# Chain of thought

> Prompting a model to reason step by step before answering — accuracy on multi-step problems improves when the working is generated, not skipped.

Asking for an answer directly makes the model guess in one jump; asking it to work through the problem makes each step inform the next. 'Let's think step by step' became famous, but explicit structure — 'first extract the quantities, then compute, then state the result' — works even better.

The caveat is that the written reasoning is a generated artefact, not a window into the model: it can be fluent and wrong. For production use the reasoning often runs hidden, with only the checked result shown to the user.

## Related terms

- https://dfieldsolutions.com/en/glossary/prompt-engineering.md
- https://dfieldsolutions.com/en/glossary/hallucination.md
- https://dfieldsolutions.com/en/glossary/llm-evaluation.md

---

Source: https://dfieldsolutions.com/en/glossary/chain-of-thought
DField Solutions — Dunakeszi, Hungary — dezso@dfieldsolutions.com
Booking: see https://dfieldsolutions.com/en/contact