Now on top of the negative externalities on people’s power, water, air, and global climate, we have another: the ceding of more & more control over critical energy & water infrastructure to Silicon Valley. I wrote about this last week in @theatlantic.com. 14/ www.theatlantic.com/technology/a... - ThreadSky

karenhao.bsky.social • 32 days ago

Now on top of the negative externalities on people’s power, water, air, and global climate, we have another: the ceding of more & more control over critical energy & water infrastructure to Silicon Valley. I wrote about this last week in @theatlantic.com. 14/

https://www.theatlantic.com/technology/archive/2025/01/openai-stargate-maga/681421/

Comments

boboinparadise.bsky.social•32 days ago

This seems like the critical subtext missing from the "discussion" about stealing water from N California.
That and that Trump Nat'l GC is in LA.

karenhao.bsky.social•32 days ago

But DeepSeek shows - in the same moment as the Stargate announcement - that the trade off that OpenAI & co frame as wholly necessary is actually not. 15/

sam044.bsky.social•30 days ago

That's not very capitalistic of china. 😆 Of course the US wants to develop as many different industries as possible to solve "problems". Doesn't matter that we create the problem. Same as the tech security industry, which is huge because we use easily compromised passwords instead of token.

sam044.bsky.social•30 days ago

Western EU has been using tokens for security (a device that generates unique codes) for 20 years. Much much harder to hack. So your well explained scenario makes a lot of American sense. Pass expense on to users, create multiple bloated industries that'll have to be replaced eventually. Thank you!

sam044.bsky.social•30 days ago

Clarifying that EU uses tokens to secure their bank accounts. They also had chips in their ATM cards in the 1990's, again, 20 years before US.

karenhao.bsky.social•32 days ago

Wait a minute. You mean to say that we don’t need to blanket the earth with data centers and coal & gas plants to maybe arrive at a future where we can wave a magical AGI wand to make all of the consequences of that go away?

Yes. This is a false trade off. Let that sink in. 16/

karenhao.bsky.social•32 days ago

As I said earlier, scaling was always more about business than science. Scientifically, there’s no law of physics that says AI advancements must come from scaling rather than approaches using the same or fewer resources. Scaling is just an incredibly easy-to-follow formula. 17/

karenhao.bsky.social•32 days ago

DeepSeek now also undermines the business case for scaling. OpenAI has been burning through staggering sums of cash to keep up its scaling paradigm and has yet to figure out how to balance its checkbooks - and it turns out it didn’t need to spend so much cash. 18/

karenhao.bsky.social•32 days ago

So it doesn’t matter if you're a company in the US, China, or elsewhere. DeepSeek should be a cue to pivot HARD toward investing in far more efficient methods of AI development. Even if you care nothing about community & climate impacts, it’s just better business. 19/

litterygeniarse.bsky.social•32 days ago

Nope - It's the emperor's new tech and the only people falling for it are our modern day Emperors

lp7.bsky.social•32 days ago

📌

boboinparadise.bsky.social•32 days ago

And other scale dependent industries should be scrutinzed.

acoonrod16.bsky.social•32 days ago

Appreciate the thread!

karenhao.bsky.social•32 days ago

And if a company does not make this pivot, that should be a major red flag for its capacity to innovate. And I don’t mean product innovation but innovation on how to develop AI models. The base of the stack. 20/

perry313.bsky.social•32 days ago

I'm a climate activist in NoVA so I'm concerned about AI and the data centers concentrated here. The cynic in me thinks that they are aware of the increases in efficiency that will likely manifest themselves in the next five years, but are using this projected demand growth to abandon climate goals.

rabin-acharya.bsky.social•32 days ago

This is fantastic thread. Many people should read it. Thank you.

theonetruepanda.bsky.social•31 days ago

Thank you Karen for your brilliant Thread! One of the best I've seen.

aspeno.bsky.social•32 days ago

maybe it's easier for China to mobilize in millions of live people to label training data

the3rdlaw.bsky.social•32 days ago

This is exactly the space our team @contellationai are working on. NGL, it’s exciting to see test runs knocking 15-25% off resource consumption for inference.

ointment666.bsky.social•32 days ago

Thank you for the thread!

fungiman.bsky.social•32 days ago

📌

kenryan.bsky.social•32 days ago

Better "business" is convincing the federal government to hand you a 12-digit pile of free money!

stretchdad.bsky.social•31 days ago

Given that generative AI and LLMs are relatively new tech, it's not surprising that there are plenty of unexplored paths for achieving the same results with fewer resources.

kevincureton.bsky.social•32 days ago

Ask AI companies to disclose their model training error rates. In my experience they could care less about errors as long as the model eventually got trained. To the tune of hundreds of thousands of training hours wasted.

bobruth12.bsky.social•32 days ago

This nugget right here in part 19 of the 🧵 is the most important piece to transmit to businesses- both producers & consumers

(All will pay higher prices. If the producer is spending more 💲 on production, cost will be passed on & consumers will spend more on consumption)

moondust1880.bsky.social•32 days ago

Thank you very much for this very enlightening thread!

gpogue.bsky.social•32 days ago

Add AI to the list of things modern capital markets can't produce efficiently. So far, housing, healthcare, healthy food, education, transportation and software. At least we have lots of flavors of soda and soap. 😞

econ-001.bsky.social•32 days ago

if you have to stamp it on silicon, seems you stopped thinking too early.

olenatheshroomlady.bsky.social•31 days ago

Excellent writing. Thank you.

oscariano.bsky.social•32 days ago

📌

trudheim.bsky.social•32 days ago

Excellent tread. Thank you.

OpenAI is snakeoil. They define success by how much revenue they can generate, not whether the product is good or whether it meets scientific or technical criteria. Ergo, snakeoil.

edenarachnelatro.bsky.social•32 days ago

First 🧵 about AI that has made sense to this boomer. Cheers.

mayamedia.bsky.social•32 days ago

They were accused of missing the AI boat, but I wonder if Apple is coming out of this in the best spot. Their deal with Chat GPT involves no payments, & it appears they’ve sunk the least R&D on development.

viennease.bsky.social•32 days ago

📌

bschiff.bsky.social•31 days ago

This reminds me of the once accepted maxim (until the mid-1970s) that GNP and energy use were inextricably linked, so electricity production had always and everywhere to increase.

emceehammerpants.bsky.social•32 days ago

Awesome thread, thank you for writing!

artemisprime.bsky.social•32 days ago

The entire plan Trump announced is now embarrassingly obsolete and a waste of money given this news.

solomonhinton.bsky.social•31 days ago

💯

dasnick.bsky.social•32 days ago

Thanks for an excellent thread! I learned a lot!

seasofsalish.bsky.social•32 days ago

The “trade off” is a lie.

We all know it’s a lie. And here we sit watching our climate worsen.

Because how do we successfully fight billionaires and their lies.

jimismash.bsky.social•32 days ago

How much of their spend goes to compute vs people working? Now I'm wondering how much their people us tools like the ones that are available to the public. How much of what goes on there is bots emailing other bots back and forth set up by employees avoiding actual work?

tironstefan.bsky.social•31 days ago

And Big Tech are not burning cash like KLF in Scotland.

How can anyone not see this as what it is: "Sorry you can't afford to own a home, we had to beat China in the race to develop a better plagiarism app" Ouch!

grimandgrimy.bsky.social•30 days ago

I hope they figure it out with two sticks, that would take the financial benefit to burn the planet to the roots and toss it to the wind. If AI costs nothing, then it can't be monopolized, nor hoarded.

megalu.bsky.social•31 days ago

📌

scotthoots.bsky.social•32 days ago

Scaling means that companies like Microsoft Azure, AWS, Oracle, HPE, Dell, etc. keep the money flowing.

brockhenry.au•32 days ago

They need to hire some of the people who made demos back in the 80s. They're probably available: https://en.wikipedia.org/wiki/Demoscene

brainrockets.bsky.social•32 days ago

I grew up with a lot of engineers and one of the things the folks I knew bemoaned A LOT was code bloat. Like in the sense that there weren't incentives to write code efficiently especially in tandem with the power available increasing. So it's not surprising that they aren't being efficient.

trudheim.bsky.social•32 days ago

If we roll back to the 90's and the Amiga/PC/C64 demo scene. Look at what *kids* could achieve by simply being good at coding and understanding the hardware they had to work with. Code usually handcrafted in Assembler and optimised *by hand* to crank out amazing things.

trudheim.bsky.social•32 days ago

We absolutely do not need these massive datacenters if the focus would be on lean code, possibly written in part in Assembler, in large parts C, to get as lean and fast as possible.
Plus, a breakthrough in *thinking* about the problem, like Deep Think has shown, just more revolutionary.

belcuore.bsky.social•31 days ago

📌

wilfordp.bsky.social•31 days ago

Thanks for teaching us the challenges of AI and the dangers to our environment they present.

Can you also report on blockchain cryptocurrency data mining?

So many stand alone facilities masquerading as "data centers".

alianaqvipro.bsky.social•32 days ago

📌

caitscheverst.bsky.social•32 days ago

I can't be the only one seeing deep seated similarities with religious Rapture fixation? 'everything will get worse and we don't need to make any effort to reduce the badness because at the end of it all the worthy will be rewarded by a magical good'

writeups.org•31 days ago

Yes, this is often summarised as TESCREAL ideology - from a Gebru/Torres paper.

ehoelsch.bsky.social•32 days ago

Hey great thread. Could you please just clarify what the acronym "AGI" means? I followed most of what you were saying throughout your broader thread but struggle with figuring out industry specific acronyms. Thanks! 😀

karencwilson.me•32 days ago

Artificial general intelligence

ehoelsch.bsky.social•32 days ago

Thank you!

onen.bsky.social•32 days ago

Large Language Models, what we call "AI" now due to marketing, is not actually intelligent, not even artificially. There's no understanding or reasoning involved whatsoever. AGI, Artificial General Intelligence, is what is used now to refer to what we used to understand as AI before

caitscheverst.bsky.social•32 days ago

You could consider LLMs to be 'synthetic imagination', but only imagination. An AGI is the sci-fi dream, we could maybe use this synthetic imagination as a part of the eventual Thing, but we're still at the point of fumbling in the dark when it comes to the eventual shape and form of an AGI.

justafrenchcat.bsky.social•31 days ago

Well if their model "learnt" from other models, didn't someone have to scale in the first place?

dibble555.bsky.social•32 days ago

📌

gchance60.bsky.social•31 days ago

This is a game changer. Maybe not total reset, but close. (Thanks for the thread!)

chuckedahl.bsky.social•32 days ago

This is incorrect. DeepSeek's ability is predicated on training on an LLM. It leverages the power-and-water-expensive AI. It can't exist on its own.

If you want DeepSeek to perform better, you need a better LLM and you're back to wasting water, greenhousing the atmosphere, and using tons of power.

bento-bongos.bsky.social•32 days ago

That requires basic elementary on economics. Which you also need to get [gasp, shock, horror] a grasp on politics. The other, dirtier side of the economic coin.

Which people refuse to do.

sysplayer.bsky.social•32 days ago

Except that its claims are exaggerated - they left out all of their GPU training - 50K of them.

Great lesson in efficiency, but this is just overblown.

charleyinma.bsky.social•31 days ago

👀

suntoadcowboy.bsky.social•31 days ago

📌

detoxponcho.bsky.social•31 days ago

📌

Comments

Posting Rules

Reply