Ask LLMs Directly, "What shapes your bias?": Measuring Social Bias in Large Language Models.