Post #1306513

Tinkering around with @PyTorch this evening, I seem to have trained a 77-parameter (not a typo) Transformer Quine that autoregressively generates a lossy copy of its own weights.