Embed Notice

Debating changing the ngscopeclient AI policy to be a bit more practically enforceable.

Basically, vibe coded / LLM generated code will continue to be prohibited as it tends to be low quality slop of unclear copyright provenance that is both legally risky and full of bugs.

But there won't be a explicit prohibition on trivial autocomplete usage, whether LLM based or classical dictionary based.

The rationale is based on practicality of enforcement of the current hard-line model: while I'm generally opposed to AI tools broadly from an ethical perspective, from a project governance and legal risk perspective I don't really care - and more importantly have no way to know - if you typed a variable name fully or if some pile of linear algebra completed it for you, or where the closing curly brace at the end of your function came from, etc.

So while I may not like your choice of tooling, if the code is correct I have no way to reject the contribution other than to rely on the author to out themself.

If the policy is instead basically "no slop" we have discretion to reject a PR and ban a user from further contributions if they submit an obviously machine generated incoherent garbage PR, but we don't have to nitpick over the ethics of whether a neural network generated curly brace is any different from one typed by a human if we didn't know which it was in advance.

I feel like ultimately any enforcement of such a policy will have to be subjective by the maintainer since there's not a reliable way to tell with certainty that a piece of code was machine generated unless they do something stupid like leaving the prompt in the code.