If labs aren't near RSI we're safe — otherwise we're at a dangerous point

moultano · x · 2026-09-11

moultano argues about recursive self-improvement (RSI) risk: he can't verify how close labs are to RSI beyond their own claims, and they say they're close. If we're not at a very dangerous point, then either the labs aren't as close as they think, or RSI has sharply diminishing returns. Everything past RSI looks Yudkowskian to him.

Related event: Google Engineer: Labs Say They're Near RSI, So We're in Danger Now(3 posts)→

Original post →

More from AGI Musings

AGI Musings channel →