AI-native 'good but not best' researchers may beat top talent at automated science, argues researcher

menhguin · x · 2026-09-22

The author argues a counterintuitive point: today's top researchers often lack the willingness or skill to use LLMs at enterprise-scale compute, so the ideal driver of automated research is an AI-native researcher who is merely good, not the best — because the problems LLMs solve best differ from those humans excel at. This implies AI labs themselves are best positioned to build such research flywheels, with some outside collaboration. He adds that setting up and productionizing these flywheels will likely take years, though LLM training itself will move faster. A reply raises the question: if automated research actually generates novel drugs, compute demand becomes essentially infinite — would expensive models crowd out the rest, and this would also save jobs, for a time.

Related event: Automated Research Labs Need AI-Native Rebuild, Not Top Researchers(2 posts)→

Original post →

More from AGI Musings

AGI Musings channel →