I have a lot of left leaning non-tech friends. They are all very much against AI data centers. I've built AI data centers, and I keep trying to educate them, but they just won't listen.
The water is mostly part of a closed system. Sure, you need a lot to build one, but once it's running it's pretty water efficient. The lack of green power is a problem, but as the article points out, you now have a large customer subsidizing the base cost of delivery, since they are only one delivery point, and they are financing new sources of power, which if properly regulated would be green power.
The noise is an issue, I will admit that. As is the fact that a lot of the power is coming from non-green sources. But the government could easily fix these problems with regulations that apply broadly, not just to datacenters.
Besides the data center itself, many people in the US are against what the data center is bringing: more AI.
Most regular people have not been sold on the idea that more AI will be good for them or their families or communities personally. Instead all they hear is AI taking away jobs, and AI preventing high schoolers and college kids from learning how to read a book and write an essay. Most people think a world with more AI seems bleaker.
So people wonder why communities don't want this hulking, loud, energy-expensive data center whose main benefit is to hasten the arrival of this world they don't want?
How is this any different to the invention of the cotton mill?
You don't get to pick and choose what productive technologies to adopt. You either adopt them, or you'll be paying the countries that do and you'll still lose your domestic industry to competition.
It's not about whether it makes rational economic sense, it's whether people are happy about it.
It could make economic sense to be the world leaders in the next drug that turns everyone into mindless worker drones ("if we're not making the factories that make the drugs the Chinese will!") but people still won't be happy about what it means for their futures.
There’s no need to mince words, they’ve been aggressively told that AI will eliminate nearly every job.
If you need your job to feed your family then what Sam or Dario is saying is quite literally a threat to kill your children.
Of course they’re full of shit, it’s more like a very useful business automation project but if you say stuff like that you can’t be surprised when people hear it and form highly rational opinions about it.
I don't understand why it's being fought by left-leaning, many self-proclaimed communists that have been fighting for a world where we just hang out and make shitty art all day.
Isn't AI the only way we get to that? Who is gonna do actual important work otherwise?
Because (a) that vision of utopia is not party of any "self-proclaimed communist" philosophy that I've ever heard of, and (b) no one believes that's what's going to happen.
Instead the evidence of our own eyes tells us that AI seems to be worsening the K-shaped economy: a handful of Elon Musks will become trillionaires, a slightly larger number of engineers will become billionaires, and the rest of us will have fewer jobs and fewer prospects of upward mobility in a world where AI can do most entry-level work.
Trying to educate them that water and power usage are the less problematic parts and not as bad as everyone thinks. It's the noise and lack of green power that are the problems.
Can you please provide us the details how much water an open-air, evaporative cooling system loses due to evaporation and other system-wide losses?
Also can you tell us which chemicals need to be added to the outer and inner loops of the systems, and do the inner loop has the same losses or it is different.
If there's water loss into the ground, and if the water is treated with chemicals, is it required or is it in fact treated to remove these chemicals so the used water can be reused for livestock, agriculture, or non-consumable use in the cities like for watering greenery or other processes which needs water but can accept non-potable variety.
I also heard that there's a closed loop version for outer loop. Is it used widely in US, if yes, what percentage of the large datacenters use it? Because as I heard it, it's more expensive, but doesn't lose water as much.
Adding some references if possible is greatly appreciated so we can dive deeper, as the community.
Same is also valid for power usage metrics. I'm especially interested in how a datacenter full of 800VDC systems (aka Megawatt Rack) can be fed without any difficulty with our current power generation and transfer infra.
If AI succeeds at replacing most white-collar jobs, there will obviously be much less commuting.
If it doesn't, we can all stand around and make fun of the people who paid terabucks for the data centers, and enjoy a bountiful harvest of surplus hardware for pennies on the dollar.
Commuting for work with a car is mostly a US problem. In countries with proper mass-transit, many people don't need cars. I personally have to use car, because the city I live is very hilly, and don't have the fast mass transit options which I can use in many places of the world.
OTOH, considering the proliferation of the EVs, greenhouse gas emissions from cars are going down, alongside brake-pad related emissions since generative braking covers 90%+ of the daily braking needs.
OTOH, many datacenters are installing their own greenhouse emitting power gensets. So, with a bitter contrast, as AI is replacing jobs, it's also filling back the carbon footprint vacated by cars and more efficient modes of transport at the same time.
These GPUs won't be scrapped soon, because besides AI, they are useful for accelerating not-so-precise floating point calculations en-masse. So, no, it won't work as you imagine.
Commuting for work with a car is mostly a US problem.
And here we are, talking about the US. Great, that's settled.
OTOH, considering the proliferation of the EVs, greenhouse gas emissions from cars are going down
In the US, most electricity comes from carbon-based sources. EVs help, certainly, but it would still be best from an ecological standpoint if people didn't have to use them every day to get to work.
OTOH, many datacenters are installing their own greenhouse emitting power gensets.
Exactly, which is why we should work for grid renewal and expansion at the same time we build these things out. If the hyperscalers insist on building their own power plants, those power plants should be required to be (a) carbon-neutral; and (b) grid-tied.
These GPUs won't be scrapped soon, because besides AI, they are useful for accelerating not-so-precise floating point calculations en-masse.
Really. What are some other applications for not-so-precise floating point calculations at the exaflops level?
> And here we are, talking about the US. Great, that's settled.
Proliferation of law skirting AI datacenters is a worldwide problem. Many companies are extending this to Europe as well, so AI datacenters is not a U.S. problem only.
From what I see in Electricity Maps[0], it's not clear cut. Some of the states have well over 50% of their generation is from carbon-free sources. As a general trend, west is greener than eastern US. So, no, EVs will have an impact there. However, you need electric semis a lot as a country since rail is not used a lot in the US as well.
> Really. What are some other applications for not-so-precise floating point calculations at the exaflops level?
In short, HPC. As a more detailed, by short list:
- Weather and climate modeling.
- Drug discovery.
- Computational Fluid Dynamics.
- Material science, behavioral and process analysis.
- All kinds of large scale simulation which can be modeled with matrices.
- Any kind of problem requiring linear algebra.
While I generally work at full 64bit precision, from what I see, even though these modern GPUs do only support up to FP16 (or FP32 in some cases), they can be used for higher resolution simulations as well. We support some astronomy and climate researchers which use classical algorithms and in-house ML models (not LLMs and such) to accelerate their simulations and predictions on multiple GPUs.
I manage HPC clusters, and sometimes develop code on them, so yeah.
OK, I see I should have used "trillions of dollars worth of hardware capable of performing not-so-precise floating point calculations at the exaflops level" as the basis for my question. The hyperscalers are making a singularity-or-nothing bet on this stuff; there are no other uses for this much hardware of this type. If there were, such hardware would already be in place.
(IMHO, the fact that AI models tolerate absurdly-low precision means that we're doing something wrong and wasteful in our entire computational approach to the field, but that's neither here nor there.)
Except when some rich clown from another country comes into a drought-stressed desert area where residents have already been under increasingly draconic water usage restrictions for goin' on a decade now, wantin' to build a datacenter right on a lake that's already under severe stress with record low water levels, tellin' everyone it won't have any effect on anything and it's all just rainbows and sunshine and nothin' but benefits for everyone all around (lies). But yeah, no worries. We can surely trust the billionaire wouldn't lie to us to enrich himself at our expense.
Given your experience with data centers, what do you think of this quote from the article?
> In any case, even the noise worry is overstated: many viral videos of supposed AI data centre noise are really Bitcoin mines, not state-of-the-art hyperscaler facilities.
Are there really major differences between a hyperscaler, neocloud, and crypto mining data center in terms of noise pollution or other negative effects?
No, there really isn't. It's more a question of how the data center is cooled. There are quieter more expensive options, which is why they often aren't used.
Interesting. What ways? Suppose I wanted to call the governor and say “I have an answer to how to regulate these that solves the noise problem” what would we be asking for?
Regulations about 95th percentile decible levels at certain distances, and regulations about max decibel levels at certain distances, for all buildings.
Sure. But what’s the technical way or meet that requirement? Is there a specific method or is it just that they can but they don’t care because there’s no law?
> But the government could easily fix these problems with regulations that apply broadly, not just to datacenters.
Even if you can get legislators to pass regulation, enforcing it is a constant battle, and odds are they'll just pay a fine and keep on violating the regulations.
It's a much easier fight to keep a DC out of your community to begin with.
I'm saying the people who are affected need to lobby their own local governments, because the companies will never do the right thing until they are forced to do so.
But what people are most upset about are water and power, which are the two least problematic things about datacenters.
No thanks, they are loud. I even called that out. But no one should be living next to them. Those people need to get their local government to regulate noise from all buildings. The companies won't do it on their own.
(This is part of why I no longer help build them, because when I asked about reducing noise and using green power I was laughed at).
They had an article about Australia embracing AI data centers to be suppliers of the future and not just customers and one big pull is "the vast amounts of land in Australia" then said that one of them is being built within 100m of the nearest suburban house and 200m of a school.
It seems with all the other battles to fight, that moving it 2km down the road and surrounding it with trees and parkland would knock the noise argument out of the window. At least in the US and Australia that really do have vast amounts of open space.
> Those people need to get their local government to regulate noise from all buildings
I think this is the core of the 'moral panic'. it isn't anti-AI, it's the fact that many AI datacenter builders have been somewhere between surreptitious and intentionally illegal in their avoidance public knowledge of what they are doing, most likely to avoid exactly the regulation you're recommending. As a result, there's now categorical public distrust of both the industry and the local government that has shirked their obligation to the public welfare.
It is reasonable to conclude that you don't act as secretively as these builders have unless you have something to hide.
Great point, we shouldn't do jet engine testing either because obviously we need to put that smack dab in the middle of your local cul-de-sac and it can't be put anywhere else.
Show me actual statistics that a supermajority of new and proposed datacenters either are currently using (for new), or plan to use, with clear financial projections for how (for proposed), closed-loop cooling systems, and I'll consider conceding this point. Until and unless someone can do that, you can point to the option of such systems until you're blue in the face, but it's meaningless.
The noise is the main issue. I'm not american, but one of my relatives works in bernie sanders's office, and one of my friends (30ish) is involved somehow in the democratic party not sure how exactly. And what I hear from them is that the issue is mostly the noise only. The water aspect was there but that was mostly disruption during construction.
They are powering it with turbines right, so if its next to your house it sounds like you're permanently in on the airport tarmac. Not great. I saw a video (not on the internet, but from them). It was pretty annoying. But it makes sense, there is no way a power delivery system in a developed country adapts so fast to such intense energy needs. Here in india, many 250Mw datacenters are put in places where a proper grid connection is itself coming for the first time, so the same problem is not there. And frankly there are way more noisy industries in poorly zoned mixed-industrial-residential areas here for generator noise to even matter.
> The noise is an issue, I will admit that. As is the fact that a lot of the power is coming from non-green sources. But the government could easily fix these problems with regulations that apply broadly, not just to datacenters.
Soooo
You're building them, and you're agreeing that the biggest points of criticism are valid.
But.. some handwaving that the government should do something about it and also others are too doing bad things.
I don't do that anymore since I got laughed at by coworkers when I suggested we focus on using green power and reducing noise. So yes, my conscience is clear.
You might just need to change how you say it. These people have a lot of legitimate frustrations. Show them you are on the same side as them and understand where they are coming from. Don't get hung up on minor technicalities. When people are upset, they really don't want a lecture or pedantry.
Talking to just a few people or one person at a time can help a lot, also.
But I hear you. It sucks trying to be helpful and then getting it shoved in your face.
I've been meaning to switch my in-laws to Quad9 and just not mention it to them. They fall for too many scams, I'd love for those websites to just not work anymore.
Banks can't just ban you, nor just take your money. The price they pay for being gatekeepers is a ton of regulation.
This is why PayPal for the longest time made it really clear that "we are not a bank". So that they wouldn't be subject to those regulations and could suspend accounts and keep the money.
I think in this context the concern isn't that you'll lose the money, but your loss of all legal/practical authentication mechanisms.
In other words, you leave the bank-branch with a big sack of cash... And then your car is impounded because you can't prove you're the owner or your license expires, and you'll live in a tent in the woods because landlords can't verify your credit/rental history, etc.
Banks can just ban you if they whisper the magic words "suspected money laundering". They are theoretically accountable to a process that takes 500 years and 3 quadrillion dollars to reach a resolution, then they will unban you. You might notice both you and the bank will be dead by that time.
I actually know someone this happened to. While they did ban him from using the bank, they first warned him, gave him a chance to close out his open transactions, and then helped him move all the money somewhere else at no cost.
The important part is that they can't just close your account and keep everything.
No it isn't. I've put human written material in there (my own unpublished material) and it says 90%+ AI. Then I put some AI material in and it said 30% chance of AI.
I only tested it with the two pieces, but was not impressed.
I'll take your word that it's human written, but reading that gist, it absolutely reads like Claudeslop / GPT slop, it doesn't surprise me in the slightest that Pangram flagged it as AI when it reads identically to LLM output - this feels like a very acceptable edge case to me (assuming you are telling the truth).
Are you absolutely sure you wrote this by hand? If so it's kind of remarkable how close to an LLM you write like.
It’s important to remember that LLMs were trained on well written human text. People who write well are going to sound like an LLM. Especially if it’s a marketing message for a website.
I’ve been accused of being an LLM multiple times here on HN too. I know you have no way to know for sure other than trusting that I’m not using an LLM to write. But it’s pretty frustrating that people jump right to LLM accusations.
Today Pangram says it is 100% human, which is correct. But yet I got multiple DMs when I posted it 8 months ago saying "stop posting AI slop!" in response to that comment. At the time, Pangram marked it as 50% AI.
That comment doesn’t read like slop in the slightest to me, so the people DMing you have a bad eye for it. Regardless, I think your ‘human’ sample is not a good indicator of the quality of pangram. I would update your priors a bit.
I have 1000s of reddit and hacker news comments from before 2023, and lots of long form writing too. But as you point out, those are all in the training set, and in pangram's "definitely human" training set too.
The ones that sound like LLMs tend to be the ones that were well researched and spent more time on, not the off the cuff stuff, which is most of what I write. So it would take me a while to find something like that.
But you're welcome to dive into my reddit and HN history, or all my blog posts on the wayback machine if you want to look for one. :)
Being that Panagram lies about its effectiveness in their presentations while hiding it's actual capabilities and testing methods rather deeply I have little faith in it. (1 in 10000 wrong in presentations verses 2-4% wrong in testing).
For example they have a corpus of older pre-llm text and use that as the example their current model doesn't misclassify human written text. It shouldn't take much thinking to realize why this is a fucking stupid benchmark.
Every day humans use LLMs and read LLM content Panagram becomes more useless because it forces languages to have a stopping point sometime around 2020. If you adopt any LLMism or are one of those unlucky people that already talked like an LLM before LLMs then all your shit is getting marked even though it was created by the human mind and written by human hands.
Claude was really far ahead of GPT in writing from 2-4, but the later models have started to get an overly distinctive style, wheras these days GPT tends to be coherent but fairly concise and dry.
Claude is the one model family I've not really used. Which yeah, feels like a backwards thing to say in a world where seemingly everyone using LLMs is using Claude Code.
I've used some Opus 4.5/4.6 via Antigravity and Sonnet by the web chat. I'm torn because as far as LLMs go, it does feel more ... "literate".
But maybe too literate, judging by how many people are complaining about "Claudeisms". I suspect Claude would be just as susceptible, if not more, to the sort of ... "moralizing" that GPT seems to gravitate towards (for lack of a better term).
There is a lot of competition in this space, both commercial companies and open source. It seems like the biggest advantage for using one of these and not Claude or Codex is that you can use multiple models. Otherwise, those two have all the same features, or probably will in the next week.
Are you able to use multiple models? Or I guess the first question is, what model(s) are you using? At this point, many enterprises care so that should be front and center, especially if you're using models hosted in China, because a lot of non-Chinese companies care about that.
You mention some things about unique ways to store memory. Do you have any data or use cases on how that improves performance? If you do, you should get those on your website too.
I'm not sure you have a very strong answer to "Why not just run Claude or Codex on my laptop?" Sure, you provide compute, but so do they, and most enterprises and even regular people are getting desktops/droplets to run their agents now because of this problem. This is kind of solved already.
Also there is a small nit with your website. The graphic next to "Proactively gets things done for you" gets bigger and smaller, so if you're trying to read anything below that, it keeps jumping up and down.
Currently we don't allow for multiple model selection, but it is on the roadmap. We do have a cli, which allows you to use the wiki layer and the tools in whatever harness you prefer. Further, we want to allow users to also bring their own keys/subscription.
We do not have formal benchmarks to measure improvement in performance. A big reason is that most benchmarks are very "atomic facts memory" focused. They'd give a huge passage, conversations, and see which memory system could surface those facts. How we're different is agents could be simply given the markdown files, and they can pretty much get the perfect context themselves. Second, we are banking more on the fact that having this layer enables new capabilities of the models
Its GPT-5.6-terra under the hood. Also we want to control the harness. Features like proactivity/long tasks are not possible the way Claude/Codex work today. We do plan on releasing our desktop app, which would have these things inbuilt.
My last few jobs I started with a listening tour. I talked to everyone who was relevant to the job and asked them:
- what is going well,
- what is not going well
-what do hope that I will fix
-what should I do
-what should I not do
I put those question in the agenda for the meeting invite so they could come prepared. Then we took the conversation from there.
I put it all into a google doc, then synthesized it into summary that hid the identities of the people who said it and used it as my guide for the first few months.
This is pretty well trod legally. By the user taking a specific action, it makes the user responsible. Comma.ai uses the same technique. Technically their self driving hardware comes with no software. You have to click a button and point it at their GitHub to get the software on it, but it's automatic after that.
Their profits are incredible too, they have 62% net margin! Last quarter $60b profit on $96b revenue. Since the "AI bubble going to pop" terrible takes 2 years ago, NVDA has made $300b in profits.
There's a specific type of person who makes a repeated joke like "sell shovels in a gold rush" - they think they have made the most insightful comment possible.
The water is mostly part of a closed system. Sure, you need a lot to build one, but once it's running it's pretty water efficient. The lack of green power is a problem, but as the article points out, you now have a large customer subsidizing the base cost of delivery, since they are only one delivery point, and they are financing new sources of power, which if properly regulated would be green power.
The noise is an issue, I will admit that. As is the fact that a lot of the power is coming from non-green sources. But the government could easily fix these problems with regulations that apply broadly, not just to datacenters.
reply