Terminology· 4 min
AI terminology cheatsheet
Fast definitions for LLM, token, context window, embedding, fine-tune, agent, MoE, quantization and more.
Key idea
Speak the language: knowing the vocabulary makes every paper, launch and product easier to follow.
Example
Context window = how many tokens the model can 'see' at once. GPT-4o Turbo: 128k. Gemini 1.5 Pro: up to 2M.
What to remember
- LLM: Large Language Model — trained to predict the next token
- Token: chunk of text (~4 chars in English)
- Embedding: numeric vector representing meaning
- MoE: Mixture of Experts — only some sub-networks fire per input
- Quantization: shrinking weights (fp16 → int8/int4) for speed