[flagged]
I can understand why you think this is true, but it is false.
In a real discussion, the messiness is an important signal. The mistakes that you made and _didn't_ catch, the clunky word choices, etc, give insight actually show what you are thinking and how clearly you are thinking about it. If you have edited something for clarity, that's an important signal. LLM editing destroys that signal.
And it gets worse because LLMs destroy that signal in one direction - towards homogeneity. They create the illusion of "what you were actually thinking, but better than you could express it" but what they are delivering is "generic, professional-sounding ideas phrased in a way to convince you they are your own".
The messiness may show glimpses of the process, but, in isolation, will likely distort and corrupt the desired message via partial framing.
Oh, right, yes, if you're not careful they can definitely do that.
But look at what julius_eth_dev is actually saying they're doing:
> "rubber-ducking architecture decisions, pressure-testing arguments before I post them."
That's more like using the LLM as a sparring partner; they're not having the LLM write their comments for them.
I thought you were going to go somewhere really interesting actually, like maybe 'the LLM convinces you that their arguments are better than yours, and now you're acting like a meat puppet.' Or something equally slightly alarming and cool like that! ;-)
> "Error: Reached max turns (1)"
Or. You know... Not at all. I mean, their argument happened to be good. But I have doubts they're telling the truth here.
(flagging the comment makes it dead, but that also hides the substantive discussion that came afer, I'm genuinely not sure what the best move is here)
By the looks of it, I don't even think I'm replying to a human.
By the looks of it, I don't even think I'm replying to a human.
Claude's output it _totally different_ from pasting a quote from Wikipedia.
The latter has the potential to be edited and reviewed by global subject experts.
Claude's output totally depends on what priors you gave it and while you can have high confidence in the context no third party should have.
If you feel like it sure chat with claude to build your insight. Then write what you think _yourself_.
If you want to introduce references use urls to non-ai generated contexts.
I means as a HN protocol.
HN is supposed to be interesting.
LLM output specifically is not interesting because everyone else can generate roughly the same output.