GPT-5 just did real math nobody had solved before

Turns out GPT-5 isn’t just spitting out essays — it actually cracked an open problem in probability theory.

‍

• Mathematicians knew that certain random sums eventually behave like a normal distribution. The mystery? How fast does that convergence happen.

• Nobody had nailed down a precise answer — until GPT-5, nudged along by experts, filled the gap.

• It delivered the first clean quantitative rate for convergence, not just for Gaussian cases but also Poisson ones.

‍

Of course, it wasn’t perfect — GPT-5 made plenty of mistakes. But with feedback, it corrected itself, proved the result, and even drafted a paper-level writeup. GPT-5 acted less like a calculator and more like a junior researcher.