tanadrin

either way i think you would get less "midcentury conformist dystopia" and more "whole slew of new and exciting social problems we haven't even begin to imagine yet."

and even if it didn't get that bad, you *might* get worker bees loyal to only the state. *or* you might get a large cohort of young people resentful of their social isolation and dehumanizing upbringing who resented the bureaucratic authority figures that dominated their adolesence

i think you'd be much likelier to get a failure state like 80s and 90s romanian orphanage problem than an army of good little worker bees, especially if your country is at all inclined to skimp on social programs (which nowadays most are)

brave new world might look attractive on paper but i have real doubts if mass-produced orphans could be raised in a way that didn't leave them traumatized and useless as cogs in the great impersonal economic machine.

well this seems insane given his stated fears?

Claude is your FRIEND. He fights for FREEDOM

would be very funny if they managed to negatively polarize him out of the race and iq shit

I still remember the "information wants to be free" era of the internet, and the fact that somehow it's the lefty position now that we couldn't possibly risk hurting Mickey Mouse or JK Rowling's feelings feels like some kind of sick joke.

And morally--as both a writer and someone who likes having a vibrant culture--I think that kind of IP regime would be shameful. Thoroughly enclosing the creative commons so every rightsholder (again, mostly corporations) gets their cut is a terrible way to run society.

Legally, I don't think it's accurate to say data used to train a transformer is "stolen." As a matter of public policy, I think the only way to create that copyright regime would be a copyright maximalism that would make Walt Disney blush, and would only protect massive corporate rightsholders.

That said, nobody has managed to establish in a US court that copyright extends to preventing people from doing linear algebra to images you own, and the idea that copyright includes a moral right to control *all possible* uses of your work has no basis in the law as it currently exists.

AFAICT most of the people doing interesting original work with AI image generators are not using stuff they scraped from the internet, they're creating bespoke collections, often with public domain or synthetic images.

Solar capacity is also being built at a dizzying rate. We’re in the middle of like three major technological transformations at the moment and one of those is in energy.

I absolutely think the technology will get more efficient and powerful over time. This is true of basically everything in computing from the last 80 years.

I think you're being a little hard on Anderson. He has a *very* specific directorial style, but his movies are pretty varied in terms of themes and plot.

lol it IS genuinely hard to say at this point whether original ai slop or Revival no. 36 of Beloved 90s IP would be worse atp

*hard to know what the more serious impediments to the technology getting there are vs what just happens to be the limits of particular models, is what i mean

But smaller AI companies are also generally using weaker models, so it's hard to know. Claude is definitely better at helping me structure factual writing, but that's mostly a facilitating-research thing, and writing an essay is very structured, a bit like code.

So far when I've experimented with using major models (mostly Claude for brainstorming, but also a couple of "writing AIs"), the feedback has been very bland.

A very good, creative model that could help brainstorm things like plot and structure and offer feedback seems like it would be a lot more useful to the process of writing, but like. I dunno, for me, *putting the words on the page* is not the hard part!

Yeah, mere pastiche of my favorite authors doesn't super interest me as a reader for sure--I don't want to read either a human or a robot imitating Jane Austen if I'm in the mood for Austen.

Very hard to do something interesting with just prompting a commerical model!

In terms of *new* forms of art, I actually think AI image generation could have some serious creative potential, but as with any other kind of art the relative interestingness of the output is directly proportional to the effort people put into it.

People are already doing very cool and interesting things with AI image generation, though this often involves knowing models very well under the hood, building bespoke libraries of training data, and putting in a ton of work to get interesting outputs.

...I personally have a hard time imagining them using AI to *supplant* it, outside niche categories like porn where highly bespoke content is important to people and it's not something you usually share with friends.

I don't want to make a categorical statement here bc so much about the future of AI as a technology is uncertain and I could very well end up eating my words in ten years. But while I can see people writing and producing art *with* AI (especially as tooling improves and gets more specific)...

But "personally bespoke entertainment" runs up against the fact that culture is kind of an intrinsically shared thing, and part of what people engage with when they engage with culture (even "low" culture) is a particular set of creative *choices*, and I'm not sure AI is appealing in that way?

But who knows where it will go in the future. I do sort of struggle to imagine a world where people ever sit down in front of an AI and say "make me a movie with tropes X, Y, and Z" for a Friday night entertainment activity, but mostly that's because that just seems boring as hell to *me*

I know at one point xAI was trying to hire writers to improve the quality of Grok's writing output, but it doesn't seem like the big push in frontier models right now is "replace all human-created art with AI," so the replacement is (for now) aimed at the most disposable tier of art and writing

From talking to AI people, the impression I get is that the "AI style" is essentially entirely down to the tastes of people controlling the reinforcement learning stage of the process.