Ofir Press: AI's jagged capability frontier will keep advancing in non-intuitive ways

OfirPress · x · 2026-10-29

Researcher Ofir Press shared brief thoughts on AI progress: the "jagged frontier" of model capabilities will keep progressing in a jagged, possibly non-intuitive way. He also named his two favorite SWE benchmarks: SWE-sweep and ProgramBench.

Related event: Princeton's Ofir Press: Millennium Problems Are the Best AI Benchmark, 70% of Open Math Problems Solved Within a Year(7 posts)→

Original post →

More from AGI Musings

AGI Musings channel →