LLemdashes
13 points by outervale
13 points by outervale
this seems to get at the heart of some of the issues that came up in the meta-discussions about a putative lobsters policy regarding LLM generated content: how do you balance the desires of eg this userbase to avoid nonhuman writing unilaterally, vs the reality that we have nothing more than subjective guesses to distinguish between human and nonhuman writing?
You move the detection up a level. If the output is full of em-dashes or bullet points or Short. Punchy. Sentences, you check whether the content actually conveys an idea or argument in a coherent manner.
If it doesn't it's slop, and it's equivalent to spam in that it's content-shaped but primarily designed to fill a marketing-shaped hole. If it does, and it's on-topic, it arguably belongs here. It might not get as many upvotes, it might get flagged, the author or the submitter might be flamed in the comments, but after a while an equilibrium will be reached about what's appropriate to post here.
In principle I agree with you, if I am reading you correctly I made similar argument earlier today. I have also found that a large portion of the user base here (and likely elsewhere on the internet) bristles at the thought of even the slightest LLM involvement. To the point that some of them call anything LLM involved slop. Where to me it seems next to impossible to put that genie back in the bottle and I rather look at how much effort the author overall put in what they are trying to share.
Agreed, but at that point it doesn't matter if a human wrote it or an LLM did. Slop is slop, why muddy the waters with a wild-ass guess about where it came from? Similarly, by this criterion you have to accept LLM-assisted writing that does convey an idea or argument in a coherent manner — and a lot of people around here would seem to disagree with that implication.
Yeah, my stated point[1] is that if it's coherent and consistent etc. it is ok - even though the author spent more time wrangling prompts than actually writing it!
I'm only speaking for myself here. I am against a blanket ban on content with any LLM "feels" because there's no clear line defining it. Not having that line makes a prohibition unfeasible, in my opinion.
But I am also for members deciding for themselves where they feel the line goes, and acting on that, by flagging, not upvoting, mentioning in comments etc.
[1] per 0vfxye
To me this reads like forcibly killing of the vibrant diaspora of human linguistic expression in favor of something that people feel like isn't AI. Of course, this new style will immediately be emulated by AI.
I totally think I can guess ai writing based on vibes.
I honestly think I can spot AI writing just from the vibe.
Which one was ai? It’s that easy imho.
Only zealots would use the em-dash criterion as a singular aspect to judge if something is AI-generated or not. In reality, this is part of a bigger picture. Reading the author's lines, it's apparent the text is handwritten, and the em-dashes are just sprinkles of punctuation.
Only zealots would use the em-dash criterion as a singular aspect to judge if something is AI-generated or not.
I'd argue it's less about zealotry and more about not being familiar with emdashes as a punctuation device outside of LLMs.
I'm sure there are plenty, plenty people who have either never encountered them or just never noticed them prior to LLMs.