pangram thinks the whole thing was LLM generated fwiw, as dodgy as AI detectors are it is probably among the best. I don't doubt the author started with their own text, but I think it's been substantially revised via ChatGPT
Here it matters much less than in generic LLMs though. There's no chance of test set leakage since the network is not general purpose / not trained on the internet.
> Not a hack, not a trick. A principled constraint that makes the architecture work at scale.