This note is about the self-awareness assumption we don't talk enough about, and which I think we need to address to ultimately understand intelligence and AI safety. RL agents have the action/observation split baked in at the architecture level. The action space A and

→ View original post on X — @nandodf,