Let's allow AI output in some restricted sense

[Palli]

Continuing the discussion from I don’t want to read LLM output:

I was going to suggest allowing AI output, and have it labelled, then that was the next thing a saw in that very long thread. [While I like reading threads, that one is getting very long… so I like some resolution.]

Full disclosure: I’ve had one post of mine hidden likely because of AI output:

[ANN] JuliaLibWrapping.jl: bringing Julia libraries to C and Python users

Test a distribution with pip install in a clean virtual environment on such a machine. I’m not involved with the project just posting what seems relevant above and: Pass privatize = false only when that isolation is knowingly unnecessary. Only the Julia runtime libraries are privatized. BLAS, Fortran, and unwind libraries may still be shared. Bundles produced by different Julia versions have not been established as mutually compatible. This behavior has been measured on Linux; macOS and …

There as always, I’m just trying to help people, most of the text was mine, like a screen-full, and the AI part only like 4 lines in the end, and I dropped them and the inferior/incomplete answer appeared again.

The discussion has been mainly about announcement posts, I don’t know the post(s) being objected to. I do know some of the great packages I see are made with AI, to a large degree, and we will see more of such. I don’t think AI use should be overused in such announcement posts, right now, the ban on AI here is more strict than for the packages you make. I think the main thing is that people know they are getting some genAI. It would be great if we could even color-code genAI parts. At least just state it, and if discourse could label as 30% is AI content etc. that would be great. First thing is do we like to allow some exceptions, like maybe up to 40% max genAI? I suppose we might want to review the policy over time… I don’t know how to make a poll on this.

[Nathan_Boyer]

This topic was expanded on and largely settled in this other thread:

Updates to the site guidelines, especially regarding gen-AI

That might indeed go into the right direction. I would simply suggest to a) appeal to refrain from AI use in direct communication with people (e.g., more or less “chatting” or “talking” with each other), b) make clearer content quality related guidelines not simply (do not use a certain tool), and c) clearly offer the poster a chance to heal the post. What about that? EDIT: A more technical solution would be “transcribed video messages” so there can be no doubt; might even be a market for suc…

Updates to the site guidelines, especially regarding gen-AI

Holy cow, I think we might have done it! @gwr: that’s the status quo, or at least the intent of it! If and how we improve the wording in the guidelines, flags and DMs is really the only question… along with the fact that I’d say that this entire site is for direct communication with people. But, yes, that is exactly our intention.