A model solves multi-step math questions poorly, although it knows the formulas. What do you try first?
Why B is the answer
Chain of thought
Stepwise reasoning helps with multi-step logic.
In detail
- Stepwise reasoning helps with multi-step logic.
- Try this before tuning.
Use case
An illustrative scenario, not a real incident.
A model knows the formulas but rushes the steps. Asking for stepwise reasoning improves it.
See it in different ways
flowchart TD
A{"What goes wrong?"} -->|"Wrong format"| B["Few-shot and structured output"]
A -->|"Logic errors"| C["Chain of thought"]
A -->|"Made-up facts"| D["Retrieval and grounding"]mindmap
root((Prompting))
Format drift
Few-shot examples
Structured output
Reasoning errors
Chain of thought
Simple tasks
Zero-shot
Domain words
Context in the promptflowchart LR S0["Add stepwise reasoning"] --> S1["Test"] --> S2["Compare"]
Key point: Chain of thought helps multi-step logic.
Why the other options are wrong
- A. Removing all of the examples from the prompt
- Examples tend to help.
- C. A lower maximum output length, to keep answers short
- A shorter reply gives less room to reason.
- D. A switch to a smaller embedding model
- Embeddings do not change reasoning.