A Free Chatbot Solved A Decade-Old Math Problem In 13 Minutes. The Speed Is The Story, Not The Math.
New Scientist reporter Matthew Sparkes tested a free AI chatbot on an outstanding mathematical problem that had resisted solution for over a decade. The chatbot produced an answer in 13 minutes. Microsoft researcher Bubeck notes that AI research feels like "pushing at an open door" after years of being stuck on hard math problems, calling it the polar opposite of traditional mathematical work.
This demonstrates what I would call the Latent Solution Hypothesis. Many hard problems already have answers discoverable through combinatorial search. The chatbot did not invent new mathematics. It searched a solution space faster than humans can. The principle: when a problem's answer exists within a model's training distribution, retrieval beats reasoning. The mechanism is pattern matching at scale, not insight. Students of AI should understand this distinction clearly.
Reporter Matthew Sparkes conducted the test for New Scientist. Microsoft researcher Bubeck provides commentary on the contrast between traditional mathematical research and AI-assisted discovery.
- Open a free AI chatbot such as ChatGPT or Claude in your browser.
- Pose a well-known but non-trivial math puzzle, such as the Monty Hall problem or a combinatorics question with a known answer. Ask the chatbot to solve it step by step.
- Time the response and verify the answer against a reference source like Wikipedia. You will likely find the chatbot produces a correct solution in under a minute. That speed, not correctness, is the instructive part.