Optimization researcher: Astra proved with one prompt what we couldn't

bremen79 · x · 2026-11-17

Optimization researcher bremen79 reveals that while his paper proved unavoidable gaps between certain function classes and convex ones, the PL case resisted proof — but his student found Astra can now do it with a single prompt using the paper's methods. Mark Schmidt: "Always possible in this brave new world." Another notable data point on frontier models tackling specialized mathematical proofs.

Original post →

More from Models

Models channel →