IIRC, High on Fire put out an AI video and it worked because it was creepy. But people hated it because it was AI. I think that’s one of the instances where the connection between these terrible people and this technology hurts the potential of art created using it. I also think it’s useful for prototypes. I’m not going to ever become a graphic designer or an artist, at least probably not. Having the ability to demonstrate an idea is useful.

But yeah, “AI” art has unfortunately also become an asthetic embraced by fascists. That’s probably going to make it toxic for a long time, and I think that’s a loss.

I think part of what we’re seeing is schismogenesis. Corporate ghouls are forcing people to use LLMs (I’m not going to legitimize the use of the marketing term “AI,” because I assume we’re having an honest conversation) in order to de-skill them and replace them. All of the arguments against these models are valid (energy consumption, data colonialism, etc). Some of these remain valid for FOSS models. Some of those ghouls are literally exterminationists who are basically trying to create an evil god.

Normal people were told that the corporations created a magic tool that can replace all human labor, and then they tried it. The thing is, this stuff doesn’t work. It makes shit up all the time. It writes fragile code that’s absolutely unmaintainable. It makes pictures and videos that are creepy. It fills the world with garbage… And people are being forced to use it. The natural response is, “fuck no.”

If people had a choice, if they were actually told what it was, then people would probably be making more nuanced decisions. But they don’t have a choice. Now they’re being told that people are getting fired over it. While that’s clearly a lie, it doesn’t change anything. People are mad.

I’m against LLMs for writing because it’s wasteful and it annihilates your voice. I’d consider it for business technical writing where you want to erase your voice, but even then I’d worry about lowering comprehension. When I’m writing a report, some of what I may say when answering questions after a readout is stuff I deleted from the report. If I never wrote it, it’s harder for me to answer. Having an LLM write for you means you’re presenting someone else’s content, and that’s always harder than presenting your own. Also, you still have to very carefully check everything because it is impossible to make sure it doesn’t hallucinate. It’s basically worthless.

I’m against LLMs for code for a bunch of reasons, one being that their code is garbage. Another being that they can’t architect. I actually think the entire paradigm of LLM code is completely backwards. Code is not for the computer. Code is for people. Why make something human readable if it’s not supposed to be read by humans? You’ve added ambiguity and complexity for no reason. That said, I’ve written frameworks with the intention of making the framework so easy code can be generated. I do think that can be useful, but it’s really about specific use cases. But people are just being sold this “AI can code anything and will replace all SDEs” line, and its obviously garbage.

GenAI art is trash because LLMs don’t have an ontology. They aren’t drawing things, they’re generating statistical translations of tokens into pixels. You will never fix visual glitching because it just doesn’t work the same way as a human does. They’re not compatible models. Now, there are some use cases where that doesn’t matter. I think it could be fun, as long as you’re not using a corporate model.

GenAI for self-driving cars and robots is absolutely bat shit insane. Cars are bad. GenAI can only be as good as humans, which is bad. It’s bad. LLM controlled robots is just a security nightmare.

LLMs don’t belong anywhere in anyone’s stack because they’re impossible to secure.

But like, I can see some use cases. I’ve used LLMs for NLP, as long as you don’t need things to be perfect (I’ve gotten 90% accuracy). They’re pretty good for shotgun social engineering attacks. But that’s not how people are using LLMs. They’re using LLMs as though they were search engines. Someone in a class my partner went to pointed an LLM at a bunch of public comments and asked how many times something was mentioned, and it lied… because LLMs can’t count.

And they’re using LLMs for mental health support, or to replace friends. Normal people do not understand enough to use LLMs, but ghouls were too obsessed with their evil god and their profits to see how horrible it would be to unleash this technology on people. They’re trying to get everyone addicted to something that is absolutely harmful when used incorrectly.

So yeah. People are big mad. They should be. I don’t feel any need to defend LLMs to people who probably shouldn’t use them. The people who are mad about it and can’t articulate why, are mad for legitimate reasons. They should never have been exposed to it, and these corporate ghouls are to blame for the reaction against the technology.

Edit: to be 100% clear, your argument is that there are a group of people poisoning normal folks against “AI.” My argument is that the ghouls that are pushing “AI” right now are so obviously evil that any technology they supported would probably be tainted by their support of it. They are trying to use LLMs to do evil shit.

People see them and want to be the opposite. There are some legitimate criticisms of LLMs (though only a small number of those apply to FOSS). People hate the ghouls, they hate how the ghouls are trying to leverage LLMs, they hate the excuses they’re using, they hate the reality it’s creating, and so they hate the technology. I personally don’t think that LLMs are inherently bad, but I think it’s important to understand why other people might think they are.