[harringdev]

This is very hard to take seriously, it feels ideologically driven.

The introduction where they claim LLMs are useless for software engineering is just incorrect. They are useful for many software engineering tasks. I do think that vibe coding is rubbish however, and more junior SWEs very regularly misuse LLMs to produce nonsense code.

The only substantive point is that the LLM may regurgitate pieces of proprietary training data; although it seems unlikely that it would be incorporated wholesale into the codebase in such as way it matters or opens them up to liability.

I do question if LLMs would even be useful for such a niche project -- but I think this should be left up to developers to figure out how it complements their workflows rather then ruling out all uses of LLMs.

EDIT: I want to point out that I think the Asahi Linux project is a jewel of engineering and is extremely impressive.

[roxolotl]

Why is something ideologically driven inherently not worth taking seriously? I appreciate maybe their understanding will be more challenging to change but people don’t necessarily arrive at ideologies for unserious reasons nor do ideologies have unserious results.

[harringdev]

You make a good point, and I will clarify what I mean.

A small disclaimer, I am not a AI-booster. I think LLMs do have issues, and one should be careful with them.

I've found that there's a large group of people who dislike LLMs strongly and claim they are totally useless or even pernicious. I think this is grounded in truth -- they are trained on copyright work, they are used for spreading misinformation, they can produce sh^t code. Although some folks take a radical/extremist approach and totally dismiss them as useless -- often without actually using LLM-powered tools in any meaningful way.

They are useless for many applications, but programming is not one of them. I think a blanket ban on LLMs has to be somewhat unfounded/radical because they do have practical applications in writing code. The tab-completion models are extremely useful for example.

For this more niche project I would think LLMs might not be as useful as they are for other projects. This said, I still think there would be a variety of use-cases where they could be helpful.

[pjc50]

Linux is, in and of itself, ideological.

[mpalmer]

I think the better word is extremist, because choosing to brand all LLMs as "Slop Generators" is just that, an extremist position with a rather self-satisfied air about it.

[nsksl]

Because decisions like these have to be taken purely on the basis of technical merits, not political ideology.

[queenkjuul]

GPL is a purely technical document?

[000ooo000]

Says who?

[josefritzishere]

Two large scale studies so far. It's worth saying that thus far there have only been two studies on the efficiency gains AI produces and they both report that it's a negative number. Perhaps later studies will rebuke those findings but SCIENCE! At this time we lack findinfs that support the contention that AI is a benefit at least when measures as work hours.