[armchairdweller]

I found one of the most interesting aspects of memory to be its non-locality. There were a lot of experiments in the 20th century (lesions etc.) showing that memory is fundamentally non-local. You could remove large parts of brains and the memories were still there. This is difficult to explain with "local" / neural-network-like theories of memory. If you lesion specific parts of GPT4, the "memory trace" will be gone.

I find this incredibly interesting. Is this still the primary view?

The hippocampus is involved in formation of new memories. Without it this process is not working at all.

[astrange]

You can build that into a neural net by training with dropout, but I don't know if it's still fashionable.