[renewiltord]

Yeah, I know the forum dislikes LLM-posted comments but I thought it was a pretty good opportunity to show how it could improve human-to-human comms. I think if we'd like we could each get a lot more information extraction ability from the world with this new machinery.

We think about the technology as accelerating development of other stuff purely through generation: code, images, video, sound. But it could actually accelerate comprehension of text because in many cases its skill at teasing out value outstrips ours.

The future is bright!

[refulgentis]

It's dumb:

- OP didn't say they didn't know what it meant

- it doesn't help improve human-to-human comms to have someone:

-- take up 80% of my screen

-- with a copy-and-pasted obvious take

-- unedited

-- from a text generator

-- that has been RLHF'd to chain of thought

I love LLMs and quit my job to work with them, but this ain't it, or an opportunity to extol about bright futures and stuff.

[Dylan16807]

> OP didn't say they didn't know what it meant

They said it makes no sense, so yes they did say that.

> it doesn't help improve human-to-human comms to have someone:

It's what the comment deserved for saying it makes no sense. Showing the dumb AI understands instantly.

And the conversation was already an overly large tangent.

[refulgentis]

> They said it makes no sense,

No, they didn't

[Dylan16807]

They compared it to a sentence that doesn't make sense, and said "makes no more sense".

If they weren't saying "flagship" makes no sense as a non-ship noun, then they used the wrong words.

Oh unless by OP you meant someone other than jvanderbot, but jvanderbot is the person that prompted the claude reply. It wasn't a reply to duxup, who didn't even mention the headline.

[refulgentis]

Ok, so you realize now you lied, and the poster never claimed it didn't make sense. That was thin gruel already, they made very clear their complaint was about usage, i.e. the part of adjectives is half their comment, that the LLM poster themselves noted.

Given that, do you still believe the huge, unedited, LLM explanation of "flagship" was responding to someone?

If so, who?

If not, I'd like to avoid normalizing spamming long, child-like, auto-generated, explanations of things to prop up straw-men in service of condescending.

A message board where that was mundane, everyday, behavior wouldn't be fun to be on.

I might enjoy it if I think someone deserved it, but I'd still balance that with the long view, because that sort of behavior is extremely corrosive.

[Dylan16807]

> Ok, so you realize now you lied, and the poster never claimed it didn't make sense.

I'm not lying. I stand by the claim that jvanderbot was saying it didn't make sense.

> Given that, do you still believe the huge, unedited, LLM explanation of "flagship" was responding to someone?

> If so, who?

Is this a real question? Do you not see how it's a reply to jvanderbot...?

> Is it helpful to normalize spamming long child-like auto-generated explanations of things to prop up strawmen? Would that be a fun board to be on?

The ship comment is just as child-like, and it came first. So while I don't think AI comments are good in general, I think in context it wasn't egregious. If we're trying to filter toward good discussion, then jvanderbot's comment should not be treated as much better than the claude comment.

[renewiltord]

It was a demo on how to use the tool. Would you like to learn how to? Try the following steps to construct a prompt on Claude Opus:

1. Type in "I saw this comment on Hacker News:"

2. Hit `>` to start a blockquote.

3. Paste in my entire previous comment

4. Type in "Is this person saying that we should copy-and-paste LLM output as comments to others? Could you explain what he is suggesting I should do if I misunderstand?"

5. Hit Enter

I just tried it and the explanation is right on the money. Not everyone can pick up on human language well, even if they're skilled at other stuff.

You can repeat that with the original sequence of comments as well.

0. Type in "I saw this sequence of Hacker News comments:"

1. Paste in Arrakeen's comment in a blockquote starting with "FWIW I agree with you"

2. Create a nested blockquote with the next response '"Flagship" means important.'

3. Create a third level of nesting with the next response Ford's battery "Important"

4. Exit the nesting and ask: "Does it appear that these people understand what "Ford's battery flagship" means or are they confused by the term?"

And it will correctly detect that they were confused as they said:

> FWIW I agree with you. The title hardly makes sense to me as well. "Ford's Battery Flagship" - even after reading the article I'm still not sure what this is.

See, he does not know what it means, but if he had asked a high-powered LLM it would immediately have gotten him there and saved him some time commenting. And it looks like you could have too! It's really pretty good stuff. Thanks for working on them.