Let me ask you my hypothetical question:
What if the code written by me is worse than the code generated by an LLM?
What if my English or Czech can express the idea better, and the LLM can transform it into the target language more precisely than I could?
When LLM contributions are not permitted...
should I not be permitted to contribute?
What if I waste more of the maintainer's time than the clanker?
1/2
From a maintainer's POV:
a) If the person is willing to improve their coding skills, does it make sense to invest time in them?
b) But what if the person is never going to code in the language the project is written in and doesn't care about learning it? What then?
The post assume the code addressing some real bug somewhere affecting one or more users.
I'm looking here only on practical aspects, ignoring the impact on the world as whole. Focusing on the local impact on the project.
2/2
LLMs likely write code much better and more efficiently than me, I waste hours on issues that LLMs could knock out in minutes really, but if they do, and I write code, even for personal fun accountable to no one, will I ever understand or be able to modify it? FOSS is for me knowing other people can write my code better, and allowing them to patch it <continued>
For english, I often fall into pitfalls of not knowing how to express myself correctly, its an eternal struggle, but to add, I have never personally felt if my grasp of a concept leaves wordless, that I can read someone else's translation and understand it better. If I am not sure my words are good enough, I cant really be certain a bot's are better for me.
- replies
- 0
- announces
- 0
- likes
- 0
@okias From a person outside of IT business. Can you maintain LLM generated code? And how? Manually, eg some human understands how it works internally on complex projects? Or with LLMs without introducing even more bugs / slop / not needed complexity? Isn't it "cheaper to tell ppl get 10x faster CPU and 2x the RAM, than to pay senior human for code optimalisation" on steroids?
@ati1 my experience:
1. you can maintain, if the new code doesn't introduce architectural changes (e.g. just filling the places already designed by REASONABLE human). You do it same way as before.
2. Maintaining code using LLM ‒ I consider that bad idea. Maybe I'll change my view over time.
3. For architectural decisions, paying principal (not senior) engineer is probably best idea.
@migratory I have answer for myself, but I'm curious what others think. Mostly would be interested in devs/maintainers view.
@okias my gut feeling is that if you can express the problem well, making an amazing issue is the way to go instead of an AI generated PR
@okias Wouldn't it be better if the maintainer would use LLMs on the raw data themselves?
Let people write their thoughts in their own language, translation is cheap (for example on-device translation in Firefox), avoids losing information for potential double translation.
Let people write about issues with their thoughts, then the maintainer can easily pipe this into their LLM of choice and iterate themselves avoiding the potential back and forth with the reporter.