"Just output probability distributions, never hallucinate": AI safety claim gets mocked

inductionheads · x · 2026-09-16

A claim that hallucinations can be prevented by explicitly limiting model output to probability distributions formatted as decisions ("think of them like enums in programming") drew ridicule. inductionheads quoted it with the jab: "lol just generate probability distributions ==> can never be wrong," highlighting that an output-format constraint hardly amounts to eliminating hallucination.

Original post →

More from Fun

Fun channel →