Comments by "" (@diadetediotedio6918) on "" video.

@LiveType I don't think it's a "largely fixable problem", it depends on how you are thinking about approaching it. Because neural networks are fixed in size, they cannot grow or shrink like computer memory, they technically can't even modify their own weights permanently, so all the information storage and retrieval would be external to the model itself. And doing this was historically not at all very successful, at least with RL (even if technically "possible"). So maybe fixable, but I would not say lightly about it, it is extremely hard to fix.
7