CoT 隐状态揭示任务求解方向:LLaMA 3.1 8B 准确率从 46% 提升至 85%
Chain-of-Thought Hidden States Reveal Task-Solving Directions for Direct Answers
AI 导读
arXiv 一项研究通过分析 CoT 隐状态,发现直接提示的表示空间中存在可将模型导向正确答案的方向。在 25 个任务上,将直接提示最后一层隐状态向该方向旋转后,LLaMA 3.1 8B 平均准确率从 46% 提升至 85%(CoT 为 69%),Qwen 2.5 7B 从 52% 提升至 86%(CoT 为 77%),且仅用约十分之一的 CoT token。
正文 · 原文
Abstract:Chain-of-Thought (CoT) prompting improves the performance of Large Language Models by eliciting intermediate reasoning steps. However, CoT is far from perfect: on some tasks, models perform worse than with direct-answer prompting, and on others they generate a few hundred tokens to reach the correct answer. We start by labelling each CoT hidden state as recovering if patching it into the direct prompt yields the correct answer, and failed otherwise. A linear probe shows that the model encodes these two classes differently. We then show that the representation space of the direct prompt holds a direction for each question that leads the model, in a short generation, to the correct answer more often than CoT does. Specifically, we rotate the direct prompt's last hidden state once, before generation, towards a direction built for each question from its own recovering and failed states. Across 25 tasks, these directions raise the direct prompt's mean accuracy from $46\%$ to $85\%$ on LLaMA 3.1 8B (CoT: $69\%$) and from $52\%$ to $86\%$ on Qwen 2.5 7B (CoT: $77\%$), with a tenth of CoT's tokens. CoT is thus not the shortest route to this accuracy, and these directions can be used in future work to train models to align their states with them.
| Subjects: | Computation and Language (cs.CL); Artificial Intelligence (cs.AI) |
| Cite as: | arXiv:2611.24598 [cs.CL] |
| (or arXiv:2611.24598v1 [cs.CL] for this version) | |
| https://doi.org/10.48550/arXiv.2611.24598 arXiv-issued DOI via DataCite (pending registration) |
Submission history
From: Houman Mehrafarin [view email]
[v1]
Wed, 18 Nov 2026 15:25:14 UTC (683 KB)
来源:arXiv cs.AI · arxiv.org