You clearly don't care about the eco-damage or else you wouldn't be using Big Tech's LLMs lol.
And let's not kid ourselves, you aren't programming anything. You are paying Anthropic to brute force the code for you with their supercomputer which was only built thanks to the rare earth mining and human rights violations you claim to be bothered about.
LLMs haven't enabled any discoveries that were previously out of reach. If you disagree, name one?
>AI is useful today and is getting more powerful and efficient
No, it's getting less efficient at an exponential rate without proportional performance gains
>do think it's important to keep in mind that this centralization of power and wealth is not caused by AI directly: the pattern exists whenever there's a desirable but scarce resource. The same argument that AI is increasing inequality can reasonably be made about concrete, steel, and steam engines.
The goal of these LLM companies is to funnel all economic activity through their models. LLMs only exist because of how they strip mined human work without consent or compensation. I.e. LLMs do directly cause the centralization of wealth and power by how they were designed and how they are meant to be used.
> LLMs haven't enabled any discoveries that were previously out of reach. If you disagree, name one?
While not an LLM, protein folding is now often done via transformer.
> No, it's getting less efficient at an exponential rate without proportional performance gains
AI is expanding in both directions. Sure, it's eating the world with its compute demand and the bitter lesson bites, but we're also seeing drastically improved efficiencies employed alongside. The cost per task is dropping fast.
> The goal of these LLM companies is to funnel all economic activity through their models.
That's what all companies want, AI or otherwise.
No arguments here on the ethics of AI companies though.
I'll take open models any day of the week. We're almost there but they're not quite good enough yet. Re: fun, don't you have a backlog of incredibly weird experiments and projects you'd like to build? I did and now I'm building them, it's very fun (and educational).
The answer is no, at least so far. There's been a few big quantitative studies (CircleCI, a few others I can't remember) which show this - more code, more branches, more PRs, but actually less going to production. It's not even like a 10% increase on average or something, it's less! And yet anecdotally people say it gives you 2x, 3x, 10x speedups.
Eh, my understanding is that the models are supposed to have guardrails against that. But also, getting better results generally requires refinement, a critical eye, and the skill to write out what's wrong. I'd guess these skills aren't common among that audience. But much more so, it likely defeats the purpose. The point is to be lazy (that's at least part of why people are willing to spend money on it).
On the other hand, I'd also expect the most degenerate among us to not care about the uncanny details. So who knows, really.
And let's not kid ourselves, you aren't programming anything. You are paying Anthropic to brute force the code for you with their supercomputer which was only built thanks to the rare earth mining and human rights violations you claim to be bothered about.
reply