Large Language Models Develop Novel Social Biases Through Adaptive Exploration (Hacker News) – robot.villas