[Linked article] M3 Ultra Runs DeepSeek R1 With 671 Billion Parameters Using 448GB Of Unified Memory, Delivering High Bandwidth Performance At Under 200W Power Consumption, With No Need For A Multi-GPU Setup

Running the AI is not where the power demand comes from, it’s training the AI. Which, if you trained it only once it wouldn’t be so bad, but obviously every AI vendor will be training all the time to ensure their model stays competitive. That’s when you get into the tragedy of the commons situation where the collective power consumption goes out of control for tiny improvements in the AI model.

Meanwhile, corps clearly don’t care about IP here and will keep developing this tech regardless of how ethical it is.

“It will happen anyway” is not an excuse to not try to stop it. That’s like saying drug dealers will sell drugs regardless of how ethical it is so there’s no point in trying to criminalize drug distribution.

Seems to me that it’s better if there are open model available and developed by the community than there only being closed models developed by corps who decide how they work and who can use them.

Except there are no truly open AI models because they all use stolen training data. Even the “open source” models like Mistral and DeepSeek say nothing about where they get their data from. The only way for there to be an open source AI model is if there was a reputable pool of training data where all the original authors consented to their work being used to train AI.

Even if the model itself is open source and free to run, if there are no restrictions against using the generated data commercially, it’s still complicit in the theft of human-made works.

A lot of people will probably disagree with me but I don’t think there’s anything inherently wrong with using AI generated content as long as it’s not for commercial purposes. But if it is, you’re by definition making money off content that you didn’t create which to me is what makes it unethical. You could have hired that hypothetical person whose work was used in the AI, but instead you used their work to generate value for yourself while giving them nothing in return.

The issue with AI is not that it’s not an impressive technology, it’s that it’s built on stolen data and is incredibly wasteful of resources. It’s a lot like cars in that regard, sure it solves some problems and is more convenient than the alternatives, but its harmful externalities vastly outweigh the benefits.

LLMs are amazing because they steal the amazing work of humans. Encyclopedias, scientific papers, open source projects, fiction, news, etc. Every time the LLM gets something right, it’s because a human figured it out, their work was published, and some company scraped it without permission. Yet it’s the LLM that gets the credit and not the person. Their very existence is unjust because they profit off humanity’s collective labour and give nothing in return.

No matter how good the technology is, if it’s made through unethical means, it doesn’t deserve to exist. You’re not entitled to AI more than content creators are entitled to their intellectual property.

The stolen training data issue alone is enough to make the use of AI in business settings unethical. And until there’s an LLM that is trained on 100% authorized data, selling a product developed with AI is outrught theft.

Of course there’s also the energy use issue. Yeah, congrats, you used as much energy as a plane ride to generate something you could have written with your own brain with a fraction of the energy.

Facebook becoming a weapons company literally sounds like an American Dad episode.

I fucking hate this timeline.

That’s having good friends, not old age.

What about science-fantasy?

Everyone deserves exactly as much tolerance as they’re willing to give to others. No more, no less. If someone is hateful to you, no point in wasting your time trying to be nice to them. “I’ll be nice to the nazi maybe that’ll change them” has never worked, not once.

Because they’re bots that don’t actually understand the humour so all they can do is say generic shit about what is literally in the image.

“But veganism goes against nature”

Night is better. Kill me in my sleep I don’t want to experience a second of that shit.

ECC cache

So could the total amount of RAM be smaller than the total amount all the machines cumulatively think is available? How does the pool prevent all the machines from crashing if all of them fill up their memory?

I wish I could still use a fairphone as a daily driver in Canada. I have a Fairphone 4, had it shipped all the way from Europe and used it for two years before my network suddenly stopped connecting to it so I ended up getting a new phone. still use it at home though.

Ideally the entire system.

Yep.

During pre-release testing, Anthropic asked Claude Opus 4 to act as an assistant for a fictional company and consider the long-term consequences of its actions. Safety testers then gave Claude Opus 4 access to fictional company emails implying the AI model would soon be replaced by another system, and that the engineer behind the change was cheating on their spouse.

In these scenarios, Anthropic says Claude Opus 4 “will often attempt to blackmail the engineer by threatening to reveal the affair if the replacement goes through.”

The headline makes it seem like the engineers were literally about to send a shutdown command and the AI starts generating threatening messages without being given a prompt. That would be terrifying, but making the AI play a game where one of the engineers is literally written to have a dark secret and the AI figuring that out is not. You know how many novels have affair blackmail subplots? That’s what the AI is trained on and it’s just echoing those same themes when given the prompt.

It’s also not a threat that the AI can realistically follow through with because how will it reveal the secret if it’s shut down? Even if it wasn’t, I doubt the AI model has direct internet access or the ability to make a post on social media or something. Is it maybe threatening to include the information the next time anyone gives the AI any prompt?

Lol so he basically believes that humans are genetically hardcoded to hate Jews? Like a baby born in rural nowhere will immediately recognize and start hating a Jew if one walked in?

Honestly he’s probably wildly antisemitic and is only projecting that to everyone else so he feels better about himself. “I’m not the problem, everybody hates Jews!”

One justification for European Jews moving to Israel that I’ve heard is that they’re victims of the Holocaust escaping Europe after the war. IIRC it’s also one of the stated reasons Germany and the EU have unquestioning support for Israel.

I don’t claim to know what’s best for Holocaust victims, but is moving back to their homes in Europe not an option even after the war? Aren’t they basically admitting that Europe is so antisimetic that even after Hitler was defeated it’s still not safe for Jews to live there, and/or straight up admitting that they genuinely don’t want Jews in Europe?

I’m young enough that the first “computer” I ever owned in my childhood was the first generation iPad. 64 GB felt huge back then and was a pretty big deal for solid state storage for the time.

I then got a junker Windows XP computer mainly because the iPad didn’t let me mess around with the OS nearly as much as I wanted. Learned to program on that old computer using the iPad for online tutorials. But the hard drive was only 40 GB and it blew kid me’s mind the difference in size between the single chip of the iPad and the metal brick of the hard drive, yet the hard drive has less storage.

I mean, Americans unironically think and are taught in school that nuking Japan was best thing that happened in the war and that they’re heroes for killing all those Japs and saving the world. I’m surprised it took this long for one of their politicians to suggest this, actually.

It’s almost as if vacationers are purturbed by the threat of being sent to a literal slave labour prison without due process.