So if someone trains a (future) LLM on your life story and everything you’ve written, do you think your survivors would find that comforting or just incredibly creepy?
@matthew_d_green Given how much these models "fake it until they make it" to fill in gaps, I suspect most people will quickly hit edge cases where the simulation is expressly not like their loved one and says something upsetting instead of comforting.
Even subtle differences have a good chance of triggering uncanny valley responses and for something like this, would likely just underscore the loss.
@matthew_d_green I think what *might* be OK is a virtual librarian that curates this catalog of the loved one's pictures/writings/videos so you can reference/replay them later. It could even do this in the style of the loved one, but with a clear dividing line that demonstrates it's simply a virtual librarian and not attempting to *be* the loved one.
