@ireneista @futurebird This reminds me of how someone actually produced a 400B. I doubt even the big companies can even run that usefully on their GPU centers even with all the crazy amounts of GPUs they throw at it. (I mean they can get it to run, yes, but not at useful capacity or speed at scale.)

I suppose they think this is good still because the large ones can be distilled down, but that's like making a lower quality MP3 out of a higher quality OGG Vorbis file -- lossy, then lossy again.

I keep wondering what's going to happen when all the VC people start finally deciding they want to see an actual return already on the incredible amounts they're shelling out for all this and all these companies suddenly find that they have sunk way too much into this and lost it all.

@nazokiyoubinbou @futurebird there's no need to guess, this has happened before with "AI". funding will dry up and it'll be a topic tech people avoid for fear of being laughed at as unserious. at least, that's the path things are currently on if nothing changes it.

(this is a public place so we should clarify that we don't claim to know when this will happen. could be years. "the market can remain irrational longer than you can remain solvent.")

@nazokiyoubinbou @futurebird in the dot-com bubble (which, for young 'uns, came decades after the AI Winter we just mentioned), people thought Yahoo was profitable because it had strong revenue...

but that revenue was mostly ads that other tech startups were buying with VC funding... so at some point the finance types stopped believing that that counts as real, and everything went poof all at once.

@ireneista @futurebird What I wonder is what keeps it afloat in the meantime? They can't be seeing truly 100% loss. I think there may be some trickery going on (perhaps even straight up scammy stuff) kind of like the day trading stuff they do with bitcoins that make them profitable for those who know how to do that (and have the capitol and lack of morals to make it work.) At least I think that's what they're doing? I'm not sure entirely how it works, just that since the costs to actually do mining are more than the actual results from doing it, they aren't making money that way...

@nazokiyoubinbou @futurebird when the Silicon Valley Bank thing happened a year or two ago, it came to light that some VCs had been telling the startups they funded to buy stuff from other startups they funded, and it was very eye-opening for us personally to realize that that's a thing.

@nazokiyoubinbou @futurebird
we of course have zero evidence about where or to what extent anything like that is happening today, we only mention it because it's a nice simple mechanism of deception that could plausibly produce some of what we're seeing. financial manipulation can take a lot of forms and we're not experts on it, we just enjoy having a concrete example.

@ireneista @nazokiyoubinbou @futurebird Ed Zitron makes a pretty strong case that what's happening is actually pretty close to what you're suggesting, where basically the only money most of these companies make on AI is from selling to each other, e.g. much of Anthropic's revenue coming from Cursor, much of Microsoft's AI related revenue coming from OpenAI, etc.
https://www.wheresyoured.at/the-haters-gui/
https://www.wheresyoured.at/openai-is-a-systemic-risk-to-the-tech-industry-2/

@csgordon @ireneista @futurebird The part that confuses me is this is a feedback loop that can't amplify. The best case scenario in such a loop would be for the funds that went in to stay the same. Except with all the costs, it's much more realistic that each time there is an overall loss that just keeps adding up more and more.

So they'd only do that if they're seeing an overall gain somewhere else somehow -- or at least the promise of one in a very near future (these people don't play the long game.)

I keep thinking there must be something more scammy going on. Maybe something like moving investments around and even possibly manipulating markets so they can do the equivalent of stock day trading in an unregulated market.

In my sci-fi story there is a massive data center and I thought about making it "10km wide" but then scaled it back a bit because that seemed absurd. (A 4km data center full of ants is still very exciting.)

This is the data center Zuckerberg wants to build.

I worry that when these guys become CEO and stop working building things and writing code they can fall for some of their own hype.

What (if anything) do you think a data center of this size could do that current centers cannot?

@futurebird

Data centers of this size allow you to take down much more compute and network power with a single drone/hurricane/flood/bomb/pick-your-poison.