AI may crack the replicable core of research with just 1-in-100 accuracy

RexDouglass · x · 2026-07-29

The post argues that AI will eventually be able to master the non-self-referential core of research if there is a rigorous, replicable structure to it.

It suggests the key threshold is not perfect accuracy, but whether the model can be right often enough to be useful: even a 1-in-100 success rate may already be worth it, 1-in-200 might still be viable, and 1-in-50 would be enough to seriously disrupt the process. The core claim is that institutions will prefer buying the actual result from OpenAI rather than funding the circular “drama” of traditional academia.

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 →