Embracing naturalistic paradigms: substituting GPT predictions for human judgments