Dwarkesh warns labs may stop external deployment during recursive self-improvement

MoonL88537 · x · 2026-09-18

Dwarkesh Patel voiced a structural concern about the recursive self-improvement (RSI) era: labs may stop externally deploying their models, going full steam ahead on the most dangerous use case while the public stays in the dark about capabilities and alignment progress. He argues this sets a path toward tremendous concentration of power—a notable big-picture judgment on AGI safety and governance.

Related event: Dwarkesh warns labs may stop deploying models at RSI stage(2 posts)→

Original post →

More from AGI Musings

AGI Musings channel →