[MikeBVaughn]

"It's obviously a bit in the future, but I look forward to the day when you can just use AI to generate new music, movies, TV shows, etc that are unencumbered so you can watch them wherever and whenever you want, even in 4k on Linux! I'll have sympathy and sadness for the individual creatives that lose the ability to support themselves, but not for the mega corps that have brought us to where we are now."

I agree that modern copyright law is atrocious and stifling, but the whole "I have sympathy for artists, but..." thing is just grimly exhausting and depressing at this point. I feel nothing but despair. I've spent a lot of time interacting with ChatGPT and image models, and when people say this, all I can do is wonder how other people interact with art and literature.

Every iteration of ChatGPT I've seen is an astoundingly bad prose stylist. At a sentence-to-sentence level, everything that burbles forth from ChatGPT's maw is joyless, flavorless garbage. It's an undifferentiated slurry that has all the verve of a corporate onboarding manual, mated with a self-help book, written by Reddit's gestalt consciousness.

Statistically aping surface form is low-hanging-fruit. Even if you just want (ugh) 'content' in the raw sense of the word - stuff that fills time between when we're born and when we're dead - I am skeptical current Large-Whatever-Models will get there. Everything they do is wholly divorced from being an agent in a physical reality. This is why image models can't even do (what-do-you-call-them, oh yeah) hands. There's a difference between regurgitating variants on two-dimensional representations of hands and drawing hands. I am similarly skeptical that a large-text-model can ever do anything that captures inner experience in a truly novel way. I just firmly do not believe that a text model trained on works written prior to Faulkner's career - regardless of how much goading you gave it - could produce "As I Lay Dying."

Wake me when a GPT model can write a thriller that delivers the rainsoaked weltschmerz and dread of Le Carre's 'Tinker, Tailor, Soldier, Spy.' Until I see an AI put out something on par with "Move. Once you stop, you never start again: there is a special stage-fright that can make you dry up and walk away, that burns your fingers when you touch the goods and turns your stomach to water. Move," I will fear the idea of AI being pushed as any sort of salvation from the overreach of copyright law.

If it does end up being the solution we get, I worry we will have to augment the free-as-in-beer and free-as-in-speech axes with free-as-in-dumpster-diving.

[freedomben]

I certainly agree as of now, but I think one of the next areas of "innovation" will be results with smaller models, which also enables higher quality assurance for the data in the model.

It totally might be 10 or 20 years away, but eventually I think we'll figure it out. There was another discussion today about "humanizing" drum machine output by subtly introducing minor timing variance and it makes it feel "human." At some point I tend to think we'll figure that out for AI art