A fascinating new study explores the "moral minds" of large language models (LLMs). Researchers presented nearly 40 different LLMs with a series of ethical dilemmas, using a method called the Priced Survey Methodology (PSM). This involved asking the LLMs to repeatedly choose from a set of answers to complex moral questions, much like choosing items within a budget. The goal was to see if LLMs display consistent moral principles, rather than just spitting out random responses based on their training data. Surprisingly, several LLMs showed behavior consistent with having a stable set of moral rules. They acted as if they were maximizing a utility function—a mathematical representation of preferences—encoding ethical reasoning. However, not all LLMs were created equal. Some exhibited more flexibility in their moral reasoning, bridging diverse perspectives, while others clung to more rigid ethical structures. This difference in "moral adaptability" became clear when researchers analyzed the statistical similarity between the models’ responses. The more adaptable LLMs acted as bridges between different clusters of moral thought, suggesting they could integrate a wider range of ethical considerations. The study suggests that while some LLMs may appear to reason morally, it's crucial to understand the nuances and variations in how different models arrive at their "ethical" decisions. This research raises crucial questions about how these AI moral minds might interact with and even influence human ethics in the future.