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. https://t.co/5J16QdzwPF

Tinkering around with

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.

3:20 AM · Feb 5, 20243.1KViews