Pretending this isn’t about a massive surveillance state part 357
Corporate bond offerings intensifies.
“Once the AI is smart enough we’ll ask it how to make a profit.”
- Sam Altman, I think.
At a current pace of what, -6 trilly?
“It will have to go beyond the thing it already doesn’t do, to things it can’t possibly do.”
That’s the real kicker. CEOs promised full AGI, full autonomy, basically the impossible, and investors invested in that. If they can’t deliver on that impossible goal then they failed their investors. It’s wonderful, except they’ll all get golden parachutes.
Does it? I thought this entire industry was completely unhitched from reality and was basically just one massive case of corporate fraud.
Some of it really is absolutely useless.
Some of it actually provides real value.
Some of it seems to provide value, but actually makes things worse.
Time will tell the proportions.
The part that provides value is dirt cheap, though.
Deepseek 4.1 and Mimo 2.6 are basically free compared to Claude, and TBH it feels like the harness/wrapper, not the LLM, is the limitation for doing useful things with them.
Pay premium rates per token just to be told it can’t do that like HAL 9000
All they need to do is get governments high on AI mass surveillance and they’ll be golden
Gouvernement doesn’t have 6 trillion to spare on mass surveillance
All they need to do is get governments high on AI mass surveillance and they’ll be golden
All they have to do is be the last AI company standing, then just absorb everyone else’s hardware for cents on the dollar.
This recent “we should slow down” talk is basically just them stalling until the collapse. They’ve spent all they dare to try and be the one on top, now it’s just waiting until the bubble pops.
Meanwhile China is out there being the black swan of the whole field. If they manage to get models with efficiency and/or utility above what these guys are doing, they’re fucked and they know it. So it’s a ticking time bomb, and there’s not that much time left.
Efficiency above doesn’t convey what China offers, their online frontier models cost 10% of what western models do, not 10% less, 90% cheaper. Also their open weights, which you can run in your own equipment (high end) can do about 80-90% of what these Frontier models do, for free.
True. The “we should slow down” also comes from the fact that there havent been any real technological advances since the transformer model. All improvements are from training bigger models better and more intricate scaffolding but you can only get so far with that.
You got causality reversed here man. They’re military infrastructure, the chatbot slop is just sugar coating
Get govs high on mass surveilance? Can’t get any higher than this.
All you need to do to reach that number is hire a major consulting firm. Arthur Anderson, PWC, Deloitte, HCL, TATA, etc. Time has proven that they are all agents of fraudulent companies who will make up any numbers you pay them to.
is there even that much money in economy that can be used for that? Where would it even come from?
Where would it even come from?
Everyone’s retirement.
So everyones future for old age will get used up so rich people can buy even more yachts.
They can already justify it: plenty of slaves to go around.
Not really. This article assumes revenue needs to cover total spend but there will be winners and losers so revenue only needs to cover the winners expenditure which is less than a trillion. It also seems to be using the wrong figure for datacenters. The amount of data center applications isnt the actual amount since they put multiple applications per data center to shop around for good deals.
Also the sheer amount of control you’d get from being the dominant AI provider is worth a lot. Its like YouTube running at a loss for 10 years now google owns TV.
How much is the cure/treatment for Parkinson’s worth? Or other intractable medical problems?
What about a device able to manipulate the containment field of a fusion reactor?
These are the reasons why everyone is buying the compute services - there are dozens of moonshot-grade problems that are suddenly possible to attempt. Many will fail, but whoever gets that one success will gain astronomical benefit.
