Glossary

AI system that perceives its environment, makes decisions, and takes actions to achieve goals. Autonomous operation. Can use tools, access APIs, execute code. Next evolution beyond chatbots.

Making AI systems pursue goals compatible with human values. The alignment problem: specifying what we want when our values conflict and change with context.

Interface for software to communicate with other software. REST APIs use HTTP requests. GraphQL queries specific data. gRPC uses protocol buffers. Abstracts complexity for integration.

AI that matches or exceeds human-level performance across any intellectual task. Unlike narrow AI (which excels at specific tasks), AGI can learn, reason, and apply knowledge across domains without retraining.

Distributed ledger maintained by a network of nodes. Each block contains transactions and cryptographic hash of the previous block. Immutable, transparent, decentralized. Bitcoin and Ethereum are blockchains.

Protocol for transferring assets between different blockchains. Lock tokens on Chain A, mint equivalent on Chain B. Security risks: bridges are frequent attack vectors. Supports cross-chain DeFi.

Continuous Integration and Continuous Deployment. Automated testing and deployment pipeline. Code commits trigger builds, tests run automatically, passing builds deploy to production. Accelerates iteration.

Protocol for nodes to agree on the state of a blockchain. Proof of Work (Bitcoin) uses computational power. Proof of Stake (Ethereum) uses token ownership. Determines security and energy efficiency.

Training approach where AI evaluates its own outputs against a set of principles. Self-critique and revision loop. Developed by Anthropic to improve harmlessness without human feedback loops.

Organization governed by smart contracts and token holders rather than traditional hierarchy. Proposals, voting, and execution happen on-chain. Coordinates resources without central authority.

Financial services (lending, trading, insurance) built on blockchain without traditional intermediaries. Smart contracts replace banks. Permissionless, composable, transparent. Total value locked exceeds $50B.

Platform for packaging applications and dependencies into containers. Containers are lightweight, portable, consistent across environments. Foundation of modern DevOps. Works with Kubernetes for orchestration.

Processing data near the source (edge) rather than centralized cloud. Reduces latency, bandwidth costs. CDNs, IoT devices, 5G networks support edge deployment. Required for real-time applications.

Vector representations of text, images, or other data in high-dimensional space. Semantically similar items cluster together. Enable similarity search, classification, and retrieval systems.

Training a pre-trained model on a specific dataset to specialize its behavior. Cheaper than training from scratch. Used to adapt general models for domain-specific tasks.

Transaction fees paid to validators for executing operations on a blockchain. Measured in gas units. Price fluctuates with network demand. High fees during congestion—major UX challenge.

Digital asset on a blockchain. Fungible tokens (ERC-20) are interchangeable currency. Non-fungible tokens (ERC-721) are unique items. Governance tokens grant voting rights. Utility tokens access services.

Query language for APIs. Clients specify exactly what data they need. Reduces over-fetching and under-fetching. Single endpoint, strongly typed schema. Alternative to REST. Developed by Facebook.

When an AI generates plausible-sounding but false information. Caused by pattern matching without true understanding. Major challenge in deploying LLMs for factual tasks.

Peer-to-peer protocol for storing and sharing files. Content-addressed (hash-based), not location-addressed (URL-based). Decentralized alternative to HTTP. Used for NFT metadata, censorship-resistant storage.

Container orchestration platform for deploying, scaling, and managing containerized applications. Automates deployment, load balancing, self-healing. Industry standard for cloud-native apps.

Neural networks trained on massive text datasets to predict and generate human-like text. Examples: GPT-4, Claude, Gemini. Scale determines capability—billions of parameters enable reasoning, coding, and multi-step tasks.

Secondary blockchain built on top of Layer 1 (e.g., Ethereum) to improve scalability. Processes transactions off-chain, settles periodically on mainnet. Examples: Arbitrum, Optimism, Base.

Staking tokens while maintaining liquidity through derivative tokens. Stake ETH, receive stETH (tradeable). Earn staking yield and use stETH in DeFi. Lido pioneered this model.

Data structure for efficiently verifying integrity of large datasets. Hash leaf nodes, combine hashes up the tree. Bitcoin and Ethereum use Merkle trees. Lets light clients verify transactions.

Asynchronous communication pattern where services send messages to a queue for later processing. Decouples producers and consumers. RabbitMQ, Kafka, AWS SQS. Supports scalable event-driven architectures.

Architecture pattern where applications are built as independent, loosely coupled services. Each service handles one business function. Scales well, but harder to operate.

Models that process multiple data types—text, images, audio, video. Examples: GPT-4V (vision), Gemini (native multimodal). Supports tasks requiring combined understanding.

Unique cryptographic token representing ownership of a specific digital or physical asset. Unlike cryptocurrencies (fungible), each NFT is distinct. Use cases: art, collectibles, credentials, property rights.

Service providing external data to smart contracts. Blockchains can't access off-chain information directly. Chainlink is the largest oracle network. Required for DeFi pricing, weather data, sports results.

Crafting inputs to AI systems to produce desired outputs. Effective prompts specify context, format, constraints, and examples. Required for working with LLMs.

Controlling the number of requests a client can make to a service in a time period. Prevents abuse, ensures fair resource allocation. Token bucket and sliding window are common algorithms.

Training technique where humans rank model outputs, and the model learns to maximize human preferences. Used to align LLMs with helpfulness and harmlessness goals.

Technique where an LLM queries external knowledge bases before generating responses. Reduces hallucination, adds real-time data access, separates knowledge from model weights.

Cloud execution model where providers manage infrastructure. Write functions, provider handles scaling. Pay per execution, not idle time. AWS Lambda, Vercel Functions, Cloudflare Workers.

Database partitioning technique where data is split across multiple machines. Each shard handles subset of data. Improves scalability. Ethereum's roadmap includes sharding for throughput.

Self-executing code deployed on a blockchain. Automatically enforces agreements when conditions are met. No intermediaries. Foundation of DeFi, NFTs, DAOs. Ethereum made them programmable.

Cryptocurrency pegged to stable assets like USD. USDC and USDT are fiat-backed. DAI is crypto-collateralized. Reduces volatility for on-chain payments and DeFi without price risk.

Smallest unit of text an LLM processes. Roughly 0.75 words in English. Models have context windows measured in tokens (e.g., 128k tokens = ~96k words). Token limits constrain conversation length.

Neural network architecture using attention mechanisms to process sequential data. Replaced RNNs and LSTMs. Foundation of modern LLMs. Introduced in 'Attention Is All You Need' (2017).

Node that proposes and verifies new blocks on a Proof of Stake blockchain. Stakes tokens as collateral. Earns rewards for honest behavior, loses stake for malicious actions. Secures the network.

Software for storing private keys and interacting with blockchains. Custodial (exchange holds keys) vs. non-custodial (you control keys). Examples: MetaMask, Coinbase Wallet, Phantom.

Binary instruction format for stack-based virtual machine. Compile C++, Rust, Go to run in browsers at near-native speed. Supports high-performance web apps. Used in games, video editing, blockchain clients.

Cryptographic method to prove knowledge of information without revealing the information itself. Privacy-preserving verification. Used in zk-rollups, private transactions, identity systems.

AI performing tasks it wasn't explicitly trained for, using only natural language instructions. Emergent capability of large models. Example: GPT-4 translating languages it rarely saw in training.