Reading · 14 min
The vocabulary map
Key takeaway: Ten terms cover 90% of AI conversations: model, prompt, token, context, temperature, embedding, RAG, agent, fine-tune, evaluation.
The ten terms
- Model — the trained network you send requests to (GPT-5.1, Claude Opus 4.5, Gemini 3).
- Prompt — everything you send: instruction, context, examples, format.
- Token — the billing and length unit.
- Context window — how much the model can hold at once.
- Temperature — randomness. Low for extraction, higher for ideation.
- Embedding — text turned into a list of numbers so similarity can be measured.
- RAG — retrieval-augmented generation: find relevant text first, then answer from it.
- Agent — a model given tools and allowed to loop until a goal is met.
- Fine-tuning — extra training on your examples to shape style or format.
- Evaluation — a repeatable test set that tells you whether a change helped.
Words people confuse
- Fine-tuning teaches behaviour, not knowledge. To add knowledge, use RAG.
- An assistant answers; an agent acts. Agents need permissions and limits.
- Parameters are the model's size; tokens are your usage.
Not saved yet — sign in to keep your progress and earn the certificate.