From language models to agent swarms: how did we get here?
Dr Lewis Tunstall
Hugging Face, Inc.
Five years ago, large language models struggled with grade-school mathematics. Today, swarms of thousands of AI agents are producing solutions to problems as formidable as the Navier–Stokes Millennium Prize in a matter of days. How did we get here? In this talk, Dr Tunstall will trace the progression from next-token prediction to systems that reason, write code, use tools, and coordinate their efforts. He’ll explain the ideas behind these advances—chiefly scaling, reinforcement learning, and test-time compute—and discuss what they mean for scientific research.