This research explores how to improve large language models' ability to make ethical judgments and adhere to societal values—critical for secure AI deployment.

• Examines moral reasoning acquisition processes in language models
• Investigates optimal learning paradigms to enhance ethical responses
• Addresses the gap between current LLM capabilities and satisfactory ethical decision-making
• Provides insights for developing more responsible AI systems that align with human values

Why it matters: As AI systems become more integrated into society, ensuring their ability to reason ethically is essential for preventing harmful outputs and building trust in AI applications.

Diagnosing Moral Reasoning Acquisition in Language Models: Pragmatics and Generalization