Deepak Nathan: Today's Models Are Too Dumb at Ambiguity, Not Too Smart

deepakns · x · 2026-09-14

Replying to a discussion, deepakns argues that today's models aren't too smart — the real problem is that they are too stupid in the face of ambiguity. Rather than worrying about superhuman intelligence, he says, what's needed is safety and guardrails to keep them from doing stupid things.

Related event: Today's Models Aren't Too Smart—They're Too Dumb at Ambiguity(2 posts)→

Original post →

More from AGI Musings

AGI Musings channel →