Supposing it's physically possible, what reasons are there to want to? I have yet to see a single one mentioned anywhere.

Downsides are plentiful:
- Cooling is challenging.
- Servicing is impossible.
- Communication is slow.
- Construction is expensive.

What could possibly outweigh all that?

Can we make the radiators much smaller by running them, say, twice as hot, and actively pumping heat from the electronics into these hotter surfaces?

Yes, of course we can, the laws of thermodynamics are fine with that. However ...

some interesting points there, but three things stood out to me about his cost calculations:
- He assumes launch costs of $1500/kg, the same as today; this basically assumes Starship completely fails to achieve its goals. He justifies this by talking about how much effort was required to refurbish the Shuttle between flights; but Starship has been designed from the ground up to avoid the Shuttle's problems (complex engines that had to be stripped down, every tile having a unique shape, boosters landing in salt water).
- He assumes each satellite must be deorbited once it's no longer at the hardware frontier. I see no reason it can't be demoted to less demanding uses, as we do on Earth.
- He assumes that satellite construction costs stay the same. I'd expect at least some cost reduction from building a million of the things!

All of these are defensible as conservative planning assumptions, but he presents them as realistic.

I found a companion piece a couple of blocks over, this one a commentary on how the detritus of capitalism has led us to turn our backs on nature.

#art #StreetArt

I hate how all the legit uses of LLMs (like in academic linguistic research, semantic analysis, and studying language) have been overshadowed by the generative bullshit ;-;

researching LLMs in linguistics used to be so much fun back in 2017 ;-;

im at a company that is extremely whiplash about it. They jump scared us with "Use AI now and were watching or get the fuck outta the way" and then followed up saying oh wait legal says there might be serious issues you cant use it. I fought as hard as i could and I continue to advocate conscious thoughtful approaches. Not taking it seriously at face value, thinking of juniors who need to develop skills. So on and so on... it feels so shitty living in this world...

After all this time I finally got a subscription and do use one frequently. I wont let it touch anything I would have wanted to do myself essentially is my rule. My hobbies are for me to experience them. But hey I have this open source code I dont really wanna mess with but it doesn't work how I want. I dont care if I keep an entirely local repo with some random crappy tweaks to meet my needs. Im never gonna put it out in public or claim it as my own. Just making my life less miserable. Versus like my art skills

So down for this new arc where we can speak out about all the issues with ai companies without denying the potential benefits of llms.

Because for all the crap and all the abuse and all the bad shit, technology is a good thing that empowers us.

@pharmafemboy got it right. "AI" sucks, but machine learning could have been so good for society if it weren't for predatory AI Bros.

NOW LET'S TALK ABOUT THOSE STUPID TRUCKS. I live in Florida which, despite Miami Vice, is pretty redneck and we see those dumbass truck-nut-sporting raised bullshit gas-guzzling trucks all the time. My son likes to make fun of the clean ones saying "They've never driven that anywhere but city streets, and they don't work construction."

I mean - technology can help. Like image recognition helping with tedious inspections of cast parts for possible casting faults with humans still making the final decisions. Astronomers also would be lost without machines analyzing petabytes of image data to help with later analysis of possible findings. But that's always *very* specific, limited, well-defined tasks.
Add autonomous driving, maybe. A machine keeping the cameras on the road and surroundings *all* the time can't be as bad and it's not insanely complex anyway. Seems entirely doable, at least. We do it as a subroutine while talking and having our brains focused on other stuff altogether.

All the hyped, generative shit merely copying the work of artists, raping them of the fruits of their labour, however? Answering machines, suggesting to be your competent friend while they're rigged for fascism, lying with utter confidence while actually being stupid af and having no actual understanding whatsoever?

Robots, devaluing labour, basically only ever enriching capitalists as their mechanical army of slaves, implied by the very name already? That also raises a lot of red flags.

Just sign me up for a #ButerianJihad any day and call me a #luddite, because luddites, after all, weren't against technology. They simply knew they would end up as slaves and they were right. They fucking knew it.

Oh no for sure, but I do dream of a post-ai bubble world where we can reap the benefits of the concept without depending on the absolute pieces of shit that currently run that show.

By my count, there oughta be a way to make llms work in a distributed manner. Realistically, nothing insurmountable is stopping us from a setup where the compute is done in actual houses where and when the weather is cold and the heat is useful.

Is "using it to do work that otherwise wouldn't get done" an acceptable limited circumstance?

I feel like many corporate environments will go to great lengths to justify their use of AI, by quantifying perceived benefits in 'hard' metrics. They're all waiting for the bill shock to land.

cyber-fascism is not cause by AI in the first place, but clearly and persistently by the cyberspace architecture and concept called internet. Internet is probably one of the worst cyberspace architecture and concept we could have in human history, and within the infinity of possible alternative cyberspaces architectures and concepts we could have.

It is the most fascist and capitalist and imperialist we could have.

AI as it exist is the direct consequence of this.

Sorry, did you mean "big ass-truck?"

xkcd.com/37/

(not a backlash, more like a touché-stone)

I actually agree with you on this.

My company has a very "measured" approach to AI. We're a bank, so they're hesitant about what we're allowed to do with it. But people do use it in limited ways. And we're allowed to.

They gave us a couple trainings on it where they're very clear "it's good at this, bad at this, and always verify and massage the output; AI is often wrong. Don't trust it." And like...that's all a fair read.

I've used it to analyze and summarize free-form responses from 1000+ customers. It saved me a LOT of time and headache.

I have a hard time blaming an individual for using the tool in that way. But if we start coding everything in it, I'm gonna struggle.

nuance? On my internet?!
*clutches pearls*
How very dare!

No but seriously there are indeed limited circumstances where they make sense and feeling guilt for using the right tool for the right job is wasteful. That emotional energy could be better directed in so many other ways.

Sensitive content Clicca per aprire/chiudere

I like your nuanced opinion, I don't agree with it on a personal level. Just like I don't think people should drive cars, public transport should be better, I don't think people should fly, I don't think people should eat meat

I do understand why people use these tools willingly, I just think that there a too many issues tied up with the use of these tools and things most won't realize until it's too late

Complimentary video essay.

After which I revised a part of my bio to state that “AI” PFP havers won't be treated seriously, but ppl that have used “AI” to save their asses (e.g. an illegal eviction attempt) will be treated seriously.

youtube.com/watch?v=y85nqc2zm7…

I've had some struggles in the past about whether or not to participate in exploitative systems...

Ethics aside, some problems I've yet to overcome:

1. AI giving wrong answers. "But it's so close, just one more prompt will get it right..."

2. The purported mental risks of psychosis and losing critical thinking skills. I don't have many to spare.

In 2021 I was really excited to use AI to explain man pages to me and help me troubleshoot, etc. But these problems stopped me.

That is sad :/
But I've read from multiple people now that some blame the people using it ... Don't like that, since it's just unfair. Most people use it because either they don't know better or are forced to by their employer. And it doesn't help to blame them, but just tell them why you don't think genAI is the way to go.

Hope you're doing well anyway :3

What bugs me is equating using LLMs with using American Megacorps.
Like, download your model from China and run it locally. Most of the arguments against it evaporate. What is left is the spam/dead internet theory problem. That is more interesting discourse.

And also yes to deleting Amazon, Google and Microsoft from this Earth, but that doesn't start and doesn't end with the use of their LLM websites.

Also, and very much IMO, if you do use AI, consider using a local model and local repo software such as forgejo. The quantised unsloth gguf Qwen3.6 versions work well locally on a workstation.

That way you're not putting money in the hands of evil tech bros, you're keeping your information private, and you're not locked into paying ever-escalating per-token costs.

coming back to this post after the necroboost to remind everyone that individualism is not the answer to societal problems, and only social political solutions will get us out of this mess.

Legislation, collective organizing, unionization, lawsuits, and (after much organizing, and when done properly) coordinated boycotts are the answer.

I wrote a bit about this fact before: riverseeber.net/blog/post/indi…

And if you'd like somebody with a bit more authority in the field on the topic, @pluralistic wrote about it too: pluralistic.net/2025/07/31/uns…