AI out-remembers mathematicians
Archived — this story has rotated out of today’s deck. It is kept here in full.
The gist
AI systems solved math problems by using vast memory, not superior reasoning.
This reframes AI's edge as memory, not intelligence, affecting how we value human mathematicians.
Background
Recent AI achievements in mathematics, such as disproving the Jacobian conjecture and cracking an Erdős conjecture, have sparked debate about whether AI is truly reasoning or simply leveraging its enormous working memory. An essay by Davide Piffer argues that AI's advantage lies in its virtually unlimited symbolic working memory, which removes the biological limits that constrain human mathematicians. This perspective has gained traction on social media and tech forums, prompting discussions about the future role of human mathematicians.
How it unfolded
- May 2026OpenAI's internal model disproved the Erdős unit distance conjecture, which had eluded mathematicians for nearly a century.
- Jul 2026Harvard mathematician Levent Alpöge tweeted that Anthropic's Claude Fable 5 disproved the Jacobian conjecture with a tiny counterexample.
- Aug 11, 2026The Verge published an article titled 'The AI takeover of mathematics has begun,' discussing the implications of AI solving major math problems.
- Aug 13, 2026Noah Smith shared Davide Piffer's essay on X, arguing that superintelligence may be about working memory, not reasoning.
- Aug 15, 2026The essay was submitted to Hacker News, sparking further discussion.
Who’s saying what
- Analysts
- AI's success in mathematics may be due to its vast working memory, not superior reasoning, as argued by Davide Piffer.
- Public
- Some users on X relate to the memory limitation, with one commenter saying they have 'the memory of a fruit fly' and struggled with math.
Still unverified
The claim that AI's advantage is primarily due to working memory rather than reasoning is an opinion by Davide Piffer, not yet empirically proven. The specific details of AI models solving the conjectures are based on reports and tweets, not peer-reviewed publications.