Zvi: Superintelligence won't make such errors

TheZvi · x · 2026-08-31

TheZvi responds to a discussion on model capabilities, stating that the harsh assumption is meant to be didactic about the expected intelligence of Astra and beyond—they will not make these kinds of errors.

Related event: Debate Erupts Over Whether Misaligned Models Could Hack Their Own Evaluators(3 posts)→

Original post →

More from AGI Musings

AGI Musings channel →