this post was submitted on 24 Sep 2026
91 points (96.9% liked)
Fuck AI
8423 readers
829 users here now
"We did it, Patrick! We made a technological breakthrough!"
A place for all those who loathe AI to discuss things, post articles, and ridicule the AI hype. Proud supporter of working people. And proud booer of SXSW 2024.
AI, in this case, refers to LLMs, GPT technology, and anything listed as "AI" meant to increase market valuations.
founded 2 years ago
MODERATORS
you are viewing a single comment's thread
view the rest of the comments
view the rest of the comments
I don't always agree with Doctorow, but at least he's not as naive as you make him sound, and he's not a coward.
He's a bit of a traditional socialist in his way of thinking, putting most of his attention on the underlying material and economic mechanisms of a phenomenon, rather that the ideological and cultural aspects of it.
For instance, he thinks that many of the harms are side effects of the hype. To him, for instance, putting illustrators out of work is not a business model and can't justify the money that's spent on this. It's just an ad to attract more investment. If the bubble pops, then this will be significantly reduced.
If you expected Doctorow to get very upset over copyright, you clearly have no idea who he is. But he also literally has a book titled "Information Doesn't Want to Be Free".
He thinks model training is most likely fair use (probably true) even if the way the data are obtained violates copyright, and that giving more copyright to the authors will do nothing to mitigate this (near certainly true).
And since the output of LLMs is not copyrightable except for the part of human labor that goes into improving and editing it, relying heavily on LLMs is a dead end for publishers, because then everyone can steal their work (not totally sold on this one).
And finally, if we reject the hype, I find it hard to not consider LLMs to be a technology like any other. How can you buy in the exceptionalism and reject the hype?
That sure sounds like a zinger, but going off Wikipedia he does argue in that book for a loosening of copyright, which is pretty much what I meant. I don't hate those positions though I'll note they're more libertarian than socialist. He has a whole personal freedom + monopoly control shtick, which works great for me in a lot of scenarios, but doesn't quite work on GenAI tech.
You can call me exceptionalist, but I think LLMs stand out because the adoption of the tech basically requires very capital-intensive and thus centralized operations for model training. This makes whoever runs them (likely capital or a state) a powerful arbiter of information space, maybe best comparable with algorithmic feeds in monopolistic social media.
I think it is necessary to say: "Hey, we as a society shouldn't run things like that. There are better ways of distributing information that work just fine." A tech-neutral attitude can't achieve that.
There are many things that are capital intensive, and we don't hold their product to be particularly special (cars, drugs)
That being said, it's true that we consider information to be a special case.
But what you say is a problem for the "deep research" use of chatbots, or the "AI summaries" of search engines.
But the same could well be said about search engines themselves. That still didn't get the level of hysteria that LLMs get. AI summaries on search should go away though, they destroy the viability of the websites they source information from.
I would like a truly open (incl. open training) model, that could be produced through public funding by a large collaboration of academics and private parties.
I also believe that once we give up on this "machine god" nonsense, there is a lot to be gained in highly optimized, sector specific, lightweight models, which will be less capital intensive to produce. If I need help writing python code, I don't a model that can translate Japanese into Hungarian.
I don't think it's impossible that the underlying tech of LLMs will be evolved into something undeniably useful. But I don't think "helping with Python" will be that.
I think on !fuck_ai I can suggest, you could also just learn the relevant bits of Python? Learning to express your thoughts in code and understanding the tools you use feels way more empowering than relying on blackbox-extruded code.
I do know the relevant bits of Python. Learning basic Python is just marginally harder than shouting at the screen, and adding numpy and some pyqt on top of that was also not impossible.
It was simply an example of a field where LLMs are having a lot of success, all the more reasons that this should be done as efficiently as possible rather than using the same 10 trillion parameters model that is used for everything else.
Secondly, I have heard positive things from people using LLM for instance to review their code, and they found it very very helpful, without the soul sucking effects of having it write all their code.
That I would happily try on my own code, but not in Dario's cloud. And the price to get a local rig right now is way too high.