The AI revolution in math has arrived

Robots ace math puzzles; commenters want trophies, not pep talks

TLDR: AI tools jumped from contest puzzles to helping prove new math, even cracking over half of a 2026 research challenge. The crowd is split between hype (“math is machine-perfect!”) and skepticism (“win a real prize first”), with bonus laughs at models needing human-style pep talks.

Math world: stunned. Internet: spicy. After AI models shocked everyone by solving 5 of 6 problems at the elite International Mathematical Olympiad, researchers started using chatty code assistants to crank out fresh proofs in days instead of months. UCLA’s Terence Tao calls 2025 the turning point—yet the comments are where the real fireworks are. One camp is hyped: “Math is perfect for machines—no lab, just logic,” cheering that models can propose, prove, and even self‑check. Another camp is savage: “Cool puzzles, bro. Call me when it wins a major prize,” demanding a Fields‑Medal‑level result and mocking the workflow as a “conversational calculator” driven by human experts.

The funniest subplot? Pep‑talk prompting. Readers cackled at reports that models did better when told “You can do this,” imagining Fields Medalists literally cheering on laptops. Meanwhile, cautious voices echo Akshay Venkatesh: keep the culture of deep understanding before we outsource it to bots. Skeptics also side‑eye the talent exodus to Big Tech and startups, while fans point to 2026’s “First Proof” challenge, where AIs solved over half of research‑level problems—no training‑data spoilers needed. The verdict below the fold: wonder meets worry, with a dash of meme energy and a scoreboard that’s still updating.

Key Points

  • In July 2025, AI models solved five of six International Mathematical Olympiad problems, drawing mathematicians’ attention.
  • Researchers began using AI to conjecture, prove, and verify results; some outputs matched professional journal standards.
  • Terence Tao and others foresee institutional and cultural changes in math practice, though AI is not expected to replace mathematicians.
  • Some mathematicians moved to industry roles at OpenAI, Google, and math-focused AI startups, reflecting growing corporate interest.
  • In February 2026’s First Proof challenge, AI systems solved over half of 10 research-level problems selected to avoid training data overlap.

Hottest takes

"Let me know when an 'AI' has earned one of them" — themafia
"Like saying ‘You can do this’ — this seemed to help" — dogscatstrees
"I wonder when AI will be able to discern the passage of time" — claysmithr
Made with <3 by @siedrix and @shesho from CDMX. Powered by Forge&Hive.