Viqus Logo Viqus Logo
Home
Categories
Language Models Generative Imagery Hardware & Chips Business & Funding Ethics & Society Science & Robotics
Resources
AI Glossary Academy CLI Tool Labs
About Contact

AI Breakthrough: Large Language Models Tackle Thousands of Unsolved Erdős Problems

AI Large Language Models Mathematics OpenAI ChatGPT Erdős Problems Formalization AI Tools
January 14, 2026
Viqus Verdict Logo Viqus Verdict Logo 9
Accelerated Discovery
Media Hype 8/10
Real Impact 9/10

Article Summary

OpenAI’s GPT 5.2 is causing a significant shift in the perception of AI’s capabilities in mathematics. Initially focused on problem-solving, the model has recently achieved a notable success in tackling over one thousand unsolved conjectures by the Hungarian mathematician Paul Erdős. This is largely due to the sheer volume of solved problems, now exceeding 15, after the release of GPT 5.2, which is considered ‘anecdotally more skilled at mathematical reasoning’ than previous iterations. The model isn’t just identifying solutions; it’s formalizing them with tools like Harmonic, even engaging with established mathematicians like Terence Tao and Tudor Achim, founders of Harmonic, who are now seeing these tools embraced by leading academics. The progress is driven by a combination of factors including formalization— a laborious process now aided by AI — and the model’s ability to leverage information from sources like Math Overflow and, crucially, to recognize patterns within the vast collection of Erdős problems. This isn’t merely about speed; it’s a fundamental reassessment of how AI can contribute to human knowledge creation.

Key Points

  • Large language models, specifically GPT 5.2, are successfully solving complex mathematical problems like the Erdős conjectures.
  • The increased number of solved problems— exceeding 15— demonstrates a significant advancement in AI's mathematical reasoning abilities.
  • Mathematicians and researchers are increasingly utilizing AI tools, such as Harmonic and ChatGPT, to formalize and verify mathematical solutions.

Why It Matters

This development has profound implications for the field of mathematics and the broader role of AI. It challenges the long-held notion of human exceptionalism in complex problem-solving, suggesting that AI could become a powerful collaborator, especially in tackling ‘long tail’ problems – those considered obscure or difficult by humans. The increased usage of formalization tools further enhances the trustworthiness and reproducibility of AI-generated results, potentially accelerating scientific progress. This isn’t just about a clever algorithm; it’s about a potential paradigm shift in how we approach intellectual challenges, raising critical questions about intellectual property, authorship, and the very definition of ‘knowledge’.

You might also be interested in