Show HN: PILF, The ultimate solution to catastrophic oblivion on AI models

This looks absolutely fantastic, please accept my meagre professional jealousy. I have long bemoaned manual hyperparam fiddling . I have on occasion dabbled with nonparametric ("genetic") methods of hyperparam tuning inspired by AutoML... but then you still have to manually tune the evolutionary hyperparams.

Finding a way to derive this from the gradients is amazing.

This is definitely not just another machine learning method. It comes from a complete cognitive science theory, rooted in a complete understanding of intelligence and consciousness.

https://github.com/dmf-archive/IPWT

:)