# More Agents Is All You Need

**Authors:** Junyou Li, Qin Zhang, Yangbin Yu, Qiang Fu, Deheng Ye  
**arXiv:** [2402.05120](https://arxiv.org/abs/2402.05120)  
**Date:** 2024-02-03  
**Summary page:** https://www.summarizepaper.com/en/paper/2402.05120v1/

## Key points

- Authors present a novel approach to enhancing the performance of large language models (LLMs)
- Method involves scaling up LLMs based on the number of instantiated agents using sampling-and-voting
- Approach is orthogonal to existing complicated methods and can be used in conjunction with other techniques
- Enhancement achieved is correlated with task difficulty level
- Comprehensive experiments conducted on various LLM benchmarks validate findings
- Results confirm effectiveness of proposed method and provide insights into its properties
- Code made publicly available for further exploration and implementation at [https://anonymous.4open.science/r/more_agent_is_all_you_need](https://anonymous.4open.science/r/more_agent_is_all_you_need)
- Research introduces a straightforward yet effective approach to improving LLM performance by increasing the number of agents through sampling-and-voting
- Offers practical implications for enhancing natural language processing tasks

## Summary

In their paper titled "More Agents Is All You Need," authors Junyou Li, Qin Zhang, Yangbin Yu, Qiang Fu, and Deheng Ye present a novel approach to enhancing the performance of large language models (LLMs). Their method involves scaling up LLMs based on the number of instantiated agents using a simple sampling-and-voting technique. This approach is orthogonal to existing complicated methods used for enhancing LLMs and can be used in conjunction with other techniques without any conflicts or interference. The degree of enhancement achieved through this method is correlated with the difficulty level of the task at hand. To validate their findings, comprehensive experiments are conducted on various LLM benchmarks. The results confirm the effectiveness of their proposed method and provide insights into its properties. The authors make their code publicly available for further exploration and implementation at: [https://anonymous.4open.science/r/more_agent_is_all_you_need](https://anonymous.4open.science/r/more_agent_is_all_you_need). Overall, this research introduces a straightforward yet effective approach to improving LLM performance by increasing the number of agents through sampling-and-voting. It contributes to our understanding of how different factors impact LLM capabilities and offers practical implications for enhancing natural language processing tasks.