AGI House · San Francisco · 2026
Automated
Discovery
at Scale
Frontier AI models have produced novel insights in mathematics, physics, and other domains. How do we expand these trickles of insight into a firehose? Join us to discuss coordinating autonomous researchers at scale to solve the hardest scientific and engineering problems.
Speakers
Who's presenting
01
Talk
An AlphaGo Moment for Numerical Methods
AlphaGo was notable not only for superhuman performance but also for making alien moves that defied expert intuition. This talk presents similar findings in the context of numerical methods. Autonomous AI agents developed convex optimization solvers that were orders of magnitude faster than existing implementations, but more importantly challenged core assumptions about algorithm convergence and primal-dual symmetry. Most strikingly, the agents' solutions suggest that the best algorithm is not a discrete choice from a menu of named methods but a continuous surface parameterized by problem structure.
02
Talk
Evolutionary Agents for Scientific Discovery
This talk explores the emerging wave of LLM-driven scientific discovery, focusing on how OSS frameworks can democratize high-level algorithmic search. We will dive into some of the technical details of CodeEvolve, and discuss how open-weight models can match proprietary performance at a fraction of the cost. Finally, we discuss the need for standardized benchmarks to reliably measure progress as we transition toward a future of reproducible, agentic scientific research.
04
Shu Liu, Mert Cemri, Shubham Agarwal, Alex Krentsel
Talk
A Flexible Framework for AI-Driven Scientific and Algorithmic Discovery
Join Us
Be part of the conversation
Apply to attend or submit a talk proposal.
Attend
Join an afternoon of talks and discussion on the frontier of automated scientific discovery.
Apply to Attend →Present
Have results in AI-driven research, agentic science, or automated discovery? Share your work.
Apply to Present →