Raw LLM Responses

Your video contains some information that might be misleading to people who don't know how AI models work and are trained. The example you used at 2:01 is very obviously an image to image example where the person used the actual image of the green eyed girl and the parameters were set to allow for little change in the output. That's hardly a fair example to use. If one uses an "image to image" AI technique then I agree with what you're asserting (i.e. there's a potential copyright issue). If the person only uses a word prompt (note that NOTHING of the original image would be used by any of the AI software systems) that's a completely different story. It's also misleading to say that "their models directly use copyrighted content". While the models may be "trained" on copyrighted content, that's as far as it goes. When someone downloads Stable Diffusion, for example on their computer, the model file contains not one single iota of the copyrighted material. Nothing at all. Additionally, the statement that you quote at 1:56 to "prove" copyright infringement doesn't really work. It's my opinion that the statement is simply a effort to head off nuisance lawsuits. I honestly don't see how this is any different from an artist training themselves by studying art from other artists and copying elements of both their technique and styles. The only difference is that the AI is MUCH better at doing so (although one could argue endless about the quality of the AI art). If one is to ban AI models from using copyrighted material to train the AI then one must also ban human artists from training on art or even looking at art from other artists. Lastly, just because it's more difficult for a human to do something "process the information through their life experience" it does not alter the situation at all. One could make the same arguement with wanting to ban Photoshop use because it makes many of the tasks of creating art easier. Does everyone have to go back to using a brush to make their art legitimate? In summary, I agree that if the AI is using "image to image" then there's a possibility of a copyright issue. If the AI is just using their model and a text prompt then the copyright issue become indistinguishable from what human artists have been doing for hundreds of years except that the AI is better and faster at it (again setting aside the discussion of quality). Sorry about the edits. Oh my grammar. 🙂