- cross-posted to:
- linux@lemmy.ml
- cross-posted to:
- linux@lemmy.ml
Imagine taking the lightweight, privacy-focused appeal of Linux and completely ruining it by hardwiring AI models straight into the OS. Nobody is switching from Windows for more forced AI features.
Fair enough. In a broader sense, all of that is true. The current implementation and business practices are completely rotten to the core. That’s why I’m not going to pay any company for AI, unless they publish a report about using renewable energy. Mistral AI is pretty transparent about their emissions, and that’s a step in the right direction. Still not good enough until they actually power their data centres with renewable energy.
Anyway, I was mainly talking about LLMs specifically, not the entire AI industry and their destructive habits. None of the current AI companies deserve any of my money, especially the American ones.
Sure, “LLMs and GPTs are just tools” is a potentially defensible take.
But, Copilot and Claude and Gemini and Apertus and every other model, “open weight” or not, that has been studied enough will (unknown the the user) sometimes generate output that infringes on the copyrights of the training data AND most, if not all, of the models with high-quality output involved human rights abuses as part of their training process (per Amnesty International).
Maybe, in the future, we’ll have models that are trained only on freely redistribute content without human rights abuses and without setting environmental goals back 10-15 YEARS. Maybe.
If that happens, I might be willing to use those tools myself, and I’ll certainly stomach their output as part of my won workflows. Until we have those ethical models, FUCK all of them and their output, they aren’t just tools; using there output is apologia for immoral acts and using the models is collaboration immoral people.
EDIT: Reworded for clarity:
fails to generate output that will not infringe the copyright terms of the training dataFirst of all, not giving them money is like pirating a Harry Potter show and then posting about how much you liked it on social media. You’re still giving it free advertising to people who will spend money on it.
Second of all, see what I said about the cognitive effects. If you don’t practice doing things without the AI, you will lose the ability to do them much more quickly than you think you will, and if you start relying on it to make tiny decisions, at best your work and at worst your thinking will slowly start to reflect the model’s biases.
More importantly, programmers who use LLMs to complete programming tasks demonstrate less complete understanding of their code, which is important because eventually you need to maintain it and even LLMs’ most ardent defenders admit that all of them are dogshit at software engineering. If you use AI to maintain an AI codebase, you’ll do great for a couple of months before it loses the ability to remember all the parts and starts duplicating work and not syncing up state with other components and before long it’s a buggy unmaintainable mess.
Those scenarios are very realistic, and I think many people will suffer from these problems. I don’t expect that people will use LLMs responsibly. If they did, many of these problems could be mitigated, but obviously that’s not going to happen.
It’s basically very similar to what happened to physical health when work was done by machines instead of muscles. Nowadays, people don’t carry hay bales or buckets of water, and it shows. Undoubtedly, the same thing will happen to the mind as people rely on LLMs to do the thinking for them. As a bonus, you’ll get voters who are even more ignorant than they currently are. What could go wrong with democracy at that point… Mussolini showed how to decommission the whole system through democratic means.