MIT study finds AI solved 35% of theory problems by mid-2026
A new study from MIT’s Neil Thompson and collaborators shows AI models now solve about 35% of theoretical problems in probability, statistics, economics, physics, and cosmology—up from 2% in mid-2024. The research tested OpenAI’s models (o1, o3, GPT-5 to GPT-5.5) on problems published after their training cutoffs, blocking internet access to ensure solutions were generated independently. The…
Key points
- MIT study finds OpenAI models solved 35% of theory problems by mid-2026, up from 2% in mid-2024
- Researchers tested models on problems published after their training cutoffs, with no internet access
- Thompson warns AI may accelerate discovery but risks undermining human verification and expertise
The study follows OpenAI’s controversial claim of solving the Navier-Stokes equations—a grand challenge in fluid dynamics—with an unreleased model. Critics, including a NYU mathematician, accused OpenAI of ‘gazumping’ collaborators at Anthropic. Since then, AI labs have announced hundreds of solved math problems, prompting calls for limits on AI use in research. Thompson’s team emphasizes AI won’t replace experts but may accelerate theoretical work, while noting models can still produce errors. Labs like Anthropic and OpenAI are now investing heavily in AI-driven scientific research, blending theoretical analysis with experimental capabilities.
The headline, key points and digest above were generated by Digest AI's editorial model from the linked sources. Automated summaries can contain errors: the sources are the record. Spotted a mistake? Tell us. Published by Martin K., who runs Digest AI and handles corrections.
More in Research
All →- Google engineer says self-improving AI remains far from reality · 1 src
- Anthropic warns small AI models may self-replicate on IoT devices by end of 2026 · 22 src
- Meta publishes ten health studies generated with its Muse AI platform · 1 src
- Researchers release Mr.Chatterbox-2.7B-Chat model trained on 19th-century British texts · 1 src
- UW researchers design AI-built proteases targeting disease proteins · 2 src
Comments
via GitHub Discussions