Po-Shen Loh, guest-posting on Terence Tao's blog, argues human experts are the brake on AI because there will not be enough of them to steer it
What's new (Terence Tao) / Po-Shen Lohhigh signalThe September 19 guest post on Tao's What's New answers the displacement question with a control argument rather than a capability one: "Driving a car faster than you can run is fine. But not faster than you can steer." Loh's claim is that advancing AI multiplies control points faster than expert supply can cover them, so labs will be forced to slow down by staffing reality, and he asks departments to weight teaching and human-facing work in hiring and tenure and to drop the stigma on mathematicians using AI as a research assistant. He cites the post-Navier-Stokes backlash by numbers: 4,000+ signatories on the Leiden Declaration, 7,000+ on Math and AI, 2,000+ on the letter opposing the Caltech Mathathon. Tao notes the prose was written entirely in vim with no AI generation, though the page layout came from Claude Code.
↳ Follow the thread