Meet a mathematician solving previously unsolvable math problems with GPT-5.6

TL;DR: A mathematician who started coding at age five used Codex 5.6 to overturn a conjecture on an algebraic surface problem that he had spent three years trying to prove. He discovered that the model can automatically spawn sub-agents to handle heavy computation, letting him focus on his favorite challenges and life. ## From Garden Walks to Code Breakthroughs When I get stuck on a math problem, I like to take a break and go for a walk in the garden. Completely lost in my own thoughts. When I come back, I let Codex handle the exploration and the technical work—otherwise those tasks would take me weeks. I started writing computer code when I was about five years old. I can barely remember a time when I couldn't program—it seems to have always been in my blood. ## An Algebraic Surface Problem Unsolved for Three Years For the past three years, we've been working on an extremely difficult problem about algebraic surfaces. We tried programming, we tried pen-and-paper derivations, we tried previous models—none worked very well. So I decided, "Alright, I'll try the new Codex 5.6." ## A Fresh Approach from Codex 5.6 Codex actually came up with a completely new approach. It helped us overturn a conjecture that we had spent three years trying to prove. The key number appeared in the derivation: 14/5 is actually larger than 8/3, so the conjecture was essentially wrong. That discovery was exhilarating, but that's why we do science, and we had a lot of fun doing it. ## Natural User Experience and Automatic Sub-Agents Using 5.6 feels natural: you set a task, the model recognizes that a lot of computation is happening, and automatically spawns sub-agents—without me even having to ask. ## AI Empowerment: Focusing on What Truly Matters I think all these AI tools are meant to empower people. Now I can focus on what I love most—difficult problems in math, spending time with my family, playing with my kids, and mowing the lawn. If you have the courage to try something really ambitious, you won't be afraid of the sheer scale of computation, because you can use the model to organize it. I think we're going to have a lot of fun and make a lot of discoveries along the way. **Source:** YouTube video link (https://www.youtube.com/watch?v=5X5CALvYvp0)