x
Causality and a Cost Semantics for Neural Networks — LessWrong