arXivDaily arXiv每日学术速递 周一至周五更新
​通知:联邦节假日调休:哥伦布日 / 原住民日(Columbus Day / Indigenous Peoples' Day),今日(13日)无更新。
arXiv 2501.04682cs.AIcs.CL

迈向大语言模型中的系统2推理:用元思维链学习如何思考

Towards System 2 Reasoning in LLMs: Learning How to Think With Meta Chain-of-Thought

Stanford University(斯坦福大学);UC Berkeley(加州大学伯克利分校)
Violet Xiang, Charlie Snell, Kanishk Gandhi, Alon Albalak, Anikait Singh, Chase Blagden, Duy Phung, Rafael Rafailov, Nathan Lile, Dakota Mahan, Louis Castricato… 展开作者

Violet Xiang, Charlie Snell, Kanishk Gandhi, Alon Albalak, Anikait Singh, Chase Blagden, Duy Phung, Rafael Rafailov, Nathan Lile, Dakota Mahan, Louis Castricato, Jan-Philipp Franken, Nick Haber, Chelsea Finn

AI总结:

本文提出 Meta-CoT 框架,显式建模生成思维链背后的推理过程,并结合过程监督、合成数据、搜索轨迹指令微调和强化学习,为 LLM 实现类系统2推理提供路线图。

摘要:

我们提出了一个名为 Meta Chain-of-Thought(Meta-CoT,元思维链)的新颖框架,它通过显式建模得出某一特定 Chain-of-Thought(CoT,思维链)所需的底层推理,扩展了传统的 CoT。我们给出了来自最先进模型的经验证据,这些模型表现出与上下文内搜索相一致的行为;同时探索了通过过程监督、合成数据生成和搜索算法来产生 Meta-CoT 的方法。最后,我们概述了一条训练模型生成 Meta-CoT 的具体流程,其中结合了使用线性化搜索轨迹的指令微调,以及强化学习后训练。我们还讨论了若干开放研究问题,包括缩放定律、验证器的作用,以及发现新型推理算法的潜力。本工作为在 LLM 中实现 Meta-CoT 提供了理论与实践路线图,为人工智能中更强大、更接近人类的推理铺平了道路。

英文摘要:

We propose a novel framework, Meta Chain-of-Thought (Meta-CoT), which extends traditional Chain-of-Thought (CoT) by explicitly modeling the underlying reasoning required to arrive at a particular CoT. We present empirical evidence from state-of-the-art models exhibiting behaviors consistent with in-context search, and explore methods for producing Meta-CoT via process supervision, synthetic data generation, and search algorithms. Finally, we outline a concrete pipeline for training a model to produce Meta-CoTs, incorporating instruction tuning with linearized search traces and reinforcement learning post-training. Finally, we discuss open research questions, including scaling laws, verifier roles, and the potential for discovering novel reasoning algorithms. This work provides a theoretical and practical roadmap to enable Meta-CoT in LLMs, paving the way for more powerful and human-like reasoning in artificial intelligence.

机构、总结、中文摘要均由AI分析而来,可能存在错误,请以论文正文为准。

↑