# Are LLMs the Master of All Trades? : Exploring Domain-Agnostic Reasoning Skills of LLMs

**Authors:** Shrivats Agrawal  
**arXiv:** [2303.12810](https://arxiv.org/abs/2303.12810)  
**Date:** 2023-03-22  
**Summary page:** https://www.summarizepaper.com/en/paper/2303.12810v1/

## Key points

- The research explores the potential of large language models (LLMs) to reason like humans across different domains.
- Experiments were conducted using existing datasets on analogical and spatial reasoning tasks, as well as more open-ended natural language questions.
- LLMs excel at analogical and moral reasoning but struggle with spatial reasoning tasks.
- This study highlights the importance of developing diverse reasoning proficiencies for LLMs in contexts where they are required to emulate human-like cognitive abilities.
- Training data plays a crucial role in shaping the performance of LLMs across different domains.
- Future development efforts should focus on incorporating more varied and relevant information during training in order to improve their capabilities for domain-agnostic reasoning skills.

## Summary

This research work explores the potential of large language models (LLMs) to reason like humans across different domains. While human reasoning abilities are multifaceted and can be seen in various forms, including analogical, spatial, and moral reasoning, among others, it remains unclear whether LLMs can perform equally well across all these different domains. To investigate this question, experiments were conducted using existing datasets on analogical and spatial reasoning tasks. Additionally, the ability of LLMs to reason like humans was evaluated by testing their performance on more open-ended natural language questions. The findings indicate that LLMs excel at analogical and moral reasoning but struggle to perform as proficiently on spatial reasoning tasks. These results suggest that while LLMs have shown remarkable progress in certain areas of reasoning, they still have limitations when it comes to other types of cognitive abilities. Furthermore, this study highlights the importance of developing diverse reasoning proficiencies for LLMs in contexts where they are required to emulate human-like cognitive abilities. The experiments shed light on the fact that training data plays a crucial role in shaping the performance of LLMs across different domains. Therefore, future development efforts should focus on incorporating more varied and relevant information during training in order to improve their capabilities for domain-agnostic reasoning skills. Overall, this research provides valuable insights into the strengths and limitations of LLMs when it comes to emulating human-like cognitive abilities across different domains. By better understanding these aspects we can continue pushing forward our understanding of how these models can be applied in real-world settings.