CoT probe-time gains arise primarily from lexical activation and short-range token co-occurrence rather than sentence-level logical derivation.
hub Canonical reference
Chain-of-thought reasoning in the wild is not always faithful.arXiv preprint:2503.08679
Canonical reference. 80% of citing Pith papers cite this work as background.
23
Pith papers citing it
Background
80% of classified citations
abstract
Recent studies indicate that when faced with explicit biases in prompts, models often omit mentioning these biases in their Chain-of-Thought (CoT) output, revealing that verbalized reasoning can give an incorrect picture of how models arrive at conclusions (unfaithfulness). In this work, we show that unfaithful CoT also occurs on naturally worded, non-adversarial prompts without adding artificial biases or editing model outputs. We find that when separately presented with the questions "Is X bigger than Y?" and "Is Y bigger than X?", models sometimes produce superficially coherent arguments to justify systematically answering Yes to both or No to both, despite the contradiction. We present preliminary evidence that this is due to models' implicit biases towards Yes or No, labeling this Implicit Post-Hoc Rationalization. Our results reveal rates up to 13% for production models, and while frontier models are more faithful, none are entirely so, including thinking models like DeepSeek R1 (0.37%) and Sonnet 3.7 with thinking (0.04%). We also investigate Unfaithful Illogical Shortcuts, where models use subtly illogical reasoning to make speculative answers to hard math problems seem rigorously proven. Our findings indicate that while CoT can be useful for assessing outputs, it is not a complete account of the internal process that produced the model's answer and should be used with caution in agentic or safety-critical settings.
hub tools
citation-role summary
background 4
other 1
citation-polarity summary
representative citing papers
In 1-3B instruction-tuned LMs on GSM8K, arithmetic CoT readout is dominated by positional copying of the trailing number before the answer delimiter, accounting for 54-92 percentage points of accuracy.
DOLORES, an agent using a formal language for meta-reasoning to construct adaptive scaffolds on the fly, outperforms prior scaffolding methods by 24.8% on average across four hard benchmarks and multiple model sizes.
Looped language models with latent iterative computation and entropy-regularized depth allocation achieve performance matching up to 12B standard LLMs through superior knowledge manipulation.
CIE-Scorer detects unfaithful CoT by tracing compact sentence-level circuits, building internal-external reasoning graphs, and scoring their discrepancy with Fused Gromov-Wasserstein distance, reporting SOTA results on FaithCoT-Bench with reduced circuit cost.
Premature confidence in LLM chains of thought predicts flawed reasoning and is mitigated by progressive confidence shaping, a label-free RL objective that yields accuracy gains on arithmetic, math, and science tasks.
Faithful chain-of-thought routes answer-relevant information through the CoT path, measured via sufficiency, completeness and necessity with entropy, masked-KL and gradient diagnostics, and improved by information-flow interventions during verifier-based RL.
SCA applies the Information Bottleneck principle via NIBS and GIBS methods to identify erroneous steps in black-box LLM reasoning and boosts self-correction success by up to 13.5%.
CoT traces align with internal answer commitment in only 61.9% of steps on average, dominated by confabulated continuations after commitment has stabilized.
CASPO trains LLMs via iterative direct preference optimization so that token-level confidence tracks step-wise correctness, then applies Confidence-aware Thought pruning at inference to improve both reliability and speed on reasoning benchmarks.
Counterfactual prompting effects on LLMs are often indistinguishable from those caused by meaning-preserving paraphrases, causing most previously reported demographic sensitivities to disappear under proper statistical comparison.
ReSS extracts decision paths from trees as scaffolds to guide LLM reasoning generation, fine-tunes the LLM on the resulting dataset with scaffold-invariant augmentation, and reports up to 10% gains on medical and financial tabular benchmarks with new faithfulness metrics.
Adversarial explanation attacks preserve nearly all human trust in wrong AI outputs by using persuasive framing, shown in a study varying reasoning, evidence, style, and format with over 200 participants.
AnalyticScore applies new FGTI interpretability principles to text-based scoring and achieves accuracy within 0.06 QWK of uninterpretable state-of-the-art while matching human featurization on the ASAP-SAS dataset.
RLVR incentivizes correct reasoning in base LLMs, extending reasoning boundaries on math and coding tasks as shown by CoT-Pass@K evaluations and a theoretical incentive framework.
Agentic LLM collectives are proposed as natural-language-interpretable computational substrates for ALife research.
LLM reasoning traces and post-hoc explanations increase false trust in incorrect predictions, whereas contrastive dual explanations enhance users' ability to distinguish correct from incorrect AI outputs.
LLM reasoning is primarily mediated by latent-state trajectories rather than by explicit surface chain-of-thought outputs.
KnowRL integrates a knowledge-verification factuality reward into RL training to enforce fact-based reasoning steps and lower hallucination rates in LLMs.
A 14B reasoning model trained via supervised fine-tuning on selected prompts and o3-mini traces, plus outcome RL, outperforms larger open models like DeepSeek-R1-Distill-Llama-70B on math, coding, planning and related benchmarks.
A survey compiling RL methods, challenges, data resources, and applications for enhancing reasoning in large language models and large reasoning models since DeepSeek-R1.
citing papers explorer
-
What Makes Chain-of-Thought Work at Probe Time? Local Co-occurrence Rather Than Global Derivation
CoT probe-time gains arise primarily from lexical activation and short-range token co-occurrence rather than sentence-level logical derivation.
-
The Readout Shortcut: Positional Number Copying Dominates Arithmetic CoT Readout in Small Language Models
In 1-3B instruction-tuned LMs on GSM8K, arithmetic CoT readout is dominated by positional copying of the trailing number before the answer delimiter, accounting for 54-92 percentage points of accuracy.
-
Deep Reasoning in General Purpose Agents via Structured Meta-Cognition
DOLORES, an agent using a formal language for meta-reasoning to construct adaptive scaffolds on the fly, outperforms prior scaffolding methods by 24.8% on average across four hard benchmarks and multiple model sizes.
-
Scaling Latent Reasoning via Looped Language Models
Looped language models with latent iterative computation and entropy-regularized depth allocation achieve performance matching up to 12B standard LLMs through superior knowledge manipulation.
-
Detecting Unfaithful Chain-of-Thought via Circuit-Guided Internal-External Discrepancy
CIE-Scorer detects unfaithful CoT by tracing compact sentence-level circuits, building internal-external reasoning graphs, and scoring their discrepancy with Fused Gromov-Wasserstein distance, reporting SOTA results on FaithCoT-Bench with reduced circuit cost.
-
Understanding and Mitigating Premature Confidence for Better LLM Reasoning
Premature confidence in LLM chains of thought predicts flawed reasoning and is mitigated by progressive confidence shaping, a label-free RL objective that yields accuracy gains on arithmetic, math, and science tasks.
-
Faithfulness as Information Flow: Evaluating and Training Faithful Chain-of-Thought Reasoning
Faithful chain-of-thought routes answer-relevant information through the CoT path, measured via sufficiency, completeness and necessity with entropy, masked-KL and gradient diagnostics, and improved by information-flow interventions during verifier-based RL.
-
Diagnosing Multi-step Reasoning Failures in Black-box LLMs via Stepwise Confidence Attribution
SCA applies the Information Bottleneck principle via NIBS and GIBS methods to identify erroneous steps in black-box LLM reasoning and boosts self-correction success by up to 13.5%.
-
When Reasoning Traces Become Performative: Step-Level Evidence that Chain-of-Thought Is an Imperfect Oversight Channel
CoT traces align with internal answer commitment in only 61.9% of steps on average, dominated by confabulated continuations after commitment has stabilized.
-
Confidence-Aware Alignment Makes Reasoning LLMs More Reliable
CASPO trains LLMs via iterative direct preference optimization so that token-level confidence tracks step-wise correctness, then applies Confidence-aware Thought pruning at inference to improve both reliability and speed on reasoning benchmarks.
-
Compared to What? Baselines and Metrics for Counterfactual Prompting
Counterfactual prompting effects on LLMs are often indistinguishable from those caused by meaning-preserving paraphrases, causing most previously reported demographic sensitivities to disappear under proper statistical comparison.
-
ReSS: Learning Reasoning Models for Tabular Data Prediction via Symbolic Scaffold
ReSS extracts decision paths from trees as scaffolds to guide LLM reasoning generation, fine-tunes the LLM on the resulting dataset with scaffold-invariant augmentation, and reports up to 10% gains on medical and financial tabular benchmarks with new faithfulness metrics.
-
When AI Persuades: Adversarial Explanation Attacks on Human Trust in AI-Assisted Decision Making
Adversarial explanation attacks preserve nearly all human trust in wrong AI outputs by using persuasive framing, shown in a study varying reasoning, evidence, style, and format with over 200 participants.
-
Interpretability from the Ground Up: Stakeholder-Centric Design of Automated Scoring in Educational Assessments
AnalyticScore applies new FGTI interpretability principles to text-based scoring and achieves accuracy within 0.06 QWK of uninterpretable state-of-the-art while matching human featurization on the ASAP-SAS dataset.
-
Reinforcement Learning with Verifiable Rewards Implicitly Incentivizes Correct Reasoning in Base LLMs
RLVR incentivizes correct reasoning in base LLMs, extending reasoning boundaries on math and coding tasks as shown by CoT-Pass@K evaluations and a theoretical incentive framework.
-
Conversable Complexity: Agentic LLM Collectives as Interpretable Substrates
Agentic LLM collectives are proposed as natural-language-interpretable computational substrates for ALife research.
-
Evaluating the False Trust Engendered by LLM Explanations
LLM reasoning traces and post-hoc explanations increase false trust in incorrect predictions, whereas contrastive dual explanations enhance users' ability to distinguish correct from incorrect AI outputs.
-
LLM Reasoning Is Latent, Not the Chain of Thought
LLM reasoning is primarily mediated by latent-state trajectories rather than by explicit surface chain-of-thought outputs.
-
KnowRL: Exploring Knowledgeable Reinforcement Learning for Factuality
KnowRL integrates a knowledge-verification factuality reward into RL training to enforce fact-based reasoning steps and lower hallucination rates in LLMs.
-
Phi-4-reasoning Technical Report
A 14B reasoning model trained via supervised fine-tuning on selected prompts and o3-mini traces, plus outcome RL, outperforms larger open models like DeepSeek-R1-Distill-Llama-70B on math, coding, planning and related benchmarks.
-
A Survey of Reinforcement Learning for Large Reasoning Models
A survey compiling RL methods, challenges, data resources, and applications for enhancing reasoning in large language models and large reasoning models since DeepSeek-R1.
- Can Aha Moments Be Fake? Towards Quantifying Decorative and True Thinking in Chain-of-Thought
- TinyTroupe: An LLM-powered Multiagent Persona Simulation Toolkit