I keep seeing comments along these lines. It doesn’t “make things up”. It outputs text under the objective function of maximizing probability of the token bei

I keep seeing comments along these lines. It doesn’t “make things up”. It outputs text under the objective function of maximizing probability of the token being outputted as “making sense” on some criteria. So, actually, it is biased towards not making things up. Occasionally, it hallucinates. I suspect we’ll see subsequent versions hallucinate less and less.