> There's no possibility that a 1B model hallucinates less than Fable 5. Sure there is. The simplest possible model that says "I don't know" 100% of the time w

> There's no possibility that a 1B model hallucinates less than Fable 5. Sure there is. The simplest possible model that says "I don't know" 100% of the time would hallucinate less than Fable 5. Scaling up to more useful models, it's just a matter of tuning false negatives vs false positives.