The next web, explained in plain English
Pretraining objective where tokens are masked and predicted; used in BERT‑style models.
Core linear algebra operation dominating transformer compute; optimized with tiling and tensor cores.
Planning algorithm that balances exploration/exploitation by sampling trajectories; used in agents.
Attention kernels that reduce memory/time complexity using tiling, FlashAttention, or linearized variants.
Set of pending transactions waiting to be included in a block; target of MEV strategies.
Set of sibling hashes proving an element belongs to a Merkle tree with a known root.
Root hash of a Merkle tree; commits to all underlying data in a block or distribution list.
A hash-based data structure that enables efficient verification of large data sets with compact proofs.
Pattern where a relayer submits a user‑signed message, allowing gasless UX.
JSON describing an NFT’s name, image, and attributes; stored on IPFS or on‑chain.
Value extracted by reordering, including, or censoring transactions during block production.
Techniques like private mempools, transaction bundles, and orderflow auctions to reduce harmful MEV.
Service that brokers blocks between builders and proposers in PBS architectures.
Authentication using more than one factor, e.g., device + passkey; common in custodial flows.
Software that sits between on‑chain data and applications, e.g., oracles, indexers, relayers.
Original term for MEV in PoW systems; replaced by Maximal Extractable Value after PoS.
To create new tokens or NFTs according to a contract’s rules.
Training with lower precision formats to save memory and increase throughput, with loss‑scaling.
Architecture where a router activates subsets of expert MLPs per token, increasing capacity efficiently.
Multi‑layer perceptron used inside transformer blocks for token‑wise transformations.
Append‑only variant of Merkle trees optimized for proofs over growing logs.
List of words (BIP39) encoding a wallet seed; must be backed up securely.
Light client that validates blocks on consumer devices and earns a share of network rewards for verification.
Reward portion distributed to Mobile Verifiers for performing verification work.
Training and controls to ensure models follow human intent and avoid harmful behavior.
Document describing model intended use, data, metrics, risks, and limitations.
Reducing numerical precision (e.g., FP16→INT8/4) to shrink models and speed inference.
Architecture that splits execution, consensus, and data availability across specialized layers.
Technique that accumulates velocity of gradients to smooth updates; basis for SGD with momentum, Adam.
Design where execution, consensus, and data availability live on one chain.
Cryptographic technique enabling parties to jointly compute a function without revealing inputs.
Trie structure used by Ethereum to store accounts and contract state with Merkle proofs.
Operating across multiple blockchains, often with bridges and messaging protocols.
Contract utility that aggregates multiple read calls into one to save RPC round trips.
Account or contract requiring multiple signatures to authorize transactions.