Conversation
Notices
-
Embed this notice
Probably $800 billion going into constructing new AI data centers in the next two years.
Chip shortage incoming.
- pistolero likes this.
-
Embed this notice
@judgedread Accurate, but it's gonna be weird/indirect, since it's specialized stuff and not a lot of manufacturers exist for some of that stuff.
-
Embed this notice
@p Yep. The fabs will delay everything else to make room for the gigantic GPU and TPU orders pouring in from the big players who will pay top dollar to cut to the front of the line.
This may explain the delays in Nvidia's desktop AI workstations. They've got bigger fish to fry.
-
Embed this notice
@judgedread Yeah, you know, you do the breakdown by nm and the maps get even more lopsided, this is going to end up as a bottleneck on the high end.
Nothing like a world-tilting shortage to trigger over-ordering in order to pop a bubble, either.
-
Embed this notice
@p If it's a bubble. Anthropic allegedly has $30 billion a year in revenue from their business plans alone. Blitzy claims to be able to rewrite 80% of old spaghetti code for modern systems, with humans handling the last 20%. If it's useless it will pop soon. If it's useful, why would it pop?
-
Embed this notice
@judgedread
> If it's a bubble.
It's a bubble. Investment is like an order of magnitude past value at least. There's a big race for the high end and people are throwing more money at it than it possibly can make. Companies have quotas for minimum amounts of code generated by machine.
> Blitzy claims to be able to rewrite 80% of old spaghetti code for modern systems, with humans handling the last 20%.
When did he say that? He disappeared from fedi when USAID cleared out, if we're talking about 11112011.
> If it's useless it will pop soon. If it's useful, why would it pop?
The internet bubble popped in 2001 because of a massive race to be the company throwing the most money into it, not because the internet was useless. Very few things have *zero* utility; a bubble pops when valuations run away. (The upside is that the machines these chips are inside are going to go *somewhere* when the bubble pops, so look forward to a glut.)
Anyway, I've talked to these things about code. They're almost comically bad at it most of the time: even the big fancy ones hallucinate failure modes that don't exist, produce code that doesn't even compile, etc. That's not their utility.
Concrete uses are scalable comprehensive NLP, the type of thing that would have been called an "expert system" in the 80s, use as a reference (as long as you can get it to avoid confidently spewing gibberish), and of course, pornography (someone please do a wellness check on OnlyFans in a year or two). These are non-zero but investment exceeds utility.
-
Embed this notice
Hold up. Are you saying Blitz was a spook? 🤔
-
Embed this notice
@Humpleupagus @judgedread I'm sure it's coincidental that he constantly expressed unwavering support for the EU, Joe Biden, and Kamala Harris, and that his fedi account went dark when the USAID fiasco started playing out.
Anyway, if you're going to force me to be serious, I'm not saying anything except that it is both possible and hilarious.
-
Embed this notice
@p @judgedread I shudder to think how many resources are wasted by online AIs repeatedly turning 6 different Stack Overflow articles with 6 slightly different contexts into one useless mess.
-
Embed this notice
@p Blitzy is a company. I don't mean the retarded eurofeeb.
Blitzy dot com
-
Embed this notice
@judgedread Ah, okay, then that changes a lot.
-
Embed this notice
@p Most dot com bubble companies had no revenue - not even clickbait ads! In contrast AI companies have billions in revenue from eager customers who upgrade to the next level to get the very finest... whatever it is they're providing. In the case of chatbots AI boyfriends for lonely women. Those sad hags are willing to pay $100 a month. That's almost five times what Netflix charges, and Netflix does pretty well revenue-wise.
-
Embed this notice
@judgedread
> Most dot com bubble companies had no revenue - not even clickbait ads! In contrast AI companies have billions in revenue from eager customers who upgrade to the next level to get the very finest...
Well, net-negative: they're still not turning a profit and they're trying to corner the market. But real estate developers have revenue and that sort of bubble can and does pop. The only thing you need for a bubble is for investment to outpace value, and we have that.
> $100 a month.
The species is fucked.
-
Embed this notice
@judgedread He stopped having a sense of humor after he got vaccinated. Many such cases.
-
Embed this notice
@p That's funny but I think Blitz left because he could not endure the dual humilations of being exposed as a meth addict (or was it opioids?) and Trump's triumphant return.
-
Embed this notice
@p I need to find some solid revenue numbers. OpenAI is now netting $12 billion annually. Of course if it costs $30 billion to serve those customers they're headed for bankruptcy, barring some game changing happening.
Like ASI...
Also, they might be turning a profit. This stuff is changing so fast if your data is a month old it could be entirely wrong.
-
Embed this notice
@judgedread
> Also, they might be turning a profit.
The signals from VC and big corp investors are that the business model doesn't have a way to stay revenue-positive. They're throwing as much money as they can into it and pushing PR pieces about how important it is and pushing low-cost tech. This means they don't know the ceiling for the market and they're trying to outspend each other to try to land at 90% when the music stops. See also: e-commerce, browsers, even smaller stuff like the coworking market or jitney cabs. Once you have an Uber and a Lyft or an iOS and an Android or whatever (competition is "bad end" for these people: it represents the failure of a single player to completely dominate the market) or you have a Google or a SalesForce or a Netflix ("good end", market cornered, one dominant player, no more surprises until you get a Disney+ and Amazon Prime and someone injects a pile of cash into Hulu).
You won't be able to get reliable hard numbers while this is the case. The bubble is over when you start seeing breathless press about IPOs, or you see the acquisitions accelerate, or you see baffling expansions into strange markets (which are done to make a monopoly look less like a monopoly in one space and more like a normal player in a much broader space).
-
Embed this notice
@p The most suspicious incident is Nvidia 'investing' $100 billion in OpenAI... which OpenAI will spend building data centers full of Nvidia GPUs.
So you see, I am not entirely naive.
Judge Dread
pistolero
?? Humpleupagus ??
EvilSandmich
All GNU social JP content and data are available under the