06. Module 12 Review — Reasoning Models

Focus: chain-of-thought, reasoning-model behaviour, test-time compute, routing, and evaluation discipline.

Review loop

1. Re-answer the self-check questions in 01_weekly_plan.md from memory.
2. Re-read the toughest sections of 02_explainer.md.
3. Use 04_daily_recall.md without notes.
4. Review 05_hands_on_lab.md and write down what the router taught you.

Conceptual checkpoint

1. Standard LLM vs reasoning model — what is the actual difference?
2. Zero-shot CoT vs few-shot CoT vs native reasoning model — when would you use each?
3. What is test-time compute scaling?
4. Why does search plus verification matter more than mere verbosity?
5. Why can a chain-of-thought be unfaithful?
6. Why is hidden CoT useful for providers but uncomfortable for auditors?
7. Why is routing a product decision as much as a model decision?
8. What does a fair reasoning eval include beyond top-line accuracy?

Applied checkpoint

1. Your SQL assistant is correct on easy tasks but fails messy business questions. How would you route requests?
2. Your reasoning model is 8x more expensive and 6x slower. When is that still a good trade?
3. Your visible CoT looks impressive, but business users still complain. What would you check?
4. A verifier catches most easy errors. Would you still keep a reasoning path? Why or why not?

Foundation checkpoint before Module 13

Before moving on, confirm you can explain these without notes: - Test-time compute — why more thinking budget can help - Cost / quality trade-off — why best quality is not always best product choice - Routing pattern — cheap first, escalate second - Eval methodology — quality, latency, cost, verifier, and human review

Self-evaluation

Section	Score	/
Conceptual	__	8
Applied	__	4
Foundation checkpoint	__	4
Total	__	16

Reflection

• Which task in your hands_on_lab truly needed a reasoning model?
• Where did a cheaper baseline perform surprisingly well?
• Which part still feels fuzzy: CoT, routing, or evaluation?
• What should feel automatic before Module 13?

Bridge

Next module — 01_multimodal_vision_systems — moves from text to vision. The attention and scaling ideas transfer. But images and video add spatial dimensions, temporal coherence, and entirely new failure modes.

Completion gate

• [ ] Weekly plan completed
• [ ] Explainer understood from memory, not only by recognition
• [ ] Assignment shipped
• [ ] Routing and eval logic feel natural
• [ ] Ready to move to Module 13