Being ‘moral’, means to have empathy. But empathy is only possible between two beings that share experiences and reality or at least some aspects of it. LLMs don’t have experiences, but it builds it’s weights from training data. It is fundamentally a computer program. Just textual information is not enough to build deep context. For example, when I say “this apple is red”, anyone reading this can easily visualize a red apple because of your experience seeing a apple. That cannot be put into text because it is a fundamental part of human experience that is not available to a computer program, as of yet.
At least that is my hypothesis. I can very obviously be wrong., which is another fundamentally human experience.
CTDummy@aussie.zone 1 week ago
Yeah these “stories” reek of blaming a failing -bordering on non-existent (in some areas)- mental health care apparatus being blamed on machine that predict text. You could get the desired results just googling “tallest bridges in x area”. That isn’t a story that generates clicks though.
ragebutt@lemmy.dbzer0.com 1 week ago
The issue is that there is a push to make these machines act as social partners and in some extremely misguided scenarios therapists