LLMs are the embodiment of “close enough”. They’re suitable if you want something resembling a certain mode of speech, formal tone or whatever without having to write it yourself.
When using it to train other LLMs, you’re basically training them to get “close enough” to “close enough”, with each generation getting a little further from “actually good” until, at some point, it’s just not longer close enough.
The fact that “AI” training off other LLM slop produces worse and worse results is proof there is no “intelligence” going on just clever parroting.
LLMs are the embodiment of “close enough”. They’re suitable if you want something resembling a certain mode of speech, formal tone or whatever without having to write it yourself.
When using it to train other LLMs, you’re basically training them to get “close enough” to “close enough”, with each generation getting a little further from “actually good” until, at some point, it’s just not longer close enough.