Hacker News Reader

Comment by croon | Hacker News Reader

> The fundamental issue is whether people are taking responsibility to communicate what needs to be communicated, in an effective manner.

While I agree with your overall distinction, I believe the scenario you portray is so unrealistic that you will never encounter it.

Yes; if I could wrangle Claude into communicating exactly as I would have, the practical difference is of course not observable. But that's never happened, and I've had an enterprise license at work for years.

The value in having a human write it is that whenever something bad/misleading/redundant is introduced, that person will learn from the feedback, but an AI will at best have some memory/skill/prompt nudge for next time that might have a shallow/localized band-aid effect, but will never stop doing similar mistakes elsewhere.

View on HN

> The value in having a human write it is that whenever something bad/misleading/redundant is introduced, that person will learn from the feedback,

For me, the value of receiving comms from a human is different: it communicates not just the idea itself, but the fact that the human already understands the idea.

The people arguing that round-tripping their thoughts through AI still means that it's their thoughts apparently also believe that the game of telephone always results in perfect comms.

It's not just AI; adding more hops to a route changes the message when each hop is a human too, it's just that we notice it more when the hop is an LLM because of just how different they write compared to humans.

I don't think you quite realize the implications of what you're saying.

Surely you don't actually believe, or even desire, for everyone to fully understand everything that they say... No room for questions, uncertainty, bewilderment, etc?

If someone is feigning competence via AI, that's the problem. Not Ai.

If they use Ai to polish a communication, be it statement, question, etc, and they do it in a thoughtful manner to ensure the final communication clearly and effectively communicates what it should be, there's no problem with that.

> If they use Ai to polish a communication, be it statement, question, etc, and they do it in a thoughtful manner to ensure the final communication clearly and effectively communicates what it should be, there's no problem with that.

My point is that it is still a problem, regardless whether they are asking an AI to write it for them or a human to write it for them.

Round-tripping a message through a 3rd party changes the message; in this specific context ("Don't Paste the AI, please"), there is no pretense of understanding or review from the human.

In some other context, sure, maybe some people are thoughtfully engaging with the AI, but I have never come across this yet. I suspect you haven't either.

> In some other context, sure, maybe some people are thoughtfully engaging with the AI, but I have never come across this yet. I suspect you haven't either.

Um, I've literally said throughout this entire thread that I regularly get Ai to assist with my writing, to great effect. It often helps me be more concise and certainly more kind - I have a tendency to be neither in writing, especially online.

It is also helpful in formatting, say, a github issue. Though, I do need to be careful with that as it is very easy for it to just appear as an Ai slop post, even if lots of thought went into it all.

To be very clear, at no point have I ever supported/advocated for thoughtlessly copy pasting whatever Ai writes for you. This thread has been about using it judiciously, as we can/should be doing for coding or anything else.

>Surely you don't actually believe, or even desire, for everyone to fully understand everything that they say... No room for questions, uncertainty, bewilderment, etc?

Nothing they said implies that.

What they said is they want the understanding of the person to be revealed in the communication. Not masked via AI.

>If they use Ai to polish a communication, be it statement, question, etc, and they do it in a thoughtful manner to ensure the final communication clearly and effectively communicates what it should be, there's no problem with that.

Yes, if that ever happens in the real world, it would be ok.

> For me, the value of receiving comms from a human is different: it communicates not just the idea itself, but the fact that the human already understands the idea

It doesn't say the (level of) understanding of the person, but that they understand the idea. They might have misspoken, but they already replied to that comment and didn't object to my framing or further clarify.

> Yes, if that ever happens in the real world, it would be ok.

Well, this entire thread is me saying this happens in the real world, and people objecting to it

They don't though; I have never seen this happen. I have seen many people claim that's what they do, but not convincingly.

> if I could wrangle Claude into communicating exactly as I would have, the practical difference is of course not observable. But that's never happened, and I've had an enterprise license at work for years.

Sorry to say, but this is a skill issue (and not necessarily with using Ai). I've used Ai MANY times to help SIGNIFICANTLY improve things that I've written - from a few sentences to many paragraphs.

As I've said in other comments, it'll rapidly and effectively remove all my worst tendencies. And, likewise, sometimes I'll have it write a first draft when I give it appropriate context/guidance that I can't be bothered to fully curate/synthesize. Then, once it gets that draft, I'll take over and tidy it up. You might say that the "can't be bothered" part is the entire issue, and I'll again say that what matters is the final result - a master chef or chief architect is not doing all the grunt work...

> Sorry to say, but this is a skill issue (and not necessarily with using Ai). I've used Ai MANY times to help SIGNIFICANTLY improve things that I've written - from a few sentences to many paragraphs. As I've said in other comments, it'll rapidly and effectively remove all my worst tendencies. And, likewise, sometimes I'll have it write a first draft when I give it appropriate context/guidance that I can't be bothered to fully curate/synthesize. Then, once it gets that draft, I'll take over and tidy it up.

Isn't this contradictory? Why do you need to take over if it's doing such a good job?

Let's fire all editors, consultants, psychologists, sous chefs, grunts, etc!

I'm not claiming an LLM writes better than me.

No, you were claiming that no one has ever used Ai to help produce better writing. I was objecting to that, because it is an absurd statement. Ai could have told you that if you asked.

> No, you were claiming that no one has ever used Ai to help produce better writing. I was objecting to that, because it is an absurd statement. Ai could have told you that if you asked.

Maybe take your own advice and ask your LLM if I did claim that.

This is what I claimed:

> Yes; if I could wrangle Claude into communicating exactly as I would have, the practical difference is of course not observable. But that's never happened, and I've had an enterprise license at work for years.

This means that if I get Claude/LLM output, I would need to edit it/write myself to make it correct/better/distilled into relevant parts. I do not get LLM output I would want to post as is without numerous roundtrips, at which point vi is faster.

You then reply to claim that as a skill issue, while also later on stating that you iterate over it or take over a draft to edit it, which I then point out is a contradiction.

Because it can’t do a great job without frequent expert input, that’s the point.

Expert types prompt into Claude , Claude produces fundamentally flawed but close output. Don’t take that output and give it to anyone — first refine it until it’s not flawed, then pass it on.

Correct, I was treating their claim as true for the sake of argument, not accepting it.

We're on the same page

>Sorry to say, but this is a skill issue

That's the same slop response as "it's a skill issue" applied to C codebases being inherently insecure due to the language's footguns/design.

Not even slightly. The burden of proof is on you, given that I regularly have AI improve my writing

Your writing tendencies must be quite bad if Claude's tendencies are an improvement. Isn't writing a skill issue?

Thanks for sharing a perfect example of something that any AI could have helped you improve

Having AI do the work for you doesn't increase your writing skills, in fact, the opposite!

Instead of seeing the errors and thinking about what changes the draft needs ('improving' your writing skills'), you're letting someone else tell you what changes to make.

Because you aren't internalizing the corrections, guess what happens next time: your writing gets worse, AI comes to help...rinse and repeat.

Reply on news.ycombinator.com