Post B6UHk7aahAHS1iRZbs by dequbed@mastodon.chaosfield.at
 (DIR) More posts by dequbed@mastodon.chaosfield.at
 (DIR) Post #B6UHk7aahAHS1iRZbs by dequbed@mastodon.chaosfield.at
       0 likes, 0 repeats
       
       Since I have seen a bunch of sneers on here about systemd & LLM agents, can anyone point me to any evidence that Poettering is using an LLM agent for coding? Because looking over his last few commits in systemd he doesn't seem to other than for fixing up prose. And the remaining use of LLMs in systemd seems pretty limited as well.Or is this just another wave of the typical systemd hatred that has seen people that have moved Linux forward more than most get death threat after death threat?
       
 (DIR) Post #B6UHk7vVRP0j4aEHD6 by SRAZKVT@tech.lgbt
       0 likes, 0 repeats
       
       @dequbed idk for code, but he uses llms pretty actively in reviews, which means he relies on its output to decide what to change. it's not quite as bad as code, but it's also quite bad as it leads to the loss of skills (specifically about reading and reviewing, which is a very hard skill to obtain and maintain, much more so than writing code)
       
 (DIR) Post #B6UHk8CsOouBwSM9Hk by SRAZKVT@tech.lgbt
       0 likes, 0 repeats
       
       @dequbed https://github.com/systemd/systemd/blob/main/.github/workflows/claude-review.yml there is even a full workflow that automatically runs on every pr. even if he doesn't use llms to generate code, this is knowingly sending every pr changes to claude, regardless of the contributor's wishes (which is an attack on labour and consent)
       
 (DIR) Post #B6UHk8X5Bh4Ix7oHmS by natty@astolfo.social
       0 likes, 0 repeats
       
       @SRAZKVT@tech.lgbt @dequbed@mastodon.chaosfield.at okay but This is arguably one of the least shit ways to use it, and inference does not imply training
       
 (DIR) Post #B6UK0U5f0Ji2M1uGB6 by dequbed@mastodon.chaosfield.at
       0 likes, 0 repeats
       
       @SRAZKVT I will give you attack on labour (and all other ethical problems associated with LLMs), I will not give you attack on consent since the fact that this will happen is not hidden from a contributor, *at all*.
       
 (DIR) Post #B6UK0UVtQmh1fOBD4C by SRAZKVT@tech.lgbt
       0 likes, 0 repeats
       
       @dequbed then i guess you're fine with a project effectively banning anyone who doesn't want their work thrown into a garbage grinder
       
 (DIR) Post #B6UK0UnGOCaUXGJ58q by dequbed@mastodon.chaosfield.at
       0 likes, 0 repeats
       
       @SRAZKVT No, I'm not? I'm *just* saying that this has nothing to do with consent! If a project only allows (upstream) contributions under specific conditions that's *their* choice. Is it ethically problematic? Yes. Am I uneasy with it? Yes. Does it violate consent? No.
       
 (DIR) Post #B6UK0V0NbR4zBwRYaO by SRAZKVT@tech.lgbt
       0 likes, 0 repeats
       
       @dequbed whatever i'm not interested in going further with this conversation or i'll just get angry over nothingthe ethical reasons should be enough, regardless of the technical ones, and i'm not interested in talking with someone who reads to me like they believe otherwise
       
 (DIR) Post #B6UK0VCmrJ0JoQFSvQ by dequbed@mastodon.chaosfield.at
       0 likes, 0 repeats
       
       @SRAZKVT that's fair, enjoy the rest of your day :)But to clarify, I'm not excusing LLM usage or waving away ethical concerns here. My problem is that the greater conversation looks like yet another unfounded smear campaign against systemd; and those have been dangerous enough in the past.
       
 (DIR) Post #B6UK0VTnq2cCfCD3Ro by SRAZKVT@tech.lgbt
       0 likes, 0 repeats
       
       @dequbed right, because systemd is a force of good that has never done anything wrong until today, and only turned evil now