Related papers: Knowing but Not Correcting: Routine Task Requests Suppress Factual Correction in LLMs

Knowing but Not Correcting: Routine Task Requests Suppress Factual Correction in LLMs

URL: http://arxiv.org/abs/2605.05957v2
Date: Fri, 08 May 2026 12:09:34 GMT
Title: Knowing but Not Correcting: Routine Task Requests Suppress Factual Correction in LLMs
Abstract summary: LLMs reliably correct false claims when presented in isolation, yet when the same claims are embedded in task-oriented requests, they often comply rather than correct.<n>We term this failure mode emphcorrection suppression and construct a benchmark of 300 false premises to systematically evaluate it across eight models.<n> Suppression rates range from 19% to 90%, with four models exceeding 80%, establishing correction suppression as a prevalent and severe phenomenon.
Score: 26.062372963777452
License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/
Abstract: LLMs reliably correct false claims when presented in isolation, yet when the same claims are embedded in task-oriented requests, they often comply rather than correct. We term this failure mode \emph{correction suppression} and construct a benchmark of 300 false premises to systematically evaluate it across eight models. Suppression rates range from 19\% to 90\%, with four models exceeding 80\%, establishing correction suppression as a prevalent and severe phenomenon. Mechanistic analysis reveals that suppression is not a knowledge failure: the model registers the error internally but task context diverts early-layer attention from the false claim as output intent crystallizes toward compliance at middle layers. We characterize this as \emph{knowing but not correcting} -- suppression occurs at response selection rather than knowledge encoding. Guided by this mechanism, we propose two training-free interventions. Correction Direction Steering (CDS) estimates a correction-compliance direction from matched pairs and injects it at middle layers before output intent crystallizes. Dynamic Payload Amplification (DPA) localizes payload tokens via attention divergence between early and late layers and amplifies their representation at the final layer, requiring no calibration data. Experiments on Qwen3.5-9B and LLaMA3.1-8B show both methods substantially improve factual strictness. CDS achieves the highest correction rate on Qwen3.5-9B (0\%$\to$58.2\%). DPA is the only method that preserves or improves reasoning capability on both models. These findings introduce \emph{factual strictness} -- the willingness to uphold accuracy against contextual pressures -- as a new dimension of model reliability.

Related papers

Guide, Not Bind: Why Defeasible Priors Fail in Augmented Lagrangian Causal Discovery [0.0]
Differentiable causal discovery methods encode expert priors as forbidden-edge constraints enforced by an Augmented Lagrangian penalty.<n>We show this design, which we call emphguide, not bind, fails for two independent, precisely characterized reasons.
arXiv Detail & Related papers (2026-09-03T06:53:34Z)
From Passive Response to Proactive Correction: Enhancing LLM Robustness Against Input Fact Perturbations [12.021107313829082]
Large language models (LLMs) produce confident yet factually incorrect responses when user inputs contain misleading premises.<n>We propose DEDUCE, a three-stage framework that transforms LLMs from passive responders into proactive error correctors.<n> Experiments on TruthfulQA, FalseQA, and our MisFactQA benchmark demonstrate that DEDUCE significantly improves both accuracy and error correction capability.
arXiv Detail & Related papers (2026-08-26T15:13:50Z)
Cybersecurity Detection Classification with Reasoning-enabled Language Models [50.72675435043546]
A major issue in Security Operations Centers (SOCs) is alert fatigue.<n>Prior work prompts or fine-tunes large language models (LLMs) to emit a triage label directly, but does not train them to reason about whether a detection is a genuine threat.<n>We train a chain-of-thought (CoT) reasoning-enabled triage classifier on real, human-labeled Windows endpoint detections.
arXiv Detail & Related papers (2026-07-30T16:22:13Z)
CaliDist: Calibrating Large Language Models via Behavioral Robustness to Distraction [51.56484100374058]
Existing calibration methods for Large Language Models (LLMs) often overlook a critical dimension of trustworthiness: a model's behavioral robustness.<n>We introduce textscCaliDist, a novel post-hoc calibration approach that directly measures and penalizes a model's susceptibility to distraction.<n>textscCaliDist consistently achieves lower Expected Error (ECE) and Brier Score compared with strong baselines.
arXiv Detail & Related papers (2026-06-04T07:27:53Z)
Overconfident Errors Need Stronger Correction: Asymmetric Confidence Penalties for Reinforcement Learning [17.384089089363382]
We identify a root cause that existing methods overlook: the uniform penalization of errors.<n>Current approaches treat all incorrect rollouts within a group identically.<n>We propose the Asymmetric Confidence-aware Error Penalty (ACE)
arXiv Detail & Related papers (2026-02-24T22:46:43Z)
Information Fidelity in Tool-Using LLM Agents: A Martingale Analysis of the Model Context Protocol [69.11739400975445]
We introduce the first theoretical framework for analyzing error accumulation in Model Context Protocol (MCP) agents.<n>We show that cumulative distortion exhibits linear growth and high-probability deviations bounded by $O(sqrtT)$.<n>Key findings include: semantic weighting reduces distortion by 80%, and periodic re-grounding approximately every 9 steps suffices for error control.
arXiv Detail & Related papers (2026-02-10T21:08:53Z)
Mitigating LLM Hallucination via Behaviorally Calibrated Reinforcement Learning [32.32593439144886]
Behavior-calibrated reinforcement learning allows smaller models to surpass frontier models in uncertainty quantification.<n>Our model's log-scale Accuracy-to-Hallucination Ratio gain (0.806) exceeds GPT-5's (0.207) in a challenging in-domain evaluation.
arXiv Detail & Related papers (2025-12-22T22:51:48Z)
Reinforcement Learning with Verifiable yet Noisy Rewards under Imperfect Verifiers [90.50039419576807]
Reinforcement Learning with Verifiable Rewards (RLVR) trains policies against automated verifiers to avoid costly human labeling.<n>To reduce vulnerability to verifier hacking, many RLVR systems collapse rewards to binary $0,1$ during training.<n>This choice carries a cost: it introduces textitfalse negatives (rejecting correct answers, FNs) and textitfalse positives (accepting incorrect ones, FPs)
arXiv Detail & Related papers (2025-10-01T13:56:44Z)
Abduct, Act, Predict: Scaffolding Causal Inference for Automated Failure Attribution in Multi-Agent Systems [20.846301581161978]
Failure attribution in multi-agent systems is a critical yet unsolved challenge.<n>Current methods treat this as a pattern recognition task over long conversation logs.<n>A2P Scaffolding transforms failure attribution from pattern recognition into a structured causal inference task.
arXiv Detail & Related papers (2025-09-12T16:51:15Z)
Selective Learning: Towards Robust Calibration with Dynamic Regularization [79.92633587914659]
Miscalibration in deep learning refers to there is a discrepancy between the predicted confidence and performance. We introduce Dynamic Regularization (DReg) which aims to learn what should be learned during training thereby circumventing the confidence adjusting trade-off.
arXiv Detail & Related papers (2024-02-13T11:25:20Z)
A Close Look into the Calibration of Pre-trained Language Models [56.998539510508515]
Pre-trained language models (PLMs) may fail in giving reliable estimates of their predictive uncertainty. We study the dynamic change in PLMs' calibration performance in training. We extend two recently proposed learnable methods that directly collect data to train models to have reasonable confidence estimations.
arXiv Detail & Related papers (2022-10-31T21:31:07Z)

This list is automatically generated from the titles and abstracts of the papers in this site.