Change "prose contributions" to "arguments and opinions"

[sffc]

There are situations where AI usage should not be allowed:

• Expanding a thought from a prompt to a paragraph. Just post the prompt!
  • Explanation: We all have access to the same LLM tools and can expand on the thought ourselves. We don't need your LLM to do it.

There are situations where AI usage is allowed:

• Machine translation
• Proofreading, so long as the proofreading avoids adding new content

And then there are situations which we haven't specifically evaluated yet:

• Can an LLM be used to do research and post the output of that research?
• Can an LLM be used to post data involving a nicely formatted graph or table?
• Can an LLM be used to triage issues; for example, taking an issue whose comments went in multiple directions and spawning multiple sub-issues summarizing the comments?

I prefer to leave the door open for such use cases, and add them to the policy when necessary. To achieve this, I am proposing to change the policy wording from the broad "prose contributions" to the more specific "arguments and opinions", which I think better captures the spirit and intent of why we originally made this policy.

[bakkot]

I intended to rule out most of those case by this policy. I'm OK with having an LLM format a chart for you, I'd classify that under proofreading, but I do not want to (for example) read someone else's LLM's summary of all or part of issue.

[ljharb]

I do not want AI to do any translation - I can do the translation myself. I only want people to write in a language they're sufficiently proficient in, so their ideas come across correctly.

[bakkot]

We do specifically allow translation in the current policy. FWIW a lot of people are uncomfortable posting comments in a language other than the dominant language of the conversation so I think it's a reasonable to allow them to translate first so they can participate without discomfort.

[ljharb]

There are plenty of non-LLM-based translation tools they can use in that case.

[bakkot]

I think most people default to Google Translate (which is LLM-based) or sometimes DeepL (which is LLM-based). What are you suggesting they use instead?

[ljharb]

oof, i didn't realize they'd made that change. fair enough, although i'd still prefer people simply gain comfort with participating in their preferred language.