OpenAI releases 722 math manuscripts from internal model, ~3 hours of compute per result
ZeroStateReflex · x · 2026-10-07
OpenAI has dumped 722 mathematical manuscripts produced by an unreleased internal model: 372 families of results from roughly 4,000 attempted research problems, with each successful result reportedly taking only about three hours of ChatGPT Pro-equivalent thinking compute.
The collection touches on topics including the irrationality of π, the Riemann zeta function, the Hodge conjecture, spin glasses, quantum magnetism, free group factors, and relativistic Vlasov-Maxwell theory. The sharer's framing: in 2023 AI made mistakes that made you question whether it understood the question; three years later it produces mathematics at the frontier of human knowledge. If this rate continues, consequences go far beyond chatbots — math feeds physics, physics feeds engineering, and AI research eventually feeds back into better AI. Scientific curiosity is becoming parallelizable.
More from AGI Musings
- Investor: millennials and Gen Z will build agentic platforms to kill spam for good — StewartalsopIII · 2026-11-22
- AI safety drama: doxxing-private-messages fight reignites ethics clash over means and ends — teortaxesTex · 2026-11-22
- Fleuret: treat AI as oracles, not explainers — human-comprehensible math will be negligible — francoisfleuret · 2026-11-22
- Writer Uses Claude to Pen a Culture Short Story — and Admits It's Obviously AI-Written — NathanpmYoung · 2026-11-22
- When Creating Life Costs Less Than Caring for It: An AI Biotech Question — PierceLilholt · 2026-11-22
- Closed models, opaque processes: will AI-made discoveries only benefit an elite? — GeorgiaChal · 2026-11-22