Great experiment — and interestingly, with all the randomness of pretrained GPT-2 model the statements make sense the more one interpret them. Something like textual pareidolia? Or just: everything makes sense?