Nah, it's totally Claude. No human writes the bulleted list in https://github.com/RokoMijic/MercurialDyson/blob/4f6cb3c0b5b..., or "The thermal management problem that naively prohibits rapid disassembly is resolved by three key insights…". The whole thing reeks of Claude, but chapters 12 and 13 are probably the clearest slam-dunks.
If you want to compare styles, https://github.com/Smaug123/gospel/blob/764996d20e11674f9221... is similarly written almost entirely by Opus (4.5 rather than 4.6) with some strong LessWrong-o-sphere background prompting and the instruction to be terse. The styles are practically identical.
> DI containers mean the call site doesn't tell you what's called. Instead: pass values in, get values out.
That one is obviously LLM.
Before I left the comment, I reviewed some of this user's commit history to about a decade back, and they genuinely write like this. Though I think this is the first long-form content I've seen from them.
For what it’s worth, gptzero.me rates it as entirely AI generated, with 100% confidence. It’s not perfect, but it’s a pretty strong signal.
Certain aspects of how it’s structured and written do also seem AI-generated—for instance, the simultaneously persuasive but explanatorily equivocal tone is pretty typical of current LLMs. Also, there are just some text formatting features that are pretty rare for humans to use—for instance, using the nice-looking Unicode 1/2 fraction glyph, which isn’t really in keeping with the otherwise unpolished maths formatting.
It’s a bit sad that AI writing is now so good as to seem almost authentic, if not for the giveaway of a few subtle stylistic quirks.
Actually, it is LLM output and I have updated the readme to show that. But I didn't just ask an LLM to one-shot this. It was a process of iteratively throwing ideas at it, doing calculations, and getting other LLMs to critique. Probably ~100-200 calls to top of the line consumer reasoning models went into this so far.
You sure about that? It really comes off as LLM output to me, in its general structure and formatting, attention-grabbing opening sentences of paragraphs ("This ratio has a profound consequence:", "This distinction matters."x2), and the classic "it's not X, it's Y" stuff ("The collector is a hybrid optical-power megastructure, not a single dense slab of ordinary powersats.", "The shell does not interact with a small number of giant launchers. It interacts with a dense distributed network.")
> Note: the written report is currently "vibe coded" physical and engineering analysis using various LLM-based AIs, with the author acting as a guide and sanity check and putting pieces together. The intention moving forward is to move calculations to code and simply report the results.
Ah, I do see that was added to the readme about 12 hours after my original comment. It does seem more heavily curated than any one-shot output would be.
If you want to compare styles, https://github.com/Smaug123/gospel/blob/764996d20e11674f9221... is similarly written almost entirely by Opus (4.5 rather than 4.6) with some strong LessWrong-o-sphere background prompting and the instruction to be terse. The styles are practically identical.