Alexandre Eremenko on MathOverflow: ‘In August 2026, at least 7’ of his ~40 collected open problems ‘were solved using AI’

Answer (score 50 when read on 13 Oct 2026) to the MathOverflow question “Soft question: an era in my research terminated by AI (which is, in a way, not so bad)”, asked on 24 Sept 2026. The exact time of the answer was not captured; it carries an addendum dated 26 Sept, so the date above is the question’s.

Eremenko writes that he had collected about 40 problems he “thought about for long time and could not solve” on his web page (https://www.math.purdue.edu/~eremenko/uns1.html). “In August 2026, at least 7 of them were solved using AI. Some of these solutions are truly exciting, and contain completely new ideas.” “I feel like I am in a fairy tale, or in a dream, where I meet an Oracle who can answer all my questions which tortured me during large part of my life”. He adds that he is a retired professor and that “young graduate students may have some other feelings. Some of them contemplate switching from mathematics to something else”.

He names two things “AI cannot do (yet)”: “Inventing new problems which are interesting to humans” and teaching, though “it is possible that teaching will be also taken over in the near future”. “The whole structure of mathematical research as a human enterprise will have to change soon (journal publications, hiring, promotion etc.).” In the 26 Sept addendum a friend “asked his AI to invent an interesting and challenging problem”; Eremenko judges the result (https://www.math.purdue.edu/~eremenko/dvi/ai-problem.pdf) “of interest to specialists”.

Which seven problems were solved is not listed in the answer; his problem page was not read for this file. One AI-assisted result with Eremenko as co-author, a counterexample to Fuchs’s conjecture (arXiv 2609.28443), is in our summer catalogue.