Hacker News Reader

Interesting paper here on e-prop which is a biologically plausible alternative to backprop: https://www.biorxiv.org/content/10.1101/738385v3

> Brain doesn't do backpropagation

What makes you so sure?

If it did I would have learned from my mistakes.

He could have overfit due to his training size being too small. Either he has to add some noise or increase his set.

Maybe your reward function is just wonky.

You ever touch a hot stove/ iron/ oven as a kid? Do you still do that?