In 1897, they built the first music synthesizer. It worked, but it took up the basement of an entire city-block sized building, so it was essentially useless. After a few decades of development, it could fit in a suitcase, and be carried around.
Data Centers are like that 1897 synthesizer. Sure, it works, but at what cost? It clearly isn’t ready for prime time. Go back to the drawing board, tweak the problems, including regulations, and maybe in a couple of decades, we take another run at the new and improved version.
But must make number go up by any means
Chanting:
Number go up! Number go up! #️⃣💨⬆️❗
Data Centers are like that 1897 synthesizer. Sure, it works, but at what cost? It clearly isn’t ready for prime time. Go back to the drawing board, tweak the problems, including regulations, and maybe in a couple of decades, we take another run at the new and improved version.
The issue with AI is not a technical or development problem. It’s not even a regulation problem. It’s a capitalism problem. Infinite growth will still be as unscalable in a hundred years as it is now no matter how good and mature the tech is.
In 1897, they built the first music synthesizer. It worked, but it took up the basement of an entire city-block sized building, so it was essentially useless. After a few decades of development, it could fit in a suitcase, and be carried around.
As per the 1890s, it probably also had to be lubricated with orphaned child blood.
And back then nobody asked why do we need a mountain sized synthesizer if the children itself make already all the funny noises.
Efficiency deserves more attention than hype.
He’s sadly right… My wife is Chinese, she watches AI Chinese dramas and she has told me that most of China watch AI dramas. And Gen Z watch AI, Everyone younger than 30 relies on AI.
Now before you downvote me, I’m 37 and I’ve worked in IT since I was 19… AI gives me nothing I can’t find for myself. The only time I’ve used AI was when chat GPT was fresh and new and I told it to be more sarcastic and snarky in it’s answers and I argued with it for 2-3 mins before I got bored.
Humans are much better at arguing with me and being sarcastic to me.
I don’t care for AI, Turn off all the air conditioners and AI burns out.
“AI gives me nothing I can’t find for myself.” True, but that’s the point.
From a worker standpoint, while you’re researching that thing you can find for yourself, someone of equal skill to you that is leveraging AI has solved 3 of those issues. And someone with more skill has solved 10. And someone with lesser has either crashed the network, or solved that same 1.
From a personal use standpoint: is the value of creation in the journey and the learning? For some it is. For others it being able to remove barriers and obstacles to create a new website you’ve been wanting to, or planning out a garden, or brainstorming an idea.
You’ve used it once 2+ years ago and wrote it off completely. Do you also still do your laundry with a wasboard and clothesline? Is the Internet just a fad also?
It’s has value. It also has large risk. But pandoras box is open. Those that refuse to leverage and learn it will be at a disadvantage.
For some it is. For others it being able to remove barriers and obstacles to create a new website you’ve been wanting to.
You’ve used it once 2+ years ago and wrote it off completely. Do you also still do your laundry with a wasboard and clothesline?
Okay but one of those is an artform and the other is physical labor.
Programming doesn’t really equate to physical labor like that.
I agree its pandoras box but I disagree about the quality of life it brings. What it is creating is larger electric bills that is strangling the lower class, killing consumer low end computers which also harms the lower class, creating a dependency which will likely increase in price, lowering childrens critical thinking (according to studies private schools are not being as affected), killing jobs, oh and the fuck ton of effect it has on climate change which also affects the lower class more.
Yes. I agree those are all terrible. And I agree with your assessment of almost all of them. But I’m not naive enough to think that we can put the evils back in and switch back to a non-ai world. We live, we grow, we adapt. This is the future, no matter how bleak it is.
I didn’t say it improves quality of life. I said it can lower barriers of entry to some things, and it can improve a workers productivity when you compare two workers of equal skill.
From a worker standpoint, while you’re researching that thing you can find for yourself, someone of equal skill to you that is leveraging AI has solved 3 of those issues.
And I’ve rejected their code for unnecessary complexity and issues they overlooked all three times so actually nothing was solved. AI does not increase productivity. At all. Study after study after study has confirmed this. AI can spew out what sounds reasonable to people of low skill. That’s it.
I don’t use AI, but I’ve watched enough other people who’s intelligence I used to respect devolve into token monkeys who understand nothing anymore.
Are you responding to something specific from this article? I feel like we might have read completely different articles.
What are AI Chinese dramas? How much is AI? Just the script, or is the entire thing AI animated? Where do you see these programs?
Yeah no shit Sherlock. AI chat bots are just a way to market and sell a dystopian surveillance apparatus to the masses wrapped in the guise of it will right your bullshit corporate emails and messages for you all the while the government and the corporations are fucking you in the ass.
It was a barely functional technology that provides convenience and laziness well hiding it’s true purpose, Machine learning algorithms for facial recognition license prints tracking making it efficient and relatively economical to spy on and control an entire population.
what makes you think advances in LLMs have anything to do with ML for Computer vision? If you wanted the latter, you would’ve gotten that way cheaper than by training on reddit text.
12 upvotes for this thin conspiracy theory, congrats Lemmy.
If you think those datacenters being built are for LLMs only, I have a bridge to sell you.
ML and LLMs both need massive compute to work with the data sets involved.
If only the mainstream media would have said this since 2023. We maybe even have dodged the bullet called the Trump second term, but now we’re heading towards a global financial collapse.
AI is the idea of putting a million monkeys in a room with a typewriter and waiting for Shakespeare.
The smart people already knew the monkeys would just starve to death. The business majors are just now figuring that out.
To be or not to be, that is…
It was the best of times, it was the blurst of times.
The ten thousandth monkey typed out the word “the” , fund them everything they’ll type out Shakespeare!
Dude, the math says it would be cheaper to fund research into necromancy to revive the actual Shakespeare than this.
Is there a petition?
Maybe a kickstarter or something?
if only AI companies optimized their AI to run on less compute (in the data centers)
Look, if you know a way to convert a PDF to text with less than 500GB of VRAM and 2000W of power used for twenty seconds, I’m all ears.
Runs on anything that runs Linux:
NAME pdftotext - Portable Document Format (PDF) to text converter (version 3.03) SYNOPSIS pdftotext [options] PDF-file [text-file] DESCRIPTION Pdftotext converts Portable Document Format (PDF) files to plain text. Pdftotext reads the PDF file, PDF-file, and writes a text file, text-file. If text-file is not speci‐ fied, pdftotext converts file.pdf to file.txt. If text-file is ´-', the text is sent to stdout. If PDF-file is ´-', it reads the PDF file from stdin.Foes it works on scanned image PDF?
.djvu
I’m…but…no…wait….
We’re not going to make it.
paywall removed: https://archive.ph/J3VLi
Now I can read the Atlantic for free while also DDOSing people. The future is now.
It isn’t about content generation at all. It’s about pattern recognition and prediction, which, in the hands of those with the most power to change the world, offers insights into our collective behavior that rulers from every age would have committed genocide to get. AI will tell them how to better build the prison the poor are being impoverished into.
It has the same flaw as every other overreaching evil. We outnumber them. A significant number on our side is willing to kill the other side.
Ima die in the crossfire for sure. But the evildoers always assume they are going to win and they literally never do. They always lose. Expensively.
It’s a dying echo. Nothing more.
Yes but we are effectively disconnected from our mutual self interest. We should already have risen up together and our complacency is a testament to their existing ability to pacify us.
Stop spreading criti-hype! Zuck didn’t invent a mind-control ray with targeted advertising, and Sam Altman doesn’t run a terminator factory with GenAI.
Wtf are you talking about? Do you even know the nature of data usage today? You can’t identify reality from science fiction?
AI enables them to better do what they have already been doing with analytics and user data. They are already doing it and have been for decades.
I’ve seen it first hand on smaller software products during analytics reviews and telemetry design discussions during preproduction through product launch and post launch. I know the questions that gets asked, the purpose of a telemetry hook for a user action, heard what they wished they could track and why. I know how they can cohort a user base, how they extrapolate and predict user behavior and user characteristics from that data to target content. There’s laws already written to prevent some data collection because of what is known can be done. That’s a small software product with a few millions users, not Amazon or Google who have billions of users, many of whom give them access to their entire phone telemetry at all times, cross device access and service wide account tracking, across decades of their lives. Location, region, timezone, battery usage, app usage, age, phone numbers, address, gender, mac ids, wifi connections and data usage etc etc etc.
With just my gender and age, you can make predictions, of some accuracy, using existing research data, about my life, who I am, what I think and how I behave. Every single piece of data more allows further clarity and breadth. That knowledge is what gives them more and more accurate predictions about me, you and all of us. Now they want cameras everywhere, microphones everywhere, OS real ID and VPNs to be banned so there is no anonymity. They want as much data as possible because they now have a data pattern recognition system(LLMs) that can effectively make use of that amount of data.
FFS, this isn’t science fiction anymore, it’s here, now. And those companies have never had your, or my, interests or well-being in mind. They will use it for power, as they always have.
“Generative” “AI” is about generation yes
“Generative” is a misnomer. It will never generate anything new, it can only regurgitate existing ideas based on patterns that already exist. It’s very good at pattern recognition and summarizing, but lacks the ability to form a distinct new idea.
It’s only good at summarizing things which have coherence to its training set. Any ability to summarize input outside of its training is accidental.
It will never generate something novel. Whether it will generate something “new” depends on your definition of “new,” which is a little more ambiguous than “novel.”
Sorry if I’m being too pedantic.
Sure, but then neither will most people.
Shhh… Let the haters hate. Hate is all they have. It’s the only thing that makes them feel superior.
Nope, pedant away. That’s a better way to convey what I was trying to say. Thanks.
If that’s your reasoning then genes and genetics fall into the same bucket.
not even close.
Well, no.
That’s an interesting thought. Genetics is largely a mixing and copying process with occasional “hallucinations” in the form of transcription errors. Most of these errors result in the termination of the hallucinated code. Hallucinations that damage the termination process result in cancer. In the larger sense of evolution, there’s a robust external “review” process. Environmental pressures, predation, and resource availability weed out most of the mistakes and selects the results most likely to succeed.
It’s generating a prediction of our behavior for them to use to better control us.
<takes another hit from the bong>
We are repeating an old pattern in computing: throw more hardware at the problem until efficiency becomes impossible to ignore. Bigger models have delivered remarkable gains, but they’re increasingly expensive. The next breakthroughs may come less from adding parameters and more from smarter architectures, better algorithms and more efficient inference.
That’s literally exactly what Chinese researchers are doing at DeepSeek and they’ve built frontier models with that philosophy
Except there likely won’t be a lot of further breakthroughs if we burn down our planet faster than we already do.
This is all an expenditure of vast amounts of energy for literally no gain for anybody except a handful of billionaires and their corporations.
DeepSeek has really led the way here, especially as they are a bit more hardware constrained. Plus they openly publish their findings and release open source models, so high hopes there.
It’s probably China’s play to pop the AI bubble, but I’m all for it (:
I wonder what all is in the deepseek code that is malicious. I’d like to try it but don’t want a million Mb/s of tracker shit across my network and can’t run it myself.
AFAIK, their open models are distributed as weights, not executables and are therefore not able to start network connections / run code. There if of course tool-calling functionality but that just works by having the model output a special pattern and having something external run predetermined commands based on that.
They are open source models, nothing malicious about them. I’d be much more careful about where you run your agents on. The wrong prompt can even make a non-malicious model misbehave.
Man I swear, AI is like astrology, people have strong opinions about it but haven’t done any research into what it actually is. So much more nuanced than ‘AI bad!’
AI is like astrology
Please keep going.
The analogy you picked really doesn’t match your point at the end, unless you somehow think astrology is nuanced (it isn’t, it’s bullshit).
Lemmy is the wrong place to be if you want nuance. Most people here aren’t accustomed to thinking for themselves.
Large Language Model (LLM) is like astrology, Mashine Learning (ML) is like astronomy. I hate that they both refer to as AI.
Neither of them really fit, but AI is a very marketable term. So basically anything that does stuff on a computer will be labeled that when someone wants to sell it badly enough.
- AI bad
- Astrology 100% superstitious bullshit
And how much research have you done into astrology? Point made.
Burden of proof is on Astrology claiming to be real.
That’s exactly what Baldur Bjarnason explains in his well-researched book “The Intelligence Illusion”. I can highly recommend it!
But it’s so good at programming if you already know how to program! Surely that’s worth burning the planet and crashing the world economy??
Actually still no
https://github.com/JustVugg/colibri
Everyone was desperate to be first because capitalism. But we are getting good models without the insane build out requirement. Which will be hilarious to leave the cunts holding the bag. Not that the planet is better for it in the end.
The engine is a single C file (c/glm.c, ~2,400 lines)
That file is almost 6k lines. The style also makes my eyes bleed. Why do people pretend stuffing 6k lines of code with almost no whitespace and meaningless variable names into a single file is a good thing? I’ve seen this a lot recently
Assuming the vars are all really short, it sounds like the same idea behind Webpack’s (et al) minification & mangling to achieve tiny performance gains everywhere it can. Which might mean there’s a dev version that isn’t squashed and ugly, but doesn’t make it to us…?
Yeah this article is already outdated and poorly researched
~1 token per second (storage bound gen4 nvme)… Some of us have places to be.
Don’t get me wrong. Its impressive that it can run at all, but honestly the usecase is exceedingly narrow. You’d have better results with a structured quantized gpu-only gemma or qwen workflow. Quality over quantity, rely on validation and a structured process: lots of cross-model review and iteration loops with spec and test driven dev. You could probably get a working alpha by the time colibri set up the environment.
Yeah I’m just beginning my local AI journey on a 5080, tried Qwen3.6 27b Q4 and was getting like 1tps because of the vram overflow. Ran it over night at it was still chewing on generating a prompt for a sub agent when I got up in the middle of the night until it simply ended in some kind of “fetch failure” lol. I think I gave it something too large to tackle, but either way 1tps is kinda garbage.
You could use the 35B MoE model, tune it a little bit and get much better results. I have a 5060 ti and 70-80 tok/s are the norm
That’s what I generally use. I wanted to see if I could use the 27b to “review” what the 35b put out. The 35b has been working pretty well, but it’s not very thorough. I asked it to make a program and then 27b was like “this is a skeleton, there are folders but no contents.” Lol
These models generally are not capable enough to do one-shot vibe coding. They are pretty good as coding assistants if you tell them exactly what you want and let them focus on a specific aspect/part of code, not the whole thing at once.
Using an agent framework (I like Kilo on VSCode but there are many others) you can start with a planning session to let the model find out what you want to build. Then you let it write that gist into AGENTS.md and double check if that is what you want. AGENTS.md will be loaded into the context automatically so the model has a solid base of understanding for everything you do afterwards. Once you have that, building in vertical slices on top of that skeleton is much easier. Another neat trick is to ask the model a few questions about the current code base (if there is any…) at the beginning of a session, e.g. "How does feature x work in the current code? ". This primes the model for what you are about to do. All this is obviously a bit more work than just vibe coding away, but it lets you keep in control of the code and helps in being alert for errors these models (and all LLMs in general) will inevitably produce.
Thanks for the tips! What I had started doing the other day was having one session where it reviewed the code and created a document explaining what the program did in technical detail and then in another session I asked it to review that document before attempting anything with the program. Agent programs and harnesses are the next thing I need to start learning for sure.
Wow I’m starting to feel bad about that time I asked AI to make a joke about scatology & eschatology sounding similar.
Is that quantized? 4 bit Qwen 3.6 can get 22tps on a 1060.
It’s the q4 quantization, but it requires 20+GB vram and my 5080 only has 16
My framework 13 with shared RAM runs qwen quite well
All for software that’ll be out of date and fashion next year!
The author seems to be confusing user scalability with performance scaling:
The problem with generative AI, in the industry’s own jargon, is that it does not scale. The cost of growing from, say, a thousand users to a million is a key factor that venture capitalists examine when they evaluate start-ups.
This is a question of whether openai can handle 1 million users asking chatgpt to write a basic html website. That can be scaled horizontally and is just a matter of building more data centers.
The author then goes on to conflate this user scaling with performance scaling:
Yet the returns are diminishing. The bigger an AI model is, the less it improves with each added parameter, and so it must be made bigger at a faster rate just to sustain steady progress. I asked a few AI researchers whether they could name any other real-world software that scales so poorly. None of them could think of any. Even outside the world of software, it’s hard to find a comparable example, given that economy of scale is the principle that has made light bulbs, cars, and clothing so affordable. By economic and engineering measures, generative AI might be the worst technology ever deployed.
This is a question of whether chatgpt can generate a full complex web app. For this there may be a limit to this bigger model approach but this is common to most technologies, performance sometimes has hard limits. You aren’t going to get a car to go 300 mph by making the engine bigger and adding more cylinders, there’s diminishing returns, that doesn’t make cars the worst technology ever deployed… maybe they are but for other reasons.
Economies of scale also isn’t about performance scaling, it’s about capacity scaling. Capacity scaling for AI does reflect economies of scale, that’s why you have these large AI companies building large data centers.
That can be scaled horizontally and is just a matter of building more data centers.
At one of my old jobs “just” was considered a bad word
One does not simply walk into Mordor.
Ok, remove the just then, the point still stands that it is a solvable problem. We know how to make data centers, it may not be easy or cheap but it’s possible just like we know how to build car factories.
Yeah and the point is that model improvements so far have meant making huge increases in size, which offsets the datacenters scale out.
The whole point is that this is futile because we will always be playing catch-up to model sizes, to our ultimate downfall. The tech needs to be smarter not larger. That’s why the whole cloud AI business is shit and not going to work. as anyone with a brain has been saying from the beginning.
Jesus Christ man, people’s homes are being sized with eminent domain for this shit. It ain’t worth it.
The performance per parameter has been improving steadily though. Gemma 4 is ~4o level at a fraction of the parameters.
which offsets the data center scale out
This is only true if everyone is always using the top line model, which most people don’t. Both because most people just use the default, which is a low or mid tier model, and because it’s expensive. The top line models are becoming increasingly niche.
The tech needs to be smarter not larger.
I agree, that’s why more focus is being put on the harness and agent orchestration these days. You can achieve better results by having a large model orchestrate a bunch of smaller model agents to do simpler tasks then trying to have the large model one shot it. This doesn’t mean the whole cloud AI business is bullshit, they’re still going to need to build out a lot of capacity for these smaller models and still going to need large models to handle the planning and orchestration, it just means the call count for these larger models are going to be lower.
So it’s probably not going to be 1 million calls to a small model turns into 1 million calls to a larger model and the capacity never catches up, it’s going to be 1 million calls to a small model and 1,000 to a large one which is more feasible to build out.
people’s homes are being sized with eminent domain for this shit. It ain’t worth it.
I don’t agree with how the data centers are being rolled out, they can and should be built out with renewable energy and consent from the community which isn’t happening. I disagree that data centers shouldn’t be built at all or that it will be an unachievable Sisyphusian task to build them out.
Bubble sort is also a good algorithm if we “solve” its inefficiency by using more powerful hardware. It may not be easy or cheap, but…
I wouldn’t separate performance scalability and user scalability as they ultimately go hand in hand together. LLMs are inefficient by design.
I wouldn’t separate performance scalability and user scalability as they ultimately go hand in hand together.
Ok think of them as different scaling factors then, maybe n for number of requests and s for size of requests and c for complexity of requests. Scaling for n can be done horizontally by building more data centers which is possible. Scaling for s or c requires building bigger models which has diminishing returns.
Scaling for n is required to make the software business model work, like the article says. Scaling for s or c though isn’t required as long as your average user keeps those constant, which is possible.
LLMs are inefficient by design.
They are less efficient when compared to what traditional computing can already do, eg. Arithmetic, structured data analysis etc. There are things that traditional computing can’t do, eg. writing an essay, that can only be compared to the human brain which is hard to do. So you can say AI is inefficient at calculating 2 + 2 , but it’s a hard case to make that it’s inefficient at writing an essay.
