Large language models (LLMs) are rapidly changing our world, but what about their moral compass? A new study dives deep into the ethical reasoning of several leading LLMs, revealing some surprising insights. Researchers put these models through a series of ethical dilemmas, from classic thought experiments to modern-day challenges. The results? Proprietary models like those from Google and OpenAI lean towards utilitarianism, prioritizing the greatest good for the greatest number. Open-source models, however, tend to favor values-based ethics, emphasizing moral duties and principles. But there's a catch: all models showed significant variability in their responses, suggesting that their ethical reasoning isn't always consistent. The study also used the Moral Foundations Questionnaire to probe the models' underlying moral values. Most models exhibited a liberal bias, aligning with values like empathy and fairness. Interestingly, Llama 2 was the exception, showing a more conservative profile. Finally, the researchers developed a new technique called SARA (Similarity-based Activation Steering with Repulsion and Attraction) to directly influence an LLM's moral compass. By tweaking the model's internal activations, they successfully steered its reasoning towards different ethical schools of thought. This breakthrough opens up exciting possibilities for shaping AI ethics, potentially allowing us to fine-tune models to align with specific values. The study's findings raise important questions about the future of AI. As LLMs become increasingly integrated into our lives, understanding and shaping their ethical behavior is crucial. The ability to steer their moral compass could be a powerful tool, but it also carries significant responsibility. How do we ensure these models reflect the values we want to see in the world? And who gets to decide what those values are? The journey towards ethical AI is just beginning, and this research provides a valuable roadmap for navigating the complex terrain ahead.