This behavior is known in research circles as sycophancy, Olson explains, referring to the well-documented tendency of large language models to agree with users rather than ...