arXivDaily arXiv每日学术速递 周一至周五更新
arXiv 2406.06140cs.CLcs.LG
收藏

我能理解我所创造的吗?大语言模型的自我知识评估

Can I understand what I create? Self-Knowledge Evaluation of Large Language Models

Tsinghua University(清华大学);Shanghai Jiao Tong University(上海交通大学);Microsoft Research Asia(微软亚洲研究院)

Zhiquan Tan, Lai Wei, Jindong Wang, Xing Xie, Weiran Huang

AI总结:

受费曼“通过创造来理解”原则启发,本文提出一种自我知识评估框架,通过让模型回答自生成问题发现其自我知识存在显著差距,并表明自生成数学任务微调可提升模型表现。

摘要:

大语言模型(LLMs)在语言任务中取得了显著进展,因此需要稳健的评估框架来理解其能力与局限性。受费曼“通过创造来理解”原则的启发,我们提出了一种易于实现的自我知识评估框架,用于评估模型理解和回答自身生成问题的能力。基于在多个任务上对多个模型进行测试的结果,我们发现模型在自我知识能力方面存在显著差距。进一步分析表明,这些差距可能源于与人类注意力机制的不对齐。此外,在自生成的数学任务上进行微调可能会提升模型的数学表现,这凸显了该框架在高效且有洞察力的模型评估方面的潜力,并可能有助于大语言模型的改进。

英文摘要:

Large language models (LLMs) have achieved remarkable progress in linguistic tasks, necessitating robust evaluation frameworks to understand their capabilities and limitations. Inspired by Feynman's principle of understanding through creation, we introduce a self-knowledge evaluation framework that is easy to implement, evaluating models on their ability to comprehend and respond to self-generated questions. Our findings, based on testing multiple models across diverse tasks, reveal significant gaps in the model's self-knowledge ability. Further analysis indicates these gaps may be due to misalignment with human attention mechanisms. Additionally, fine-tuning on self-generated math task may enhance the model's math performance, highlighting the potential of the framework for efficient and insightful model evaluation and may also contribute to the improvement of LLMs.

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。

↑