Hear from the greatest minds in Maths and Computer Science – introduced by Tom at the 2026 Heidelberg Laureate Forum.
Jacob Tsimerman – 2026 Fields Medal
Torsten Hoefler – 2024 ACM Prize in Computing
Jeffrey Ullman – 2020 Turing Award
Yu Deng – 2026 Fields Medal
Yu Deng: Critical Problems In Statistical Physics and PDEs
Natural systems can be classified by scaling into subcritical, critical, and supercritical ones. This philosophy applies to both deterministic and random systems. In general, subcritical problems are simpler, and critical ones may have very different behaviour. In this lecture I will discuss recent progress on critical problems in different contexts including particle systems, wave turbulence, random data problems and SPDE.
Torsten Hoefler: Can We Build an AI Scientist?
Large language models have proven to be powerful knowledge interfaces, yet their full potential for scientific discovery remains largely untapped. While current AI research often focuses on suppressing inaccurate outputs, this talk argues that the true frontier of AI for science lies in building an ”AI Scientist” – a closed-loop system capable of autonomous hypothesis generation, rigorous verification, and iterative knowledge integration. We outline an architecture for an AI Scientist that moves beyond simple factual retrieval to active scientific reasoning. By integrating LLM-based hypothesis generation with literature grounding, physical simulations, formal methods, and experimental feedback, we demonstrate how AI can accelerate the scientific method. We examine climate science as a challenging case study, highlighting where data abundance necessitates sophisticated reasoning systems. Our central claim is that the primary bottleneck for AI -driven discovery is not hypothesis generation, but the development of scalable, domain -aware verification loops that connect imaginative proposals to rigorous testing.
Jacob Tsimerman: AI Safety and Mathematics
What does it mean for an AI system to be safe? There are of course deep questions and complicated coordination problems regarding (cyber)security and policy. But even if we get that right, it turns out to be extremely complicated to design and direct a super intelligent system such that it acts in ways we would consider ”safe” or ”good,” or even just to avoid catastrophic consequences. The plan is to chat about this problem, why it’s hard, and why mathematics might play a big role in attempting to solve it.
Jeffrey David Ullman: AI and Industrial Revolution
There is a reasonable comparison between what happened 200+ years ago during the industrial revolution and what is happening today with the ”AI Revolution”. Is it possible to get the advantages without the disadvantages this time?
