Erdos Problem 1196: Can AI now solve maths that no human can?

Episode Summary

Executive Summary: The episode examines how AI unexpectedly produced a workable solution to Erdős Problem 1196, a long-standing pure mathematics puzzle about primitive sets. It profiles mathematician Jared Duker Lichtman, who had studied the problem for years, and AI user Liam Price, whose prompt to ChatGPT yielded a candidate proof verified by another model and accepted by Jared as correct. The segment argues that AI can accelerate discovery, but mathematicians remain essential to interpret and validate results.

Main Topics: AI and the surprise solving of a pure math problem (Priority: 5/5): The central story is AI's apparent solution to Erdős Problem 1196, surprising mathematicians because it had resisted human effort for decades. What Erdős problems are and why they matter (Priority: 4/5): The episode explains that Erdős problems are a large collection of difficult mathematical questions, many still open, and part of a broader ecosystem of famous unsolved problems. Primitive sets as the mathematical setting (Priority: 4/5): The transcript briefly defines primitive sets and frames Problem 1196 as asking how a score grows when all numbers in the set are larger than x as x approaches infinity. Jared Duker Lichtman's long engagement with the problem (Priority: 4/5): Lichtman describes years of personal fascination and work on related Erdős problems, showing the depth of human investment before AI entered the picture. How Liam Price used AI to generate and verify a proof (Priority: 5/5): A 23-year-old amateur mathematician used an OpenAI model, then another model for verification, before sending the solution to experts for review. What AI means for mathematical work (Priority: 5/5): The episode argues that AI can be a powerful collaborator and intuition engine, but human mathematicians are still needed to understand, judge, and use the output.

Key Arguments: AI can generate candidate proofs for hard mathematical problems much faster than humans in some cases, as shown by the 80-minute solution attempt for Problem 1196. A proof is only meaningful in mathematics if experts can inspect and verify it; AI output still needed Jared Duker Lichtman's judgment. The success does not eliminate the need for mathematicians; rather, it shifts their role toward interpretation, validation, and follow-on work. Mathematics differs from fields like journalism or photography because the general public cannot independently assess the correctness of AI-generated results. AI is already functioning like a collaborator or trusted colleague, producing ideas that can bear fruit when guided by human expertise. Even if AI solves some open problems, the deeper mathematical ecosystem of understanding, proof, and extension still depends on human researchers.

Data Points: Erdős problems: 1200 - The website launched in 2023 collected roughly 1,200 Erdős problems for mathematicians to track and solve. AI attempt time: about 80 minutes - Liam Price says the OpenAI model produced a candidate solution after roughly 80 minutes of thinking. Human work on Problem 1196: 7 years - Jared Duker Lichtman had been thinking about Erdős Problem 1196 for seven years before the AI-generated solution arrived. Human work on related problem 164: 4 years - Lichtman says he solved Erdős Problem 164 after four years of persistent work. Fermat's last theorem resolution: 1994 - Katie Steckles notes that the theorem was resolved in 1994, centuries after Fermat's original note. AI verification: another model found no errors - Liam Price reports that a second AI model reviewed the proof and could not find errors in the argument.

Pivotal Quotes: "I received a message saying that this essentially amateur mathematician had run GPT 5.4 Pro on this problem and received output that he thought could be a candidate solution to this problem." — Jared Duker Lichtman: Jared describes first learning that AI may have solved the problem he had studied for years. "It was pretty clear that it was correct. And the idea was very nice." — Jared Duker Lichtman: His assessment after reading the AI-generated proof and concluding it was valid. "At the moment, right now in 2026, we're at a point where we're kind of at a collaborator kind of feedback." — Jared Duker Lichtman: He characterizes AI as a collaborative tool rather than a replacement for mathematicians.

Implications: AI may increasingly assist pure mathematics by generating proofs and insights, but human experts will still be needed to verify, contextualize, and extend them. The role of mathematicians may shift from sole problem-solvers to collaborators and evaluators.