8b-instruction-tuned-backbones-training-lora-corrector-diagnose-typed-executable CLM-0351For 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.
Capability: Fixing its own mistakes
Observed on
Llama-3-8B-Instruct and Qwen3-8B; six in-domain and seven transfer benchmarks; a few thousand constructed transition pairs; gains smaller on factual multiple-choice (MMLU) where SC.
Sources
- Validated by: In-domain average 67.5 vs 64.3 for the strongest same-backbone baseline (DiffCoT) and 63.7 for SCoRe over GSM8K, MATH, HumanEval+, MBPP, MMLU, PIQA, five seeds, Holm-corrected; transfer averages 76.4 (Llama-3, base 67.7) and 80.2 (Qwen3, base 72.9); ablations drop to 79.1/80.6 on GSM8K for operator-only and adjacent-state-only.. Measured on 13 datasets with two 8B backbones, five seeds, same prompt/extractor per benchmark, and component ablations; baselines are published correction methods. [truncated]
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
- For 8B instruction-tuned backbones run in an iterative typed-error-operator correction loop, accuracy gains concentrate in the first few iterations and then plateau rather than decline as the denoising budget grows, so a small step budget captures nearly all the benefit.Fixing its own mistakes · reviewed by AI
- In UI-to-code generation, where a rendered screenshot gives genuine external feedback, iterative self-critique still drifts below the initial draft over ten rounds, because a code edit changes the rendering non-locally and a fix for one region breaks regions that were already faithful; restricting each round to one typed, scoped repair target and carrying past targets forward as an avoid-list reduces that drift.Generating and editing working code · reviewed by AI
- For a frozen LLM run in a generate-critique-revise loop with no external feedback, the accuracy gain is concentrated where single-shot reasoning is already unreliable: on clean multi-step reasoning (BBH, ~74% single-shot) depth makes no difference beyond the confidence interval, while on harder math (MATH, 26.2% single-shot) the same loop gains +14.2 points, with GSM8K in between (+4.2).Fixing its own mistakes · reviewed by AI
- In open-ended editing tasks with no unique correct output (multi-slide design editing), a separate judge model that scores the structural diff between the original and edited artifact against the instruction, and returns its critique verbatim as the next-turn instruction, raises instruction-following over a single pass, but the gain is selective: most tasks pass on the first attempt, and a minority of the ones that enter the loop get worse unless an earlier iteration is restored when the judge's own score declines.Fixing its own mistakes · reviewed by AI
- Used zero-shot as step-level action critics in long-horizon tool-calling tasks, frontier models are over-pessimistic — flagging a large share of correct actions and pushing the actor into revision loops that lower success below no critic at all — whereas small models fine-tuned on action-level verification rationales flag far fewer valid actions and raise repeated-run reliability.Using the tools it is given · reviewed by AI
Notes
Drafted from the paper by a model and filed unreviewed. Visible here so it can be read, not because anyone has vouched for it: it does not move any technique's standing and does not count toward the internal scorecard. Drafted confidence: low. Falsifier as drafted: A replication where a free-form self-revision baseline matches the operator-based corrector on the same backbones and benchmarks, or where the operator-only objective alone reaches the full method's accuracy. Drafted stance toward self-critique-without-grounding-unreliable: supports -- Free-form self-revision baselines gained little over base CoT while structured, applicability-checked operator edits gained more, consistent with unconstrained self-critique being a weak signal.