# HuggingGPT: Solving AI Tasks with ChatGPT and its Friends in HuggingFace

**Authors:** Yongliang Shen, Kaitao Song, Xu Tan, Dongsheng Li, Weiming Lu, Yueting Zhuang  
**arXiv:** [2303.17580](https://arxiv.org/abs/2303.17580)  
**Date:** 2023-03-30  
**Summary page:** https://www.summarizepaper.com/en/paper/2303.17580v1/

## Key points

- Achieving Artificial General Intelligence (AGI) in AI requires solving complex tasks across different domains and modalities.
- Large Language Models (LLMs) could act as controllers to manage existing AI models and solve complex tasks.
- Language could serve as a generic interface to empower LLMs in this role.
- HuggingGPT is a system that leverages LLMs such as ChatGPT to connect various AI models from machine learning communities like HuggingFace to solve AI tasks.
- ChatGPT conducts task planning and selects appropriate models based on their function descriptions available in HuggingFace when receiving a user request.
- The system achieves impressive results in language understanding, generation, interaction, reasoning, vision, speech, and other challenging tasks.
- This approach provides a new way towards AGI by enabling LLMs to act as controllers for existing AI models and solve complex tasks across multiple domains and modalities.

## Summary

In the field of Artificial Intelligence (AI), achieving Artificial General Intelligence (AGI) requires solving complex tasks across different domains and modalities. While there are many AI models available for various domains and modalities, they often struggle with complicated tasks. To address this issue, researchers propose that Large Language Models (LLMs) could act as controllers to manage existing AI models and solve complex tasks. Language could serve as a generic interface to empower LLMs in this role. To demonstrate this philosophy, the researchers introduce HuggingGPT, a system that leverages LLMs such as ChatGPT to connect various AI models from machine learning communities like HuggingFace to solve AI tasks. When receiving a user request, ChatGPT conducts task planning and selects appropriate models based on their function descriptions available in HuggingFace. The system then executes each subtask with the selected AI model and summarizes the response according to the execution results. By leveraging the strong language capabilities of ChatGPT and abundant AI models in HuggingFace, HuggingGPT can cover numerous sophisticated AI tasks across different modalities and domains. The system achieves impressive results in language understanding, generation, interaction, reasoning, vision, speech, and other challenging tasks. This approach provides a new way towards AGI by enabling LLMs to act as controllers for existing AI models and solve complex tasks across multiple domains and modalities.