Post B48itzC55tZyZJl5fc by mgorny@social.treehouse.systems
 (DIR) More posts by mgorny@social.treehouse.systems
 (DIR) Post #B48CZ11ht7dtdcrYDw by mgorny@social.treehouse.systems
       1 likes, 0 repeats
       
       The key takeaways from the early part of the #chardet thread (I didn't read beyond the ~30 first comments, I have my limits).1. People there love cosplaying lawyers. Except when the other side also starts cosplaying lawyers, in which case they suddenly divert to suggesting asking professional lawyers.2. Almost nobody there is concerned with ethics or morality.3. There's a lot of GPL haters there. Like, they seem the kind of people who don't really care about licensing at all, just used MIT in their projects because it was cool and they heard something about license incompatibility and now bash at everything that's (L)GPL.4. People don't get that LLMs are statistical models and can't build anything from the ground up. All they can do is remix, which implies they use existing code for inspiration.5. The maintainer who did the rewrite is a total asshole, and is perfectly aware of it.Honestly, I'm truly waiting for the subsidizing to end and companies start charging obscene amounts for the use of LLMs. Of course, the reality is that we're totally fucked. We have a lot of projects that adapted a lot of #slop, and people who are being increasingly addicted to this shit. The moment they can't afford it, we'd be left with lots of broken code nobody wants to maintain.And I definitely don't want to put my effort into packaging crap if its maintainers don't even bother trying.https://github.com/chardet/chardet/issues/327#AI #LLM #NoAI #NoLLM
       
 (DIR) Post #B48D9vtLwS2mQZYhUm by lanodan@queer.hacktivis.me
       0 likes, 0 repeats
       
       @mgorny Subsidy part makes me wonder, aren't some of those self-hostable?And that's with me wanting it to blow up before it causes further damages (specially given how nasty and long-term the last ones have been), so for once I'd rather not have something self-hostable.
       
 (DIR) Post #B48itzC55tZyZJl5fc by mgorny@social.treehouse.systems
       1 likes, 0 repeats
       
       @lanodan, kinda. You can run a local model, but practically it will be much smaller (and therefore less powerful) than whatever BigTech are throwing at you. As my coworker once explained, the difference will be like: you can tell Claude what a program is supposed to do, and it will spew a whole program; with a local model, you have to describe said program function-by-function. And we're talking about the kind of hardware companies can afford, I suppose it will be even worse for unpaid individuals.
       
 (DIR) Post #B4X1QVshYm4IBMOMWe by clacke@libranet.de
       0 likes, 0 repeats
       
       @lanodan @mgorny I would like to see what an LLM trained on Public Domain datasets and running offline would be able to do and at what cost.The main arguments against LLMs are: - They're built on unethically sourced material and their output is legally dubious - The claims that they save money are based on loss-leader subsidiesThe main arguments for LLMs are: - They're so useful and make people productive - They save money compared to hiring expertsIf an offline, ethical LLM actually allows me to make complex changes to a codebase in reduced time, I'd happily use it. If it's hopelessly bad or costly, it's an admission that its current supposed value is a mirage propped up by smoke and mirrors, that it can't work without plagiarizing and/or doesn't provide net value to society.
       
 (DIR) Post #B4X1QW6WjN7wsErP4i by mgorny@social.treehouse.systems
       1 likes, 0 repeats
       
       @clacke @lanodan, no, these are not arguments "for" LLMs. "Productivity" is just useless metric of increasing liability (see https://andrewmurphy.io/blog/if-you-thought-the-speed-of-writing-code-was-your-problem-you-have-bigger-problems). And not hiring experts is just an asshole argument to avoid paying people while expecting people to have money somehow to live.