arXivDaily arXiv每日学术速递 周一至周五更新
arXiv 2410.14166cs.CLcs.AI
收藏

LLM 天才悖论:语言与数学专家在简单基于单词的计数问题上的挣扎

LLM The Genius Paradox: A Linguistic and Math Expert's Struggle with Simple Word-based Counting Problems

University of Southern California(南加州大学)

Nan Xu, Xuezhe Ma

AI总结:

本文通过多组评估验证了 LLM 在简单单词计数任务上失败并非源于 tokenization、架构或训练数据等固有缺陷,并证明引导推理比微调和上下文学习更能稳健高效地提升计数表现。

摘要:

有趣的是,LLM 仍然在一些人类认为微不足道的基本任务上存在困难,例如计算单词“strawberry”中字符 r 的数量。关于 LLM 在简单基于单词的计数问题上存在缺陷的原因,有几种流行的猜想(例如 tokenization、架构和训练数据),它们都持有类似的信念,即这种失败源于模型预训练,因此在部署期间可能不可避免。在本文中,我们精心设计了多个评估设置,以研究流行猜想的有效性。同时,我们测量了从专用 LLM 到简单计数任务的高级数学和编码推理能力的可迁移性。尽管专用 LLM 也受到计数问题的困扰,但我们发现关于 LLM 固有缺陷的猜想是无效的,并进一步寻求从 LLM 中引出对计数任务有益的知识和能力的机会。与通常用于增强新任务或挑战性任务性能的微调和上下文学习等策略相比,我们表明,参与推理是帮助 LLM 更好地感知任务并给出更准确响应的最稳健、最有效的方式。我们希望我们的猜想验证设计能够为研究 LLM 未来的关键失败模式提供见解。基于将高级能力迁移到更简单任务中的挑战,我们呼吁更多地关注模型能力获取和评估。我们还强调了在模型预训练期间培养“先推理后响应”意识的重要性。

英文摘要:

Interestingly, LLMs yet struggle with some basic tasks that humans find trivial to handle, e.g., counting the number of character r's in the word "strawberry". There are several popular conjectures (e.g., tokenization, architecture and training data) regarding the reason for deficiency of LLMs in simple word-based counting problems, sharing the similar belief that such failure stems from model pretraining hence probably inevitable during deployment. In this paper, we carefully design multiple evaluation settings to investigate validity of prevalent conjectures. Meanwhile, we measure transferability of advanced mathematical and coding reasoning capabilities from specialized LLMs to simple counting tasks. Although specialized LLMs suffer from counting problems as well, we find conjectures about inherent deficiency of LLMs invalid and further seek opportunities to elicit knowledge and capabilities from LLMs that are beneficial to counting tasks. Compared with strategies such as finetuning and in-context learning that are commonly adopted to enhance performance on new or challenging tasks, we show that engaging reasoning is the most robust and efficient way to help LLMs better perceive tasks with more accurate responses. We hope our conjecture validation design could provide insights into the study of future critical failure modes of LLMs. Based on challenges in transferring advanced capabilities to much simpler tasks, we call for more attention to model capability acquisition and evaluation. We also highlight the importance of cultivating consciousness of "reasoning before responding" during model pretraining.

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。

↑