Amateur armed with ChatGPT solves an Erdős problem
scientificamerican.com
April 25, 2026
4 min read
🔥🔥🔥🔥🔥
72/100
Summary
Liam Price, a 23-year-old without advanced mathematics training, solved a 60-year-old problem previously unsolved by mathematicians using ChatGPT Pro. Recent advancements in artificial intelligence have led to solutions for several conjectures known as Erdős problems.
Key Takeaways
- Liam Price, a 23-year-old without advanced mathematics training, solved a 60-year-old mathematical problem using ChatGPT Pro.
- The solution involves special sets of whole numbers called "primitive sets," where no number can be evenly divided by another.
- Price's method for solving the problem appears to be novel and may have broader applications in mathematics.
- Experts caution that many AI-generated solutions to Erdős problems have been less original than they seem, highlighting the complexity of evaluating AI's mathematical capabilities.
Coverage & discussions
Hacker News · scientificamerican.com
Community discussions
Community Sentiment
Positives
- The ability of LLMs to make novel connections across different fields is a significant advancement, showcasing their potential to contribute to scientific breakthroughs.
- The successful proof of an Erdős problem by an amateur using ChatGPT indicates that AI models are becoming increasingly capable and sophisticated in tackling complex mathematical challenges.
- This milestone demonstrates that LLMs are not only improving in performance but also revealing new insights and connections that were previously overlooked by human mathematicians.
- The collaborative idea of creating a GitHub repository for unsolved problems highlights the potential for community-driven AI applications to accelerate mathematical discoveries.
Concerns
- Some users experienced limitations with the free version of ChatGPT, indicating that access to advanced features may hinder broader participation in solving complex problems.
- Concerns were raised about the depth of understanding required to evaluate the proofs generated by LLMs, suggesting that reliance on AI may not fully substitute for human expertise in mathematics.
