AI may crack the reproducible core of research with only a 1-in-100 hit rate

RexDouglass · x · 2026-07-29

The post argues that AI may eventually master the non-self-referential parts of research if there is a rigorous, replicable core to find.

The key claim is probabilistic: if a system only needs to be right 1 in 100 times, it may already be useful; even 1 in 200 could still matter, while 1 in 50 would be enough to make many jobs vulnerable.

Related event: Scholar Slams AI Evaluation Double Standards, Says 1% Hit Rate Can Drive Science(5 posts)→

Original post →

More from AGI Musings

AGI Musings channel →