We're Shipping Expert Work Faster Than We Can Check It
OpenAI's 722 AI-written maths papers in a day is the headline that should scare builders more than it comforts them.
OpenAI's 722 AI-written maths papers in a day is the headline that should scare builders more than it comforts them.
Many proofs shipped with Lean versions a computer can check line by line, which is real and useful. But the volume itself was called a 'slopocalypse' by mathematicians who objected to the process, and none solved a Millennium Prize problem.
The lesson isn't about maths. It's that AI can now produce expert-level work faster than we can verify it. The bottleneck was never generation. It was verification, and it still is.
Celebrate the capability, but build the checking. A world that can't verify what it generates is a world that ships confident nonsense at scale.
I treat every AI draft as a first draft from a very fast, very confident intern. I still read the work.