Stop "reinventing" everything to solve alignment

Integrating some non-computing science into reinforcement learning from human feedback (RLHF) can give us the models we want.

alignment computing everything feedback human human feedback reinforcement reinforcement learning rlhf science solve