Wait, the obviously wrong paper was written by GPT-4? https://arxiv.org/pdf/2309.05689 (from the comments on the original blog)