Kinetix14h ago
Oli

A surge in new datacenters, each with the power demand of 100,000 households and a cooling water demand of 1,000,000 m³ per year to train AI models on material obtained without consent on hardware now unaffordable to consumers so fascism-adjacent tech billionaires can sell us the idea that any skill is now worthless and in doing so creating the largest economic bubble ever while simultaneously destroying society and environment.

I think that about sums it up.

#genai #llm

Feb 7 at 7:01amWeb
Show threadOliFeb 7
Yes, yes, every tool can be used for good or bad purposes, but if making a hammer required human sacrifice, you would have to find some pretty important nails to justify it as "good". Especially if the hammer drills a hole or puts up drywall in 40% of the cases.
Show threadAmorpheusFeb 7
@oli Society endured a lot of questionable developments. Society cannot be destroyed that easily. Crucial is what we build on top of the remains of failed models... especially the insights gained once the bubble pops.
Show threadEQFeb 7

@oli Thanks for todays belly-laugh.

Now to get rid of the chilling knowledge that this is indeed just facts, not a joke. Got anything for that?

Show threadさよなら皆さんFeb 7
@oli assuming this is in ref to LLMs agreed 100
Show threadLeendaalFeb 7
@oli i still disagree that any of this IS a tool.
Show threadTJFeb 7
@oli sounds basicly like the premise of most cyberpunk stories, but missing a lot of VR and body enhancement tech... 🤔
Show threadAndyFeb 7
@oli Very well said, and how can I not follow you after that?
Show threadOliFeb 7
@AGT
Because my rants and musings are usually in German 😂
Show threadConny NaschFeb 7
@oli @AGT just as well I speak German also...
Show threadBenFeb 7
@oli Celebrated and funded by the same people as cryptocurrency, which itself was designed only to kill the planet faster, help criminals, further the wealth gap, bribe politicians, divide society, harm the ability of democracies to adequately manage economies, AND do all of this in as stealthy a way as possible that is annoyingly difficult to stop and makes monsters of the people who join its cult.
Show threadBenFeb 7
@oli Forgot to mention that it also turns young men into gambling addicts at alarming rates.
Show threadtshepangFeb 10
@ben444 how so
Show threadBenFeb 10
@tshepang Crypto doesn't create any value. It's a giant zero-sum game where everybody is betting against everyone else. And I'm not even talking about all those meme-coins, those are basically glorified pump&dump schemes. Plus, there's a general trend towards gambling, particularly in the US, where everything is a bet now.
Show threadDobrochnaFeb 7
@oli The Resistance discovers plans of a new Imperial weapon capable of destroying whole planets.
The only hope is a young slave living on one of unstable and dangerous worlds, who turns out to have an oddly powerful connection with Force.
Meanwhile the Siths are creating a clone army
Show threadOliFeb 7
@dobrochna
I had the same thought. And I keep wondering who is Emperor Palpatine or Darth Vader in this scenario when everyone acts like Jar Jar Binks cosplaying as Jabba the Hutt.
Show threadnico o. s.Feb 7
@dobrochna @oli i sent a friend a screen cap of this post and they did this https://starwarsintrogenerator.com/scroller?u=2aw7c38r
I made you a STAR WARS intro sequence

Click below to watch my message…

Show threadOliFeb 7
@nicoyarz
😂
@dobrochna
Show threadbobschiFeb 7
@oli @punissuer you made a little mistake when you wrote “fascism-adjacent” instead of “fascist”
Show threadJimDFeb 7
@bobschi @oli @punissuer still a good term though. Collaborating, facilitating, driving, capitalising and manipulating also come to mind as well as adjacent.
Show threadWhiskers 🇦🇺Feb 7
@oli You are too kind 😆
Show threadTAVAloniaFeb 7
@oli *nods in commiseration , emitting a deep sigh*
Show threadNibor4000 🇬🇧 ⓋFeb 7
@oli
We are being encouraged to use AI where I work; this post may get printed out and put on a noticeboard 😀
Show threadMatthias KrämerFeb 7
@oli Forgot to mention Jeffrey Epsteins role in bankrolling the technology.
Show threadAhoFeb 7

@oli but ain't 80% of the data they train on nowadays already generated by earlier generation of #LLM ?

Internet suxx as source nowadays, just filled with LLM generated text, low quality auto translations, white power shit and fake news pushed by russian, chinese, iranian botnets.

Show threadudo m. rader ☕ 🇪🇺 🇺🇦 🐧Feb 7

@oli I have to disagree on one point: it takes surprisingly little hardware resources to run #GenAI efficiently for say 20 ppl. The hardware required for this is now quite affordable for a reasonably sized company as well as for enthusiasts (excluding the current crazy RAM prices). And decentralised operation neither requires a nuclear power plant nor an ocean for cooling, it works perfectly well with renewable energy.

Decentralisation is the key, one could perhaps even call it democratisation.

Show threadOliFeb 7

@riaschissl I'd argue that you can't run infra for 20 ppl as efficiently as infra for 10 million people, resulting in even worse environmental net impact. And your small rig may be able to run inference but not training.

My main problem is that we're throwing non-deterministic "solutions" at deterministic problems in the first place.

Show threadudo m. rader ☕ 🇪🇺 🇺🇦 🐧Feb 8

@oli Yes, using our existing DC infrastructure may not be as efficient as once again surrendering our sovereignty to the same tech bros we regularly complain about. And I would argue that more than 90% of today's typical workload is inference rather than training.

And to your last point: give any two software devs the same task and you'll get two different results. Have them do it again and you'll get another set of different solutions. Absolutely non-deterministic. No different with LLMs.

Show threadDimitris ZervasFeb 7
@oli @riaschissl even for 20 people, what would run and on what hardware? even deepseek ~0.5T parameter model needs a tremendous amount of vram (around 1G/B params) and the price of it is in the ballpark of "let's buy a house that will be useless in 5 years"
Show threadudo m. rader ☕ 🇪🇺 🇺🇦 🐧Feb 8
@dzervas @oli the main question is: what is good enough for a given task? As we know since GPT 4.5, pure parameter count isn't the main enabler for "better" LLM results. And then we have Unsloth's Dynamic 2.0 quantization, which is extremely helpful to bring down the VRAM requirements while still keeping the accuracy at a very high level. So yes, pack two or three RTX 6000 Blackwell GPUs into a decent server, and you're good to go at a decent price.
Show threadDimitris ZervasFeb 8

@oli @riaschissl I agree with you that pure parameter count is not what we're after but "a few rtx 6000" translates to roughly 30-40k + supporting hardware around it + electricity. That's nowhere near what 20 people would spend on AI subscriptions and the performance of the model you'll be running is simply worse - maybe by a lot maybe by a little, currently is by a lot but that could change soon

and in 5 years you would still have to buy new hardware, doing a similar (or bigger) investment.

I fail to see how that makes any kind of economical sense

Show threadudo m. rader ☕ 🇪🇺 🇺🇦 🐧Feb 8

@dzervas @oli take the cheapest Claude Max subscription for 100 per month, this sums up to 24k per year for 20 devs. And if you just take the ordinary team subscription, it's more than 7k per year. So the economical sense is there, plus not having to worry about what happens to your data.

And in a corporate environment, replacing hardware every 5 years is not unusual, if not the norm.

Show threadftmFeb 7
@oli it totally does
Show threadPetra van CronenburgFeb 7
@oli May I ask for a source of these figures? (I collect such facts for arguments).
Show threadOliFeb 7

@NatureMC These are rough estimates and rounded figures from a lot of research. Base figures: A household requires ~4 MWh/a, a 50 MW datacenter requires 50x365x24=438 GWh/a -> 100,000. NTT is building a ~500 MW DC in Nierstein/Mainz Area (D) -> 1,000,000 households. Cooling reqs are hard to come by.

Google was unable to build their planned DC in Neuenhagen as the required 1.4 million m³ of water could not be provided.

Benchmark for a 100 MW hyperscale DC is 2.5 million m³/a.

https://www.heise.de/news/Rechenzentrum-Projekt-von-NTT-in-Nierstein-nimmt-naechsten-Schritt-11168911.html
https://www.deutschlandfunkkultur.de/tesla-gigafactory-gruenheide-wasser-100.html
https://www.techzeitgeist.de/der-stille-durst-der-ki-wie-rechenzentren-wasserreserven-zehren/

Bebauungsplan für NTT-Rechenzentrum in Nierstein steht

Die Planungen für die 482-Megawatt-Anlage von NTT Data nehmen die erste rechtliche Hürde. Der gesamte Bau soll rund zehn Jahre dauern und 2027 starten.

heise online
Show threadPetra van CronenburgFeb 7
@oli Great, thank you for the sources!
Ich suche immer nach Beispielen, wie man Fachfremden ganz einfach erklären kann, was sie mit ihrem Whats-app-Witz oder ihrem ständigen Prompting so rausblasen (inklusive der Kosten /Ressourcen fürs Training) - da hilft das weiter.
Show threadOliFeb 7
@NatureMC Schwierig. Meiner Erfahrung nach blocken Leute bei diesen Informationen ab, weil es zu viel kognitive Dissonanz erzeugt. Und da haben wir noch nicht mal angefangen darüber zu reden, dass Bildgeneratoren unter anderem auf CSAM trainiert wurden und irgendwelche Klickworker vorher den schlimmsten Dreck aus den Trainingsdaten rausgefischt haben. Am Ende läuft es entweder auf "Aber MEINE KI ist ethisch sauber" oder "Tja, so isser halt, der Fortschritt" hinaus.
Show threadPetra van CronenburgFeb 7
@oli Ja, das ist wie mit anderen schwierigen Themen auch.
Ich vermeide in solchen Gesprächen eigentlich all das Fachzeug oder trockene Fakten.
Ich agiere da eher so nach Art des Kinderfernsehens. Bringe es also durchaus fertig, vor so einem eifrigen Witzler Wasserflaschen abzufüllen und vor den zu stellen. Wenn dann ein erstauntes Stirnrunzeln kommt, sag ich: "Das war dein Witz jetzt grad wert. Hat sich das gelohnt?"
Also mehr so subversiv ... 😉 Und irgendwelche Zauderer bekommt man immer.
Show threaduniFeb 8
@NatureMC @oli nicht all zu viel wenn man einigen relativ groben rechnungen glaubt, sehe so ca 0.3-10 wattstunden. problematisch werden halt eher vollautomatische sachen wie content farms die vielfach aufwendiger videos generieren, vibe coding mit tausenden tokens hintereinander etc
Show threadPetra van CronenburgFeb 8

@uniwuni Du musst das Training mit einrechnen, alle anderen Kosten ... es ist komplex und viele Zahlen verbergen die Betreiber. Empfehle @gerrymcgovern zu folgen, der kennt sich da aus.

@oli

Show threadNeil MoffattFeb 7

@oli Very succinct. Necessarily damningly so.

What a horrible state of the human race. A key driver of civilisation collapse is instability - and accelerating complexity in all spheres of life will pretty well guarantee that.

Show threadPaul Wilde    Feb 7
@oli just add the bit about the power demands being so unachievable that tech bro-llionaires have thousands of GPU chips in storage because there isn't enough energy available to distribute to them and we're about there, yep.
Show threadPutinBibiMBSownTrumpFeb 7
@oli Thanks.........
Show threadravelinFeb 7

@oli don't forget the obvious utility of the tool as a easily manipulable conduit for spreading propaganda through biased training, data sets, and/or prompt frames. If you can get people to consider "Asking AI" as equivalent to researching a topic, you can tell a lot of intentional lies and hide a lot of truths. The ease of lying with this tech is a feature to them.

This technology is a means of control.

Show thread🎹 Your Masto in April 🎹Feb 7
@oli Hopefully once the bubble bursts, we can repurpose the land used for the data centers in solar farms:
https://www.youtube.com/watch?v=KtQ9nt2ZeGM
You are being misled about renewable energy technology.

Let's learn and grow. New things are cool!Links 'n' stuff down below. Lots of links.First, the "clean version." Please pass that around.https://youtu.be/Zgxb...

YouTube
Show threadBo TallyFeb 7
@oli Very well summarized
Show threadM.Feb 7
@BoTally54 @oli same response
Show threadLarge Format ProjectionistFeb 7

@oli

I am absolutely certain my math is wrong, but 1M m^3 water/year may be underestimated.

Using US stats because we waste so much water:
314 liters/person/day
113K liters/person/yr
342M US pop
133M US occupied houses

US pop/houses= ~3 peeps/house
113K liters * 3=339K liters/house/year
339K liters=339 m^3/house/year
339 * 133M= 45B
45,000,000,000 m^3 water/all houses/year

Pacific Ocean=714K k^3 or 714B m^3

Fun exercise, even if wrong. Well, the Pacific Ocean value is ok.

Show threadLord Caramac the Clueless, KSCFeb 7
@oli OTOH this might exactly be the kind of bubble to end all bubbles that could destroy the entire capitalist world economy in one huge crash and show the whole world that Capitalism is over, that growth is over, and that all the 1960s dreams of robot servants and cities on Moon and Mars are over and gone. An economic crash so horrible and devastating that the whole idea of makets and private property generating wealth just collapses. In order to do that, people must absolutely lose all hope in a glorious future, they need to see that the Industrial Age has gone horribly wrong and is going down now.
Show threadIho-aNnFeb 7
@oli @AudeCaussarieu
Show threadartfulrobotFeb 7
@oli brilliantly put!
Show threadScooter Feb 7
@oli "Nice data center you've got there. Be a shame if something happened to it..."
Show threadMichael ThingnesFeb 8
@oli all of this. Plus it makes software development pay to play and sucks all the joy out of it.
Show threadjuniFeb 20
@oli ACCELERATE
Show threadWulfy—Speaker to the machinesFeb 23

@oli

And what are you doing about it besides micro-dopamine inducing posts?

Show threadOliFeb 23
@n_dimension Joined a party, campaigned, got elected to the local council, started fighting against the planned data center on the outskirts of town. How about you?
Show threadWulfy—Speaker to the machinesFeb 23

@oli

Good on you!
That makes you a rarity!

Wrote letters to politicians, Wrote an AI Senate submission, initiated my state political party AI policy and contributed to the federal one.

#regulateAi #peoplepower