Ask HN: Time to discourage "I asked ChatGPT" comments?

21 points by fire_lake ↗ HN
There is a trend of people on HN responding to posts by asking ChatGPT and putting its response. I don’t think this adds much to the conversation since anyone who cares to can easily do this themselves.

Is it time to discourage such posts in the HN rules?

Note: this is a different problem to people posting generated content for astroturfing purposes etc. The author is not hiding the fact that ChatGPT has been used.

Note 2: please don’t ask ChatGPT for me.

15 comments

[ 3.4 ms ] story [ 42.9 ms ] thread
Anytime I see it, they are ranked in the bottom of any post so people doing it are losing karma. That's enough.
So let me see if I'm understanding; currently there are two scenarios on HN:

- People declare that they used ChatGPT/et al for commenting.

- People who hide that they used ChatGPT/et al for commenting.

You expressly want to go after the former group with new rules that would punish them for those declarations, essentially forcing them into the second group? And you think this will somehow improve the site?

I don't understand this at all. You're suggesting punishing honesty and rewarding people who misattribute their contributions. Talk about a perverse incentive.

There is also a 3rd, vastly more popular scenario - those who dont use ChatGPT to make their comments.

As we are all capable of using ChatGPT, what does regurgitating a ChatGPT response add to any conversation?

It is often used as a way for those who don't understand a topic to participate, and is nearly always self-evident, whether declared or not. It usually results in a downvote from me, but I can appreciate why others would want a rule that can be pointed at.

Yes, but I would prefer that they either:

1. Add their own commentary, or

2. Don’t post at all

I don’t think they would move to the latter group since they have already gone to the effort of signposting their usage of ChatGPT.

The problem with banning ChatGPT/et al entirely is that it is unenforceable. We've already witnessed multiple "AI detection" offerings that have such high false-positives that they're harmful (or human moderators that use their gut to ban stuff based on tone/or being too high effort).

I myself have been victim of this, I used to post high quality bullet summaries of articles that took quite a while to create (I did the unthinkable: read the articles), I was doing this pre-LLMs for years but then got banned from three different popular subreddits within a few months after ChatGPTs initial release for posting "AI generated content" in spite of my account history. There is no way to prove something is hand-created.

So you can create a rule to ban it, which you cannot enforce, and then you can create a rule to injure people who declare its usage but all you'll end up with is ChatGPT/et al generated content anyway and the honest dealers being pushed underground/replaced by dishonest ones.

I'm sure HN will have a rule against AI generated content eventually. But except in cases where they outright admit it, it won't be possible to enforce, or they'll enforce it incorrectly on human content.

There are lots of things we can’t practically enforce but still make sense to uphold as norms.

- Don’t tell lies. - Don’t post in bad faith. - Don’t cut and paste from Wikipedia as if it is your own writing. - Don’t post ChatGPT responses as if you wrote them. - Don’t post ChatGPT “opinions” even labeled as such.

This is interesting. I have been wondering if we are going to get towards a reality like that movie surrogates where we are going to "insist" people to come up with best answers (and explicitly with gpt) instead of declaring GPT out of honesty or trying to do it all themselves! At that point "declaring" is just noise no? We might as well start assuming that very little content is going to be 100% non-AI generated. I thought the OP was suggesting - banning those titles and not GPT use itself. So something like "Dammit just post something interesting, dont tell me if it is GPT or not".

  > Is it time to discourage such posts in the HN rules?
No, because sometimes they can be useful. If you don’t like a particular comment, downvote it or ignore it.
Think about what you are asking ChatGPT, and what a reasonable answer would be...

ChatGPT answers confidently 100% of the time, but if you know that, it is an amazing heuristic to get work done. It can give you 80% of a job instantly; say you are writing a script or some email, it will do that for you.

Every once in awhile it will output a nonsensical answer, which then if you question it, it will fix or admit it gave a nonsensical answer.

For example, it said I should connect modules together to create a graph to import, but this would clearly be an incorrect result, you have to be able to spot this kind of crazy in the result.

Otherwise, it is amazing for outlining ideas and getting started on projects, and for revising text.

These comments are already actively discouraged and some of the fastest flagged comments around.
(comment deleted)
The community already enforces this. The question is if a rule is necessary for clarity/fairness.

I don't know, probably no. Just like AI-generated images were everywhere for a while, and now they have a negative impact on the reputation of anyone using them.

Low-effort posts are already discouraged. https://news.ycombinator.com/newsguidelines.html

I don't know how many would agree with me, but I would equate "I consulted an LLM and it said/spewed..." with a low-effort post by a human who didn't use an intermediary.

It would be enough to define these as low effort posts and apply existing rules.