新研究表明，尽管人类和大型语言模型（LLMs）都会根据问题的难度调整解决时间，但其内部机制却存在显著差异。人类倾向于放弃那些他们认为困难或可能出错的问题，而大型语言模型则会在更难的问题上花费更多的计算资源，但这常常导致错误。这种“审议分配”上的分歧表明，大型语言模型在困难任务上延长处理时间源于不确定性，而非像人类那样进行战略性投入。 AI
影响 强调了大型语言模型和人类在处理复杂问题方式上的一个关键区别，表明当前大型语言模型的推理策略可能与人类的灵活性不完全一致。
排序理由 多篇 arXiv 论文讨论大型语言模型的推理能力和局限性。
在 Hugging Face Daily Papers 阅读 →
AI 生成摘要 · Google Gemini · 来自 15 个来源。 我们如何撰写摘要 →
arXiv:2606.28186v1 Announce Type: cross Abstract: Predicting human item difficulty is central to educational assessment, where reliable estimates support fairness and effective test construction. Existing methods often depend on costly human calibration or item-level textual repr…
Predicting human item difficulty is central to educational assessment, where reliable estimates support fairness and effective test construction. Existing methods often depend on costly human calibration or item-level textual representations, providing limited evidence about the …
arXiv:2606.27103v1 Announce Type: new Abstract: Humans flexibly adapt their reasoning strategies to the requirements of a given problem. Large language models (LLMs) have performed well on many cognitive tasks, however, it is unclear whether this accuracy is a result of pattern m…
arXiv:2606.26108v1 Announce Type: new Abstract: Larger language models consistently outperform smaller ones on reasoning benchmarks, yet the reasoning differences underlying this gap remain underexplored. Across benchmarks in mathematics, physics, chemistry, and programming, we o…
arXiv:2601.03388v3 Announce Type: replace-cross Abstract: Earlier research has shown that metaphors influence human decision-making, raising the question of whether metaphors also influence large language models (LLMs)' reasoning pathways, given that their training data contain a…
arXiv:2606.26797v1 Announce Type: new Abstract: Supervised fine-tuning (SFT) on a small, high-quality set of long reasoning traces is an effective approach for eliciting strong reasoning capabilities in Large Language Models (LLMs). However, existing methods for curating high-qua…
arXiv:2606.26502v1 Announce Type: new Abstract: Large reasoning models (LRMs) take longer on harder problems, just as humans do. This surface similarity hides an opposite pattern within items. When an LRM gets a problem wrong, it spends more tokens than when it gets the same prob…
Epi2Diff framework transforms LRM reasoning traces into cognitive episodes to predict human item difficulty more accurately than existing methods.
Humans flexibly adapt their reasoning strategies to the requirements of a given problem. Large language models (LLMs) have performed well on many cognitive tasks, however, it is unclear whether this accuracy is a result of pattern matching from training data or flexible reasoning…
Supervised fine-tuning (SFT) on a small, high-quality set of long reasoning traces is an effective approach for eliciting strong reasoning capabilities in Large Language Models (LLMs). However, existing methods for curating high-quality SFT data rely heavily on strong reasoning m…
arXiv:2509.22193v2 Announce Type: replace Abstract: Distilling reasoning traces from strong teacher models has become the standard recipe for building capable small language models. Yet reasoning traces are 5-20$\times$ longer than standard instruction fine-tuning (IFT) outputs, …
Large reasoning models (LRMs) take longer on harder problems, just as humans do. This surface similarity hides an opposite pattern within items. When an LRM gets a problem wrong, it spends more tokens than when it gets the same problem right; humans do the reverse, spending less …
Large reasoning models (LRMs) take longer on harder problems, just as humans do. This surface similarity hides an opposite pattern within items. When an LRM gets a problem wrong, it spends more tokens than when it gets the same problem right; humans do the reverse, spending less …
<p>Imagine you are asked to solve a complex math problem. If you answer immediately, you’ll likely rely on intuition or a guess. But if you are given ten minutes to scratch out ideas, double-check your logic, and correct your mistakes before speaking, your accuracy improves signi…
When should a reasoning model quit a problem it probably can't solve? Conformal Thinking sets stop/continue thresholds for test-time compute by distribution-free risk control, holding the error rate under a target you pick. It adds a lower threshold that gives up early when confi…