AutoGen: Enabling Next-Gen LLM Applications via Multi-Agent Conversation
Qingyun Wu, Gagan Bansal, Jieyu Zhang, Yiran Wu, Beibin Li, Erkang, Zhu, Li Jiang, Xiaoyun Zhang, Shaokun Zhang, Jiale Liu, Ahmed Hassan, Awadallah, Ryen W White, Doug Burger, and Chi Wang
TL;DR
AutoGen is a versatile open-source framework that enables the creation of complex multi-agent LLM applications with customizable interactions, demonstrated across diverse domains such as mathematics, coding, and decision-making.
Contribution
It introduces a flexible infrastructure for building multi-agent LLM applications with customizable behaviors and interaction modes, supporting natural language and code-based programming.
Findings
Effective in diverse application domains
Supports complex multi-agent interactions
Demonstrates versatility and robustness
Abstract
AutoGen is an open-source framework that allows developers to build LLM applications via multiple agents that can converse with each other to accomplish tasks. AutoGen agents are customizable, conversable, and can operate in various modes that employ combinations of LLMs, human inputs, and tools. Using AutoGen, developers can also flexibly define agent interaction behaviors. Both natural language and computer code can be used to program flexible conversation patterns for different applications. AutoGen serves as a generic infrastructure to build diverse applications of various complexities and LLM capacities. Empirical studies demonstrate the effectiveness of the framework in many example applications, with domains ranging from mathematics, coding, question answering, operations research, online decision-making, entertainment, etc.
Peer Reviews
Decision·Submitted to ICLR 2024
- The approach defines a generic design for agents that can use LLMs, human inputs, certain tools or combination of them. LLM agents can use capabilities such as role playing, progress making from conversation history, proving feedback. Human involvement can be configured at different levels e.g. frequency and conditions for when to request human input. Tools agents can execute code/functions (suggested by LLMs). Combining these agents in different configurations can result in powerful agents w
None
(1) The proposed framework simplifies the overall complex LLM workflows and enables automation. By using conversable and customizable agents, it supports conversational modes for complex workflows. It provides a collection of working systems with different complexities. These systems cover a wide range of applications from various domains and complexities. (2) The proposed method allows developers to use a fusion of natural and programming languages to define agent behaviors and conversation p
(1) The paper does not address the issue of context length, which may become too long as the number of conversation turns increases. This could affect the performance and efficiency of the LLMs and the agents. (2) It would be better to consider the cost issue, which is important for practical applications. The experiments are conducted on GPT-4 and GPT-3.5, which are expensive and not widely accessible. How would the framework perform on open-source LLMs with lower capacity? (3) In my opinio
+ Open-source multi-agent programming framework is definitely an interesting project. + Authors include some empirical results to showcase their performance of the tools. + Authors provide extensive documentation to explain different use cases.
- Not much takeaway in terms of scientific learning. I don’t find as a reader what lessons or results we can get from the paper. The main message is that we have a tool that can help developing multi-agent conversation. In my personal opinion, developing it based on a mature LLM agent is neither time-consuming nor scientifically challenging. - The results are not convincing. I found the comparison of results is not rigorously evaluated and not convincing. For example, the paper shows that by re
Code & Models
Videos
No videos yet. Explain this paper in a talk, walkthrough, or lecture? Add one.
Taxonomy
TopicsMulti-Agent Systems and Negotiation · Natural Language Processing Techniques · Semantic Web and Ontologies