AI: Hard Fork in Mathematics
Insight
Insight
Beyond Calculation: When AI Reinvents the Rules
We've spent decades teaching computers to follow our math. Now, they're starting to tell us our math is wrong.
I was recently watching a keynote by Mo Gawdat—former Google X exec—and he dropped a bombshell about matrix multiplication. For 56 years, we’ve used the same standard method. It’s the bedrock of code.
That single invention resulted in a 26% performance boost and slashed hundreds of millions of dollars in energy costs for Google. This isn't just "solving" math; it's radical architectural reinvention.
The Putnam Miracle
Axiom’s AI scored 120/100 on the hardest math competition in the world. Last year’s median human score? Zero.
GPT-5 Pro Preview
Solved Yu Tsimurutsu’s 554th problem in 15 minutes. No internet. Just pure abstract algebra reasoning.
AIbox.ai
Build your own AI tools without a single line of code. Prompt it, and it links the models for you.
The Erdos List: Problems Solved Since Christmas
Out of 15 "unsolvable" Erdos problems cracked recently, 11 explicitly credited AI tools.
The "Harvard" Connection
When software engineer Neil Simani fed a 2013 unsolved problem into ChatGPT, he watched it do something human: it went digging. It found an old Math Overflow post from a Harvard mathematician, grabbed a logic thread from a related (but different) problem, and adapted it to solve the "unsolvable" puzzle.
Terence Tau—essentially the GOAT of modern math—thinks AI is going to tackle the "overlooked" problems. The ones humans find too boring or tedious. AI doesn't get bored. It doesn't need coffee. It just searches through thousands of possibilities until the logic clicks.
As we look at these breakthroughs in pure reasoning, it's clear: math is just the training ground. The real-world implications for engineering and medicine are right around the corner.
