# OpenAGI: When LLM Meets Domain Experts

**Authors:** Yingqiang Ge, Wenyue Hua, Jianchao Ji, Juntao Tan, Shuyuan Xu, Yongfeng Zhang  
**arXiv:** [2304.04370](https://arxiv.org/abs/2304.04370)  
**Date:** 2023-04-10  
**Summary page:** https://www.summarizepaper.com/en/paper/2304.04370v2/

## Key points

- Pursuit of Artificial General Intelligence (AGI) is a challenging task
- Recent developments in Large Language Models (LLMs) have shown remarkable learning and reasoning abilities
- To achieve AGI, it is crucial to equip these models with the capability to harness various domain-specific expert models for complex task-solving
- OpenAGI project has been developed as an open-source AGI research platform that offers complex multi-step tasks accompanied by task-specific datasets, evaluation metrics, and a diverse range of extensible models
- OpenAGI formulates complex tasks as natural language queries that serve as input to the LLM
- The LLM then selects, synthesizes, and executes external models provided by OpenAGI to address the task at hand
- To improve the LLM's task-solving ability further, the project proposes a Reinforcement Learning from Task Feedback (RLTF) mechanism that uses the task-solving result as feedback
- Equipping large intelligent models with domain-specific expert knowledge is essential for achieving AGI
- OpenAGI demonstrates how LLMs can be used as controllers for selecting external expert models in solving complex tasks while RLTF provides feedback for self-improvement AI
- The authors have open sourced their codebase along with benchmarking and evaluation methods on GitHub
- Their work presents an exciting step forward in developing AGIs capable of solving increasingly complex problems.

## Summary

The pursuit of Artificial General Intelligence (AGI) is a challenging task, and recent developments in Large Language Models (LLMs) have shown remarkable learning and reasoning abilities. To achieve AGI, it is crucial to equip these models with the capability to harness various domain-specific expert models for complex task-solving. The OpenAGI project has been developed as an open-source AGI research platform that offers complex multi-step tasks accompanied by task-specific datasets, evaluation metrics, and a diverse range of extensible models. OpenAGI formulates complex tasks as natural language queries that serve as input to the LLM. The LLM then selects, synthesizes, and executes external models provided by OpenAGI to address the task at hand. To improve the LLM's task-solving ability further, the project proposes a Reinforcement Learning from Task Feedback (RLTF) mechanism that uses the task-solving result as feedback. This paradigm of LLMs operating various expert models for complex task-solving is a promising approach towards AGI development. The authors of this paper are Yingqiang Ge, Wenyue Hua, Jianchao Ji, Juntao Tan Shuyuan Xu, and Yongfeng Zhang who propose that equipping large intelligent models with domain-specific expert knowledge is essential for achieving AGI. Their work on OpenAGI demonstrates how LLMs can be used as controllers for selecting external expert models in solving complex tasks while RLTF provides feedback for self-improvement AI. To facilitate long-term improvement and evaluation of AGIs' ability in solving increasingly complex problems, they have open sourced their codebase along with benchmarking and evaluation methods on GitHub. Overall their work presents an exciting step forward in developing AGIs capable of solving increasingly complex problems.