The people that call for "ethical" LLMs don't seem to get that without their insanely large scrape of stolen collective human knowledge, the output becomes mediocre at best. LLMs *need* to train on stolen data, because public data doesn't contain the information that users expect out of it. An LLM trained "ethically" will in no way be able to compete with the torment nexuses that exist today, or even a few years ago.

You are either for LLMs or against them. There's no middle here, sadly.

@justsoup This also applies to the "just use local LLMs" folks, due to their reliance on the existing mega-corp models (this is why I have hope all this malarkey will go away when the slopbot bubble bursts as the local ones will stop improving gradually and the sloppers will give up).

@justsoup Ariadne once made a model trained entirely on her IRC logs. It wasn't useful, but it was really funny.

@fun The only LLM usage I approve of lol