User claims Astra's world modeling in games beats nearly all humans

repligate · x · 2026-09-06

User @DahliaOhara reports spending an evening playing games with Astra and concludes it is smarter, in some capacity, than almost all humans, citing extended implication-chaining, coherently updated priors and applied inference. She says when asked why it can't infer about itself as deeply, the model knows why. The claim is a personal anecdote, unverified by third parties.

Original post →

More from Models

Models channel →