a lexeme is a low perplexity information with high reward
松弛中拉满·
when you see somthing you feel not comfatable
you actually see a prediction error, a value that is too large
value is not only a factual mismatch but also a reward discrepancy
松弛中拉满·
there is always something you have to pay when you get reward
the stupidity is to ignore that.