CultureMore in Culture→
AI’s Scientific Reach Expands Beyond Mathematics
Key Takeaways
- MIT-linked researchers tested frontier AI models on theoretical problems with later-published solutions.
- The best systems rose from about 2% to about 35% correct between mid-2024 and mid-2026.
- The study was limited to theory questions not requiring simulations or experiments.
DE
DT Editorial Team··via wired.com
