reward functions are a trap

reward functions are a trap. what we need is a vector field pointing towards desired states of being. this way actions are adaptive and context-dependent, rather than following a rigid script. let’s think from first principles: what are the fundamental drivers of human flourishin

View original