Module 01 Summary

What this module established

Fluency is not evidence. A language model produces confident sentences whether or not it has read your documents, and the only mechanical difference between a real answer and a plausible one is a source a reader can open. Retrieval is how you move the answer into a source you control - it is not a technique for making models smarter.

Carry forward

• RAG fits documented facts. It does not compute aggregates, make judgements, or retrieve what was never recorded.
• Retrieval running is not the same as retrieval being used. A model handed irrelevant chunks answers from its weights and looks identical.
• The boundary document - what it answers, refuses and escalates - is the artefact everything else in the course is tested against.

Before moving on

Move on when you can take fifteen real questions, sort them by kind, and say which ones this system should decline before any code exists.

Mark complete