Mathematicians say ChatGPT solves 60-year-old Erdős problem
Updated
Updated · Futurism · May 2
Mathematicians say ChatGPT solves 60-year-old Erdős problem
9 articles · Updated · Futurism · May 2
Liam Price, 23, posted the GPT-5.4-assisted proof on erdosproblems.com, and experts including UCLA's Terence Tao and Stanford's Jared Lichtman said it appears genuine.
They said the AI used a known formula in an unexpected way after human researchers had made the same early misstep, though experts still had to reconstruct and verify the poor raw output.
Tao called it a new way to think about large numbers but said its long-term significance remains unclear, after earlier AI claims on Erdős problems proved overstated or relied on existing solutions.
What risks might arise if flawed or unverifiable AI-generated proofs become widespread in mathematics and beyond?
Could AI breakthroughs like the Erdős problem solution signal the end of human-dominated mathematical discovery, or will collaboration always be necessary?
Erdős Problem #1196 Solved by ChatGPT-5.4 Pro and 23-Year-Old Amateur, Confirming Primitive Set Sum Bound
Overview
In early 2026, 23-year-old amateur mathematician Liam Price solved the 60-year-old Erdős Problem #1196 using ChatGPT-5.4 Pro in an 80-minute session. Price’s intuitive prompting helped the AI bypass traditional human thinking traps, enabling it to generate a novel proof that combines number theory, probability, and fluid dynamics. Leading mathematicians Terence Tao and Jared Lichtman validated and refined this elegant 'Book Proof' and began formalizing it with the Lean proof assistant. This breakthrough not only offers a new way to understand integers but also signals a shift where AI acts as a conceptual guide, accelerating discovery and targeting major open problems in mathematics.