Agentic AI Summit
Accelerating Algorithm Autoresearch
Date, time, and room will be added once confirmed.
Talk overview
Optimizing software with agentic AI is more complex than it seems. Autoresearch, which uses agents to iteratively modify and test code, faces challenges like reward hacking and overfitting. This talk explores these issues and how to turn autoresearch results into production-ready solutions, using examples from open-source GPU kernel competitions and infrastructure.
