Post

I appreciate that you're reflecting on this critically, but personally I'm not moved by calls for nuance on this issue anymore, because I don't think the criticisms are unnuanced. I don't think the concepts of statistical inference of vector databases are what anybody fundamentally takes issue with.

It's the societal harm, the theft and devaluing of art, the worker disenfranchisement, the mental health effects, the baffling use in warfare, every available means of entertainment online getting infested with vacuous slop. And yes, the environmental impacts. It just never ends.

For one, there are individual issues here that would prevent me from trying to benefit from the same tech that causes them, given the choice. And that's even if I was convinced that I would benefit, which is a separate question. I don't see much evidence that software as a craft is better off.

For two, the sheer laundry list has become so exhaustive and overwhelming that it feels silly to me to run through it and then stop short of saying that LLMs are holistically objectionable. At least as currently constituted.

The nuance I'm willing to extend at this point exists in hypotheticals. There could have been a timeline in which we all got a couple more plugins for our IDEs and some natural-language search tools, where this didn't encroach on every corner of our lives, and we might not have minded so much.