y'all I used to mess with generative AI all the time back in like 2020-2021 before chatgpt was a thing and it generated entertaining nonsense
i had a paid subscription to inferkit for my amusement. it was a whole thing
Thing is. Since I played with LLM's back when they were in their early stages. I know how they work. specifically i am intimately aware of why they aren't, and can't be, "intelligent," and why there's no way they could possibly work well for the tasks they are being used for
you see, the earlier models offered sliders to manipulate the amount of "randomness" in the result and how the "randomness" was applied. Set the sliders low, you get endless loops of repeating words and sentences; set the sliders high, and you get nonsense gobbledeygook
furthermore, chatGPT has a sort of framing device where all the outputs are formatted in "person talking to another person" form; earlier models didn't use this and they just generated "more" of whatever you input
it made it very hard to miss that it is a statistically likely sentence assembling machine