Particle.news
Download on the App Store
5 ARTICLES
·
9h ago

OpenAI Retracts GPT-5 Erdős Breakthrough Claims After Backlash

Reporting now indicates GPT-5 surfaced prior solutions in the literature rather than original proofs.

Mathematician Terence Tao points out that current AI systems often miss fundamental mistakes in mathematical proofs that humans would intuitively catch, highlighting their lack of a "mathematical sense of smell." AI struggles particularly when pursuing incorrect solution paths, as it does not possess the human instinct to judge which approaches are promising or flawed. Systems like AlphaZero suggest that combining neural networks with symbolic search can help AI develop forms of strategic intuition, which some experts see as a step toward more human-like mathematical reasoning.
Image
Image
Recent AI Success for Thorough Advanced Research

Overview

  • In a deleted post, OpenAI VP Kevin Weil said GPT-5 had solved ten Erdős problems and made progress on eleven more.
  • Thomas Bloom, who runs ErdosProblems.com, said the items marked open reflected his awareness rather than the state of the field.
  • OpenAI researchers deleted or revised posts and acknowledged the model located existing papers instead of generating new proofs.
  • Google DeepMind’s Demis Hassabis called the episode embarrassing and Meta’s Yann LeCun mocked the overstatement.
  • Mathematicians including Terence Tao framed the takeaway as AI’s usefulness for difficult literature search rather than autonomous discovery.