NYU talk: Mathematician Tristan Buckmaster sees AI agents as parallel test-time compute scaling

yoavgo · x · 2026-09-26

Sharing a slide (around 19:40) from mathematician Tristan Buckmaster's NYU talk, Yoav Goldberg notes an interesting framing: Buckmaster views "AI agents" essentially as a device to scale up LLMs by easily running many instances in parallel — for math, it's all about scaling test-time compute.

The talk itself is mostly on the "boring" Navier-Stokes / PDE mathematics, but the AI-for-math remarks and Q&A contain other interesting bits worth watching.

Related event: Mathematician frames AI agents as parallel test-time compute scaling(2 posts)→

Original post →

More from AGI Musings

AGI Musings channel →