self-critique-without-grounding-unreliable CLM-0145mechanismsingle paperReviewed by AI and Russ Miller
Without an external, ground-truth signal — a failing test, a compiler error, a verifier's output — a model's own critique of its reasoning is not a reliable improvement signal, and asking it to review and revise a correct answer often turns it into a wrong one.
Evidence for: Boyd's law of iteration (narrows), Linus's law (narrows), No one should be judge in their own cause (holds)
Capability: Fixing its own mistakes · Reasoning, Math
Observed on
2023, GPT-3.5/GPT-4 class reasoning benchmarks.
Model cards: GPT-4
Sources
- Intrinsic self-correction (review-and-revise with no external feedback) did not improve accuracy on GSM8K or CommonSenseQA and often degraded it.
Either way, bring evidence: a link, or what you saw.
Disagreeing is the most useful thing you can do here. Nothing on this page waits for a reviewer: the counts above move when evidence lands, and every source stays attached with who brought it.
Related claims
- Prompting a model to generate its own feedback on a draft output and revise accordingly improves quality on open-ended generation tasks (dialogue, code, review-writing) over a single-shot attempt.Fixing its own mistakes · contested
- For 8B instruction-tuned backbones, training a LoRA corrector to diagnose a typed, executable error operator (error type, step location, corrupted and repaired proposition) and apply its inverse as a localized edit raises accuracy over free-form self-revision baselines (Self-Refine, RISE, SCoRe) on math, code and commonsense tasks, and dropping either the operator-prediction loss or the adjacent-state reconstruction loss loses most of the gain.Fixing its own mistakes · reviewed by AI
- Reflect-and-retry raises task success when the feedback in the loop is a genuine external signal — a failing test, a compiler error, an environment outcome — but not when the "feedback" is the model's own unaided critique. The grounding, not the reflection step, is what does the work.Fixing its own mistakes · contested
- On simple HLS kernel generation tasks (85 designs from CHStone, MachSuite, Polybench, Rosetta, C2HLSC), giving a modest open-source model (gpt-oss-120b) a shell with a C++ compiler for self-verification and iterative refinement saturates the benchmark at pass@10 (100% through synthesis) and exceeds 90% at pass@1, so this class of task no longer separates models; repeated sampling helps the weaker model far more (+60 points synthesis pass rate at k=10 for gpt-oss-20b vs +11 for the 120b).Generating and editing working code · reviewed by AI
- On verifiable instructions such as length and format constraints, strong models still fail a meaningful share, and failures grow when several constraints apply at once.Following an unfamiliar procedure
Notes
Single-paper strength is honest here — this is widely cited and discussed, but I don't yet have an independent replication specifically isolating the no-feedback condition the way this paper does.