chezmoi is a great tool, and I admire this project taking a strong stand. However I can’t help but feel that policies like this are essentially unenforceable as stated: there’s no way to prove an LLM wasn’t used to generate code. In many cases it may be obvious, but not all.
This is dumb. Llms are a tool, a very useful one. Bad PRs should be rejected always no matter the source, but banning a tool because some people can't use it is not what engineering is about.
This sounds limiting. I compare LLM generated content to autocomplete.
When autocomplete shows you options, you can choose any of the options blindly and obviously things will fail, but you can also pick right method to call and continue your contribution.
When it comes to LLM generated content, its better if you provide guidelines for contribution rather than banning it. For example:
* if you want to generate any doc use our llms_doc_writing.txt
* for coding use our llms_coding.txt
This false equivalence with autocomplete is a red herring. You can't just dismiss the very real problems of slop coding the maintainers were forced to deal with[1] by comparing it with something tangentially related that has none of the problems.
I'm minimally exposed to vibecoding, but already finding it immensely useful. That said, one thing I don't want to do, is to touch that autogenerated code, hardly opening in an editor.
Anyone feeling the same? That they're not for humans to see?
What's interesting is the change in the policy. Old policy:
> If you use an LLM (Large Language Model, like ChatGPT, Claude, Gemini, GitHub Copilot, or Llama) to make a contribution then you must say so in your contribution and you must carefully review your contribution for correctness before sharing it. If you share un-reviewed LLM-generated content then you will be immediately banned.
...and the new one:
> If you use an LLM (Large Language Model, like ChatGPT, Claude, Gemini, GitHub Copilot, or Llama) to make any kind of contribution then you will immediately be banned without recourse.
Looking at twpayne's discussion about the LLM policy[1], it seems like he got fed up with people not following those instructions:
> I stumbled across an LLM-generated podcast about chezmoi today. It was bland, impersonal, dull, and un-insightful, just like every LLM-generated contribution so far.
> I will update chezmoi's contribution guide for LLM-generated content to say simply "no LLM-generated content is allowed and if you submit anything that looks even slightly LLM-generated then you will be immediately be banned."
> Isn't it more reasonable to explain in excruciating detail what kind of contributions you will allow?
No, it's not. You can read the rule as "If it's obvious enough your code has been LLM-generated, you will get banned" if you feel like the conciseness of the current rule makes you uneasy about using Copilot.
Besides, I suspect in the maintainer's case, banning unreviewed LLM contributions is effectively congruent to banning all LLM contributions.
If you think the rule is unfair towards LLMs because they can do such good, feel free to open a good, clean, useful PR clearly stating how you used the LLM to generate code.
14 comments
[ 3.1 ms ] story [ 42.5 ms ] threadAnd corresponding discussion: https://github.com/twpayne/chezmoi/discussions/4010
When autocomplete shows you options, you can choose any of the options blindly and obviously things will fail, but you can also pick right method to call and continue your contribution.
When it comes to LLM generated content, its better if you provide guidelines for contribution rather than banning it. For example:
[1]: https://github.com/twpayne/chezmoi/discussions/4010
Am I now assured that the copyright is mine if the code is generated by AI? Worldwide? (or at least North America-EU wide)?
Do projects still risk becoming public domain if they are all AI generated?
Does anyone know of companies that have received *direct lawyer* clearance on this, or are we still at the stage "run and break, we'll fix later"?
Maybe having a clear policy like this might be a defense in case this actually becomes a problem in court.
Anyone feeling the same? That they're not for humans to see?
> If you use an LLM (Large Language Model, like ChatGPT, Claude, Gemini, GitHub Copilot, or Llama) to make a contribution then you must say so in your contribution and you must carefully review your contribution for correctness before sharing it. If you share un-reviewed LLM-generated content then you will be immediately banned.
...and the new one:
> If you use an LLM (Large Language Model, like ChatGPT, Claude, Gemini, GitHub Copilot, or Llama) to make any kind of contribution then you will immediately be banned without recourse.
Looking at twpayne's discussion about the LLM policy[1], it seems like he got fed up with people not following those instructions:
> I stumbled across an LLM-generated podcast about chezmoi today. It was bland, impersonal, dull, and un-insightful, just like every LLM-generated contribution so far.
> I will update chezmoi's contribution guide for LLM-generated content to say simply "no LLM-generated content is allowed and if you submit anything that looks even slightly LLM-generated then you will be immediately be banned."
[1]: https://github.com/twpayne/chezmoi/discussions/4010#discussi...
Then before submitting if I ask an LLM to review my code and it proposes a few changed lines that are more efficient. Should I then
- Leave my less efficient code unchanged?
- Try to rewrite what was suggested in a way that’s not too similar to what the LLM suggested?
No, it's not. You can read the rule as "If it's obvious enough your code has been LLM-generated, you will get banned" if you feel like the conciseness of the current rule makes you uneasy about using Copilot.
Besides, I suspect in the maintainer's case, banning unreviewed LLM contributions is effectively congruent to banning all LLM contributions.
If you think the rule is unfair towards LLMs because they can do such good, feel free to open a good, clean, useful PR clearly stating how you used the LLM to generate code.