Skip to content
Understanding AI· Dan Kagan-Kans·Indexed by this site · Originally published· Opinion →· Its place in the scan record →

I interviewed seven physicists who are grappling with the AI revolution

I spoke to seven physics experts grappling with the AI revolution

AI summary

Understanding AI author Dan Kagan-Kans interviewed seven theoretical physicists, documenting AI's impact on the field. Physicist Suvrat Raju and others built the TheoPhysEval benchmark using questions never previously made public; GPT-6 Astra actually answered 49 of 50 questions correctly, at an average compute cost of less than $3 per question, and the benchmark was saturated before release. A Turkish physicist posted on X that Claude Opus 5.5 spent 10 hours reproducing the core of his doctoral thesis, which had taken him a year and 500 pages of hand calculations; submissions to the hep-th category on arXiv from January to October rose about 30% year over year, far exceeding the previous two-decade maximum annual increase of 8%. Several interviewees worry that young physicists will lose the path of building intuition and reputation through calculation; Raju said he will not take on doctoral students until the outlook is clearer; in October, OpenAI announced that its model proved a mathematical proposition about pure Yang-Mills, and a Lean version of the 1147-page proof was reviewed by an expert panel and deemed correct.

Related papers
Read originalunderstandingai.org