Raw LLM Responses

A comprehensive study was recently released and found that the incidence of near copies in output from prompts is exceedingly rare, far more rare than your video implies. Anecdotally I have done over 20 thousand generations and have never run into one. They do happen, but very rarely. Also, these systems can absolutely exist without the work of artists in the dataset. The share of the dataset containing art is fairly small. The majority of the training data are photos of people, care, houses, objects. This is what constitutes the core of what brings coherency to imagery. Artist's works simply add a layer of style on top, the cherry, if you will. You can still get outstanding photoreal results without the art within. The small amount of copyright free art existent would also allow humans to train models evolutionarily , but passing down traits and so expanding upon the novel style library that way. Why would it be unethical to copy a style? I'm an artist, you don't think we copy styles? Do you think 'Sam Does Arts' happened upon his style by chance? Not by looking at manga? We artists copy styles all the time. Art wouldn't exist without copying. That is why fair use exists. They're not going to ask for permission. What use would that be? Stable Diffusion is out, it's free, easy to install and everyday users all over the next build dozens and dozens of new novel models and fine-tunes. This isn't stopping. It will actually accelerate. The issue with the argument for prices going to 0 is that AI is nowhere close yet to affect the cost of energy and critical goods & services which house and feed us. Those are the ones we need, but AI is more likely to drive everything else to 0 first and still keep us poor.