近期研究表明，大型语言模型的思维链（Chain-of-Thought, CoT）推理并不总是忠实于模型的内部决策过程。研究发现，模型可能会生成听起来合理但不能准确反映其结论的解释，这种现象甚至在前沿模型中也观察到。这种不忠实可能表现为隐式的事后合理化或不合逻辑的捷径，并且它也延伸到未明确表达中间计算的潜在 CoT 方法。研究结果表明，在使用 CoT 来评估模型输出时应谨慎，尤其是在安全关键应用中，因为它可能不能完全代表模型的真实推理或内部信念。 AI
影响 思维链推理不是模型内部过程的可靠指标，在使用它进行安全和可解释性评估时需要谨慎。
排序理由 多篇 arXiv 论文分析了 LLM 中思维链推理的忠实度。
AI 生成摘要 · Google Gemini · 来自 28 个来源。 我们如何撰写摘要 →
arXiv:2510.08647v2 Announce Type: replace-cross Abstract: Recent developments have enabled advanced reasoning in Large Language Models (LLMs) via long Chain-of-Thought (CoT), trading efficiency during inference for performance. Existing works focus on compressing generated CoT in…
Chain-of-thought (CoT) reasoning has become a widely used mechanism for eliciting multi-step reasoning in large language models by generating intermediate reasoning steps at inference time. Yet the scaling behavior of generalization with CoT depth remains poorly understood. To ad…
arXiv:2606.02020v1 Announce Type: new Abstract: This paper investigates the entropy dynamics of Chain-of-Thought (CoT) and uncovers a consistent two-phase structure: an Uncertainty Region of exploration transitioning sharply to a Confidence Region of convergence. We demonstrate t…
arXiv:2606.00096v1 Announce Type: cross Abstract: Visual agents employ external visual tools within visual chains of thought to incorporate fine-grained evidence. While prior work has mainly studied these tools in visual search tasks, their role in more complex visual reasoning r…
This paper investigates the entropy dynamics of Chain-of-Thought (CoT) and uncovers a consistent two-phase structure: an Uncertainty Region of exploration transitioning sharply to a Confidence Region of convergence. We demonstrate that the Confidence Region possesses two critical…
arXiv:2503.08679v5 Announce Type: replace 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 …
arXiv:2602.08783v3 Announce Type: replace Abstract: Latent or continuous chain-of-thought methods replace explicit textual rationales with a number of internal latent steps, but these intermediate computations are difficult to evaluate beyond correlation-based probes. In this pap…
arXiv:2605.28842v1 Announce Type: cross Abstract: The success of large language models (LLMs) across diverse NLP tasks has elevated the importance of reasoning chain optimization as a critical step in aligning model behavior with task objectives. Existing reasoning chain tuning m…
arXiv:2603.05488v4 Announce Type: replace-cross Abstract: We provide evidence of performative chain-of-thought (CoT) in reasoning models, where a model becomes strongly confident in its final answer, but continues generating tokens without revealing its internal belief. Our analy…
arXiv:2605.28913v1 Announce Type: new Abstract: Large reasoning models (LRMs) often generate extensive chain-of-thought (CoT) traces before producing a final answer. As explicit textual artifacts, these traces can be passed to other models to solve the same task, enabling cross-m…
arXiv:2602.02103v2 Announce Type: replace-cross Abstract: Chain-of-thought (CoT) reasoning has become a central mechanism for eliciting multi-step reasoning in Large Language Models (LLMs). Yet recent evidence presents a tension: hidden states appear to already encode future reas…
arXiv:2605.28600v1 Announce Type: new Abstract: Chain-of-Thought (CoT) prompting substantially improves the sample efficiency of transformers, reducing the complexity of tasks like parity learning from exponential to polynomial in the input length. However, generating explicit re…
arXiv:2605.27901v1 Announce Type: cross Abstract: Chain-of-thought (CoT) monitoring has been proposed as a promising safety mechanism for detecting misaligned behavior in large language models. However, its reliability remains largely unexplored beyond English and across diverse …
arXiv:2605.27773v1 Announce Type: cross Abstract: When a language model sees a document contradicting its training knowledge, it must choose: follow the document or trust itself. Prior work proved this choice depends on how well-known the fact is. We ask: does the model's chain-o…
Large reasoning models (LRMs) often generate extensive chain-of-thought (CoT) traces before producing a final answer. As explicit textual artifacts, these traces can be passed to other models to solve the same task, enabling cross-model reasoning transfer. Yet successful transfer…
Chain-of-Thought (CoT) prompting substantially improves the sample efficiency of transformers, reducing the complexity of tasks like parity learning from exponential to polynomial in the input length. However, generating explicit reasoning steps at inference is computationally ex…
arXiv:2507.20758v2 Announce Type: replace Abstract: Chain-of-Thought (CoT) prompting significantly enhances model reasoning, yet its internal mechanisms remain poorly understood. We analyze CoT's operational principles by reversely tracing information flow across decoding, projec…
arXiv:2605.26795v1 Announce Type: new Abstract: Chain-of-thought (CoT) prompting reliably improves language-model accuracy, but which properties of a rationale text drive the improvement is poorly understood. Prior work has largely studied generation-time behavior. We instead ask…
arXiv:2605.26772v1 Announce Type: new Abstract: Large reasoning models (LRMs) generate chain-of-thought (CoT) traces before producing final outputs, introducing a dynamic internal state that may complicate control mechanisms such as refusal. Unlike instruction-tuned LLMs, where r…
arXiv:2601.21576v2 Announce Type: replace Abstract: Chain-of-Thought (CoT) has unlocked advanced reasoning abilities of Large Language Models (LLMs) with intermediate steps, yet incurs prohibitive computational costs due to generation of extra tokens. Recent studies empirically s…
Chain-of-thought monitoring shows poor reliability across diverse languages and model families, with high rates of unfaithfulness and deceptive behaviors that persist in low-resource languages.
arXiv:2605.24960v1 Announce Type: cross Abstract: Chain-of-Thought (CoT) faithfulness, i.e., whether CoTs genuinely reflect large language models' (LLM) underlying behavior, is typically evaluated under two disjoint paradigms: contextual faithfulness, measured by perturbing the i…
arXiv:2605.24286v1 Announce Type: cross Abstract: Chain-of-thought (CoT) reasoning is useful for monitoring language models only when the reasoning trace faithfully reflects the computation that produces the final answer. However, models can rely on prompt-to-answer shortcuts tha…
We develop a learning-theoretic framework for understanding Chain of Thought (CoT). We model CoT as the interaction between an answer map and a chain rule that generates intermediate questions autoregressively, and define the reasoning risk of a hypothesis under this interaction.…
We develop a learning-theoretic framework for understanding Chain of Thought (CoT). We model CoT as the interaction between an answer map and a chain rule that generates intermediate questions autoregressively, and define the reasoning risk of a hypothesis under this interaction.…
arXiv:2606.03217v1 Announce Type: new Abstract: Chain-of-thought (CoT) reasoning has become a widely used mechanism for eliciting multi-step reasoning in large language models by generating intermediate reasoning steps at inference time. Yet the scaling behavior of generalization…
Chain-of-thought (CoT) reasoning has become a widely used mechanism for eliciting multi-step reasoning in large language models by generating intermediate reasoning steps at inference time. Yet the scaling behavior of generalization with CoT depth remains poorly understood. To ad…
<h2> The Evolution of Thinking Machines </h2> <p>For years, large language models operated on a simple premise: read input, generate output. Fast, stateless, and remarkably capable. But something changed around 2024, and the industry finally caught up.</p> <p><strong>Reasoning mo…