OpenAI data: agents reshape research but experiments-per-researcher up only 1.68x

CatAstro_Piyush · x · 2026-11-20

OpenAI's second 'jagged frontier' report quantifies AI's effect on its own research: agent tokens per labor cost up 2-3x in pre/posttraining, zero-intervention success on 2-4h tasks doubled to 46%, but experiments per researcher rose just 1.68x. It calls on labs to publicly track recursive self-improvement indicators, with safety/alignment headcount growing fastest.

Original post →

More from AGI Musings

AGI Musings channel →