LLM (ChatGPT) output as forum posts

[tkrunning (Thomas K. Running)]

@sj-to-pt

I think the “easily verifiable research” use case is still kind of a no-go, personally.

My point was that it’s fine to use it as a tool to dig up relevant sources (which should be verified before sharing), but not rely on or necessarily even share its output in these cases.

@cleversprites

Your details proposal is probably the right balance considering ease of implementation, but ideally I would like:

• A new tag that is LLM-specific, such as [ai-text]
• A summary field to help the reader decide whether to expand the text or not, which could be as simple as the LLM prompt string
• Expanded LLM text rendered in a different font and/or color

so for example:

[ai-text=“new law grants instant citizenship to nomad gate members”]
…AI slop…
[/ai-text]

Ideally we’d have a dedicated tag for it with a UI element users could click to add it. However, I don’t think there are any existing plugins for this forum software that adds anything like this. And I don’t think it makes sense to spend time/resources on developing and maintaining such a plugin. Or maybe I can ask ChatGPT to do it?

So let’s use the Hide Details feature for now. Instead of remembering the markup you can also insert it with two clicks in the UI:

@cleversprites

Finally, I’m assuming that not properly enclosing LLM text would be considered a flag-able “offense” to complete the new policy.

Yes, please flag instances of it. I’ll see if I can add a custom flag for it, for now just use Inappropriate or custom message.

There’s now an option to flag a post as containing unmarked LLM/AI output: