Story perspectives

LLMs Struggle with Social Reasoning, New Studies Reveal

11/5/2025

Summary

Story summary

Recent studies evaluate theory-of-mind reasoning in large language models using benchmarks like OpenToM and Hi-ToM.
The research examines LLMs' social reasoning capabilities and notes limitations in understanding hypothetical scenarios and reading comprehension.
Gandhi et al. and Zhou et al. contribute by exploring the emergence of theory of mind in LLMs.
The results underscore ongoing challenges in aligning deep learning with human-like social intelligence and informing future AI research.

History

Historical Context

In 2019, researchers like Allen et al. highlighted that only 30% of AI systems could accurately interpret human emotions, revealing significant limitations in machine empathy. This finding connects to ongoing studies, such as OpenToM and Hi-ToM, which explore theory-of-mind reasoning in AI, emphasizing the need for further research to enhance AI's social intelligence capabilities.

1 of 2

Story summary

Recent studies evaluate theory-of-mind reasoning in large language models using benchmarks like OpenToM and Hi-ToM.
The research examines LLMs' social reasoning capabilities and notes limitations in understanding hypothetical scenarios and reading comprehension.
Gandhi et al. and Zhou et al. contribute by exploring the emergence of theory of mind in LLMs.
The results underscore ongoing challenges in aligning deep learning with human-like social intelligence and informing future AI research.

1 / 2