Then, sorry, but you don't understand today's NAS market. All it takes is looking at the products QNAP, Minisforum, TruNAS, UnRAID and many more have been putting out for years - and it sells well. Nobody wants more infrastructure than they need and a well powered NAS is a great place for homelabbers and SMB to invest.
I just recommended an N5 Air [0] to a friend because it's got a powerful CPU, can expand to 96GB of RAM, has a great mix of NVME and HDD and has a PCI slot they can run an Intel A310 in for local transcode. It's the perfect Proxmox host to run everything they'd want to do including local services but also fault tolerant storage for a reasonable price.
But that's not what they want to see. They want to see that AI is a savior for all. The psychosis of what AI is, especially in these comment walls, is unfortunately and spectacularly real.
Anyone who's done any amount of investment can see this through the lens of history and I'm right there with you. This is the Lehman Brothers stage of the game and SA could be that "one". But Bro, we're still up 80%! - even WSB isn't filled with this level of ignorance.
A fund went up 400% by making a few highly leveraged, concentrated bets. The market moved against its leveraged positions and it was forced to quickly sell to a rival fund, at only an 80% profit.
Nothing about that is meaningfully the same as Lehman Brothers other than the most superficial aspect of "A financial thing went down fast"
Honestly I have no idea what you are even replying to. My comment was essentially "The fund bet big and lost, but don't read it as him going bankrupt" as I saw many people interpret it that way.
I find the emotional charge in your message quite strange, and the "psychosis" might be you projecting something.
"They" don't "care" about anything. It is a stateless computational run across thousands of semiconductors. There is no objective this software has other than the computational function completing. To care would mean the model would have a level of discernment that goes along with sentience.
The supposed "magicality" of human consciousness is hard to reconcile with evolution -- where on that long path from amoeba to homo sapiens did God (PBUH) implant our connections to the soul realm?
Even evolution-deniers run into difficult-to-justify self-contradictions when trying to explain how the supposed supernatural part of our consciousness is both affecting the natural world and affected by the natural world, but not part of it.
Dario is more of a threat to the US, in terms of advancements in AI, than China. In Dario's mind anything that can't be controlled competitively is a threat to Anthropic, so he positions his FUD strawman so that Dario doesn't have to worry about the competition. And then he can artificially inflate token costs so his IPO can happen. Dario doesn't actually care about ethics, alignment or availability of LLMs - he just likes to use those words to sound like he does. Yet we've all seen how Anthropic actually acts vs what they say.
The scary part very few are talking about is that every compute device is Turing complete. So everything from the phone in your pocket to a DGX Spark is a threat to national security now since, technically, every device can run any model (how well is not a question of concern when you start to argue hardware should be gated just the same as Dario likes to gate models). I mean, along these lines of thinking Linux should not be available to the masses! What if someone runs some code that's not approved by the benevolent dictator for life, Dario? People will say: that can't happen, but the reality is it already is. If everyone has reasonable access to compute to run models that are mostly capable comparative to burning Anthropic tokens, why wouldn't they? It's risk reduction and price protection. Yet we can't buy those systems because of future production already being purchased by these organizations.
But back to the models themselves... We played this game with Metasploit back in the day: many who had no clue claimed exploit tools should be regulated and only available for use by those blessed, illegal elsewhere (I believe the closest this got was the Wassenaar delegation in the US, but only through collateral inclusion of "cyber weapons "). Except in that timeframe the authors of these tools weren't advocating for protection. Today the world is fine, systems improved because of security FOSS tooling. The same thing will happen with LLMs. Unless, that is, Dario gets his way. I'm not a fan of Altman but I think he's standing back watching this play out knowing what Dario is doing: either he succeeds and OAI benefits or Dario ends up the Chicken Little of AI and Anthropic fails to launch (their IPO).
The reality is Dario is only doing this because this is a real risk to his business. China's constraints in building competitively have given them an advantage: they are doing more with less. And if you think that their distilling from US models was in any way anti-competitive or illegal, then I guess maybe "deal with it", much akin to Anthropic, Google and OAI's response around taking the (copyright) content in the first place with no repercussions.
People who don't work in the AI bubble don't care at all about any of these people. They could all be gone overnight and the world would continue to innovate, probably in a much more productive manner, without them.
> And if you think that their distilling from US models was in any way anti-competitive or illegal, then I guess maybe "deal with it", much akin to Anthropic, Google and OAI's response around taking the (copyright) content in the first place with no repercussions.
Exactly. The cries in favor of distillation regulation from the US AI companies ring hollow and fearful.
OpenAI and Anthropic didn't realize that distillation was going to be so (a) effective and (b) un-technically-stoppable at scale.
Now they're seeing their IPOs at risk and clutching at governmental straws.
Dario's argument is transparently working backwards from {protect Anthropic's economic model} <- {need government regulation} <- {justify government regulation via AI fears} <- {we love open models, but so sorry they can't pass regulation}.
If the rise of the web in the 90s taught us anything, it should have been that companies that take economic reality as it exists thrive, while those that predicate their value on regulation fail.
If distillation at scale works and is technically feasible? That's reality. Deal with it.
This is akin to saying "we replaced the tires on the car and the average user didn't notice". If the entire car was replaced, then maybe we get a bit more excited.
However if "replacing small sections of code" is heralded as "the most amazing achievement in software development" then we've all lost the plot a little bit somewhere.
It makes sense both ways, no?
The reality is, of course, most people aren't going to notice. If the functionality and performance is 1:1 why would they? Is it impressive? A bit, but not how you're positioning it. People seem to forget LLMs are good at what they've learned from training data. And the LLM is good at compressing time. The only notable thing about these types of marketing spins is that a rewrite was accomplished in a small time frame that was hard to pull off before LLMs. The actual act of the move is less so.
Did the codebase improve? Is it more performant? I've seen nothing to really stake those claims with any objectiveness. Lastly: what was gained?
We're building more data centers for a technology that's ramping global warming even faster. Water consumption, heat generation and tons and tons of CO2 emitting concrete to house these things.
So we're doing something, but those things are counterproductive to our planet's general well being. Apparently billionaires don't need an inhabitable Earth to live on.
Growth trend matters more than absolute present value.
Data centres generally had been seeing energy use growing at 12% annually, or with a doubling time of 5.8 years (about 5 years 10 months). AI is accelerating that trend.
In the United States, power consumption by data centres is on course to account for almost half of the growth in electricity demand between now and 2030. Driven by AI use, the US economy is set to consume more electricity in 2030 for processing data than for manufacturing all energy-intensive goods combined, including aluminium, steel, cement and chemicals. In advanced economies more broadly, data centres are projected to drive more than 20% of the growth in electricity demand between now and 2030, putting the power sector in those economies back on a growth footing after years of stagnating or declining demand in many of them.
>Growth trend matters more than absolute present value.
Only if the current absolute present value is enough that the resulting trend matters.
It was only 2 weeks ago when someone posted an article claiming data centers worldwide use 24x less water than golf courses worldwide.
And I mean, you need somewhere to collate all that climate data? You need telecoms to let scientists communicate?
>Driven by AI use, the US economy is set to consume more electricity in 2030 for processing data than for manufacturing all energy-intensive goods combined, including aluminium, steel, cement and chemicals. In advanced economies more broadly, data centres are projected to drive more than 20% of the growth in electricity demand between now and 2030, putting the power sector in those economies back on a growth footing after years of stagnating or declining demand in many of them.
If horse use trends continue at their current pace, London will be covered in 20 meters of horse manure by the end of the decade.
> In advanced economies more broadly, data centre's are projected to drive more than 20% of the growth in electricity demand between now and 2030
20% of the growth isn't much overall either is it? That's likely assuming that the million and 1 spruiked projects actually get off the ground.
Exponential growth, a/k/a constant periodic percentage growth (per day, month, year, etc.), increases rapidly. Differences (or errors) in magnitude measurement disappear rapidly: a factor of two in a single generation, a factor of ten in about three generations, a factor of 100 in less than 7, a factor of one thousand in fewer than 10.
What matters is that the growth rate continue.
Which of course, it usually doesn't. Most growth trends follow a sigmoid rather than exponential curve (initially), or more complex longer-term dynamics.
The point of your infamous horse quote was less to show that the Earth would be covered in horses (or their effluvia), than that the observed current trend could not reasonably continue. And as some of us are aware, it didn't, due to the adoption of ... the petroleum-fueled automobile ... which has some bearing on this particular HN post itself.
Had the automobile not come along, or petroleum reserves proved much more limited as was widely believed at the time[1], large horse-congested cities such as London, New York, and Chicago would likely have seen sharply slowed or even reversed growth under negative hygiene pressures. It was the multiple factors of growth in fossil fuel use (displacing horses, and their effluvia), development of sewerage systems, solid waste disposal, food safety and purity, refrigeration, and public health which turned cities from death zones to attractive living places, rather than merely drawing desperate immigrants through high wages amidst rampant rural and foreign poverty.
But the point remains that the sustained growth rate matters far more than the starting point.
2030 is three and a half years away. That's soon.
________________________________
Notes:
1. It wasn't until the 1930s with the discovery of the East Texas Oil Field in the US and of vast fields in Saudi Arabia that it seemed likely that oil was more than a short-term resource. In marked contrast to, say, the short-lived Indian natural gas boom of the late 19th and early 20th centuries: <https://en.wikipedia.org/wiki/Indiana_gas_boom>. Daniel Yergin's The Prize is an excellent history of the industry's growth, though I disagree strongly with the author's strong allegiance to it.
>Exponential growth, a/k/a constant periodic percentage growth (per day, month, year, etc.), increases rapidly. Differences (or errors) in magnitude measurement disappear rapidly: a factor of two in a single generation, a factor of ten in about three generations, a factor of 100 in less than 7, a factor of one thousand in fewer than 10.
Exponential Growth in datacenters is extremely unlikely.
I doubt it could be sustained for half a generation before Colo became cheaper than air.
>But the point remains that the sustained growth rate matters far more than the starting point.
If you add "Sustained" in sure, but its unsustainable, leading back to the starting point mattering more than the rate.
Theres not enough RAM for now. You double colo space worldwide, and there's simply not going to be computers to put in there. That's just the canary, there's indicators of other component shortages. But like, pretending this is exponential will quickly burn out other things. Steel. Powder Coat. Fibre.
Not to mention that funding for data center expansion is already drying up and projects are being cancelled. I honestly dont believe in a single local project that isnt branded NextDC or Equinix. Outside of them I wouldnt be surprised if 90% of projects failed to materialise.
I dont know what world you live in where you think theres an exponential Data Centre growth going on but its not this one. We are already up against the wall. I guarantee you will have a new industry you just learned about to get anxious about by 2030. Just how Bitcoin didn't exponentially grow to consume the whole planets energy supply by 2025 as predicted.
Again: the point remains that the sustained growth rate matters far more than the starting point.
And again: it usually doesn't. Most growth trends follow a sigmoid rather than exponential curve.
You're arguing against points I'm not making.
The point I am making is to keep an eye on the growth trend (and whatever might constrain that, you've given some credible concerns here), rather than the starting magnitude.
The question of how much societal resource information services can plausibly consume over a long term is an interesting one. Looking at biological analogues, the human brain accounts for roughly 25% of our own metabolic output, which I suspect is the present record.[1]
I get 8% for total present use in commercial computing, with the US EIA projecting a pretty steady growth to exceed all refrigeration, lighting, space cooling, and ventilation, by 2050, at about 350% of present levels. It looks as if that analysis presumes a 5% annual growth rate. The full report does incorporate AI in its considerations.
We assume AI servers will increasingly skew more energy intensive, the installed stock of AI servers grows exponentially through at least 2040, and computational efficiency will increase over time. In our High Electricity Demand case, we assume the exponential growth in AI servers will continue through 2050, and the rate of improvement in server computational efficiency will reflect historical trends.
1. Apparently among vertebrates typical metabolic fraction is between 2--8%, though there may be some tree shrews which approach humans as a percentage if not total CNS metabolic magnitude. "Ratio of central nervous system to body metabolism in vertebrates: Its constancy and functional basis" (1981) <https://www.researchgate.net/publication/15916066_Ratio_of_c...>.
Socially, our computational energy is on the high side for vertebrate animals.
I don't either, but it's an extant one, and one in which we can collect a fair bit of largely-unbiased data reasonably immune to short-term manias, unlike, say, stock markets or technology hype cycles.
Over about 4 billion years, and under immense selective pressure, maxing out energy expenditure (biology's chief currency) at roughly 25% of the whole to information processing (input, storage, processing, output) seems to afford maximum benefit. It's also quite possible that genus homo's brain size has decreased over the past 200,000 years or so; h. sapiens neanderthalensis had slightly larger brains than modern h sapiens sapiens, though that size may not have been as intellectually or linguistically capable.
The other possibly interesting data series is of the magnitude of impact of information technology on social capabilities given rates of improvement of the former. Robert Solow, "You can see the computer age everywhere but in the productivity statistics". Computer capabilities have increased roughly 33 millions-fold since the 1950s, but productivity hasn't followed that same curve. Even looking at information technology spend vs. growth shows limited gains (computer capacity has become vastly cheaper over the same period, the same outlay buys far more capability than financial values would indicate). It seems likely that AI will follow a similar trend.
Though there've also been some marked transitions where capabilities have exceeded some limit or new applications / formats have become possible. Business computing beginning in the late 1950s, small businesses and universities by the 1970s, personal computing by the early 1980s, general networking in the 1990s, mobile devices in the aughts, algorithmic ad-tech / fin-tech / social media in the 2010s, and AI in the 2020s. There've also been capability-based advances including precision machining, automated equipment and systems controls, digital media (audio/video), accounting, navigation controls (marine, air, rail, land), re-usable rockets (largely from the ability to re-land boosters), massive satellite swarms, and LLM / gradient descent / synthetic annealing artificial intelligence. What the net societal and economic impacts will be isn't clear, and a fair bit of that is already clearly negative, as is the question of how those fruits will be distributed.
All of which affect the question of how big AI's energy footprint can grow. Though I'll note again it's growth rate and duration which matter, not present magnitude.
(Hope for clarity and accessibility that's both sufficiently nuanced and insufficiently verbose, though I've my doubts on the last.)
No, in ~6y they will count for 1% of present day consumption. The rest of the quote compares it to where we expect steel manufacturing projections. It all reads like something that seeks to influence rather than inform.
> TSMC’s manufacturing success comes at an ecological cost. The company accounts for around 150,000 tonnes of water per day and 6.4% of the national electricity consumption.
It would be great for you to experience living close to one of these data centers [0]. Maybe you'd have a consideration outside of your own myopic view then.
Yeah, except those aren't being built in areas where people already live. You do realize that a refinery lasts 50+ years and the OECD already has surplus capacity? So why would there be significant spend here? Meanwhile DC starts totaled $25B+ in January of this year [0].
Maybe you should try providing some data given your comment, I'd be interested in hearing about all of these refineries and steel mills being produced!
It feels like you're being sarcastic but "AI data centers are what, half a percent of the world electricity consumption" -- 0.5% of the global total is huge!
If you think that every data center is now a closed loop system enjoy your propaganda.
Colossus 1 and 2 still use open loop systems. And with the case of blow down water that's "returned" it's not usable due to the use of concentrated salts (1,200 to 6,000 mg/L).
In your mind this is a closed loop system of clean water. It is not. And as DCs have less and less discharge there is more and more solid waste produced.
Your assertion that "water consumption is a red herring" is utter bull shit. I'd suggest enlightening yourself [0][1][2][3].
Yes, but they _could_ be a closed loop system if governments demanded it. And they should.
The could, and should generate their own electricity as well. And perhaps even provide free power to the surrounding neighborhood to make up for the noise and inconvenience of having one nearby.
Have you seen that some of these things are so loud they’re damaging to health? But they’re loud at frequencies we can’t hear so it’s not easy to tell without specialised equipment
Unfounded comments that are nothing but a lie are OK. Yet, calling it out directly goes against the rules.
In this case the response is warranted. Until people start to see what actually goes on in these data centers with respect to pollution (of many kinds) I don't have a problem being direct.
I'm doing so both staying within the guidelines and specifically calling out the idiocy displayed, though without relying on that particular word.
Calling out Guidelines violations does not imply disagreement with what's been said otherwise. It is a call for more substantive participation, and probably strengthening your own arguments.
You've been here long enough that all of this should be quite familiar to you.
(I'm not a mod, which is why I'd initially linked a mod's own admonition.)
Very interesting how your enlightening comments about dirty AI data centers are being downvoted. Pro-Musk, pro-AI bots and shills trying to keep the smoke screen going?
Except on small memory / processing devices (like ESP32) it's not actually that simple. Processing ADS-B on an ESP32 would, likely, be easier.
Part of the reason I've come to understand how finicky TLS is on ESP32 is because there have been a ton of TLS issues with these devices in Home Assistant [0]. The linked issue just being one of them that's still not fully resolved.
This is my Monday morning dance of Bluetooth. 99% of the time I'll join a call and everything is connected, line levels are set and yet... No audio. I know people complain about Pulse but MacOS is just as bad and, seemingly, getting worse.
I just recommended an N5 Air [0] to a friend because it's got a powerful CPU, can expand to 96GB of RAM, has a great mix of NVME and HDD and has a PCI slot they can run an Intel A310 in for local transcode. It's the perfect Proxmox host to run everything they'd want to do including local services but also fault tolerant storage for a reasonable price.
[0] https://youtu.be/y8WFoV4ZSTg
reply