I push my bash history, attach a couple files, paste some doc and web references, and then lightly edit the result. Beats not writing the blog post.
The tricky part is not getting fooled by the fact that it looks decent, and really put in the effort to think each sentence over to catch the misleading "fringe hallucinations" (as I call them, not full on hallucination, but mildly misrepresenting the actual intent/meaning).
The problem is that it wastes the reader's time to parse the bullshit. Now of course a lot of humans write bullshit too, especially in areas that are at the peak of their current hype cycle ;)
Actually, you did make a totally baseless accusation and the fact that the person above you is being downvoted is evidence that the HN userbase is pretty awful at recognizing what "bases" are.
> For LLMs to be effective, they require a few things:
And then that list are just a dead giveaway.
I'm not nearly as cynical about GPT as many are (in fact I use them myself), but I think text written by it should be marked as such.
edit: The author expressed that it was indeed not written using any GPT.