rotting bones wrote: ↑Thu Sep 24, 2026 3:12 pmI do hate the AI industry.
Those are one and the same, you cannot separate them.
Ares Land wrote: ↑Fri Sep 25, 2026 2:50 amMy own position on AI is decidedly contrarian. I don't think it's an important issue one way or another.
Ok Ed Sheeran.
WeepingElf wrote: ↑Fri Sep 25, 2026 3:38 amAI has legitimate and useful applications, such as predicting which way a protein folds. But creating artwork is not really one of them.
Those are two quite different kinds of AI.
I think we need to separate the types of AI there are, the impact they have on society, the benefits they bring, and the downsides they have. When talking about "AI", most people mean LLMs + generative AI. Nobody is talking about the AI that scans medical images and detects early-stage carcinoma, or the type of AI that can predict protein folding. That type of AI existed before LLMs, and has quite different needs in terms of data centers etc.
So, LLMs. LLMs are amazing predictors of tokens, and, at least initially, passed the Turing test with flying colours (even though they're basically a Chinese Room). LLMs are useful, but unreliable, and rely on a vast database of knowledge that's scraped from the internet and paper books. There is genuine concern about copyright violations when gathering the data, but it's less obvious than with generative AI. Use of LLMs is ubiquitous. Companies use them for chatbots, auto replies, redaction of writing, drawing up contracts, and so on. People use them as a vast knowledgebase, replacing Google search, to ask medical questions, holiday destinations, partner choice, child rearing, scientific topics, and home work, and so on and so forth. It's the one-stop partner for any question you have, critical thinking not needed. Even the "anti-AI" kids use ChatGPT to do their home work and write their essays. Almost nobody says "AI slop" meaning LLM text output. The downside: data center need (see below) and the fact it makes people
dumb. Critical thinking skills are out the window first, but thinking skills in general are affected badly. And don't tell me anyone likes talking to an AI "agent" when you have a customer complaint. Oh, and the elephant in the room: rogue agents. If we manage to destroy civilization, it's no doubt because some band of rogue agents destroys critical infrastructure.
A special type of LLM is the code generator. Copilot and Claude Code are the most used for general purpose coding, and then there are app generators like Lovable that craete full apps for you. "Vibe coding" is indispensible in the modern software industry. Yes the code is inefficient and bloated, full of bugs and security holes, and unmaintainable, but let's be honest - so is most code of the average software developer. The downside isn't the lousy software, but the sheer volume of code it can produce. Lousy developers can do only so much by themselves, given them a code generator and they easily produce 10-100 times the shit. There's less ethical concerns here, I think, as software can't be copyrighted, and most code generators are trained on publically available source code.
Generative AI is the type of AI most people (except some "prompt bros") hate. It produces bland, easily recognizable-as-AI pictures and videos, nobody except your grandma on Facebook likes to see. It's also the type of AI that has the most problems: it's trained on copyrighted material, but more importantly, it can recreate its training data almost flawlessly. It replaces graphics artists and designers, formerly working on company logos and art work, training material, website design and images, and so on. Why hire an expensive artist for your yoghurt cup if AI can create something passable? But the worst this type of AI has to offer is deepfakes. Famous people's heads have been added to porn actors bodies, deadly outrage has been created by fake videos showing people doing things they never did, and so on. And the other side of the medal: people don't believe proof of genocide anymore, because "it's AI-generated" (even if it isn't). Reality will never be the same. And of course, this type as well needs large data centers.
So, these types of AI have quite a bit of downside, and almost no upside. And there's another set of problems they all share. First, AI needs huge, huge data centers to be able to operate. For training, but also for regurgating their training data when people ask questions or ask for an image or video. Data centers are springing up everywhere, in a time when energy is scarce and we have a global warming crisis. They need clean water (and in practice hardly recycle it), huge amounts of power and produce a lot of heat. Also, they need computers, especially ones with expensive "graphics" cards and lots of memory, having caused a global shortage in supply, and making it impossible for consumers to buy computers with good graphics and enough memory. The economy behind it all is unmaintainable. The investments currently made exceed 2 trillion a year, which means to break even, every person in the world needs to spend about $250 a year on AI (including all third-world countries). NVidia is the richest company in the world, because it lends the money to buy their graphics cards to the corporate buyers of said cards. When the AI bubble bursts (and it has to burst sooner or later), we'll have an economic crash the likes we've never seen before. And of course, all of this stuffs the pockets of a small elite of billionaires that seem to have the end game of destroying the planet and living in a bunker till they die.
Oh, and before someone starts about Luddites, remember they didn't fight technology. They fought rich industrials that first took over their business, then forced them to work in the factories in extremely bad conditions for a measly pay.
JAL
