Yale NLP/LLM Interest Group Talks
Empowering Biomedical Research with Large Language Models and AI Agents
We envision 'AI scientists' as collaborative agents capable of reasoning and coding that empower biomedical research. I will first demonstrate MedAgents, the first multi-agent framework in the medical field. Med Agents examines how large language models (LLMs) can be used as collaborative experts to enhance medical reasoning through multi-round discussions. Then, I will briefly present our works on benchmarking existing LLMs on bioinformatics code generation and agents on repository-level data science tasks. Finally, while LLMs' capabilities are promising, I will highlight vulnerabilities and potential risks associated with their misuse and emphasize the need for safety measures.