x
Research agenda for training aligned AIs using concave utility functions following the principles of homeostasis and diminishing returns — LessWrong