Hi Gwern, I'm the owner of 7goldfish, the developer in question, and a fan of yours via reputation and the occasional comment I run into on Reddit.

First I want to acknowledge that draft one of this was LLM written, by Claude, though it reflects a pretty detailed outline of an experience pretty accurately. As you point out the quotes from both Mr. R and myself were also mostly spat out by the LLM as well (though not the quotes from external entities)

Mr. R signed off on the draft before posting, and well, it was me. I tended to think of it more as a "movie treatment" than a technical post-mortem, so I wasn't really worried about it. I also was only expecting this to get circulated within my own small/medium sized community so, in general, wasn't really worried about it.

That said, I definitely love using LLMs to write. To be perfectly honest they write considerably better and faster than I do (as you noted, lots of typo-o's and similar. I was still spelling at a 6th grade level when I graduated from Uni with CS degree), though I still feel like I have both ideas and experiences worth sharing. If you click any of the earlier stuff you will probably see the clumsy results that take about 10x the time.

I waffle on the idea of how much disclaimer of "written via LLM, but with multiple revisions and actual thought" vs "just don't bother saying anything" it's worth including. I'm curious if you consider having a ghost writer to be lying, or a cinematic re-enactment. I notice as I say that that it sounds defensive, and I want it to be a genuine question, as I share your concern about living in a media world where it feels like "basically nothing can be trusted".

For what it's worth, the numbers should be about right, though there is only so much energy we were willing to spend on the post-mortem. If there is some informational reason you would like to get into it deeply I would be happy to share the post mortem docs privately.

Nice.