arXiv:2603. 05167v2 Announce Type: replace-cross Abstract: Large language models (LLMs) are increasingly used as judges of chain-of-thought (CoT) reasoning, yet it remains unclear whether they can reliably assess process faithfulness rather than merely answer plausibility.
arXiv:2607. 11266v1 Announce Type: new Abstract: Chain-of-Thought (CoT) prompting has significantly advanced the reasoning capabilities of Large Language Models (LLMs), yet it often incurs substantial computational costs due to over-reasoning: the generation of redundant, verbose, or irrelevant steps.
arXiv:2602. 20094v2 Announce Type: replace Abstract: As large language models (LLMs) witness increasing deployment in complex, high-stakes decision-making scenarios, it becomes imperative to ground their reasoning in causality rather than spurious correlations.
arXiv:2606. 11211v1 Announce Type: cross Abstract: The ability of large language models (LLMs) to express calibrated uncertainty is important for safe deployment.
arXiv:2608. 03291v1 Announce Type: cross Abstract: Chain-of-thought (CoT) reasoning improves large language model (LLM) performance while also providing an observable interface to the model's reasoning process.
The paper investigates whether the text of chain‑of‑thought reasoning steps actually reflects their true importance for a model’s final answer. By defining step importance as the advantage in expected reward when a step is included, the authors use Monte Carlo rollouts to estimate ground truth and then test whether large language model judges can identify high‑advantage steps. They find that capable LLMs can beat a prevalence baseline but still fall far short of a noise ceiling, and that fine‑tuning a step‑level critic improves detection for incorrect responses but remains distant from the ceiling for correct ones, indicating that step importance is only partially recoverable from the reasoning trace text.
The paper introduces ICE (Intervention-Consistent Explanation), a framework that evaluates the faithfulness of large language model explanations by comparing them to random baselines of equal size across multiple intervention operators. It demonstrates that faithfulness varies with the chosen operator, with significant differences observed when switching between deletion and retrieval infill operators. The study evaluates seven LLMs on four tasks, revealing that operator changes can cross the positive-evidence threshold in 18% of configurations and that random baselines uncover anti-faithfulness in nearly a third of English deletion setups, findings that also hold across six non‑English languages and two attribution methods.
GRACE is a step‑level benchmark for evaluating the faithfulness of chain‑of‑thought reasoning over context. It provides human annotations for each step in CoT traces from 10 models across 4 datasets, labeling faithfulness, error category, and natural‑language explanations. The benchmark introduces a data‑driven taxonomy that splits errors into GRACE‑Inference (deductive) and GRACE‑Grounding (factual) tracks, each with four categories, and demonstrates that incorporating step‑level faithfulness signals can improve downstream accuracy and reasoning reliability.
The paper introduces LSR‑Ben, a benchmark designed to evaluate process reward models (PRMs) on scientific and logical reasoning tasks, addressing a gap left by existing math‑focused benchmarks. Experiments on 22 models reveal that PRMs and LLMs perform poorly in non‑mathematical domains, with LLMs tending to over‑identify errors while PRMs tend to overlook them. LSR‑Ben aims to spur research that broadens PRM applicability and improves LLM reasoning.
arXiv:2608. 12585v1 Announce Type: new Abstract: Improving reasoning LLMs requires the ability to judge the quality of long reasoning traces for effective reasoning data curation, strong training signals during reinforcement learning, and an in-depth understanding of reasoning behaviors during model performance evaluation.
arXiv:2601. 13735v2 Announce Type: replace Abstract: Probabilistic confidence metrics are increasingly adopted as proxies for reasoning quality in Best-of-N selection, under the assumption that higher confidence reflects higher reasoning fidelity.
The paper introduces an adaptive triggering mechanism for bias correction in large language model (LLM) reasoning. By framing bias intervention as an online change‑point detection problem, the authors update a CUSUM statistic at each step using either a white‑box next‑token probability signal or a black‑box LLM judge signal, and inject corrective prompts only when the accumulated evidence exceeds a calibrated threshold. Experiments on gpt‑4o‑mini and six open‑weight models show that adaptive black‑box triggering restores most of the accuracy lost by fixed‑interval interventions while reducing the number of corrections, whereas the white‑box signal improves ambiguous‑item accuracy but can hurt disambiguated‑item accuracy due to difficulty distinguishing stereotype reliance from correct evidence.
ReFIne is a training framework that augments large reasoning models with three trustworthiness properties: interpretability, faithfulness, and reliability. It combines supervised fine‑tuning with GRPO to produce structured, tag‑based reasoning traces, explicitly disclose decisive information, and provide self‑assessments of soundness and confidence. Applied to Qwen3 models, ReFIne improves interpretability by 44.0 %, faithfulness by 18.8 %, and reliability by 42.4 % on mathematical benchmarks.