Why Context Matters
An LLM only knows two things: what it saw in training, and what you put in front of it right now. Good output needs good input context. Supplying that context on purpose is a craft — it’s called grounding.
What a Model Can’t Know
Left to itself, a model has three gaps that no amount of clever prompting closes:
- Knowledge cutoff — training froze on a date (Module 5), so it can’t know anything newer.
- No access to your data — your private files, your database, today’s numbers: none of it is in the model.
- It fills the gap anyway — asked something it doesn’t know, a model tends to make a plausible answer up.
Give It the Facts
Grounding flips the default. Instead of hoping the model already knows, you fetch the relevant information and include it in the prompt — so the answer is built on real sources, not on the model’s guess.
The Payoff
Grounding buys you three things a bare model can’t give:
- Fewer hallucinations — the facts are in front of it, so it has less reason to invent.
- Current and private data — today’s numbers and your own material become usable.
- Traceable answers — you can point to the source each claim came from (Module 8’s honesty and verification).
Grounding Is a Menu
There isn’t a single way to ground a model. RAG — retrieval-augmented generation, the classic technique you met in GenAI‑101 — is one option among several, each suited to a different situation. The rest of this module walks the menu.
Build It
How to implement: list the facts your tool’s answers depend on that the model can’t already know — and, next to each, write down where that fact actually lives. That list is your grounding plan.
- Weekly AI Tasks tracker — “what did I do last week?” can only be answered by grounding the model in your task data, never in its training.
- Personal brand site — every claim about the person must be grounded in their real material, or it’s invention.
What you learned
A model knows only its training and what you put in front of it. Grounding supplies the missing facts — cutting hallucinations, unlocking current and private data, and making answers traceable. RAG is one technique on a menu this module now explores.