I think you’re in a way conflating two problems. One is policy and private investment in AI and the other is personal LLM use.

The first is stupid irrespective of the underlying economic system. They’re literally using environmental resources that are scarce and our taxes to fund these unnecessary data centers.

The second is something different. I think of it as being able to hire someone to do something for you. You could hire someone to do your homework for you which would be really stupid because it’s essential you do that. Most people didn’t do it not because they had restraint but because it cost too much. With LLMs you just reduced the cost.

I honestly think if the public subsidization of LLMs stopped and policy made them actually pay taxes and environmental regulation fines or for mitigation measures, we would see the actual cost and most frivolous use will stop.

It’ll probably be limited to institutional use and helping doctors summarize notes etc.

The other aspect is probably correct safe guard for the tool. You don’t need to ban calculators for ever. You just need to ban them at stages of education where the students need to practice the underlying operations to learn, i.e. probably up to the end of high school.

Unfortunately I don’t think caution is a virtue that is rewarding in most circumstances to most people. New tools need to be extensively and rigorously tested before being used.

I don’t even think it’s a individualism/capitalism thing unfortunately. I’ve been in cultures/societies that are not either and both still use these tools to further their goals. It’s just power at the end of the day.

It’s like the nuclear bomb. It doesn’t really matter what the underlying economic system of US or USSR were, they still used it to further their goals.

I think the insidiousness is in the power of the tool. For most people it’s just too powerful to not use. I can be an excellent photographer or artist and not make a dime if I don’t engage in social media.

For me that’s the sad thing. Self-hosted small models have been extremely useful to me to perform selective tasks that completely changed how things work. It’s allowed me to manage my research and information processing so much better. But I also know most people don’t put any limiters on it and use it for anything and everything.

It’s a powerful tool that people are using without restraint. I think this to be expected in the first few years after any new powerful tool is found. Humans will find a way to mess it up.

See radium cosmetics and ideas to dig the Panama canal using hydrogen bombs. Social media is probably as much or even more dangerous than LLMs.

Maybe it’s just the colorblind ones?

I think it’s sort of a big problem to solve because there’s no corporate backing or down stream derivatives contributing a lot back, making a fully curated repo hard to achieve. I wouldn’t mind a fully vetted AUR either but I don’t think the community can get there with the current resources.

Nix honestly sounds really awesome. When I tried it long, it kept on running into edge cases for things that weren’t really working well like properitery drivers or old ass packages that needed a lot of workarounds. It sounds like it would be amazing on a server/vm.

You want little bit of the ability to customize everything without having to compile from scratch every time? You want a well documented system that has excellent documentation and a community that digs into most edge cases? You want upstream fixes ASAP because your team of volunteers are struggling to backport fixes?

It has its place. I’m saying this as someone who uses Debian and red hat derivatives extensively.

To me slow roll, tumbleweed and fedora were good contenders for the same slot but none of them let you completely fuck around with the design like Arch does.

I was hoping that testdisk would show you something funky going on with the partition table based on the parted error. No luck I guess.

My next two ideas are,

1. As far as I know wipefs just tries to wipe what blkid sees and it might have screwed up something. What if we just dd the whole drive with zeros and let your enclosure take it from there?
2. What does smartctl say? Might be worth it to run the long and short tests to make sure that the drives themselves are okay.

Total shot in the dark but what does testdisk say?