*Before I start the post, a small update: Yes, I completely forgot I was supposed to do promo yesterday. This thing — between the infection and the anti-biotic, has given me a preview of being ninety. Which I don’t like. I give it no stars and am trying to send it back. Mostly it does so by being too tired to think.
That said, today I woke up better. We’ll see if that survives the antibiotic dosage I just took. But hopefully it will.
OT and separately, the clanker song of the day: I’m insufferably proud of the style of the video, which I came up with to circumvent the whole “birth” thing which otherwise would get weird. OTOH I could just be crazy. It’s a choice. Anyway A Night Of Years, first song of the third book of No Man’s Land.
And that’s it. Today, in fact very shortly I have to leave for an appointment with a specialist. So I guess Promo Post will be on Sunday. I promise I’m trying to get over this thing as fast as humanly possible. – SAH.*
When did you start suspecting there was an anti-AI psy ops going on?
For me it’s the “Water that goes into data centers just evaporates on contact” thing going on in Twitter.
Do data centers use disproportionate amounts of water? I don’t know. And neither do you. What we do know is that any number of people on X (and other places) assure us the water use is apocalyptic.
It’s hard to figure out how this can be. I know what the water is used for — cooling — and a lot of industrial processes use it the same way.
When I pointed out that the water doesn’t simply vanish, someone on X assured me that it’s then not in a state to be returned to reservoirs. To which my answer is “Why not?” And in the same way, why would they be using a lot of “new” water. I believe the normal way of using water as coolant is to run it through a closed system, where it’s run somewhere to lose the heat (usually underground) and then back up to cool. Am I wrong in this? I know a lot of you are far more knowledgeable on this stuff than I am.
The problem is that this isn’t being treated as a “new process that might have these problems, we need to make sure the plants don’t use too much water”. It’s being treated as the beast of the apocalypse and we’re enjoined to banish it to the outer darkness.
And I can tell you when things like that are done, and uniformly pushed, particularly by a band of foreigners and bots or foreign bots, I get mulish and go “No.” And fix my little hooves on the path and can’t be moved. Because the psy-ops are never in your best interests.
I don’t care if psy-ops are telling you that going outside once a day and eating an apple is good for you. You’ll find they want you to to run outside starkers or to eat radioactive apples. Because it is never in your best interests.
Now I think a lot of the psy-ops is being run by places like China and people who wish us ill. And I think part of that is that they have an inflated idea of what AI can do for our country and therefore, of course, would like to defang us.
Inflated? Oh, come on. A lot of you will tell me that no, really, AI is all that and a jar of peanut butter. And maybe it is, but right now I think most of what it does is be a force multiplier for the highly competent.
I’m not deriding this as an advantage, particularly in the short term. As I have said before, we are facing a massive crisis of competence, so allowing our competent people to do more is a bridge to getting out of this bind. Provided, of course at some point we train MORE competent people. (The path to that is… well, the indications are mixed though I’m immensely chuffed by the rise in homeschooling.)
However I suspect most of the reason the psy-ops was unleashed was the overestimation of what AI can do and how amazing it is.
It is amazing mind you, but we’re going to try to use it for all sorts of things it can’t do. How do I know that? Because we’ve done that with every single technology when it’s new. Who can forget putting radioactive buttons in kids’ clothing so you could follow them with a geiger counter? Okay, so that’s a Disney comic, but at the time it was clearly viewed as a completely sane thing to do. Not to mention doing x-rays of your feet to fit shoes. Or a dozen others highly inappropriate uses of radiation.
And this happened for every single technology. So why not for AI. And it’s happening to the extent that corporate managers are telling people they have to use AI even when it makes no sense.
Now, I’ll admit since I’m not a programmer my use case for AI is much more reduced. I mostly can get it to do things like bring me up to speed on the setup of series I’m trying to finish after years of lying dormant. So if I hit a wall I can go “What does Kyrie’s earring look like?” And it spits it out. Very handy. (Yes, I do have to verify. it can be wrong very confidently. But it’s much easier to look for “feather” than to look for earring, which might not be called that at that particular point, say.)
But I know it’s more useful to a ton of other people.
I think it is being overestimated, particularly in its defense/war capacities, and that is causing the psy-ops.
Or maybe it’s not.
At any rate we shouldn’t let it be decided for us by allowing foreigners or bots, or foreign bots to stomp us into unreasoning fear.
Water and energy use? Well, those have known solutions. (Including nuclear.)
We should also, of course, try to curb unreasoning blue-sky uses.
The key though is unreasoning. Go back up there where I said AI is most useful for leveraging the competency of highly competent people.
The best we can do is create a lot of very competent people by teaching the new generations as thoroughly as we can.
Since that’s the same recommendation as for “merely surviving as a species” it’s a no brainer.
Ignore the psy-ops and move on, using your reason and abilities, and passing those on as well as you can.
According To Hoyt 
This is an excellent summation, with a number of things I was not aware of concerning the relationship between AI data centers, utility companies, and regulators the psy-op is taking advantage of.
https://open.substack.com/pub/amuseonx/p/the-anti-data-center-hysteria-is?r=6em1o&utm_campaign=post-expanded-share&utm_medium=web
LikeLiked by 3 people
What’s “interesting” is that there’s the Pro-AI psy-op and the Anti-AI psy-op.
One tells us that AIs are going to lead us into a Great Future where people Won’t Need To Work.
The other tells us that AIs are EVIL and allies of Skynet.
I doubt that either of the above are true. [Wink]
LikeLiked by 3 people
Both are nuts, yes.
LikeLiked by 1 person
Both have achieved perfection in their innocence of economics.
LikeLiked by 1 person
Over on “Hot Gas”, David Strom makes the case that the Donks are training and selecting for Cluster B Disorder personalities. (…/2026/05/28/…).
The article he links to at the Cleveland Clinic says that some CBPD cases may be due to mental abuse and cruelty. Is it possible that the combination of double-think and endless hatred in the Donk world actually grooms for Cluster B?
LikeLiked by 2 people
I wouldn’t be surprised at all. So many of them seem to positively delight in hurting people… The left attracts that kind of person and then systematically (whether on purpose or not) goes about removing the guardrails that keep so many more from turning into that kind of person.
LikeLike
I just read Jack Williamson’s The Humanoids, and it’s fascinating (and fun, albeit in a very dystopian sense). The “Humanoids” of the book are tools, but they are tools that humans can choose to cede responsibility to. Given such power, these tools do what tools will do. If you cede control to a hammer, the hammer will treat every problem as a nail.
LikeLike
Though there are some on the pro-AI side (Sam Altman comes to mind; and seriously, his last name is “alternate man”?) that I suspect would happily fulfill all of the worst nightmares of the anti-AI people, if they had the ability to do so.
LikeLiked by 3 people
Reality is written by Kojima. Change my mind.
(Also, kudos for saying in one sentence what it took me three paragraphs to say.)
LikeLike
The Davos crowd and their adjacents are working feverishly to capture this new technology or at least bend its implementation towards their own ends, I do know that. They’ve spun up at least one AI-focused version of the global leadership school (or whatever it’s called) that produced so many of the indoctrinated globalist “leaders” that plague us. (It’s not hidden; the websites for their seminars and trainings are out there to be found by anyone, if not attended by just anybody; would link, but as usual I forgot to bookmark everything.)
LikeLiked by 1 person
How long until those “Great Reset”-style planning documents are written by LLMs? At least hallucinations are par for the course…
LikeLiked by 2 people
Oh noes warm water. Akin to the complaints about the slightly-warmer-than-ambient water piped offshore from the Diablo Canyon nuclear power plant, which the enviros predicted would be catastrophic but the fish and crabs and other sea life love, clustering around the outflow.
So put the data centers in rice country – rice seedlings like them some warm water. Higher rice yields, oh noes.
LikeLiked by 1 person
The Florida Manatees -love- the waste heat outflow from Florida’s nuclear plant. They congregate in winter to enjoy the warm water. Even a few degrees can make the difference between survival and doom.
The AI issue is where there is an evaporative leg (Cooling tower). That water isn’t going back into the aquifer unless it recharges locally. Many do not.
Now the clever folks that tapped wastewater treatment outflow for non-potable cooling water are rather smart. The “waste” water would otherwise be dumped in the nearest river or stream. -Maybe- some would be used in part to water some community lawns the way Melbourne Florida pipes reclaimed water for sprinklers. By evaporating the wastewater, they also capture any missed untreated crud instead of outflowing it.
LikeLiked by 2 people
My gut feeling is that, at the end of the day, AI is just another technological tool, like fire, steam engines, the internal combustion engine, computers, and nuclear energy. It will prove remarkably useful in some hands. And undoubtedly damaging in others, especially for those who use it unwisely (students using it write papers, rather than learning how to think for themselves, for example.) But it’s here. And there’s no going back.
LikeLiked by 3 people
RE: Closed vs. open loop. The last time this came up, someone said that closed loop cooling is doable but open loop is cheaper, so many of the proposed data centers are pushing for open loop. (Anyone who follows this stuff can correct me if I’m wrong.) Not that it should matter, for the reasons you state.
The whole debate is complicated by the fact that the AI CEOs are their own worst advocates. There’s a strong case to be made for AI. There are valid concerns about AI. And then some Silicon Valley executive comes barging in bragging about how his product will put everybody out of work and how we have to restructure all of society to make his company more money.
There’s a huge disconnect between the upper ranks of the tech companies pushing this stuff and the people they’re trying to convince. They’re already sure that AI is the future, that it’s worth every cost, and that everyone else is on board too, so now it’s just a matter of logistics. They seem to be completely unaware about how that looks to other people or how much resentment they’ve incurred.
Not that there isn’t an anti-AI psyop. I’d almost be insulted if China weren’t trying it. But the people who should be AI’s greatest champions are busy hyping up models as too dangerous to release, announcing the end of work, or wiring LLMs into Notepad. Which means the sane AI advocates are fighting a war on two fronts. It’s very strange.
LikeLiked by 2 people
I just saw a headline today that had Anthropic and OpenAI CEOs walking back their “AI will replace you all” rhetoric.
Although TBH, there are a lot of people in jobs that could be done far better by AI models. The issue with any AI replacement would be who is responsible when it screws up?
LikeLiked by 2 people
Maybe replace the average CEO with an AI? (grin)
Could handle that, just Stay Alert, Trust No One, and Keep Your Laser Handy…. (grin)
LikeLiked by 1 person
Oh, great, the Pointy Haired Data Center….. 8-)
LikeLike
Maybe on three to five fronts.
This particular series of disputes would have hit differently at a significantly earlier point in time.
By now, we have a lot of fairly cracked people eager to treat this as a single good-or-evil political argument of pivotal significance.
Well, if you have a hundred ‘pivotal’ events per century, some of those were not actually pivotal. Some things have niche consequences, some things do not cause a large motion.
We’ve basically lost some sense of proportion after recent gambles.
LikeLiked by 2 people
My grumble with the data centers being built out here is 1) putting the power draw in before adding the necessary generating capacity [already have occasional brownouts/black outs when there’s too much pull on the system] and 2) the initial pull of water even for closed-loop systems. The data center people phrase it as “less than X% of what farming uses,” but they are pulling from municipal networks. The PR folks won’t answer questions about the percentage of urban water that they will need to get started, closed or open loop. That reticence makes me twitch.
Too, one of the big companies building stuff out here has already undergone “reorganization.” I have a sense of a looming crash that will leave the cities/counties on the hook for the sites, water, and so on. I hope I’m wrong
LikeLiked by 3 people
Fly by night developers would never leave fixed investment type victims stuck that way.
I’m a realtor specializing in bridges, and I have some hot deals for you, that will have massive returns.
LikeLiked by 2 people
Saw an article today with the CEO of BlackRock telling people they Really Must invest their pensions and 401Ks in AI investments.
No.
LikeLike
Oh.
I forgot about the EU.
Their ‘elites’ are having a big green mad about perceived US tech advantage. (I mean jealousy, not the environmentalism that is part of why they screwed their domestic production.)
For example, their Draghi report. Which is this one academic in government who did some maybe legitimate study of their problems, and came up with a proposal to fix them. Which includes gems like noting that their engineering faculty academics are patenting stuff, but not doing start ups, while at the same time saying that the EU intended to continue the fight against income inequality.
The EU leadership is a bunch of forty to sixty year old men who got top marks in a bunch of plodder studies degrees, for regurgitating a bunch of consensus models on human behavior. Anyone who actually liked and enjoyed machinery in general or computers in specific left their circles decades ago.
I mean, I’m not saying anyone senior in American government is actually good on this. (Though, there is some evidence that this Brandon Herrera fellow actually likes the firearms. He can still lose the general, so might not count yet.)
One of the selling points the EU is trying now, is the fear of the tyranny of technocrats with too much freedom. “Look how dangerous freedom is”, they say. Well, yes, inside the narrow views of the elite EUrophiles, they don’t see that Americans have remedies. Americans do have remedies.
The EU’s ‘thinkign’ political class feels that they have a choice between being a vassal of Brussels, of Moscow, of Beijing, or of DC. They also perceive this AI technology as being part of a bigger technical picture of geopolitical dominance.
Their tech policy in general, and their AI policy in specific is all just symbolic magic to them.
Again, not saying that the US peeps are good. I really was not impressed by some of the relatively recent cybersecurity talk, for instance.
LikeLiked by 1 person
Apologies, this was /not/ intended to be posted as a reply to TXRed’s thread. I screwed up by posting it after my actual reply.
LikeLiked by 1 person
The main source for the negative polling for Herrera seems to be a Dem PAC (thehousemajoritypac.com), so those “within striking distance” numbers are exceedingly likely from gamed push polling. I would be skeptical.
LikeLike
“Vassal of…” has been the default European situation for millennia. The ones who seriously reject it are mostly here in the USA.
LikeLike
Yeah, the details of some of these proposals are another red flag. My general impression is that water use is a non-issue on a macro level, but at the micro level, it depends on the location. The municipal vs. farm water distinction is another aspect I hadn’t considered.
Tech companies should be as pure as snow on this one. Their business model depends on public goodwill in terms of adoption, regulation, construction, etc.. Instead, we’re getting anti-human pronouncements from the CEOs and sketchy/obfuscated data center proposals. C’mon, guys. Make it easy to support you.
LikeLiked by 1 person
For my part, I think programmers being told “learn to plumb” is the height of hilarity.
LikeLiked by 2 people
Power is my big red flag – unless they have their own power via coven plant or the modular nuke plants, a rural power grid will not suffice. On the other hand, a natgas or nuke plant that puts excess out onto a beefed up local grid would be a plus.
The water thing is a distraction from the power thing.
LikeLiked by 1 person
Heh. COGEN plant. Witches not required, but hey, a few on staff couldn’t hurt.
LikeLiked by 4 people
Say you’ve worked in semiconductors without saying you’ve worked in semiconductors…
(My old roommate claimed he used a rabbit’s foot and a crow’s feather to get the machines to behave. I’m not saying what I did…)
LikeLiked by 2 people
How many of the Great Old Ones were involved? 8-)
LikeLike
[Nudges tentacles behind cabinet.] Er, what did you ask?
LikeLike
Great Old Ones? none. Virgin Goats?? No comment
LikeLike
“If it’s stupid and it works it isn’t stupid.”
LikeLike
Computers are NOT a dark art. There are sound technical reasons why you have to sacrifice that black goat every new moon.
LikeLiked by 1 person
Black? goat. Shoot. Now I know what I’m doing wrong!
LikeLiked by 1 person
No, no, no — there are sound technical reasons why you should not shoot the black goat.
LikeLiked by 1 person
She has a Thousand Young, and all of them will be wanting revenge….
Yes, I played way too much Call of Cthulhu….
LikeLike
that should last you a few decades, then
LikeLike
No, no, no. You have to wave the dead chicken over the keyboard! (Source: The New Hacker’s Dictionary. Look it up. Online! catb dot org)
LikeLike
Dead ducks work better. Duck blood with your duck tape.
LikeLike
Is it okay if it’s still in the Chik-fil-a wrapper when you wave it?
Asking for a friend…
LikeLike
You want the kind with bones so you can cast the augeries afterward.
LikeLiked by 2 people
I have, on several witnessed occasions, restored PCs and servers to operation by giving them a Fonzie Thump.
LikeLike
I had an Inspiron that responded to a Gibbs Slap. For a while.
The official-ish way to fix it entailed mucking with inconveniently located jumpers. That worked the first few dozen times, then nada. Didn’t have any C4, so it went into the county’s eWaste slot. I did salvage the RAM (if I get another desktop that can use that config) and minor gubbage. Fairly nice heatsink…
LikeLike
The reason there’s two big plastic bins marked “computer parts” out in my garage:
“Ooh that’s a cool heat sink/bracket/gizmo/whatever!”
LikeLike
Right now everyone is power limited and there is a huge backlog of chips which can’t be turned on.
They are building their own power, which moved part of the bottleneck to gas turbine production and part of the crying to noise complaints.
LikeLiked by 2 people
Just realized I kind of said the same thing you did. :) GE, which which touches just about all the infrastructure-related business, is doing gold-rush levels of business. Looked at YOY revenue figures for them the other day, and…wow.
LikeLiked by 1 person
It seems a fair number of data center projects are building their own generation sources in addition to their utility connections, and many of them could produce more power than the data center itself strictly requires. Trump has made noises about getting all of them to do this. Onsite generation may not be feasible or necessary in many cases, but there’s a big opportunity here for private enterprise to do things that benefit EVERYBODY in terms of building out electrical infrastructure and increasing capacity, without pushing the cost onto the general public. Could maybe even decrease the long-term cost of energy. If they manage it right, that is. Will they? I dunno…
LikeLiked by 3 people
For the one that may or may not happen in my area, I am mostly concerned about the potential power bill hike. That and the fact that they can’t be arsed to put out any engineering docs to the county govt newer than a preliminary study from 2019.
LikeLiked by 1 person
A lot of this seems bubbly, like the myriad residential solar companies that sprouted like weeds when the subsidies were there, only to fade away when those expired.
(Similar, too: $CITY offers a subsidy for $COMPANY to do business. Goes well until the subsidy expires, at which point it’s A1-Mover time. F-Falls fell for that, with two companies ghosting it after the grift was done. I suspect Company #3 got a great deal on the office buildings…)
Yeah, I can see Joe-Bob’s AI center going toes up in a few years. Kind of like the businesses that showed up given a huge tax break + incentives, only to disappear once the goodies expired.
The 45,000 acre proposed data center in Utah makes me wonder if we’re hitting peak bubble. OTOH, a largish center that came bundled with a power plant (nat’l gas or mini-nuke, take your pick) could be useful.
LikeLike
Or chip factories. Interesting on how that happened locally. Property tax deferment during build and 5 years operational. Six year, shutdown and mothballed. Know of a number of locals who trained under dislocation (timber workers) program, only to be out of a job, again, five years later.
LikeLike
The thing with municipal water is that it is an industrial product. Which renders the water objections moot before you even get to all the other lies they are based on.
LikeLiked by 1 person
There’s actually a big question about whether these data centers are operational or not. Quite a few seem to be in some kind of development limbo – https://www.wheresyoured.at/where-are-all-the-data-centers/
But I agree that there’s a huge data center AI bubble going on. As someone said on Discord we can hope the bubble lasts long enough to get mini-nukes up and running everywhere to solve the power problem
LikeLiked by 1 person
I’ve also never understood why they can’t just use wastewater or gray water. It’s not like the server farms can get sick from e. coli.
LikeLike
They do. Which the activists then use to say the water is contaminated, carefully not mentioning that the water was contaminated before ever reaching the DC.
LikeLiked by 2 people
Ah.
LikeLike
It’s a race for the first movers to cash out before the fact that none of these companies make any money becomes common knowledge. Much as the internet is still here whilst many of the boom internet companies are not combined with a level of psychopathy and grift that impresses even me who spent their life dealing with, well, psychopaths and grifters. The AI bubble is the greatest grift of all time however it actually turns out. I suspect it’ll be like the introduction of automated spreadsheets and search engines, but that’s just me.
LikeLiked by 1 person
and BTW. I’m not sure that any of the gigawatt data centers have actually been completed. Some are partially done, and more are spoken about, but actual evidence of any operating at capacity is lacking. I haven’t found any, and I’ve been looking. Elon seems to have gotten furthest along, but he’s running gas turbines and that’s not sustainable. They need data centers now, but they take two to four years to build and the cost is astronomical, trillions. Something to think about. There might not be any there there.
LikeLike
According to https://cleanview.co/data-centers/us the 612 total current operational data centers capacity is ~18.8 gigawatts, and individual centers look to top out just into the 700 megawatt range. There are, however, planned centers up to 9.7 gigawatts capacity listed as a couple years out, but my question would be whether those would need on-site nukes.
LikeLike
And are they bigger than one of the currently deployed modular reactors generating capacity, which means either more than one, or a full scale nuke. Personally, I’d rather stack SMRs because they would be easier to scale, either up or down.
LikeLiked by 1 person
Yes, the modular part certainly implies the capability to stack. That would be extra useful compared to the humongous all-or-nothing reactor designs of the past – one down for maintenance of some sort, the rest stay going.
LikeLike
Yes, this and all of this!
That point “none of these companies make any money” is hardly a secret but massively under-appreciated. (Space X is an exception; but the money is from Starlink and launch services, the “xAI” side of the company is a huge sucking black hole — which the IPO will doubtless spotlight).
Literally nobody knows at the moment whether this industry, as presently constituted or in some yet-undreamed-of configuration, can ever connect paying customers and its product/service well enough to turn a profit. Any profit, at all.
And yet the hype and blather have been flourishing, or festering, literally for generations. It’s easy enough to come up with examples from the 1940s — the Manhattan Project era — e.g., Jack Good.
LikeLike
I think the disconnect comes from the economics.
The AI companies are generally pursuing the “grow fast to own the market” strategy. Some markets vastly reward the first mover and they think they’re in that market.
So the CEO is doing everything to push the company to get bigger.
This costs lots of money, since the models are really expensive.
They’re borrowing that money, so they need to persuade the lenders/investors that they will get a high return.
So they promise lots of things that might or might not work out, but they’re all about how people will pay the company lots of money in the near future for that.
They’re trying to come up with scenarios whereby they will be able to recoup the money that’s already been spent, quickly.
We’ve seen this lots of times before. Sometimes it pays off, but often it doesn’t.
E.g. Iridium flew a bunch of satellites in polar orbits to allow for communication everywhere. But they couldn’t close the loop with what they spent and it went bankrupt.
Someone else came in and currently operates them. I presume they sloughed off the debt incurred in getting it running. I note that it’s running, but it took them 15 years to start flying more.
LikeLiked by 2 people
Yeah, that tracks with what I’ve seen. Lots of hype/optimism aimed at investors and enthusiasts, without a lot of thought on how that plays to the public. It feels short-sighted, but nobody’s throwing billions of dollars at me, so what do I know?
It doesn’t help that tech types tend to treat the value of technology as self-evident. Why wouldn’t you want an app for each restaurant you go to? Why wouldn’t you want Copilot integration for Notepad? Why wouldn’t you want thousands of AI drawings crowding out human artists? It’s not that there isn’t a case for this stuff. It’s that they tend not to recognize there is a case against it, so their arguments come across as self-serving and vaguely inhuman.
LikeLiked by 1 person
Oh boy you need to read Ed Zitron. There is so much incestuous financing and funny math going on
LikeLike
most ceos are stupid and have no idea what LLMs actually do. they’re just following;g the herd and the hype. Don’t get me started on the AI companies, circular financing and accounting fraud are just the tip of the iceberg.
LikeLiked by 2 people
Stupid? No.
Not informed on the latest improvements in LLMs? Sure.
More interested in listening to the stock analysts than their own engineers? Likely. Since in most cases their compensation is directly related to what the stock analysts think, this is a rational decision. At least in the short term.
LikeLiked by 1 person
A couple of separate thoughts on AI/Data Center stuff…
My son has to use 100,000 Claude tokens per week, or else he gets a warning. There are things he does with the AI that are useful, like going through all his emails and finding the ones that have meeting requests, and checking against his schedule to find out where there are conflicts, or summarizing Teams threads. But, his company’s business CAANNOT have company information exposed to the outside world (financial company), and he’s in logistics anyway, so he ends up using his work account to do mundane research for his personal use so he can use up all the tokens.
There’s a Data Center proposal for a Superfund site near me, and the opposition stinks to high heaven like astroturf. Huge numbers of large, professionally printed “Say NO!” signs all over the area. I’ve never bothereed to research it, but maybe I will look up their website, and then see who owns it, and maybe, if I’m lucky, find out who’s funding it.
As for energy/water use, color me utterly unimpressed. I grew up in the Pittsburgh area in the 1960s. Steel mills are HEAVY industry. A single electric arc furnace uses as much electricity as a city of about 50,000. The waste heat made sure the rivers NEVER froze, ever, even in the coldest winters. (How do I know it was the steel mills? In the decade after they essentially all shut down, the cities were driving dump truck full of snow out onto the frozen rivers to get it off the streets.) These data centers are TINY by comparison.
I can see the pro-AI psy-ops, and where they are coming from, as simply the people making the AI and associated infrastructure need it have universal adoption. The anti-AI, including the anti-data center, people? I suspect many are external enemies opposed to something that is now very useful to us, and will be more so in the future.
And now you can see that our gracious hostess is not the only one around here who has thoughts randomly strewn all over the floor!
LikeLiked by 3 people
Having actually lived through a number of IT metrics disasters, I’m getting really strong “lines of code per day” vibes here.
Pointy haired boss…..
LikeLiked by 3 people
Oh yeah, Jim’s son would be wise to start very quiet feelers for another job. Any company that measures input metrics instead of output metrics probably has other dysfunctions.
For anyone needing the explanation: lines of code produced by the programmer, or AI tokens consumed, is an input metric because those things are not the actual goal. The actual goal is a product or service that the customers want to pay for. Lines of code can be high-quality ones that improve the product, or they can be low-quality ones full of bugs and maintenance burden (“technical debt”); any experienced programmer can tell you that the best days at work are the ones where you left the project with fewer lines of code than it started with, and yet it was better (new feature, some bugfixes, even a refactoring that reduces the complexity, because that means less maintenance burden in the future and less chance of introducing new bugs because of not understanding what the code was doing).
Similarly, AI tokens can be spent on new features (and there, the skill and experience of the human writing the prompt really matters: AI code can be really good, but it takes an experienced programmer to spell out in detail what the code is supposed to do, and to tell the idiot AI “no, you fool, there’s already a library for that, just call the library instead of reinventing the wheel”). Or the AI tokens can be spent on bugfixes or refactoring, improving the product that the customers pay for. Or… they can be spent on “what is the average wingspeed of an unladen swallow?” research, producing no value to the company.
Telling him “you must spend at least 100k tokens per week” is like telling him “you must spend at least $2.50 of the company’s money every week” without checking on what he uses it for. He could spend it on new whiteboard markers which he’ll actually use in his job, or he could spend it on a new “motivational” poster for the office which doesn’t actually improve anyone’s work.
Measuring the wrong thing is a temptation many companies face. And when you find yourself working for one of them, sometimes you have to stick with that job because you can’t find anything better. But it’s still a good idea to make sure your resume is up-to-date and you have quiet feelers out there looking for anything better.
LikeLiked by 3 people
And then you get into definitions: what is a ‘line of code’? For languages that use them, you can sort of count semi-colons for statements. Is a line/statement in C++ comparable to a line of assembler (no).
LikeLiked by 1 person
“Does it have to be working? How about deployed?” Lots of paths for gaming.
LikeLike
THIS. Man, do I feel lucky to be working where I do right now. AI is everywhere in the company, but ownership & management have made it very clear that it’s a TOOL, the purpose of which is not to replace people, nor to be used for its own sake, but to help people do their jobs better and faster.
Right now, the models my team has access to aren’t good enough to do anything faster or better than I can do it my own self to the standards we require (tooting my own horn, I would describe myself as highly competent, perhaps hypercompetent, in my particular line of work).
BUT…I do wonder sometimes if what I think of as not good enough is just not having the effort into properly setting up an AI agent. I kind of want to try and also kind of don’t; the not doing it is made easier by the fact that I am already swamped trying to keep up with deadlines…paradoxically, the very thing that would become easier if an AI agent *could* shoulder more of the process.
LikeLiked by 1 person
One tip I’ve seen for getting more useful work out of AI tools is by defining a goal they are supposed to reach (all the integration tests pass), the way they’re supposed to measure it (to run the integration tests, run `pnpm playwright-tests` inside the `frontend` directory), and what action they’re supposed to take if the goal isn’t reached (read the error messages from the failing tests, analyze the cause of the error, fix it, then try again). Then you tell them to go away and work until the goal is met. That way you won’t ever see code that doesn’t pass the tests. And your integration tests can be a decent (though never perfect) set of guardrails to prevent the AI from getting too far off-base in what it tries to write.
This does result in a lot more tokens consumed, of course. But it also demands less attention from the human, which might be a good tradeoff depending on your particular job.
LikeLike
I’m not doing anything at all with code (crosses self; heaven forfend), but that sounds like a solid technique you could use in any situation. Writing is my thing, actually: technical marketing. But we’ve got very clear written standards and guidelines for what good work looks like to us (words we use and don’t use; voice & tone, with examples; a detailed editorial style guide in Sharepoint; many exemplars of known-good writing; etc., etc.), and we could have the AI iteratively test its output against them.
It’d actually be really cool, if also unsettling, to be able to feed a campaign brief into an AI agent and get written output that doesn’t take as long to prompt, re-prompt, and revise as it would to write from scratch. Now, finding the time to do that…
LikeLike
It works… sort of.
You end up having to do additional testing to make sure the thing actually obeyed the instructions, but yeah – the better you define what’s expected, the better result you get.
I’m currently writing a decent sized project in VBA, a language I never really learned, but if I pseudocode it out in detail, GPT is pretty good at faithfully turning it into syntactically correct code
LikeLike
That tip, BTW, came from this link that was posted in last night’s Daily Tech News over at AoS: https://arps18.github.io/posts/claude-code-mastery/
I haven’t used Claude Code myself yet, but I probably will soon: my current employer is willing to pay for a subscription for me, and all reports say that Anthropic has some of the best models for code generation. (Currently, at least. Who knows where that will be in six months, the LLM landscape is changing fast right now).
LikeLike
Never underestimate Management’s ability to measure the wrong thing.
One of the most retarded claims about AI is that it’s going to give the companies infinite power and leave everyone else in the dust……… except that it’s individuals and tiny teams who are actually doing valuable things, and the corporations are mostly setting money on fire.
Which would have been the expected result to anyone who doesn’t found their worldview on marxist poison.
LikeLiked by 4 people
Yet another story: Uber pushed AI for the worker bees, including a leader-board for who used it most. And they spent the year’s AI budget in 4 months. Got a chuckle reading that on Insty.
LikeLiked by 3 people
My son has to use 100,000 Claude tokens per week, or else he gets a warning
Not going to lie, that’s insane.
One the other hand I know some AI video artists that would be ecstatic to have that much to use.
(… I think 100000 is a lot. I don’t actually know, never having used a paid engine)
I was highly amused by the article on Insty (or AOSHQ?) that claimed companies are starting to realize that the AI is more expensive than the humans they used to have and causes more down time.
LikeLiked by 1 person
For reference your comment is about 142 input tokens. API costs are sold by the million tokens (different prices per million input / cached input / output tokens) and are a few cents to a few tens of dollars depending on the model. If you are doing anything serious burning 100k tokens isn’t hard. I know people who are spending or expensing thousands of per month — sometimes per week or day — on productive work.
LikeLike
That’s thousands of *dollars*
LikeLike
Whoof. For that price they’d better be getting a lot of paying work done.
LikeLike
That’s the people at the extreme end of usage. In some cases they are using so much that they hit account limits and have to run multiple accounts. One discovered that there is a limit of 40 million tokens PER MINUTE.
And yes they are producing things. I know people singlehandedly or in tiny teams doing projects for clients in weeks which would have taken large teams years. Because it’s so fast they can afford to go step on every dependency and fix that too, recursively, and fix it correctly.
ESR burned through all of the technical debt in his 60-odd projects in about two months, including untangling a piece of legacy code which no one understood anymore.
I’m just using the basic $20/m OpenAI subscription, but I’m completely rebuilding a new piece of control software for a client. Much slower because I’m slower and have limited time, but that still faster than the norm before. And I have multiple personal projects which had always been Someday(tm) because of their size and domain knowledge requirements. They are starting to progress as I have time.
LikeLiked by 2 people
Abstruse bits of code that are hard to understand may be a good place for AI tools to help. In the C# project I’m working on, my colleague needed to work around a bug in an upstream library until the PR he’d submitted was accepted. The bug in the upstream code was in an initialization method that was ignored on nearly all platforms but would be run during the AOT compilation step on Android, and would fail at runtime because of one constructor that was being looked up by reflection with an incorrect type — they changed the constructor’s type signature but missed the reflection, perhaps because they had used a “Foo” string instead of nameof(Foo) and so the rename refactoring missed the string.
After multiple, multiple attempts to work around the bug, my colleague ended up doing two things: one, submitting a PR to upstream to fix the incorrect reflection type (which has been approved by an upstream maintainer but not yet merged as of this writing), and two, using an AI tool to write code pulling in the Mono.Cecil library at build time, only for the Android build, and monkeypatching the generated IL to replace the incorrect constructor type signature with the correct one.
(For the non-programmers who are not yet shuddering in terror, that’s like swapping out one bolt for a different bolt in your jet engine while you’re 10,000 feet in the air. You need to know exactly what you’re doing or the results won’t be pretty).
My colleague said that without the AI tool writing that monkeypatch code, he wouldn’t have done it. Too abstruse, too much time spent digging through obscure manuals to find exactly the right command. But the AI tool write that in just a couple hours, and it was easy to verify that the code was correct: because if it wasn’t correct, the Android app would fail to start. (And the monkeypatching code, though hard to write, was easy to read and verify that it was doing what it should).
I did the code review on his PR, and I ended up approving it, because although monkeypatching the IL at compile time is hairy and extremely brittle, upstream has already approved the bugfix, so once the bugfix is merged and a new NuGet package gets released, we can rip out the monkeypatch workaround. So both my colleague and I expect the monkeypatch to be necessary only for about 3 months, not enough time to become the nasty kind of tech debt that “charges interest”.
And that bug was one of two bugs that had been blocking me from doing a pretty major overhaul to some of our dependencies, so I was happy to see the workaround. (The same PR fixed the second bug as well, and finished the dependency overhaul for me).
LikeLike
AI:
LikeLike
Does he work for Uber? Because Uber did something like that and blew its entire annual AI budget spend in 4 months
Uber Burns Its 2026 AI Budget In Four Months On Claude Code
LikeLike
I saw that somewhere else (I usually don’t read Forbes; not sure where I found it). In Uber’s case, they were doing a leaderboard thing for the teams. The more they used, the higher the attaboys. However, it didn’t sound like there were limitations to actually do productive things with the AI use.
Setting the clock back a few decades, that would be like rewarding people for commenting/moderating/meta-moderating on Slashdot. (Shudders at the memory of the “goatsex” image. IYKYK…)
LikeLike
In case you’re at all interested, the company my son works for is run by the owner, handles retail financial transactions, and creates money (LOTS of money) out of thin air. The photos he has of the 2025 company Christmas party are insane. The entire dining area was a designed set with dozens of tables, and each table had a candelabra, that is, a living model dressed as a candelabra with about 15 lit candles. They were NOT the craziest thing there – not by a long shot. The company probably doesn’t even notice the AI spend.
LikeLiked by 1 person
My advice to your son (in a different comment) would still stand: enjoy working there for as long as it’s enjoyable, but make sure his resume is up to date and that he’s got VERY quiet feelers out for another job. Because management that tracks input metrics instead of output metrics doesn’t give me a good feeling, nor does management that spends ridiculous amounts of money on silly stuff. That job may pay well, but he shouldn’t plan on its lasting forever: it will probably go away when the company hits a run of bad luck that their management didn’t account for.
For example, my first job was for a data-analysis company that did data modeling — this was years before AI tools — to analyse other companies’ data for them on a consulting basis. One of their customers was a credit-card company whose name you would definitely recognize, who wanted to know if they could predict, based on people’s spending patterns, who was likely to default on their credit-card debt. When the dotcom bubble burst and the US economy slid into recession, many companies decided to pull back on consulting spending and do things in-house. My employer lost a client that represented about a third of their income, so they laid off about a third of their employees. My manager told me that he had wanted to keep me — I was fresh out of college but I was doing good work — but the decision was made at upper levels to just cut by seniority, and fire the most recently-hired one-third of the company. I had been there only nine months, so I got my first pink slip. I didn’t mind too much: the pay hadn’t been all that great, but it did teach me a lot about handling very large data sets so I was overall glad I’d gotten the job. But it also taught me the importance of keeping a resume up to date, and having options for moving to a new job. I hadn’t been doing that, and it took me about nine months to find a new job in that recession. Unemployment insurance kept me above water, but it was tight.
LikeLike
Aye. I got caught in Dot-Com bust V1.0 after 22 years at HP/Agilent. Trying to find a new job (especially since separation was 9/1/2001) was no fun. Mercifully, a vendor knew that I was familiar with the product line she sold, and I ended up consulting for the manufacturer. (That gig died in Sept 2002 along with the client company.)
The final decision after that, was to finish remodeling the San Jose house and beat feet. It worked. (SJ house is now 3X the “market price” than when we sold out in 2003. Speaking of bubbles….)
LikeLike
Hmm, human candelabra? AFAIK, that’s seriously kinky stuff. I’ve run across fetish images like that (and left at hypersonic speeds).
LikeLike
Obligatory:
LikeLiked by 1 person
The human candelabras were 1) something you can order up from the local Actors Equity shop (I kid you not, there is a planning catalog with those actors and their rates), and 2) it took a while to figure out they were living humans doing that job – definitely not a kink kind of thing. Just…bizarro.
LikeLike
Another wrinkle is the long-term economics of data centers. ESR had a few tweets on this recently, but I’ll only link one to avoid the filter. The gist is that pricey tech gets crowded out by cheaper, “good enough” tech. The better LLMs get, the easier it is to run a small, good-enough model locally, which cuts into the profits you can make from the big, expensive ones in the data centers.
I’m not sure how it will play out, but I hope we get some nuclear power plants before the music stops.
https://xcancel.com/esrtweet/status/2057948773297303928#m
LikeLiked by 1 person
From what I’ve read, current theory is that most of the AI companies will go bankrupt before long anyway, leaving only one or two that managed to figure out the trick to making money.
LikeLike
If I can get a truly local, and I mean doesn’t have to access the Internet local, I would probably use that. Once I hit a cople hundred gigabytes of stored stuff on my computer, finding that one thing I know is here somewhere gets tedious. I am NOT, however, going to let some AI train on all of my life sitting on my computer.
LikeLiked by 1 person
You can.
LikeLike
Do you have any recommendations, specifically for code-generation models? I came across someone saying that GLM-5.1 was really good, and he was planning to switch to it from Claude… but searching for GLM-5.1, I couldn’t find any instructions on how to get it set up and running. Perhaps that model, although open-source, is still large enough that you need dedicated server hardware rather than a standard desktop machine, i.e. it’s good for companies who want to pay $20,000 up front for a machine with 4TB of RAM and then run the model for free (well, for just the cost of electricity etc) thereafter, but it’s not useful for individuals. (That’s speculation, and the RAM requirements are completely made up for the sake of example, they’re not something I know is real).
I’ve been able to find instructions for getting Stable Diffusion models to run locally, for image generation. But so far I haven’t found the equivalent instructions for setting up local code-generation models. Do you have any pointers for getting started?
LikeLike
GLM 5.1 looks like it is 754b parameters in BF16 native. That is gargantuan (on the order of 1.5Tb just for weights, and I don’t see smaller distillations yet but those will probably come. Trying to run something that size is going to require the trick of multiple mac minis with unified ram.
For the general question: the state of the open weights models stays 6 months to a year behind the closed flagship models, continually improving along the way. So the current crop of open weights are getting to where they can do useful difficult work even at the 30-100b scale.
I haven’t yet used local for much in the way of serious work. I have a bunch of personal projects for building custom tooling around this idea but the last year has been too personally chaotic to make much progress. I did use qwen3-coder:30b Q4 a couple months ago to do analysis of legacy code files to explain what was going on in them.
LikeLike
“analysis of legacy code files to explain what was going on in them.”
That must have been enlightening – any stories you can share?
LikeLike
It wasn’t anything spectacular: new codebase I wasn’t familiar with, written in C++ which I’m not familiar with.
Booted up an ollama server (at that time running on a pair of P40s), pasted in the file and asked for a summary of what it does overall and what each function did.
LikeLike
This may be helpful – https://www.jaredwatkins.com/posts/2026/05/smb-inference-stack/
LikeLike
When it comes to setting up your own neural net, for some specialized application, the one dude whose work I really know much about the inside of spent years working with python, and reading papers on arxiv. He eventually made a library in ‘jax’ to do his final recipe quickly. (Answer to a question you did not ask.)
My search ‘seems’ to have given me some answers for GLM.
The full GLM-5.1 model is over one and a half TB, so is maybe not a self host yet.
There is a software called unsloth which can be applied to reduce a bunch of the requirements. (GGUF is a related term.)
codersera seems to have a description that looks plausible.
https://codersera.com/blog/run-glm-5-1-locally-on-cpu-and-gpu/
The other search answer involves ‘how do I set up a server on linux’, and Ollama appears to be a server software that models can be installed on top of. It has a github page.
I don’t think this is what the other guy I follow online is doing, but I don’t remember.
Edit: qwen is definitely one of the models I have seen someone discuss.
LikeLike
Ars has a few articles about local AI models. Here’s one describing Microsoft’s(!) mini model, Phi-3-mini. There are also links to Ollama articles in this one.
https://arstechnica.com/information-technology/2024/04/microsofts-phi-3-shows-the-surprising-power-of-small-locally-run-ai-language-models/
LikeLike
Ollama or LMStudio is a good place to start for running an LLM locally. (There might be options that are better-suited for coding, but I haven’t gone looking.) Picking a model is trickier, but LMStudio guesses which ones will run on your machine, so it’s easy to experiment.
LikeLike
I have Claude writing python scripts to have Gemma 4 series models (in LM Studio server mode) do a chapter by chapter review of manuscripts that are too big for my Claude account. :D
LikeLike
It’s wild how you can chain these things.
LikeLiked by 1 person
Thanks for the codersera link, that looks useful. I currently have two laptops, a Linux laptop with no video card (I don’t play games, and I bought this just before the AI craze hit, so it only has the embedded GPU that comes with the motherboard) and 96 GB of RAM (which I got just before the AI craze hit, so it was still reasonably priced). And a Windows laptop with “only” 32 GB of RAM and an 8 GB video card from the 4000 generation (I think a 4060). Looks like I won’t be self-hosting code generation models until my next hardware upgrade. (I’ll probably wait until prices return to sanity, perhaps two or three years, and then buy a video card for my Linux laptop — and if you saying “what do you mean, buy a video card for your laptop?”, well, it’s a Framework 16).
LikeLike
Tying a couple of threads together, the same son I mentioned was talking last night (around the family campfire) how his company is going to decomission their servers (upgrades), and will possibly have a rather large amount of terabytes of “older” memory (DDR4 era) to get rid of. They may be offering it to employees for nominal cost. We may end up with a family LLM machine.
I’m old enough to remember when terabytes was something you dreamed of for a hard drive.
LikeLiked by 1 person
I am hoping the AI hype drives Fusion Power to market. (grin)
Meanwhile, we may get back on track with fission reactors.
LikeLike
There’s one factor in AI that I haven’t yet seen addressed: the Panopticon Effect.
Right now, we literally have no anonymity from either private or government snooping. If the government wants to avoid the whole “probable cause” effect, they can just buy all kinds of data that they physically couldn’t have gotten access to before because your commercial transactions, where you went, who you met, etc. were all considered as part of your “private papers” under the 4th Amendment. If the cop couldn’t see it through the windows with his unaided eyeballs, he had to have a reasonable suspicion of a crime. That started eroding with the advent of the internet. Then it accelerated through cell phones and “smart devices”, but there were still limitations because a human being had to think to ask, and then analyze the data with their own human brains and most importantly, endurance.
Once data mining became a thing, it was obvious where this was headed. Now, with Agentic AI, we have a 24/7/365 army of snoops to crawl through the data looking for any hint of non-conformity. And there’s effectively NO requirement even for suspicion. Just ask the clanker to spit out anyone’s name at all who meets the “criteria”, regardless of bias in the AI and it’s training data. Heck, this blog comment is probably a red flag. And once a flag goes up, our three felonies a day legal system will provide Beria with a long list of crimes.
It would be a good thing if President Trump would put out a regulation requiring probable cause and a warrant before commercial data can be accessed for legal purposes, because it’s part of the 4th Amendment “private papers”.
Better would be for Congress to say that in a formal law, and / or SCOTUS to confirm.
Best would be to add an Amendment saying the 4th Amendment “private papers” includes electronic records from any source.
LikeLiked by 2 people
It’s rather a tell that this isn’t right up at the top of objections list, isn’t it?
I think it’s cool tech, that can be a useful tool for a number of things. Many of which, I cannot foresee.
But the hype wildly exceeds the benefits. It’s a bubble, and inflating economic bubbles is not sound policy.
Further, I’m not at all in favor of my electricity prices doubling (again). Or the details of my life becoming even less private. Or being pissed on and told it’s raining. (Our Governor is boasting of how they won’t increase our electricity prices, when no new generation capacity is being added.)
I’m very confused as to why the centers are being located out in the boonies, rather than near the telecommunications and utility infrastructure trunks they would seem to rely on. Especially when there are so very many sites that could much more easily be retrofitted near those trunks. (I get not wanting to put them in Detroit, but even my red state suburb has vacant warehouses near substations.) There’s a game being played, but I didn’t know what it is.
Really, the major reason I’m neutral, rather than opposed, is how insultingly fake the opposition movement is.
LikeLiked by 1 person
Why rural for datacenters? Cheaper land and resources. Lower taxes on same. Fewer nuts/heroes to stage protests. They require minimal staffing, so low population is not a problem. They can be a largish fish in the small pond, unlike in a major city where they are a 7-11 or two in headcount.
LikeLike
(Needed to read down further. Sigh.)
Makes somewhat sense. But even rural area communities/towns have underutilized (empty) areas where data centers can be staged that already have infrastructure. Old mills, PNW this would be old log, plywood, and paper mills where the old infrastructure (buildings and equipment) have been decommissioned, and leveled. Power lines are still there.
LikeLike
Not to mention that data centers have upsized quantities of energy to keep in check, the contents don’t recycle cleanly by burning, and the downwind plume footprint won’t mean as many people to evacuate if a fire starts
LikeLike
Agree. Why aren’t the data centers being put within easily obtainable infrastructure? Why take up farming or ranching land?
LikeLike
At least some of it is them looking for places where there’s not a lot of regulations to hold them back.
LikeLike
I don’t know where you are looking, but people screaming about surveillance 24/7 is only beaten by the retards screaming about water for how common it is.
There is zero benefit to putting new build….. anything really…. in cities. This isn’t the ancient world or the 19th century anymore. Cities are only relevant in cases where you are looking for a certain kind of gullible person who still thinks all the important stuff is in cities.
It’s effectively impossible o build power infrastructure if there are enough people around. And any power plant is going to be located out in the sticks anyway for the same reasons.
Burying fiber is cheap and easy. Especially when that never really got locked into cities the way other systems did.
Warehouses aren’t much more useful than random office buildings or apartment blocks if you are building a dense compute DC. The amount of retrofitting to make the building fit for purpose would be your classic regulation build: demolish all but one wall of a building, build the new building, demolish the remaining old wall. Totally the same structure, as far as the law is concerned anyway.
There is only a game being played if you refuse to consider any of the relevant questions for citing a facility.
LikeLiked by 2 people
Wasn’t it Steve Jobs old partner who recently told a graduating class, That, and I am paraphrasing, ‘You all have AI. Actual Intelligence’. Not to mention all the companies dropping AI because it is not ready for the real world.
LikeLike
The Woz was the smart one on the initial Apple team.
LikeLike
And I think part of that is that they have an inflated idea of what AI can do for our country and therefore, of course, would like to defang us.
I am slightly too off my game to make a sarcastic comment about what the PRC subjects think about surveillance states, and about what improved automation can actually accomplish in that sort of enterprise.
Items:
1. PRC leadership would have learned some things from the rocket stuff with the USSR, and from their own stuff with the internet.
2. They seemingly view AI as an existential necessity for their own internal surveillance, and this is part of the chip stuff that was starting to happen prior to the latest AI push.
3. On the one hand, stressed ignorant regimes. On the other hand, almost none of anyone has a good mental toolkit for understand this question of what a novel machine can and cannot do.
LikeLiked by 2 people
It often sounds as if they’re just search-replacing old anti-nuclear talking points.
Presumably paid for by the same people.
LikeLiked by 2 people
Probably Chicoms rather than Soviets… but essentially true
LikeLike
It often sounds as if they’re just search-replacing old anti-nuclear talking points.
Presumably paid for by the same people.
LikeLiked by 1 person
I want to say I miss Grok Imagine. I understand why they moved it over to the paid side, but getting so many pictures over a single prompt was amazing. (And then, it wasn’t as safe as it should have been putting real people into “spicy” videos). Grok itself, at least the free version, doesn’t do nearly as much.
I still think we are headed into an AI crash, but I don’t think it’ll be of long duration, and we’ll come out of it with saner expectations on both sides.
I’m am hopeful that we might even get cheap nuclear energy out of it. A girl can dream, anyway.
If, like I’ve seen people try to state, they are releasing brown nasty water into the ecosystem, there are laws already dealing with that. Enforce them. You don’t need new laws.
And if it’s a foul libel, well, enforce that too.
LikeLiked by 3 people
THIS. On everyone.
LikeLiked by 1 person
We’re in the “drama” phase of the tech curve. I’ll be curious to see how things balance out. (And fingers crossed for nuclear!)
LikeLiked by 1 person
I am reminded of C. S. Lewis talking about the balance of science and religion, and how every now and again, something new comes up, and there is a bunch of fuss and bother. Then the dust settles, and leaving everyone about where they were before. (Heavily paraphrased. Lewis might not recognize it)
I think we are in the fuss and bother stage.
LikeLiked by 1 person
Two anecdatas. A local used car place is called AIcertifiedusedcars. Never buy a bad car again. Ours are AI certified. Uh huh.
I trusted Trunk Monkey more.
Someone who is in position to know said companies that let people go because AI….are now rehiring for those positions
LikeLiked by 1 person
Yep. And they’re hiring H1Bs and recent college grads. The institutional knowledge is long gone.
LikeLike
Trunk Monkey!
Trunk Monkey Complete Commercial Collection
LikeLike
That brown water is commonly known as “a construction site”.
This is one of the common ways to lie about water and DCs.
LikeLike
The Left seems to have ditched the Small Lie strategy for the Big Lie strategy. Another example of the Big Lie is the claims about AR-15s that they liquify bone and can decapitate people. Or that the Jews train dogs to rape Arabs.
I’m assuming that they want to win on the issues, so it must have some effect helping their side.
LikeLiked by 2 people
It provides talking points to encourage random violence among their street fighters.
LikeLiked by 2 people
And on that note, Treasury is having something to say about street fighter funding:
https://x.com/nicksortor/status/2060072884261867911
LikeLiked by 1 person
The water use talking point is specious. Data centers will fill up their system with water, and then they’ll run it through chillers and cooling towers to recycle it back to the proper temperature to cool the computers (or probably more to the point, the package A/C units running in various places in the building to keep them cool). After the first fill, they’ll only have to add more to address loss. Otherwise it will be reused constantly.
They don’t “one and done” the water through the system, and even if they did, it would simply go back into the river or wherever it came from, probably having been filtered and cooled and cleaned up since that’s what is expected these days. But these fools act like the water will go in, be destroyed forever, and then replaced, ad infinitum. It’s not how it works.
LikeLiked by 3 people
It is the initial draw that’s the problem here. The regional aquifer is already declining, and that first pull may hit ag and domestic wells around the centers pretty hard. [Law of Capture applies to ground water in Texas, so he who has the biggest straw wins.]
LikeLiked by 1 person
Sounds like a loverly case for an SMR powered desalination plant built over some of the salt water typically associated with oil domes, or on the Gulf with a pipeline to each data center for fill and replacement. When it’s not topping off data centers, I’m sure someone’s thirsty somewhere in the area.
LikeLiked by 1 person
Yes. The wonderful thing (in this context) about nuclear is it makes lots of “waste” heat — and at temperatures pretty close to what you need for (multistage vacuum) distillation. That’s pretty efficient, and you get 2.5 to 3x as much rejected (potentially not-waste) heat as you did electricity, to do stuff like boil brackish or sea water (note again it doesn’t even have to be raised to boiling point, and the vacuum can be maintained easily with steam-jet pumps; actually those can make chilled water as a by-product too).
Big nuclear (and coal) plants can do that already too, but they need massive cooling right where they are; small modular reactors are aimed more at going (more or less) where the big loads are. This way, they can make their own fresh cooling water, and then some…
The big problem with “waste” heat schemes like “district heating” was always getting that heat from the (huge) power plant to the end-use point, cheaply enough to do you any good. With small modular reactors, the power plant just about comes to you.
LikeLiked by 1 person
After that big tsunami in the vicinity of Thailand back in what, 2004?, we had some aircraft carriers go over for assistance. I was told by a Navy friend of mine who was on one of those that the primary thing they did was desalinization, no doubt using the power plants they have on board.
Aircraft carriers make a lot of fresh water when they want to.
LikeLike
Ah, the Christmas Tsunami!
I am not the friend you knew, but — that’s most of what we did.
Yeah, we were a five star emergency medical hospital. A ton of resume workers. Food. A floating ER.
But our biggest effect was pumping out thousands of gallons of totally safe drinking water.
Later on, we organized liberty times by “who can use a big economic boost?”
LikeLiked by 1 person
Osmotic desalination seems like a good use for wind and solar. You can store large amounts of water cheaply, so you can run on non-dispatchable power and run your plant at 10% or so from the grid off-hours if your storage drops below, say, eighty percent. Assuming you have enough non-evaporating storage.
LikeLiked by 1 person
A solution I’ve heard is taking discharge from water treatment plants that is not good enough to make enviro folks happy with it being put into the river, and using that, then discharging after the secondary heat treatment.
LikeLike
At that point……… just pay a couple truckers to bring in some treated water.
And if this is Texas we’re talking about, sorry, but I have even less sympathy to water issues than normal: sure you are a dry state which would normally buy you a lot more sympathy, but you are also a state ON THE FREAKING OCEAN that prides — oh boy does Texas ever pride — itself on being pro business, had dirt cheap energy, and yet you aren’t running massive desalinization projects.
LikeLike
My region is 550-650 miles from the coast, 3500-5000’ of elevation.
Back in the 1970s there was talk about building 15-20 nuclear power plants to run pumps that would move water from the Mississippi River into Texas (supply the Metroplex) and then boost the rest of the water up here. It never went past the initial plans, for a gazillion reasons. It was going to be sold as “flood prevention.”
LikeLike
California and the remarks about the Columbia River, which by default includes the Willamette, Snake, and all the tributaries.
The remarks:
“All that wasted water getting to the ocean …”
Never mind that the Columbia River acts as a major water superhighway with the biggest port in Portland, 100 miles (80 nautical) upriver from the Pacific.
LikeLiked by 1 person
Given the scale involved (Texas being freaking huge), I suspect the primary cost there would be getting the desalinated water to customers, rather than the deal itself.
LikeLike
The outside loop of a cooling tower is the evaporative side, thus non-recirculating. (not directly anyway)
Many aquifers do not recharge locally. The input may be hundreds of miles away and no necessarily downwind.
Many western deep aquifers are functionally “fossil” water. They don’t meaningfully recharge on human lifespan time scales.
LikeLike
This is something that’s becoming a problem where I live. The regional aquifer filled up at least 12,000 years ago and recharges on a timeline in the neighborhood of one thousand years, as best anyone can tell. 120 years ago, it produced artesian wells. Those simmered down in the ’50s-’60s, iirc, and now the wells have to be hundreds of feet deeper to tap any water at all. A lot less water usage now than there was 20 years ago, even with a growing population, which is good…but the aquifer is still declining, and nobody knows how long it’ll last. It’s a very worrisome question. Anything that needs a big slug of water to get started probably shouldn’t (and won’t be) located out here.
LikeLike
Listening to some music while I prepare to plant flowers, and The Moody Blues is singing about this very AI subject…from a 1970 perspective.
“In The Beginning” from “On The Threshold of a Dream”
https://www.youtube.com/watch?v=5z_P_In0Dxw
LikeLike
I had that on the car’s music the other day. “You’re magnetic ink.”
LikeLike
AI AI AI AI AI Wipeout!
LikeLiked by 1 person
Just posed all of :checks note: ten minutes ago.
LikeLiked by 3 people
Go back up there where I said AI is most useful for leveraging the competency of highly competent people.
:facepalm: And THAT is (part of) why Tony Stark had JARVIS in the MCU as a virtual butler. You can’t get much more “highly competent” than Tony Stark when it comes to his strengths of business and tech. It’s when you get to his people skills that JARVIS hampers rather than helps because JARVIS has better people skills, and he’s an AI, dang it!
LikeLiked by 2 people
“Too, one of the big companies building stuff out here has already undergone “reorganization.” I have a sense of a looming crash that will leave the cities/counties on the hook for the sites, water, and so on. I hope I’m wrong”
Me too. I have the feeling that the data centers will be way overbuilt.
LikeLike
What if we are trying to close the Tofu dregs gap with China?
LikeLiked by 2 people
Fortunately they are still valuable after their builders die. And while H100s are more[1] valuable than when they came our several years ago I hope for a massive shutdown and selloff of Ampere GPUs onto the secondary markets.
[1]: seriously when does *that* ever happen in computing?!
LikeLike
I really like both the clanker and the song lyrics. However, listening to the song and reading your first paragraph led me down a peculiar rabbit hole. “I give it no stars and am trying to send it back,” for certain values of “it”, along with Charon’s barge and the whole dancing Greek chorus about life, led me to….
Well, just remember if it’s life you’re returning, you have to send it back in the original packaging, which can be…, problematic.
LikeLike
I write the lyrics. The music is clanker.
LikeLiked by 1 person
I know, and, if you excluded the very brief references to Elly, this could be a universal poem/song. Still the Greek urn motif of the video was very effective as well.
LikeLiked by 1 person
Oh, and of course there is the music prompting and sometimes wondering what Suno is on.
LikeLiked by 2 people
Sarah, I know this is way late, but I really, really, really want to see someone use this black figure animation to make “The Odyssey”. I’m a computer sub-literate fogey, but I’m presuming it could be made in someone’s basement for less Cristopher Nolan’s new movie spent on toilet paper. Any comment on how much your video cost and how long it took to make?
LikeLike
Oh. I could do that! It would just take me long.
Time, a couple of hours. cost: I pay the pro level on midjourney, which is $60 a month. Justifiable for covers alone. And yeah, I PROBABLY could do that, if I had time. It would just take me like six months, and I have books to write.
LikeLiked by 1 person
Thank you so much for your reply. I certainly wasn’t suggesting you do it yourself – your time is valuable.
I was just wondering about the cost in time and money of getting it done (by someone tech literate). Costs should be low – the original Greek script has been around for about 2800 years, and Wikipedia lists 40 modern English tanslations that should be in the public domain, so no worries about royalties. Use a clanker narrator and you don’t have to pay SAG.
I’m retired, so time is no problem. Currently, I can about use Windows 666 or whatever the current version is, but maybe it’s time to get off the stick and try to do it myself. Any suggestions on where to start?
Thanks again for your time. Now get back to work. ;)
LikeLiked by 1 person
I shall. It really is NOT arcane. And there are probably better programs than Midjourney. And honestly, if you can get someone to read the lines properly, the thing would sell itself.
I wish I were twenty (laments) and had time to learn the whole thing and see where it leads. :D I mean this movie making thing.
LikeLiked by 1 person
IK,R? My guess/hope is that within 10 years “Hollywood” will be a ghost town. Sturgeon’s law will remain in effect, but if you can make a feature film that looks better than the latest studio $500,000,000 *event* for $10,000, that 10% is going to be amazing.
LikeLiked by 2 people
My husband says investigate also open AI’s Sora and Google Gemini Studio. I should confess that Midjourney would not do black figure for me, and so I did those in gemini nano banana, using the subscription my husband has. I didn’t know they also had animation.
LikeLiked by 1 person
So I typed to Grok: you’re an app running on a bunch of servers in one/many places; describe the cooling technology used in one of those locations, with emphasis on water use
Thought for 5s
xAI’s Colossus supercluster in Memphis, Tennessee (the primary large-scale deployment powering models like me) relies on a hybrid direct liquid cooling system combined with facility-level water loops and cooling towers. This setup prioritizes high-density GPU cooling for hundreds of thousands of NVIDIA GPUs while addressing water use through reclaimed wastewater.
Server/Rack-Level Cooling (Direct Liquid Cooling)
• Servers (e.g., Supermicro 4U systems with 8 GPUs each) use cold plates on the GPUs/CPUs and 1U coolant distribution manifolds between servers.
• Each rack has a Coolant Distribution Unit (CDU) at the base with redundant pumps. This circulates a specialized coolant (often a water-glycol mix or dielectric fluid in closed loops) directly to the chips for efficient heat removal.
• Rear-door heat exchangers capture additional heat from any air-cooled components.
• Heat transfers from the internal closed-loop coolant to the facility water loop via the CDU. The internal coolant recirculates with minimal losses or replenishment. This direct-to-chip/liquid approach is far more efficient than traditional air cooling for high-power AI workloads. It reduces overall energy for cooling and allows denser packing, but still relies on water for the broader system.
Facility-Level Cooling and Water Handling
• Closed internal loops handle server cooling with recycled coolant.
• Open-loop facility water absorbs the transferred heat. Warmer water goes outside to chillers or cooling towers, where a significant portion evaporates to reject heat (typical for evaporative cooling towers). The cooled water recirculates.
• Evaporation drives most water “consumption” (losses to the atmosphere), not the liquid touching the servers directly.
Emphasis on Water Use and Reclamation
• Scale: Early projections for the site were in the range of 1–5+ million gallons per day (varying with load and expansion to 100k–200k+ GPUs, with plans for more). The open-loop portion specifically needs roughly 3 million gallons/day of makeup water to replace evaporation.
• Reclaimed wastewater focus: Instead of drawing heavily from Memphis’s drinking water supply (Memphis Sands Aquifer), xAI built the Colossus Water Recycling Plant (an ~$80 million facility). It treats effluent from the nearby Maxson Wastewater Treatment Plant using advanced filtration (e.g., ceramic membrane bioreactor) to produce suitable non-potable water for cooling.
• This diverts water that would otherwise go to the Mississippi River, protecting local drinking supplies and reducing aquifer stress. xAI pays for the treated water. The plant aims to supply the bulk of needs (targeting millions of gallons daily), with plans to share excess with nearby facilities like a TVA power plant.
• Benefits and trade-offs: Liquid cooling slashes evaporative losses compared to pure air-cooling towers, and using reclaimed water avoids potable draw. However, evaporation still occurs in the open loop, and expansions (e.g., Colossus 2) may rely more on municipal sources temporarily. Overall, this is more water-efficient and sustainable than many traditional data centers.
In short, the system combines cutting-edge direct liquid cooling for performance with smart water sourcing (reclaimed wastewater + closed loops where possible) to keep servers running reliably while minimizing impact on local freshwater resources. Other xAI locations likely follow similar principles but scale with local climate and infrastructure. Data centers evolve fast, so designs continue optimizing for efficiency.
For the primary Colossus facility in Memphis, the net daily evaporation loss is approximately 2.55 million gallons per day (about 85% of the 3 million gallons/day open-loop makeup water).
===
Why, it’s almost as if someone had asked the question before!
LikeLiked by 1 person
And a followup …
No, the xAI Colossus Water Recycling Plant in Memphis is not in full operation.Current Status (as of late May 2026)
LikeLike
Elon is smart enough to think things like that through. I has my doubts about a number of the other players, and not even Elon can think of everything.
LikeLiked by 1 person
That was a brilliant idea.
LikeLike
“…you will tell me that no, really, AI is all that and a jar of peanut butter. And maybe it is, but right now I think most of what it does is be a force multiplier for the highly competent.”
“A force multiplier for the highly competent” actually has a name, and has for a long time: IA.
For “intelligence amplification” — and what it really is, in the end, is far more an attitude towards the “AI” (cough!) group of technologies, and a way of employing them, than a different “thing.”
“Artificial Intelligence” sounds like (cue sound clips of “AI” boosters tallking this up) it’ll make us a mind-in-a-box substitute for human intelligence, and/or turn intelligence itself into something you can churn out from a “data center” like megawatts from a power plant — a product not a human attribute, measured in “gigawatts of compute” by power consumed as input, not results put out.
“Intelligence amplification” implicitly assumes a synergy between man and machine, so that you get a product measured in enhanced human productivity — like how much land a man can plow, vs. how much he can plow using a mule or four, vs. how much he can plow using an eight-gang behemoth behind a large-ish (only -ish, these days) tractor.
IA, like AI, might entice you to fire 3/4 of your technical writers. Unlike AI, it’ll never entice you to fire all of them at one go — that kind of stupidity it doesn’t amplify. (Compare and contrast “AI”.)
LikeLiked by 1 person
Data use increases demand more facilities for storing, and transferring data. While many don’t like them, I don’t see many people putting down their hand-held cellular computers to alleviate the demand.
I don’t know much about the newer, and proposed, data centers, except they’re being designed, and built, with somebody fronting billions for the costs. Maybe some will turn into money pits, but there must be more than just a theory for the amount of money being invested.
LikeLike
Never forget that not all rewards are physical. Is total surveillance capacity worth x amount of “investment”? And never forget that we haven’t yet fully tested how much surveillance is possible. Hint: more than you think.
This is a scene from David Weber. It’s also EXACTLY the blueprint we’re using with current AI in Iran.
24/7 data intake, married to 24/7 data analysis. Panopticon.
LikeLike
If this were about surveillance it would have been a few dozen built and then done. Barely even noticed in the constant builds by services like AWS or Azure which are needed for paying customers.
You know. Like the NSA already did decades ago.
Surveillance is an explanation which has the least possible justification. People who think it is the only possible reason demonstrate that they have no sense of scale for either the needs of surveillance, or of the economy.
Why do we need all these gigantic car factories!? I’ve never owned more than one car at a time. Why, they *must* be planning to throw thousands of cars at my house with trebuchets! It’s the only possible explanation!
LikeLiked by 1 person
You could at least try to respond to what I said in response to who I was saying it to. Original comment being responded to:
My response (emphasis added; I forgot you need crayons):
Both of those would say to anyone honest: There’s always factors involved we can’t measure, and we haven’t nearly seen all the ramifications of this technology. Move past the obvious.
And of course, I provided an example of what was considered science fiction, followed by a link as opposed to your assertions, to someone with an example in real time, and who’s had the expertise for meaningful input.
LikeLike
This is the “near future” approach that the two wildlife conservation officers, with their FBI cohort, take to determine something sketchy is going on in a particular warehouse district (built around a particular house basement, but they don’t know that) in San Diego area. Over the years, since end of WWII, large group of displaced people groups (NAZI, South African Whites, etc.) enter and never emerge. Certain elite members of specific corporate families associated with the warehouse district disappear for months, or years, into the area, almost like they live there … “Conquistadors” S. M. Stirling. Following up on this data mine causes a lot of consequences.
LikeLike
My current hypothesis (excuse me, while I adjust the tinfoil*) :
Both pro- and anti-psyops are run by the same crew of oligarchs who desire a techno-feudal future.
“Let’s you and him fight” is the modus vivendi of the current mass media and soft-power maneuvers.
I reasonably expect to be wrong, btw. I do not expect that the hoo-rah either pro- or anti-issues sweeping mass media will be what comprise our real wins and disasters.
We do (using the pronoun extremely loosely), after all keep falling for the feints.
(*Yes, I was wrong that the clotshot was not a Frankenvaccine funded by NIAID escaped into the wild from Wuhan’s CCP-level bio-safety protocols. Good story, idea, though)
LikeLiked by 1 person
When the NSA facility in Utah was floated, they told the public that the facility would pinky promise absolutely recycle the water, with a one time use of something like 30 million gallons. In a desert.
After it was finished, water rates skyrocketed and someone did a foia request, only to find that the facility was drawing something like 6 million gallons per month. And to rub salt in the wound, they weren’t even being required to pay for it. To get the contract, the city agreed not to charge them for the water. Which means they knew it was not going to be recycled.
Why should I trust that THIS TIME they’re going to do what they say?
LikeLiked by 3 people
Governments are not beholden to the population.
LikeLike
It is almost certain that some data centers are being built in places where the water and/or electrical infrastructure isn’t up to it. Some of those places will suffer because the authorities will screw up and not permit appropriate remediation. But many problems can be solved with small modular nukes and my hope is that data center demand will drive widespread deployment of the same.
I’ve been harping on this in many places (e.g. https://ombreolivier.substack.com/p/ai-mania?r=7yrqz ), but I find a really useful metaphor for AI (and for that matter any number of other tools) is whether it is used like a bicycle or like a train.
When you use a bicycle it makes your own efforts more efficient so you can go faster for longer that you can on your feet, but you are still providing the base motive power. If you stop pedaling you stop moving (ignoring downhills and even those are not infinite).
On the other hand when you use a train you just pay for the ticket and hop on. No further effort is required. You are contributing nothing to the process and have no control over it beyond deciding when to get off.
AI as train is bad. AI as bicycle is “leveraging the competency of highly competent people”
LikeLike
Personally, I find that jumping from a moving train is unhealthy….. 8-)
LikeLiked by 1 person
That is another problem with trains.
Mind you jumping from a moving bicycle also hurts. I have personal experience of this
LikeLiked by 2 people
There’s a metaphor for AI in there I just know it. 😁
Wilder’s AI piece on Wednesday is called “Your AI is cute. Theirs is a chained God”
IFF he’s right, the quasi-eco-disaster power-and-water needs of the new toll are a feature, not a bug.
To be serious, if this new tech is anything we can influence, rather than merely adapt to (and we had best be doing that) here’s a possible angle of approach: How did the liberating power of the OG internets – we were there, right? Listservs & dialups replacing mimeo-graphed fan rags – get subverted?
Ideally the pro-AI people should be going hard on the potential disasters and the Anti-, vice-versa.
LikeLike
William Lehman had part of his recent substack post on AI and concerns. (I’m taking some of his advice/dismissal about surveillance with a small grain of salt; there’s a lot of data that can leak, and curtailing anti-social media is only a part of it.)
Link to it: https://wlehman.substack.com/p/useful-idiots
LikeLike
That may be the best use for LDMs, but I suspect the most common use will be as force multipliers for the abysmally stupid and incompetent, churning out gigabytes more of their bullshit. Sturgeon’s Law applies here too.
LikeLiked by 1 person
Somewhat off topic (but only somewhat, given Space X owns xAI), multiple sources saying that Blue Origin’s latest New Glenn rocket just blew up on the launch pad during a static fire test. See for example:
https://xcancel.com/visegrad24/status/2060167627637465508?s=20
(and simply flip it back to x dot com if you want the original-original). Damage to the pad may be quite extensive, it was pretty much one big fireball; and they only (had?) the one pad operational.
Blue O. has been trying to pick up the pace recently, obviously this will not help. Also not sure if they’ve lost (another) customer payload in that ‘anomaly’ there.
LikeLiked by 1 person
If they’re still doing static fire tests, I would hope no payloads are involved other than packaged sand (maybe) to simulate the weight and mass of a future payload.
LikeLike
Yeah, rocket engines are very hard to make work properly at all.
The AI side of whatever it is that Musk thinks he is doing, and the rocketry side are not fundamentally distinct.
You don’t plug an LLM directly into an airplane, and prompt it to have fun.
But the software side of aerospace engineering is not purely straightforward and done correctly once thirty years ago and reused without changes ever since.
There are a lot of interesting questions in that area. Beyond that, the non-software side can produce oodles of data to reduce and to try to draw sense from.
Frankly, if Musk had started up the AI company solely to avoid being screwed by the other AI providers once it became clear how wild things are, it would not be the least sensible thing that he has done.
LikeLiked by 1 person
It looked like anything that stood above ground was leveled or obliterated.
Word was, no payload was on board. They were perhaps overly confident, but not that arrogant.
The X videos on Insty were awful (including Awe-full in that word). Yikes!
LikeLike
A serious case of kinetic disassembly. (Or a lot of parts suddenly leaping into Brownian Motion.)
LikeLike
OT, Jeff Bezos Blue Origen rocket blew up on the pad tonight. Noone was hurt, which is pretty miraculous when you see the video. If I had seen it in person I would have had hysterics. The mushroom cloud, don’t y’know. (Yes, I know mushroom clouds don’t automatically mean a nuclear explosion, but I’m a child of the ’60s).
LikeLiked by 1 person
No flash, no heat pulse, not much of a bang really, indicating a large fast burn rather than an instantaneous HE-like explosion, just dramatic nighttime conflagratory (totes a word) big badda boom graphics. Dramatic enough that apparently miles and miles and miles of folks up and down the Space Coast called 911 as a result.
My concern has always been a ground strike of a meteor big enough to make all the nuclear detonation effects – flash, heat pulse, mushroom cloud – resulting in public panic. Eventually the experts would be on the news say stuff about “no radiation detected” but that would be too late. It would be nice if someone had produced one thing from H’wood over all these years making the point that not every flash-bang-mushroom cloud is automatically a nuke, but that story wasn’t pot-stirring scary enough during the Cold War, and then the USSR fell, history ended, and nobody cared anymore.
LikeLike