Aishan Maoliniyazi

Towards Better Value Principles for Large Language Model Alignment: A Systematic Evaluation and Enhancement

As Large Language Models (LLMs) advance, aligning them with human values is critical for their responsible development. Value principles serve as the foundation for clarifying alignment goals.Multiple sets of value principles have been proposed, such as HHH (helpful, honest, harmless) and instructio…