How do you interpret the coefficients of a linear regression, and why is it dangerous to extrapolate beyond the training data?

Interview question asked to Data Scientists interviewing at Cognizant, Coursera, Medallia and other companies. Original question asked: How do you interpret the coefficients of a linear regression, and why is it dangerous to extrapolate beyond the training data?.