Very true. Can’t argue with that.
And as to _j’s point, Yeah I was wondering if some kind of fine-tune with formatting or something else had an unintended side effect.
I think a major problem is that usage for these models vary so greatly, let alone individual usage, that what might seem like help in one field might limit another. The balance for creative flexibility and accuracy is a very, very fine line that isn’t easily observable.
Regardless, I am curious to know how they are going to actually solve the current problems right now. I’m starting to notice its performance dip too. OpenAI has acknowledged it, most of us are aware of it. The question is how are they going to resolve it, if they even identified the problem yet?