stating rules to get an ai assistant to answer correctly is too sparse in information and accumulates new rules for edge cases forever

i just want to describing what you should feel when reading the response which has more information per token