Google SLAMS OpenAI's GPT-5: This Is EMBARASSING!

OpenAI researchers' celebratory claim that GPT-5 solved previously 'unsolved' mathematical problems was debunked as merely a sophisticated literature search, leading to public embarrassment and highlighting AI hype.

• OpenAI's GPT-5 did not solve new math problems, but rather found existing solutions through advanced literature search.

• The incident highlights the dangers of AI hype, where impressive information retrieval is conflated with revolutionary breakthroughs.

• Skepticism and consultation with domain experts are crucial to accurately evaluate AI advancements and avoid premature, exaggerated claims.

OpenAI faced public embarrassment after its researchers claimed GPT-5 solved numerous 'open' Erdos mathematical problems, sparking celebration and recruitment announcements. However, Thomas Bloom, the maintainer of ErdosProblems.com, clarified that GPT-5 merely found existing solutions that he was unaware of, not generating new proofs. This incident underscores the current AI hype cycle, where companies often overstate capabilities, confusing impressive information retrieval with groundbreaking problem-solving.

OpenAI's Misleading Claims

• 00:00:26 OpenAI researcher Mark Selk announced that GPT-5, with thousands of queries, found solutions to 10 'open' Erdos problems and made significant progress on 11 others, even claiming an error in Erdos's original paper. This led to celebratory tweets from other OpenAI researchers like Sebastian Bubeck, who framed it as a scientific acceleration and a recruitment opportunity, touting AI's ability to solve problems that had stumped mathematicians for years.

• 00:03:07 Demis Hassabis, head of Google DeepMind and OpenAI's main competitor, publicly dismissed OpenAI's claims with a terse 'This is embarrassing.' Thomas Bloom, who owns and maintains ErdosProblems.com, further clarified that the problems were not 'unsolved' by humanity, but rather that he, as the database maintainer, was personally unaware of existing papers that had already solved them. GPT-5 performed an advanced literature search, retrieving these existing solutions, not creating new mathematical proofs.

• 00:06:02 OpenAI's actions are considered embarrassing for several reasons, including professional AI researchers misinterpreting basic research principles like literature reviews and prematurely making public announcements on Twitter. The company used the false claim as a recruitment tool, was publicly called out by a competitor's CEO, and failed to verify information with the accessible domain expert, Thomas Bloom, before making grand claims.

• 00:07:29 This incident exemplifies the pervasive hype cycle in the AI world, where companies often overstate AI capabilities in a race to appear most groundbreaking. While GPT-5's ability to search vast academic literature and retrieve relevant papers is genuinely useful for researchers, there is a critical distinction between finding existing solutions and achieving revolutionary breakthroughs by solving previously unsolved problems. Skepticism is essential, and domain experts' insights are crucial to discerning actual progress from inflated claims.