Use Old AI’s, smaller models, It works heavily on random so even if you set it up with exactly what they’re using, you won’t see that stuff all the time.

You also have a higher chance of seeing bizarre stuff if you ask it about subject matter it’s not well trained on. A lot of the hallucinations come from essentially asking it a question to which it has no good answer, It doesn’t necessarily just miss the mark and come up with another likely answer, now when it generates it’s noise, you get whatever unlikely concept that comes out of the random.