🧠 AI just helped crack a 25-year-old math problem

🧠 AI just helped crack a 25-year-old math problem A Microsoft researcher revisited a theoretical problem in wireless communications that had been open since the early 2000s. The challenge: how can a receiver recover the exact original message from a noisy signal when dozens of signals are interfering with each other? GPT-5.6 reportedly produced an initial proof in around 30 minutes. 🤯 Claude Fable 5 then independently suggested a different approach, and the two models were used to challenge, simplify and improve the arguments. The final proof uses a surprisingly simple idea: make an initial guess from the noisy signal, then repeatedly make small corrections until the exact message is recovered. What makes this important is that the new approach reaches the same fundamental recovery limit as the theoretically optimal method, but without its enormous computational cost. AI didn’t invent an entirely new branch of mathematics, it did something arguably just as interesting: it connected existing ideas in a way humans had missed for 25 years. @aipost 🏴