[electrograv]

Do they think Copilot/ChatGPT are usernames of contributor accounts which can be banned from a project/repo?

It seems the underlying issue here is bad code submitted ultimately by human contributors. Consistently thorough code review and testing will always be necessary.

[nimih]

How do code review and testing address the licensing concerns identified by the project? If you read their policy (which is two paragraphs, one of which is quoted directly in the article), it's pretty clear that the quality of the code is besides the point.

[electrograv]

I’m not claiming that quality or licensing concerns are irrelevant, but that:

1. It will not be possible to reliably detect whether code was LLM assisted or not.

2. Humans are not always 100% truthful.

3. All the concerns cited here also applies to human-written code anyway.

So attempting to treat LLM coding assist tools as a special case here is going to be a losing battle. To solve these issues, we’re gonna have to come up with code review processes and tools that apply to ALL code up for review.

[nimih]

Those same quibbles applied to the policy before the addition of the LLM section: how does the NetBSD project detect if I copy & paste a bunch of code from my day job into a patch submission (and then lie about it)? Obviously, they can't. I, personally, don't feel like it's a failure of the policy if it relies on your contributors acting in good faith, because:

a) many people are acting in good faith, and their behavior will change as a result of this policy;

b) if someone wants to be a jerk and use an LLM after they were told not to, and is at some later time found out, it makes it easier for the org to act quickly and in a fair and consistent manner;

c) [more speculative as to the motives of the NetBSD project] normative statements by well-regarded institutions are useful in setting an example for other organizations to follow, so there is some political utility regardless of the practical efficacy of these rules.