Raw LLM Responses

Training on copyrighted work is iffy legally and could go either way (generally speaking the AI companies seem to be winning so far). Producing content that includes other peoples copyrighted work on the other hand is a straight up legal victory for the aggrieved party every time. So is it legal to train your AI on Disney movies? Maybe. Is it legal to create pictures of Disney characters for your paying customers without a licence? Absolutely not. In effect what you have here is generative AI reaping the results of all that content they stole for want of a better word. It's all well and good training on other peoples IP but when there is no way to filter it from the output you're in a lot of trouble. I suspect the ChatGPTs and so on will see the same, as they increase model size and the models get better at reproducing the input verbatim expect to see the lawsuits roll in. I think there was one study that said an older model could reproduce something like 48% of the Harry Potter books. You can bet your ass the copyright owners are waiting for that number to creep up a little before landing the mother of all lawsuits.