
ZebPaycrypto: ZebPay is one of India’s oldest cryptocurrency exchanges, offering mobile trading, secure custody, and fiat on-ramps for digital assets.
Zelcorecrypto: Zelcore is a multi-asset wallet and interface for the Flux ecosystem, supporting DeFi trading and staking directly from custody.
Zelliccrypto: Zellic performs deep cryptographic-protocol audits, specialising in zero-knowledge proofs, consensus algorithms, and DeFi smart-contract implementations.
ZenGocrypto: Zengo is a non-custodial wallet using multiparty computation to remove seed phrases while maintaining cryptographic self-custody.
Zengo Guardcrypto: Zengo Guard combines MPC wallet infrastructure with on-chain-risk monitoring, preventing phishing and malicious-approval exploits.
Zephyr Protocolcrypto: Zephyr Protocol issues privacy-preserving algorithmic stablecoins secured by over-collateralised vaults and zero-knowledge proofs.
Zerioncrypto: Zerion is a DeFi dashboard and wallet enabling portfolio tracking, swaps, and NFT management within one unified interface.
Zerion Analyticscrypto: Zerion Analytics aggregates portfolio and transaction data, visualising user behaviour trends across decentralised wallets and NFT activity.
Zero-knowledge proofcrypto: A way to prove a statement is true without revealing why — proving you are over 18 without showing your birthday. Used in crypto for privacy and for compressing huge amounts of computation.
ZeroLendcrypto: ZeroLend is a decentralised borrowing protocol using isolated markets and risk-tranching mechanisms for safer collateral exposure.
Zeta Marketscrypto: Zeta Markets provides Solana-based derivatives trading with unified cross-margin and institutional execution tools.
Zetachaincrypto: Zetachain is an omnichain interoperability network enabling native cross-chain message passing and Bitcoin-compatible smart-contract functionality.
ZetaChain Bridgecrypto: ZetaChain Bridge connects multiple layer-ones and layer-twos, supporting unified cross-chain liquidity and message execution.
Zignalycrypto: Zignaly is a social-trading platform connecting retail investors with professional crypto traders through profit-sharing investment pools.
Zilliqacrypto: Zilliqa is a sharded layer-one blockchain achieving parallel transaction processing and smart-contract execution efficiency.
Zircuitcrypto: Zircuit is a modular rollup combining zero-knowledge proofs with optimistic fraud prevention to improve safety and settlement efficiency.
ZK Aggregatorcrypto: ZK Aggregator merges proofs from multiple rollups, enhancing throughput and reducing gas costs through recursive zero-knowledge composition.
ZK Rollupcrypto: A ZK Rollup bundles transactions off-chain and posts succinct validity proofs on-chain, enabling high throughput and strong security guarantees.
ZK-IDcrypto: ZK-ID uses zero-knowledge proofs to verify identity attributes privately, supporting compliant yet privacy-preserving digital authentication systems.
Zk-SNARKcrypto: Zero-Knowledge Succinct Non-Interactive Argument of Knowledge proofs verify computations privately, essential for privacy-preserving and scaling applications.
Zk-STARKcrypto: Zero-Knowledge Scalable Transparent Argument of Knowledge proofs provide similar privacy guarantees with greater transparency and post-quantum security properties.
ZkEVMcrypto: A zkEVM reproduces Ethereum’s virtual machine within a zero-knowledge environment, providing scalability while maintaining full Solidity compatibility.
zkKYCcrypto: zkKYC applies zero-knowledge proofs to verify identity compliance without revealing personal information on-chain.
zkSynccrypto: zkSync is a ZK-rollup scaling Ethereum with low fees and instant withdrawals, supporting EVM-compatible smart-contract deployments.
ZKSync Eracrypto: ZKSync Era is the mainnet implementation of zkSync’s rollup, offering fast, secure, and low-fee transactions for Ethereum users.
ZKValidatorcrypto: ZKValidator operates staking infrastructure and promotes zero-knowledge technology adoption across privacy-enhancing blockchains.
Zodia Custodycrypto: Zodia Custody is a regulated institutional crypto custodian operated by Standard Chartered Bank ensuring compliance and asset protection.
Zonecrypto: Zone is a gamified DeFi ecosystem offering play-to-earn mechanics, prediction markets, and social staking features.
Zone Protocolcrypto: Zone Protocol gamifies on-chain engagement using rewards and achievements, fostering community interaction across DeFi ecosystems.
Zoracrypto: Zora is an NFT protocol providing on-chain marketplaces, creator royalties, and minting tools for digital media distribution.
Zora DAOcrypto: Zora DAO governs NFT protocol upgrades and community funding initiatives using on-chain governance mechanisms.
Zora Marketplacecrypto: Zora Marketplace enables creators to mint and trade NFTs directly on-chain with programmable royalties and transparent fee structures.
Zora Networkcrypto: Zora Network is an Ethereum-compatible layer-two focused on NFT minting and media distribution using Optimism’s OP Stack technology.
Zunami Aggregatorcrypto: Zunami Aggregator routes stablecoin liquidity across DeFi pools, maximising yield through algorithmic allocation and risk diversification.
Zunami DAOcrypto: Zunami DAO governs stablecoin aggregation strategies, adjusting allocations algorithmically across multiple liquidity pools for optimal yield.
Zunami Protocolcrypto: Zunami aggregates stablecoin yields across DeFi protocols, optimising risk-adjusted returns through automated liquidity rebalancing.
Zyficrypto: Zyfi allows gasless transactions on zkSync networks, enabling users to pay fees with stablecoins through sponsored transaction relayers.
Zero-shotai: When a model performs a task from instructions alone, without being shown any worked examples in the prompt beforehand.
Zero Redundancy Optimizer (ZeRO)ai: A method that removes duplicated data across devices by sharding model states, greatly reducing the memory needed to train large models.
Zero-noise extrapolation (ZNE)quantum: An error mitigation method that deliberately increases noise, measures the effect, then extrapolates back to estimate the noise-free result.