protologue

Self-Verification

Also called backward verification.

Self-verification is a technique in which a model, after producing candidate answers with chain-of-thought, checks each one by working backward, predicting the original conditions of the problem from the answer, and keeps the answer that is most consistent.

Description

It turns verification into a separate prediction task rather than asking the model whether it is right, which reduces the tendency to approve its own mistakes. It applies most naturally to arithmetic and logical problems whose conditions can be masked and recovered.

Sources

  1. Weng et al. (2022). Large Language Models are Better Reasoners with Self-Verification.

Cite this entry

Protologue. (2026). Self-Verification. In Protologue: A Taxonomy of Prompting and LLM Techniques (v1.4.0, PTL-0118). https://protologue.com/t/self-verification/

BibTeX
@misc{protologue_self_verification,
  title = {Self-Verification},
  author = {{Protologue}},
  year = {2026},
  howpublished = {Protologue: A Taxonomy of Prompting and LLM Techniques, v1.4.0},
  note = {Entry PTL-0118},
  url = {https://protologue.com/t/self-verification/}
}

Markdown JSON