My favorite quote from this article:
I realize this is a big, serious newsletter, where I write serious financial analysis, but I am so, so tired of this bullshit. Nearly a trillion dollars promised over four fucking years, and everybody is walking around saying “yeah, sure, sounds fine to me” as if there’s even enough money in the world to support it.
That would already be ridiculous and fantastical, but then you add that all of this is in pursuit of making more LLMs. There is no innovation here, no magic, fuck, there’s barely even a future associated with this, just more growth, more GPUs, more data centers…and for what? So more people can use ChatGPT to do… something that OpenAI can’t explain, and neither can anyone else?
No, really, putting aside the money — why are we doing this? To what end? What is the endpoint of all this capital? What is the point of making these data centers?
It’s like Universal Paperclips IRL.
But also the torment nexus, at the same time.
I believe you, but looks like the article was updated as the quote is no longer found.
It’s in the paid section of the article. The preview is very extensive.
Can you upload the full article to archive.is?
Sure thing, let me pull it out of my pocket
Oh and all of our electricity and freshwater
Counter offer… I also do nothing meaningful, but for only 2 trillion dollars.
Scaling is important. I’ll be glad to join you in this effort.
“Guys, OK so hear me out. All I need is 5% of global GDP. That’s it! And thats down from our original estimate of 46% earlier today.”
The concept of leasing a GPU is hilarious. They fully depreciate in like 3 years.
People roll negative equity on car leases all the time.
I await the invention and then subsequent collapse of the suprime GPU leasing market.
If we are real lucky, they’ll figure out how to collateralize those loans for maximum financial damage.
I thought they already have that? At least i remember reading about coreweave or similar using their GPU’s as collateral to buy more GPU’s like a year ago.
Oh god I had not heard of that.
Well, lets hire some accountants and actuaries and get that all the way up to CDOs or a GPU-Lending futures based ETF on the DJIA/Nasdaq, I guess.
After all, its not the state of the economy after the bubble bursts that matters, its the profits you made along the way!
I think it was this one from 2022 that mentions coreweave borrowing 29 billion backed by GPUs.
especially nice given how fast they depreciate.
You could give me a tenth of that and I’ll be a billion times more useful than any generative AI.
So, does that mean they won’t fail for another four years until the bill is due?
God, I hope not. I can’t endure those stupid techbros and their idiotic promises for another four years…
No. No they don’t.
They must be stopped before they successfully build ASI
AGI, at our current tech level, is arguably impossible. None of our LLMs can “think” or “reason” - they’re basically just really advanced predictive text, like when you start typing on your phone and the phone “predicts” what word it thinks you want next based on statistics. We’ve emulated VERY, VERY simple minds - bacteria, invertebrates. Emulating a thinking, living, INTELLIGENT being is a whole other order of magnitude harder. We barely even know how thought works, or physically how memories are stored, or a multitude of other things that would be very very important to actually building a thinking non-human intelligence.
Besides, what they think is going to happen is incredibly dumber. These techno-cultists think that simply by dumping enough data into their LLM model, somehow, through some measure, it’ll just spontaneously develop intelligence. By dumping enough processing power and data into what is a glorified chatbot, it’ll somehow manifest as either some form of incredibly intelligent life, or a real actual Machine God, omnipotence included (because we all live in a Simulation, and this Machine God will basically be the Architect, Neo, and Agent Smith rolled into one).
At the rate things are going, the amount of data and money in the world may not be enough to build ASI yet
OK I mean no offense but all of you are gravely underestimating the potential of ASI and you will pay dearly for it. Some of you I see don’t even know the difference between AGI and ASI. This is ridiculous, I thought this group was anti AI but the last comment j put saying we need to stop them before they develop ASI has dislikes for no reason. Are you all made I’m anti AI or that I said theses tech guys need to be stopped? I’m confused.
I understand the potential of ASI if someone actually manages to build it, I’m just highly skeptical that the current technology we have for doing AI can even scale to proper AGI, let alone ASI. I’m more concerned that these tech companies will waste a ton of resources trying to build it out of a tech that doesn’t have what it takes to get there through just throwing more energy and data at it.
