A model predicts house prices. One prediction is off by $500,000 while most are off by $10,000. Which metric is more sensitive to this large error?