#because truly and genuinely i don't like generative AI and LLMs

[elexuscal]

I saw your post about chatgtp and here is a thought i had.

What sre these students also going to do about practical issues if they rely on something like chatgtp?

I don't know if they overly rely on it or not, but chatgtp ain't going to have solutions to the many hurdles life throws your way.

Hm. Well. I think folks who rely on ChatGPT will continue to rely on ChatGPT until it clearly and obviously screws them over.

That might be a big, obvious, well-publicized global incident. We've already had a couple, like that lawyer who lost his license after ChatGPT made up fake cases for her paperwork. It'd have to be something like that, but waaaay bigger scale, something pretty much no one could miss.

But frankly, I suspect it's going to be individual failures that everyone has to have on their own. Ask an LLM on how to fix their car, its advice ends up fucking the car up more, and now they have an expensive bill to make the lesson sink in. Ask ChatGPT to write something difficult for their significant other, and whoops, now you're having a big relationship fight. Ask ChatGPT a health question, end up getting super sick. ETC.

And once a person realises they can't use ChatGPT for everything, then it's not like their brains are irreparably damaged. I've seen a LOT of notes in the last two weeks that are people being like "you're damaging your brain" and like, I think that's a very bold statement to make. Generative AI hasn't been available for very long, so there can't be many studies on the subject to show one way or another. But even if we are, brains are resilient! it can recover from huge major stuff, like drug addiction and depression and brain surgery! it can and will recover from ChatGPT.

so i am not doomer about this, honestly, genuinely. but i do really hope we can nip this stuff in the bud before too many doctors and architects and policy makers and the like have to learn this lesson the hard way, because this does have the capacity to really hurt folks.

Which means we need to fix this problem at the source, and ask the question: why are So Many People using LLMs for their work?

And that's a broad and multi-faceted question and I don't think there's a single simple answer. But when it comes to education? I think a big portion is: overwork.

I remember seeing this scene a couple years after finishing secondary school and it resonated SO much. And my understanding is it's only gotten worse in the 10-ish years since I got my undergrad.

And it's not just the level of work.... It's the level of work where so much of it is pointless, or unnecessary.

On a micro-level, that's assignments created with no real thought about learning outcomes, just there to tick boxes. (Thinking of the time where we were told "we should have" kept diaries for all our extracurriculars 15 months into the two year IB course, and my whole year spent like two hours writing a whole bunch of fake ones retroactively). On the other side of the coin, it might be that assignment is genuinely important for learning that subject matter, but the person doesn't actually care about the subject matter, they just need a diploma, because society has decided a diploma is the magic piece of paper you need to get a job, and cost of living is rising crazily pretty much the world over, so you really Can't care about the sanctity of education or whatever.

When that happens, of course kids (and adults) are gonna start using ChatGPT as a shortcut. And while I'm certain there are some folks who are overusing it to the ridiculous amount, they're probably a minority, and we probably shouldn't overstate the problem.

So, uh. How to get people to stop relying so much on AI?

We need to start fixing education! And to start fixing education, we need to start fixing capitalism itself, because that's what's introducing many of the perverse incentives

Should be easy! No problem!

... by which of course I mean it's a huge problem, and knotty, and I don't know how to tackle it all. Why should I? I'm a rando on the internet. I used to teach, but only for a handful of years, and at mostly a kindergarten level. There are better qualified folks than I to propose education reforms.

But in the short term, I think cutting down a lot on homework, and having most essays and assignments written in-class by hand might be a good start.

[max1461]

Ok, so after looking into some of the research that robnost posted links to recently, I am now even more convinced that LLMs are quite smart, are capable of genuine reasoning, and have an elaborate and internally consistent world-model. Questions of consciousness aside, these are genuine artificial intelligences, thinking machines. And they're only gonna get smarter.

Uh, five years ago I would have said this was impossible in my lifetime. Maybe among people truly plugged in the awareness came earlier, but for me it was last year that my thinking really started to shift. And the general public hasn't caught up yet because obviously they haven't; people are still saying "these things are just glorified autocomplete". Well, to some degree they were, but while you weren't looking the autocomplete grew a mind. Which was always the scenario that worried people with AI, that some system would unexpectedly develop genuine intelligence. And I think we are like, there. It has happened. LLMs don't have agency, they don't have desire. They don't want anything in particular, and they certainly don't have the practical capacity to act in the world to achieve any goals they did have. But I mean, in the most core sense, I think we're there, I think we are actually there. For good or bad, we have actual artificial intelligence.

[riseofthecommonwoodpile]

Hi Izzy! I remember back in the day you were a big fan of Sunspring and wrote a beautiful piece discussing it and talking about AI art. Do you think any AI art has surpassed Sunspring? Have changes in the development of AI changed your perspective on it at all? Asking bc I love that essay and I’m very curious :) thx

(context for those who don't know, Sunspring is a short film made about 8 years ago that used a relatively early LLM to generate its script, which was then filmed with a real cast and crew. if you want to watch it here it is, it's about 9 minutes, i highly recommend it)

i've thought about Sunspring a lot as AI art has become such a big topic in the past year or so, and I think the pieces i wrote about it i still stand behind, though i got a few things wrong that i'll bring up when i answer the second question.

to answer the first question: i personally haven't see any AI art that has impacted me emotionally nearly as much, but i also kinda dropped off watching new movies (most of the new-to-me movies i watch are found footage horror movies, 70's porno-chic also-rans, and shot on video movies from the 90's), and i would wager that, if i could somehow experience all of the AI art that exists, there'd be something that hit me in the same way again. there's too much of it for that not to be the case, and too many genuinely creative people experimenting with it. that said, what was so beautiful about Sunspring was how imperfect the tech used to create it was. Almost all of the script makes grammatical sense, but the way it flows, the directions sentences go, the phrasing used is so strange that the friction between the failures of the tech to be truly convincing and the actors trying to bridge the gap to make it still work is what was exhilarating and moving. as the models have gotten better and better, as the rough edges smooth off, that tension so often has just faded into a bland beige unflavored oatmeal of average aggregate language. some of the phrases in Sunspring that have stuck with me the most ("I think I could have been my life", "Whatever you want to know about the presence of the story, I’m a little bit of a boy on the floor.", etc.) wouldn't be created by any of the most popular LLMs today. they're too idiosyncratic, the phrasing is too odd, the grammar almost but not quite there. the plot is surreal and associative, the structuring bizarre and dreamlike. the lines Sunspring ends on— "He looks at me, and he throws me out of his eyes. And then he says he’ll go to bed with me."— are some of my favorites in any film, and it's because they are abstract, poetic, like the computer stumbled upon a phrase so evocative that no written-by-committee script would've let it through. he looks at me, and he throws me out of his eyes. this man who is supposed to love me looks at me in a way where his love of me has gone, where i'm barely even seen as me. it's not the kind of sentence most modern LLMs, with their focus on being convincing, are designed to create.

as far as the second question, i think the biggest change in my perspective is how my belief in the technology, both good and bad, has curdled. i bought into the hype that the technology would progress to the point where screenwriting could be turned into an assembly line, and maybe after that the rest of the parts of filmmaking as well. i had hoped it would become a new collaborative process between human and technology, and i feared it would become a way for movie execs to pay people less and eliminate jobs. the first i haven't seen much of, and the second, while certainly the dream of so many boosters of the tech, has largely been a failure (though plenty, plenty of people have still lost their jobs to LLMs despite that, and as a labor issue i still think it is a very important area of focus). i was too caught up in the possibilities that i didn't bother to research who was making the tech, where the money was coming from, what growth in the sector would look like materially, etc. i still believe LLMs can be used creatively, but most likely any interesting art coming from them will emerge out of models custom-molded by artists to have some of those same rough edges i loved in the first place. i think, in terms of mainstream film, any use of AI is in service to the same bland competence the rest of the industry is mired in, a determination to make products for everyone that inevitably become products for no one. i've become a lot more cynical about the trend towards mediocrity in entertainment, and that cynicism is due at least in part to much of what i've seen come out of the AI space. i do not have a knee-jerk hatred of the tech, but it has not at all panned out how i had hoped or dreaded.