Module 10 Summary
What this module established
Write prompts whose answers can be checked mechanically, and give every one an explicit way to say the model cannot tell. Ask only what pixels can answer - for anything else the model produces a confident completion identical in form to an observation.
Carry forward
- Counting, small-text reading and spatial relations are known weak points answered fluently. Measure them on your own images.
- Require a region with every visual claim. It makes the answer checkable in seconds instead of minutes.
- Compute the refusals you can before calling the model, and enforce policy refusals in routing.
Before moving on
Move on when your prompts all have an unsure option, and you have measured counting and small-text accuracy on your own images.
