Post #4116099
2026-07-25 20:30 UTC
@severin@hci.social I'll disagree on at least three points here :
- Thinking that LLM allow to build "excellent" open-source tools only look at the code production part, not at the maintenance. If an LLM produces code that no one understand and no one can maintain in the long term, it's not an "excellent" open source tool, just a time bomb. The best free software project are never just the code, but about the community, the interactions, the design process and how we collectively improve that common (in Ostrom's sense).
- I also want to live in a world where people own their tech stack and do not depend on the whims of a few capitalistic actors, such as all the actors with the power to (re)train LLMs (anyone with enough money to build a datacenter and influence to bypass/ignore whatever IP law that may exist anywhere). Normalizing that code production can depend on those actors is already capitulating, and depending on those tools for producing *anything* is the contrary of empowering, it's saying that learning is useless.
- "Being a solid coder" is a craft and asking the (mostly volunteer) maintainers, who master that craft, to accept whatever output anyone who is feeling empowered by today's trillionaires thinks is an improvement to their software is asking too much of them. I can mentor someone who wants to learn, but I cannot mentor anyone who already thinks that learning is useless.
Replies (2)
-
@FlashMobOfOne@mstdn.social 2026-07-25 23:49
@mc@social.sciences.re @severin@hci.social "I cannot mentor anyone who already thinks that learning is useless." >>> Ultimately, this is the crux of the issue.
-
@shaedrich@mastodon.online 2026-07-26 12:20
@mc@social.sciences.re @severin@hci.social My two cents: – Readable code is not only nice, I believe, that it is imperial for open source: Maybe not for the one producing the code because they probably use the LLM to update this very code. But for others to read, learn form it, and being able to participate without an LLM – Also, in the sense of FLOSS, using (proprietary) LLMs makes the resulting code also proprietary in a sense that it is now depending on a closed-source source which kinda defeats the purpose of FLOSS.