LLM Learning Hub

workspace/llm-course/home

Verification

Verification is the reasoning-model step of checking candidate answers for correctness — via self-evaluation, unit tests, or a separate verifier model — before returning a final answer.

What is Verification?

Verification is the reasoning-model step of checking candidate answers for correctness — via self-evaluation, unit tests, or a separate verifier model — before returning a final answer.

Verification is the reasoning-model step of checking candidate answers for correctness — via self-evaluation, unit tests, or a separate verifier model — before returning a final answer.

Where is it used?

OpenAI o1's hidden verifier, AlphaCode's test-case filtering, and self-consistency voting all verify outputs; `SelfCheck` and `Self-Refine` papers formalise verification prompts.

How to build it

Generate N solutions, prompt the model with `"Check this solution for errors: {solution}"`, keep solutions marked correct, and return the majority-vote answer among verified candidates.