ChatGPT’s math skills have been criticized by experts

A statement by OpenAI researchers about the alleged “mathematical triumph” of GPT-5 has sparked a heated debate online as it turned out to be untrue. Earlier on Social Network X, manager Kevin Wale claimed that the model had “identified solutions to ten unsolved Erdes puzzles” and had made progress on eleven others.

Another claim by OpenAI researchers was that the model had “solved ten unsolved Erdes puzzles” and had made progress on eleven more.

This publication quickly went viral, but was removed after criticism from academics and industry experts. It turned out that GPT-5 was unable to solve the unsolved math problems on its own.

Mathematician, Thomas Bloom, creator of erdosproblems.com, clarified that by “revealed” he meant only the lack of a solution in its database, not the scientific field as a whole. In fact, the AI simply discovered published scientific papers that Bloom was not aware of.

The head of DeepMind, Demis Hassabis, characterized the episode as a “disgrace,” while Meta** AI director Yann LeCun added that OpenAI had “succumbed to its own hype.” After a flurry of criticism, OpenAI admitted its mistake and removed the original statements.

This incident resonated in the scientific community for another reason. How could it be that leading researchers made such a high-profile claim without checking its factual basis? A mathematician, Terence Tao, stressed that the true value of AI to science is not in finding solutions to unsolvable problems, but in optimizing everyday processes and systematizing knowledge.

It is not about finding solutions to unsolvable problems, but about optimizing everyday processes and systematizing knowledge.

*Meta* has been recognized as an extremist organization in Russia and banned

*Related to Meta, it has been recognized as an extremist organization in Russia and its activities are banned