Inverse-RLignment: Large Language Model Alignment from Demonstrations through Inverse Reinforcement Learning
This addresses the problem of aligning LLMs more efficiently and effectively for developers and users, though it appears incremental as it builds on existing reinforcement learning frameworks.
The paper tackles the challenge of aligning Large Language Models (LLMs) for safety and utility by introducing Alignment from Demonstrations (AfD), which uses high-quality demonstration data to overcome issues like noisy labels and high costs in existing preference-based methods, and demonstrates strong empirical performance on Harmless and Helpful tasks.
Aligning Large Language Models (LLMs) is crucial for enhancing their safety and utility. However, existing methods, primarily based on preference datasets, face challenges such as noisy labels, high annotation costs, and privacy concerns. In this work, we introduce Alignment from Demonstrations (AfD), a novel approach leveraging high-quality demonstration data to overcome these challenges. We formalize AfD within a sequential decision-making framework, highlighting its unique challenge of missing reward signals. Drawing insights from forward and inverse reinforcement learning, we introduce divergence minimization objectives for AfD. Analytically, we elucidate the mass-covering and mode-seeking behaviors of various approaches, explaining when and why certain methods are superior. Practically, we propose a computationally efficient algorithm that extrapolates over a tailored reward model for AfD. We validate our key insights through experiments on the Harmless and Helpful tasks, demonstrating their strong empirical performance while maintaining simplicity.