Amateur Solves 60-Year Math Problem With One AI Prompt

A Single Prompt Changed Mathematical History

A 23-year-old student has done what elite mathematicians could not accomplish in six decades — and he did it by asking an AI chatbot a single question.

Liam Price, an amateur mathematician, used OpenAI's GPT-5.4 Pro to solve a longstanding conjecture from the legendary Paul Erdős, one of the most prolific mathematicians in history. The solution, which concerns special number-theoretic structures, was posted on erdosproblems.com, a website dedicated to tracking the hundreds of unsolved problems Erdős left behind before his death in 1996.

As reported by Scientific American, the AI didn't just assist Price — it generated a complete proof using a method that no human mathematician had previously developed.

What Makes This Breakthrough Extraordinary

Erdős problems are notoriously difficult. They represent some of the deepest open questions in combinatorics, number theory, and graph theory. Many carry cash bounties that Erdős himself famously attached to his hardest challenges, and they have resisted the efforts of Fields Medal winners and entire research teams for generations.

What sets Price's achievement apart isn't just the result — it's the process. According to the report, Price is not a professional mathematician. He is a 23-year-old student who crafted a well-structured prompt and fed it into GPT-5.4 Pro, OpenAI's most capable reasoning model. The AI returned a novel proof that introduced a method previously unknown to the mathematical community.

The fact that a single prompt — rather than months or years of dedicated research — yielded a valid solution raises profound questions about the future of mathematical discovery and the role of human expertise.

AI as a Mathematical Collaborator — or Replacement?

The result lands in the middle of an intensifying debate within academia: are large language models genuine reasoning engines, or are they sophisticated pattern matchers that occasionally stumble onto correct answers?

Skeptics have long argued that LLMs lack true mathematical understanding. They point to well-documented failures where models confidently produce incorrect proofs or hallucinate nonexistent theorems. Yet Price's result complicates that narrative significantly. GPT-5.4 Pro didn't merely verify an existing approach — it invented a new one.

'This is not the AI regurgitating something from its training data,' the implication of the discovery suggests. The proof method appears to be genuinely novel, meaning the model engaged in something that at least resembles creative mathematical reasoning.

For the professional mathematics community, this development is both exciting and unsettling. If an amateur with the right prompt can outperform decades of expert effort, it challenges fundamental assumptions about how mathematical progress works.

The Erdős Legacy Meets the AI Era

Paul Erdős published more than 1,500 papers during his lifetime and posed hundreds of open problems, many of which remain unsolved. The erdosproblems.com website currently tracks these challenges, and the community around it has traditionally consisted of professional researchers chipping away at these questions using classical techniques.

Price's AI-assisted solution marks a turning point for this community. It suggests that LLMs could become powerful tools for attacking the remaining Erdős problems — and potentially open problems across all of mathematics.

This isn't the first time AI has made headlines in mathematics. DeepMind's AlphaProof and AlphaGeometry systems demonstrated strong performance on International Mathematical Olympiad problems in 2024. Terence Tao, widely regarded as one of the greatest living mathematicians, has spoken publicly about using LLMs as 'co-pilots' for research. But those efforts typically involved world-class researchers guiding AI systems through carefully designed workflows.

Price's case is different: a non-expert, a single prompt, and a problem that defeated the best minds in the field for 60 years.

What This Means for the Future of Research

The implications extend well beyond mathematics. If LLMs can generate novel proof strategies for problems that have resisted human efforts for decades, similar breakthroughs could emerge in theoretical physics, computer science, and other formal disciplines.

However, experts urge caution. A single spectacular result does not mean AI can reliably solve hard open problems on demand. LLMs still struggle with consistency, and the mathematical community will need to rigorously verify Price's proof before it is fully accepted.

There are also thorny questions about credit and authorship. Did Price solve the problem, or did GPT-5.4 Pro? How should the mathematical community attribute discoveries that emerge from human-AI collaboration — especially when the human contribution is limited to formulating a prompt?

The Bigger Picture

Regardless of how these philosophical questions are resolved, one thing is clear: the barrier to entry for serious mathematical research just dropped dramatically. Tools like GPT-5.4 Pro are democratizing access to capabilities that were previously reserved for a tiny elite.

For OpenAI, Price's result serves as a powerful demonstration of its latest model's reasoning capabilities, arriving at a time when the company faces fierce competition from Anthropic, Google DeepMind, and open-source alternatives.

For the rest of us, it's a vivid reminder that the AI revolution isn't just about chatbots and image generators. It's beginning to reshape the frontiers of human knowledge itself.