AutoGen: Next Gen LLMs with Multi-Agent Conversation
And how does this affect the current LLM applications?
So essentially,
"AutoGen allows LLM apps via multiple agents to converse and accomplish tasks!"
Paper: AutoGen: Enabling Next-Gen LLM Applications via Multi-Agent Conversation
Researchers from Microsoft Research, Pennsylvania State University, The University of Washington, and Xidian University are interested in understanding and evaluating how multiple agents can collaborate and different strategies of communication.
From previous research, we know that multiple agents can help encourage divergent thinking (Liang et al., 2023), improve factuality and reasoning (Du et al., 2023), and provide validation (Wu et al., 2023). The main question researchers wanted to answer is how can we facilitate the development of LLM applications that could span a broad spectrum of domains and complexities based on the multi-agent approach.
From the paper, their methodology involved:
Experimentation with various solutions to compare the effectiveness of AutoGen including creation of conversation agents and conversational programming.
The usage of a strict format requirement for output format evaluation creates a new combination of natural language and programming which can be used for agent conversations
Evaluation of output similarity to all valid action options using the BLEU metric.
For evaluation, they used the following metrics:
BLEU metric to evaluate the similarity of the output to all valid action options.
Comparison of AutoGen's performance on the Natural Questions dataset (Kwiatkowski et al., 2019) to obtain comparative evaluation metrics for the system's performance.
AutoGen is open source and features a unified conversation interface among the agents, along with an auto-reply mechanism, which helps establish an agent-interaction interface that capitalizes on the strengths of chat-optimized LLMs with broad capabilities while accommodating a wide range of applications.
Their future work would involve exploring the effective integration of existing agent implementations into the multi-agent framework and investigating the optimal balance between automation and human control in multi-agent workflows. Also developing and refining agent topology and conversation patterns for the most effective multi-agent conversations while optimizing the overall efficiency, among other factors would be useful.
So essentially,
"AutoGen allows LLM apps via multiple agents to converse and accomplish tasks!"