The complete AI terminology reference
Every AI term you'll encounter — foundational ML, LLMs, transformers, RAG, agents, fine-tuning, safety, and more — explained with real examples and cited sources.
AI Terms Guide is the internet's most comprehensive reference for artificial intelligence terminology. Every term has a dedicated page with a clear definition, real-world examples, cross-references to related concepts, and citations to primary sources. Whether you're a developer integrating an LLM, an executive evaluating an AI strategy, a student learning ML from scratch, or a journalist covering the field — you can find the vocabulary you need here.
The 2,000+ terms are organized into 18 categories below. Each category has its own hub page with the full list of terms in that area, plus editorial context on how the terms relate. Popular starting points: Transformer, RAG, MCP, LoRA, extended thinking, and MoE.
Browse AI terminology by category
Every term is organized by what it does, not just alphabetically — so you can see neighbors, prerequisites, and next steps.
Foundational ML
Neural networks, gradient descent, backpropagation, loss functions.
80 termsLLMs & Language Models
Context windows, tokens, prompting, temperature, streaming.
100 termsDeep Learning Architectures
Transformer, CNN, RNN, Mamba — building blocks behind every model.
60 termsVector Databases & RAG
Embeddings, chunking, semantic search, rerankers.
70 termsFine-tuning Methods
LoRA, QLoRA, DPO, GRPO, PPO, RLHF — how to adapt a model.
50 termsAI Agents
MCP, tool use, ReAct, agentic loops, orchestration.
50 termsMultimodal & Vision
Diffusion models, CLIP, VLMs, image and video generation.
50 termsAI Safety & Alignment
Constitutional AI, red teaming, interpretability, jailbreaks.
40 termsPrompt Engineering
Chain-of-thought, few-shot, ReAct, DSPy, prompt injection.
40 termsEvaluation & Benchmarks
MMLU, HumanEval, GSM8K, MT-Bench — what benchmarks measure.
40 termsTraining & Optimization
ZeRO, FSDP, gradient checkpointing, mixed precision, Adam.
70 termsInference & Serving
vLLM, PagedAttention, KV cache, speculative decoding.
70 termsModern Architectures
Mamba, State Space Models, RWKV, Retentive Networks.
40 termsMixture of Experts
Sparse MoE, Switch Transformer, expert routing.
20 termsFoundation Model Families
GPT, Claude, Gemini, Llama, Mistral families.
60 termsReinforcement Learning
PPO, DQN, MCTS, policy gradients, AlphaZero-style.
30 termsSpeech & Audio
Whisper, TTS, voice cloning, Suno, Udio.
25 termsDatasets, Infra & Companies
The Pile, FineWeb, H100, TPU, HBM.
110 termsThe 10 AI terms every practitioner should know
Deep entries with historical context, technical detail, code examples, and cross-references throughout.
Transformer
The neural network architecture built around attention that processes sequences in parallel. Foundation of GPT, Claude, Gemini, and every modern LLM.
Read the full entry → ArchitectureAttention Mechanism
The core computation that lets a model weigh which parts of the input to focus on. Introduced by "Attention Is All You Need" (2017).
Read the full entry → LLMsLarge Language Model (LLM)
A transformer-based model trained on massive text corpora to predict the next token. Powers ChatGPT, Claude, Gemini, and every conversational AI product.
Read the full entry → RAGRetrieval-Augmented Generation (RAG)
Give an LLM access to external documents at inference time — retrieve, then generate. Reduces hallucinations and lets models answer from current data.
Read the full entry → TrainingFine-tuning
Adapting a pretrained model to a specific task or domain. Compare with LoRA, RLHF, and prompt engineering.
Read the full entry → Fine-tuningLoRA (Low-Rank Adaptation)
Parameter-efficient fine-tuning that freezes base weights and inserts small trainable matrices. Cuts compute cost by orders of magnitude.
Read the full entry → PromptingChain-of-Thought (CoT)
Prompting technique that asks the model to reason step by step before answering. Foundation for modern reasoning models.
Read the full entry → VectorsEmbedding
A dense vector representation of text, images, or other data. Foundation of semantic search, RAG, and clustering.
Read the full entry → FoundationsNeural Network
The core computational structure of modern AI — layers of connected nodes trained via backpropagation. Everything else builds on this.
Read the full entry → LLMsTokenization
How text is broken into pieces the model can process. Determines context window math, pricing, and model behavior on edge cases.
Read the full entry →Terms don't live alone — they connect to models, tools, and concepts
Models
Every AI model with specs, benchmarks, and pricing.
Tools
800+ AI tools reviewed and compared.
Concepts
Deep-dive tutorials — how things actually work.
Compare
Head-to-head comparisons with clear "which to pick".
Providers
The companies building AI — profiles and portfolios.
Use Cases
AI applications organized by profession.
Learn
Structured learning paths from zero to expert.
Questions about Terms
Terms are organized into 18 categories covering foundational ML, LLMs, architectures, RAG, agents, fine-tuning, prompting, evaluation, and more. Every term also has an alphabetical entry so you can browse both ways. Cross-references connect terms to related concepts, models that use them, and tools that implement them.
Existing definitions get quarterly full audits. New terms are added within 48 hours of a major paper or product announcement. See our update cadence for details.
No — each term entry uses progressive depth. The opening TL;DR is written for anyone. Later sections go into technical detail with code examples. Every term also links to prerequisite concepts, so if you get stuck on a word, click through and learn it first.
Yes. Email us via the contact page with the term and one or two authoritative sources. Reader suggestions routinely become new pages.
Every claim links to a primary source — a research paper, official documentation, or trusted benchmark aggregator. See our sourcing philosophy.