Paul Riechers: Understanding AI, Unlocking Optionality | Intelligence at the Frontier 2026

Sat, May 2, 2026, 12:00 AM

We are building systems we don't understand. Paul Riechers thinks that's the core problem, and that it's fixable. Paul, co-founder of Simplex and research lead at the Astera Institute, brings a physics background to AI safety. His central argument: without a mathematical understanding of what intelligence actually is, the debate between AI optimists and pessimists has no ground to stand on. Everyone has strong feelings. Nobody has a shared truth to point to. His research finds that transformer models develop internal world models with universal geometric structure, the same representations emerging across different architectures. That's evidence that interpretability is possible and that we can eventually understand what these systems are doing, not just observe what they produce. Speaker: Paul Riechers (Simplex / Astera Institute) This session was part of Intelligence at the Frontier, held March 14–15, 2026, at Frontier Tower in San Francisco. Learn more at https://www.fundingthecommons.io