An Amateur Just Solved a 60-Year-Old Math Problem - by Asking AI

Why It Matters

The result demonstrates that large language models can generate original mathematical insights, potentially reshaping research workflows and accelerating discovery in pure mathematics. It also raises questions about verification and credit in AI‑assisted breakthroughs.

Key Takeaways

• ChatGPT-5.4 Pro solved a 60‑year‑old Erdős conjecture
• 23‑year‑old Liam Price prompted the solution with a single query
• Solution concerns primitive sets where no element divides another
• Experts see AI as a new collaborator for open mathematical problems

Pulse Analysis

Artificial intelligence has moved beyond data analysis into the realm of creative problem‑solving, and large language models like GPT‑5.4 Pro are leading the charge. By interpreting natural‑language prompts and accessing vast mathematical corpora, these systems can synthesize novel arguments that rival human ingenuity. Recent months have seen a surge of AI‑driven proofs, signaling a shift where researchers treat models as research partners rather than mere tools, accelerating hypothesis generation across disciplines.

The breakthrough centered on an Erdős‑type problem involving primitive sets—collections of integers where no member divides another. For six decades, mathematicians have struggled to determine the maximal density such sets can achieve, a question that sits at the intersection of number theory and combinatorics. Price’s single prompt coaxed the model into constructing a proof that not only met the conjecture’s criteria but introduced a previously unseen combinatorial technique. The solution, posted on erdosproblems.com, was vetted by experts who confirmed its validity, underscoring the model’s capacity to navigate deep, abstract reasoning.

The implications are profound for both academia and industry. Universities may integrate AI assistants into graduate curricula, allowing students to explore conjectures faster while honing verification skills. Publishers and funding bodies will need new standards for attributing credit when AI contributes substantively to a result. Moreover, the episode fuels debate about the reliability of machine‑generated mathematics, prompting calls for rigorous peer review pipelines. As AI continues to mature, its partnership with human mathematicians could unlock solutions to problems once deemed intractable, reshaping the future of mathematical research.