protologue

Self-Correction

Also called intrinsic self-correction, self-revision.

Self-correction is the general practice of having a language model review and revise its own output, either from its own judgment alone (intrinsic self-correction) or using external feedback such as tool results, tests, or retrieved evidence.

Description

Research has found that intrinsic self-correction does not reliably improve reasoning accuracy and can make answers worse, because the model often cannot tell which of its answers are wrong. Self-correction works best when the critique is grounded in something the model cannot fake, such as running code, executing a search, or checking against a reference.

Sources

  1. Huang et al. (2023). Large Language Models Cannot Self-Correct Reasoning Yet.

Cite this entry

Protologue. (2026). Self-Correction. In Protologue: A Taxonomy of Prompting and LLM Techniques (v1.4.0, PTL-0117). https://protologue.com/t/self-correction/

BibTeX
@misc{protologue_self_correction,
  title = {Self-Correction},
  author = {{Protologue}},
  year = {2026},
  howpublished = {Protologue: A Taxonomy of Prompting and LLM Techniques, v1.4.0},
  note = {Entry PTL-0117},
  url = {https://protologue.com/t/self-correction/}
}

Markdown JSON