Mastodawn

RE: https://hachyderm.io/@janvhs/116184719405207871

We’re quickly approaching a bifurcation in FOSS with projects including or not including LLM generated code.

We’ll have to come up with a name so we can talk clearly about it.

I desperately hope the non-LLM side is going to be large enough to include a full stack, otherwise there’s a lot of work ahead of us for no good reason.

—

At least until we have truly free and open, non-exploitative, consentually sourced, sustainably powered tooling, which is a bubble or two away at least.

Show thread

DNAvinci Mar 7

@janl
I guess I don't care how the code was written, so long as the author can answer questions about it, and the code is of sufficient quality that I can read and maintain it.
Usually one or both of those are not the case when someone makes a LLM PR.

If they want to make the chatbot refine from something that "definitely executes" into something that is robust, minimal, and follows a human-like pattern, honestly I might not be able to tell that it's bot slop. Be less stingy with your tokens?

Show thread

Jan Lehnardt

@dnavinci you don’t care that the models are full of stolen data, use exorbitant amounts of energy and water, make the objectively worst people on the planet even richer and put the power to create software into the hands of three companies?

(“No” is a fair answer)

Show thread

DNAvinci Mar 10

@janl
Naturally I care.
But they already did it anyways, so do you want the improvement or do you want to throw away that electricity?
Obviously that should be discouraged, but most of these folks aren't going to read your PR policy before having it written.

I'll level some nuance here -
I use self hosted models on my $500 gaming GPU.
It's still stolen data, but I pirate a lot too.

Also check out Hank Green's nuance on the water use.

Happy to dismantle capitalism as well.