Daily Arxiv

This is a page that curates AI-related papers published worldwide.
All content here is summarized using Google Gemini and operated on a non-profit basis.
Copyright for each paper belongs to the authors and their institutions; please make sure to credit the source when sharing.

Theory of Mind in Large Language Models: Assessment and Enhancement

Created by
  • Haebom

Author

Ruirui Chen, Weifeng Jiang, Chengwei Qin, Cheston Tan

Outline

This paper provides a comprehensive analysis of the Theory of Mind (ToM) capabilities of large-scale language models (LLMs), i.e., their ability to infer the mental states of themselves and others. We review methods for assessing ToM in LLMs, focusing on recently proposed and widely used story-based benchmarks, and provide an in-depth analysis of cutting-edge methods for enhancing ToM in LLMs. Furthermore, we suggest future research directions for further developing ToM in LLMs and making them more adaptable to realistic and diverse situations.

Takeaways, Limitations

•
Takeaways:
◦
We present a systematic methodology for assessing and improving ToM abilities in LLM.
◦
We empirically analyze the ToM abilities of LLM students using a story-based benchmark.
◦
We introduce and compare and analyze various strategies for improving ToM ability in LLM.
◦
It contributes to the development of ToM ability in LLM by suggesting future research directions.
•
Limitations:
◦
There may be a lack of discussion on the limitations and directions for improvement of the currently proposed story-based benchmarks.
◦
Comparative analyses of LLM's abilities on different types of ToM tasks may be inadequate.
◦
There may be a lack of concrete discussion of real-world applications.
👍
Made with Slashpage