Okay, so after that police disaster, you’re probably ready to throw your chatbot into a lake. We get it. But before you totally give up on our silicon friends, check this out, because the 'hallucination' vibe took a 180-degree turn into 'absolute genius' territory." How?
Well imagine a puzzle so brutal that the world’s elite math brains have been scratching their heads for 30 years. Now imagine a robot solving it in the time it takes you to fold a load of laundry. That just happened.
Over the last weekend, former quant researcher, and startup founder Neel Somani copy-pasted a legendary math problem into the new GPT-5.2, grabbed a snack, and came back to a complete solution. We’re not talking about a lucky guess or a "close enough" answer, this was a verified, bulletproof proof that actually outperformed a solution a Harvard mathematician found back in 2013.
And get this; Fields Medalist Terence Tao (aka the Michael Jordan of math) personally checked the work. And his verdict? It’s legit.
Why this is a "Holy Crap" moment:
Since Christmas, AI has been on an absolute heater, cracking 15 legendary "Erdős problems." Now, these aren't your "solve for x" homework questions; these are the kind of brain-teasers Paul Erdős (the GOAT of discrete math) left behind when he passed in '96.
The highlight reel:
Autopilot Mode: GPT-5.2 rattled off complex theories like Legendre's formula and then actually improved on old Math Overflow posts from a decade ago.
The Competition: Google’s AlphaEvolve (powered by Gemini) actually started this party in November. Now, both models are in a "solving spree," with 11 of the 15 recent solutions crediting AI.
The Scorecard: Terence Tao is tracking this on GitHub, noting 8 problems where AI made real autonomous progress and 6 where it found "lost" research and leveled it up.
So, are mathematicians out of a job?
Not exactly. Before we start worrying about robots taking over the chalkboard, Tao says these are the "easiest" of the hard problems.
Think of AI like a super-powered intern:
The Good: It’s incredible at organizing the "Lego pieces", handling the boring, soul-crushing calculations that take humans weeks.
The Gap: GPT-5.2 scores a massive 77% on math competition problems, but only 25% on the deep, creative "vibes-based" math that requires brand-new ideas.
The Bottom Line: AI isn't replacing mathematicians; it’s becoming their Exosuit. Tools like Aristotle (by Harmonic) are now using a system called "Lean" to double-check every step with a "super-strict teacher" vibe.
And when university professors with reputations to protect start publicly admitting they’re using these tools? You know the game has changed.
