Frontier models are felling decades-old open problems across math, physics and chemistry
IgorCarron · x · 2026-10-15
Igor Carron has been collecting cases where frontier models (Claude, ChatGPT) helped or solved problems humans had been stuck on for decades, and compiled the full map.
- Broad spread: mathematics, physics, chemistry, chip design, radio, cryptography, accounting, literature — no domain seems out of reach
- Math is ground zero: the 59-year-old Sylvester conjecture fell, Mathieu's conjecture resolved, Erdős Problem 579 formally verified in Lean, Erdős–Sós conjecture proven, KLS conjecture, and more
- Core observation: what struck him wasn't any single result but the spread and who's solving — no PhD increasingly required, though taste still matters; semi-experts can now join in
Related event: Blogger Rounds Up Frontier Models Cracking Decades-Old Unsolved Problems(2 posts)→
More from AGI Musings
- Jeff Ladish: nearly all my hope routes through coordinating to stop uncontrolled intelligence explosion — JeffLadish · 2026-10-15
- OpenAI's Quasi-Riemann proof sparks chain of results, Bhargava proves primitive root theorem — Singularitarian · 2026-10-15
- Alleged 2021 Anthropic doc shows 1T-parameter models, RL, scaling laws all planned early — akbirthko · 2026-10-15
- Google Workshop Report Maps Agentic Privacy and Security Open Problems with 50+ Experts — dl_weekly · 2026-10-15
- Katja Grace: Are Altman, Musk and Amodei bringing their A game to AI safety? — NathanpmYoung · 2026-10-15
- ML researcher Andrew G. Wilson: open letters are 'worse than nothing' — andrewgwils · 2026-10-15