Recently, the claims by OpenAI's GPT-5 in the field of mathematics have sparked widespread controversy. Yann LeCun, Meta's chief AI scientist, described the incident as "self-inflicted," while Demis Hassabis, CEO of Google DeepMind, said it was "embarrassing."
Image source note: The image is AI-generated, and the image licensing service is Midjourney
The incident began when Kevin Weil, OpenAI's vice president, claimed in a deleted tweet that GPT-5 had solved 10 previously unsolved Erdős problems and made progress on 11 other problems. Erdős problems are famous conjectures proposed by the renowned mathematician Paul Erdős. However, Thomas Bloom, the mathematician who maintains the Erdős problems website, refuted this, stating that although these problems are listed as "open" on his website, it does not mean that GPT-5 has truly solved them. He pointed out that the so-called "solutions" were actually references that GPT-5 found, which he was unaware of, and these references contained the solutions.
Subsequently, OpenAI researcher Sebastien Bubeck also admitted that GPT-5 only found existing literature with solutions, but he still considered this an achievement worth acknowledging, as the difficulty of literature search should not be underestimated.
This incident has sparked heated discussions in the tech community, with many questioning the true capabilities of AI in the field of mathematics. Although GPT-5 demonstrated some ability in literature searching, the claim of solving unsolved problems was clearly seen as exaggerated.
Key points:
🔍 GPT-5's mathematical achievements are being questioned, and the tech community has reacted strongly.
📄 The vice president of OpenAI claimed that GPT-5 solved 10 Erdős problems, but this was criticized as exaggerated.
🧩 Experts say GPT-5 only found literature with solutions, but did not truly solve the unsolved problems.
