Large Language Models for Behavioral Economics: Internal Validity and Elicitation of Mental Models