REVIEW 12 cited by
LLMs cannot find reasoning errors, but can correct them given the error location
Not yet reviewed by Pith; the record is open.
This paper has not been read by Pith yet. Machine review is queued; the pith claim, tier, and objections will appear here once it completes.
SPECIMEN: schema-true, not a live event
T0 review · schema-true
One-sentence machine reading of the paper's core claim.
pith:XXXXXXXX · record.json · timestamp
Signed reviews
read the original abstract
While self-correction has shown promise in improving LLM outputs in terms of style and quality (e.g. Chen et al., 2023b; Madaan et al., 2023), recent attempts to self-correct logical or reasoning errors often cause correct answers to become incorrect, resulting in worse performances overall (Huang et al., 2023). In this paper, we show that poor self-correction performance stems from LLMs' inability to find logical mistakes, rather than their ability to correct a known mistake. Firstly, we benchmark several state-of-the-art LLMs on their mistake-finding ability and demonstrate that they generally struggle with the task, even in highly objective, unambiguous cases. Secondly, we test the correction abilities of LLMs -- separately from mistake finding -- using a backtracking setup that feeds ground truth mistake location information to the model. We show that this boosts downstream task performance across our 5 reasoning tasks, indicating that LLMs' correction abilities are robust. Finally, we show that it is possible to obtain mistake location information without ground truth labels or in-domain training data. We train a small classifier with out-of-domain data, which exhibits stronger mistake-finding performance than prompting a large model. We release our dataset of LLM-generated logical mistakes, BIG-Bench Mistake, to enable further research into locating LLM reasoning mistakes.
Forward citations
Cited by 12 Pith papers
-
The Calibration Floor: Format Repair Can Masquerade as Self-Correction at Small-to-Mid Scale
Across 0.8B-12B models plus a frontier arm, apparent self-correction effects are dominated by format-recovery and format-loss artifacts, with near-zero content-level change at capable scale.
-
CXXCrafter: An LLM-Based Agent for Automated C/C++ Open Source Software Building
An LLM-driven agent, CXXCrafter, automatically builds 587 of 752 C/C++ open-source projects (78%), beating default build commands (39%) and bare LLMs (32 to 38%).
-
Judge Decoding: Faster Speculative Sampling Requires Going Beyond Model Alignment
Judge decoding replaces strict token-alignment verification in speculative decoding with a learned correctness classifier on target embeddings, accepting about three times more tokens and enabling up to 9x speedup on ...
-
Multi-Turn On-Policy Distillation with Prefix Replay
ReOPD offline-distills multi-turn agentic LLMs via teacher-prefix replay plus step-decay sampling, matching online OPD accuracy at ≥4× speed with zero tool calls.
-
The Self-Correction Illusion: Role Relabeling Gates Explicit Error Flagging in Large Language Models
Relabeling an identical erroneous claim from the model's own thought role to an external chat role increases explicit correction rates by 23-93 percentage points across 13 model-domain cells, indicating a chat-templat...
-
MINERVA: Evaluating Complex Video Reasoning
MINERVA provides 1,515 multi-step video QA questions with human reasoning traces; frontier models score far below humans and fail mainly on temporal localization and perception.
-
Self-Correction Makes LLMs Better Parsers
A treebank-rule-guided self-correction method improves LLM constituency parsing F1 by up to 10.12 points on PTB, CTB5, and MCTB across multiple LLMs.
-
Reflection-Window Decoding: Text Generation with Selective Refinement
Selectively refining uncertain windows during decoding improves text quality over greedy and beam search, with a theory formalizing why greedy decoding can miss the joint-probability-optimal response.
-
Refining Answer Distributions for Improved Large Language Model Reasoning
RAD iteratively refines a distribution over answers by marginalizing over refinement samples, improving accuracy on six arithmetic benchmarks over self-consistency and hint-based prompting.
-
Mitigating Think-Answer Mismatch in LLM Reasoning Through Noise-Aware Advantage Reweighting
S-GRPO reweights GRPO advantages by an assumed noise level p to down-weight unbalanced groups, claiming improved and more noise-robust math reasoning training.
-
Not All Errors Are Created Equal: ASCoT Addresses Late-Stage Fragility in Efficient LLM Reasoning
ASCoT claims later reasoning errors are more harmful than early ones and uses a position-weighted verifier to prune and correct CoT steps, but its key evidence is internally inconsistent.
-
Critic-V: VLM Critics Help Catch VLM Errors in Multimodal Reasoning
A DPO-trained VLM critic that critiques and iteratively refines a reasoning VLM improves accuracy on several multimodal benchmarks, with large gains on MathVista and RealWorldQA.
Discussion (0). Continue with ORCID to comment.