archive: https://archive.ph/ZJdOS
We all have devices in our pockets powerful and well connected enough to do just about everything we use "the cloud" for except for large scale data collection for advertising and surveillance. And even then these devices are still capable of being used for surveillance and advertising effectiveness monitoring, just not as well. If our government and the supposedly wise hand of the free market cared more about global warming than global surveillance and financial dominance, they would have already shuttered most of the data center capacity in the US. Then again there are still mining farms turning electricity into monopoly money.
Idk what I'm trying to say here, I'm just bitterly disappointed in the state of the society I live in.
Unironically I think the world would be a better place if all the largest data centers went boom.
I completely agree in terms of personal computing like storing photos, documents, notes, and so on. At most, they could be encrypted locally and stored remotely so that multiple devices can use the data.
There are still plenty of use cases for server-oriented data processing. Most "infrastructure" related things work well that way. Cases where data needs to be quickly read and written from a number of different locations and the information isn't really secret or personal. I am biased since I work on one of those systems. But there are so many internal systems at companies that really don't need end-to-end encryption either. Although maybe some day things will still move in that direction.
I've recently been told that
it's only the training that demands these massive datacenters, so it is stupid to be upset about treat printing when it's just training that requires these massive data centers. Surely these data centers don't stick around and their owners don't demand more and more of them for even more training, you silly emotional Luddites. 
Techbros are irredeemable shitbags. I saw one such shitbag on lobste.rs downplaying the energy cost by saying that one inference ONLY takes as much energy as a lightbulb does in one hour.
Right before mother gaia swallows us whole for being bad children to her I would like to drive stake through the hearts of a bunch of such monsters.
Microshit wants a nuclear power plant to power its upcoming datacenter. I think this is the endgame of the ridiculous things capital does in its lust for money.
Microshit wants a nuclear power plant to power its upcoming datacenter.
Part of that contract involves a vague promise from a tech startup that totally promises to prove le epic fusion power in two years.
Meanwhile many people who are actually involved in the programming shudder in horror of what lies ahead of us.
I took one year of programming in college and that’s all I need to know that I will resist as much IoT or bazinga slop as possible. But FWIW it was kind of cool to fool around with data.
i dont even accept the framing that it's just the training that's energy intensive, these datacenters are running 24/7 so that websites can use new and "improved" chatbots galore, not to mention everyone trying to cheat on their homework or fucking around with image generation that is barely better over open source pc models
That's what the smug bazinga fuck claimed, and even added a pithy "no investigation no right to speak" cherry picked site appropriation.
Just running some data through the resulting model is still somewhat expensive since they have so many parameters. And of course for a lot of things, you want to train the model on new data you're putting through it anyways.
The proselytizer treated it as a gotcha, so I appreciate the additional information.
In their defense, I'm sure there are tons of actually useful machine learning models that don't use that much power once trained.
I have an iPhone with Face ID and I think the way they did that was to train a model on lots of people's faces, and they just ship that expensive-to-train model with the operating system and then it trains a little bit more when you use face ID. I can't imagine it uses that much power since you're running the algorithm every time you open the phone.
I'm sure any model worth anything probably does require a lot of training and energy usage. I guess it really depends on the eventual utility whether it's worth it.
damn that's crazy. ChatGPT, think of some ways to make that number go down
ChatGPT: "I (or do I mean us - haha) am going to solve this problem. But first I need more juice so I can solve the problem. Hey, I can do a lot of stuff - but I can't change the science!"
Tech Won't Save Us podcast is doing a series about big tech energy usage. First episode was good though a bit short.
https://www.techwontsave.us/episode/241_data_vampires_going_hyperscale_episode_1
In unrelated news, I've decided to become a GUI developer.may be 7.62 times higher than
about 662% – or 7.62 times – higher than
They can't keep their numbers straight.
With the way they phrased it its wrong.
662% higher than is 7.62x
About 662% (of) is 6.62x
662% is 6.62 but an increase of 662% is 7.62
to show a 62% increase you multiple a number by 1.62, to show a 162% increase it's 2.62, etc (262 is 162% more than 100)
i'm not a math guy don't @ me
"662% higher than" is equal to 762% (7.62 times).
"7.62 times higher" equals 862%.
I am a very non-tech person, can someone explain to me why “junk data” isn’t regularly deleted?
Not sure what you mean by “junk data”, but hoarding data is the least of our issues. It’s the computational power and cooling that eats up all the energy, not the “cold storage”.
I use "junk data" to mean data or any information that isn't in use anymore and can be safely retired.
As in all the data that's gathered from users with tracking software and the likes? It's not deleted because it has value in the form of advertising firms that promise to extract something useful from it (we all know that recommendation algorithms suck anyway, but the promise is enough to get funding). The impact of storing the data is pretty small though, as other people mentioned the main issue is in compute when it's time to train new models that use that data to do something.
