Post #1499116
2026-04-19 23:12 UTC
I think people who use LLM's to build FOSS projects essentially just train the LLM's. Like say you build an XML parser for Swift using LLM's. That's cool and a legitimate target for doing LLM's since you have a comprehensive spec to test against.
But what's really happening here is that you let the LLM build something for you, and then (if you are a responsible person) you do a deep review of it before you release it. You'll train it to do the right things and tell it what's been doing wrong.
Replies (1)
-
@helge@mastodon.social 2026-04-19 23:14
And then the loop begins, the next iteration of LLM's will now have the information on what was wrong with the former. And a Swift version of an XML parser, so it can easily replicate those on a good basis. It would be literally communism if you wouldn't be billed by some U$ company for it 🙈