But what if it turns out that human+LLM can produce more "thoughtful, curious discussion" than human alone?

That's the dichotomy: Do we prefer text with the right "provenance" over higher quality text?

[Perhaps you'll say that human+LLM text will never be as high-quality as human alone. But I'm pretty sure we've seen that movie before and we know how it ends.]

That said, you're right that because human+LLM is so much more efficient, we'll be drowning in material--and the average quality might even go down, even if the absolute quantity of high-quality content goes up.

I think, in the long term, we will have to come up with more sophisticated criteria for posting rather than just "must be unenhanced human".

▲

Avicebron 7 hours ago | parent | next [-]

I think "must be unenhanced human" is probably the most sophisticated criteria even if it's simple. I don't think there's much value in trying to optimize the perfect "thoughtful, curious discussion", why would there be, it implies some ideal state for "thoughtful and curious" vs the reality that discussions between living breathing people is interesting by default as long as folks loosely follow some guidelines.

▲

altairprime 7 hours ago | parent | prev | next [-]

> what if it turns out that

HN need not offer itself up as a Petri dish for AI writing experimentation. There are startups in that space, and at least one must be YC-funded, statistically speaking. Come back with the outcomes of the experiment you describe and make a case that they should change the rule. Maybe they will! As of today, though, they are apparently unconvinced.

> the average quality might even go down

We have a recent concrete analysis of Show HN indicating support for this possibility, resulting in the mods banning new users for posting to Show HN (something they’ve probably been resisting for close to twenty years, I imagine, given how frequent a spam vector that must be).

> Perhaps you’ll say that human+LLM text will never be as high-quality as human alone

Please don’t put words in my mouth, insinuating the tone my reply before I’ve made it, and then use that rhetorical device to introduce a flamebait tangent to discredit me with. I’ve made no claims about future capabilities here and I’m not going to address this irrelevance further.

> in the long term, we will have to come up with more sophisticated criteria

Our current criteria seem sophisticated already. Perhaps you could make a case that AI-assisted writing helps avoid guideline violations. This one tends to be especially difficult for us all today:

”Please respond to the strongest plausible interpretation of what someone says, not a weaker one that's easier to criticize. Assume good faith. Eschew flamebait. Avoid generic tangents.”

▲

GMoromisato 6 hours ago | parent [-]

> Please don’t put words in my mouth, insinuating the tone my reply before I’ve made it, and then use that rhetorical device to introduce a flamebait tangent to discredit me with. I’ve made no claims about future capabilities here and I’m not going to address this irrelevance further.

I apologize--the "you" I meant was the person currently reading my post, not the person I was replying to. I was merely trying to answer a common objection that I've heard.

> HN need not offer itself up as a Petri dish for AI writing experimentation.

I'm not sure HN has a choice. I don't think we can prevent posters from experimenting with LLMs to post on HN--even if they adhere to the guidelines. For example, can I ask the LLM to come up with the strongest argument it can and then re-write it in my own words? That seems to be allowed by the guidelines. Would someone even be able to tell that's what I did? [NOTE: I did not do that.]

I think you're arguing that we should not encourage even more use of LLMs on HN. I get that. But I feel like that this community is uniquely qualified to search for better solutions.

> Our current criteria seem sophisticated already.

I hope you're right! That implies that you believe the current guidelines are sufficient to keep HN as the place we all love despite the assault from LLMs. I'm skeptical, but I've been wrong plenty of times!

▲

altairprime 6 hours ago | parent [-]

> I don't think we can prevent posters from experimenting with LLMs to post on HN

And yet, she persisted, we will still set guidelines; so that people know they’re unwelcome to do so when they do, so that they can’t argue that they didn’t know, so that we as a social club can strive towards the standards we argue about and accept from the organizers. The point of guidelines is not that they prevent malicious intent; the point is that they inhibit those behaviors that exceed the defined boundaries, however vague or precise they may be. Prevention of malice is an impossibility in all human social affairs, whether guidelines are defined or not; one must find other reasons for rules than prevention to understand why rules are at all.

▲

GMoromisato 6 hours ago | parent [-]

> And yet, she persisted, we will still set guidelines

I'm not sure if you're including or excluding me from the "we". If you're excluding me, then I feel our conversation has come to an end.

But if you're including me, then I think the guidelines need to evolve to deal with LLMs. Maybe not right now--maybe the current guidelines are sufficient for the next year or two or three. But I think we as a community are uniquely qualified to design and influence the future of internet social clubs in the face of LLMs.

	▲	altairprime 5 hours ago \| parent [-]
		> I'm not sure if you're including or excluding me from the "we". “We” here refers to individual human beings that are members of the human social-entity constructs (‘social clubs’) that precipitate naturally out of human groups, both in general to all such groups and in specific to the group under discussion here today, HN participants. Whether or not you’re a member of “we” HN participants is conditional on whether or not you are honoring the policy of no AI-assisted writing at HN that is in effect as of whenever you saw this post or the new guidelines. I have no judgment to offer you in that regard, and in any case you’re readily able to decide that for yourself. Separately, I’m not engaging with discussion about future policy; perhaps you should start a top-level thread about it, or write a blog post and submit it (after a few days have passed, so it doesn’t get topic-duped and so that passions have cooled somewhat).

▲

davebranton 7 hours ago | parent | prev [-]

It doesn't matter.

The guidelines are perfectly clear, no matter the outcome of your thought experiment. Hacker News wants intelligent conversation between human beings, and that's the beginning and the end of it.

If you want LLM-enhanced conversation then I'm sure you will find places to have that desire met, and then some. Hacker News is not that place, and I pray that it will never become that place. In short, and in answer to "Do we prefer text with the right "provenance" over higher quality text?".

Yes. Yes, we do.