Multiple Choice

In the cat classification example, if training error is ~1% and dev error is 11%, what is the estimated variance?