Kevin Weil of OpenAI posted on X that GPT-5 had "found solutions to 10 Erdős problems that were previously unsolved." He also stated that progress had been made on an additional 11 problems.
However, mathematician Thomas Bloom immediately countered this claim. He explained that the label "open" on his website simply meant that he personally did not know the solution, not that the problems themselves were unsolved. GPT-5 had merely presented existing research that Bloom had overlooked.
DeepMind CEO Demis Hassabis called the incident "embarrassing." Yann LeCun of Meta AI also pointed out that OpenAI had fallen victim to its own hype.
The original post has been deleted, and the researchers have acknowledged the error. This incident has strengthened the impression that OpenAI is acting carelessly under pressure.
In reality, GPT-5 proved useful as a research tool for tasks such as searching for related papers. Mathematician Terence Tao views the greatest value of AI not in solving difficult problems, but in accelerating tedious tasks such as literature searches.
Source: OpenAI researcher announced GPT-5 math breakthrough that never happened（HN 430pt・234コメント） (HN Search (backfill), 2025-10-19)