Compute scale-ups undercut algorithmic RSI, researcher calls on labs to release FLOP numbers

1a3orn · x · 2026-09-07

In a discussion on recursive self-improvement (RSI), the author argues that both Astra and Mythos were, by most accounts, large compute scale-ups. The more capability gains come from compute scaling, the less likely an intense algorithm-fueled RSI becomes.

He concedes this doesn't rule out some software-level RSI, but argues labs releasing actual FLOP numbers would let outsiders reason far better about the question.

Related event: Opinion: Compute-Scaled Astra and Mythos Suggest Algorithmic RSI Is Overrated(2 posts)→

Original post →

More from AGI Musings

AGI Musings channel →