Personalized Adaptation via In-Context Preference Learning
This addresses the need for scalable and efficient personalization in large language models for users, though it appears incremental as it builds on existing RLHF and in-context learning techniques.
The paper tackles the problem of aligning language models with individual user preferences, which existing RLHF methods often neglect, by introducing the Preference Pretrained Transformer (PPT) that uses in-context learning for dynamic adaptation, achieving superior personalization and significantly reducing computational costs in a contextual bandit setting.
Reinforcement Learning from Human Feedback (RLHF) is widely used to align Language Models (LMs) with human preferences. However, existing approaches often neglect individual user preferences, leading to suboptimal personalization. We present the Preference Pretrained Transformer (PPT), a novel approach for adaptive personalization using online user feedback. PPT leverages the in-context learning capabilities of transformers to dynamically adapt to individual preferences. Our approach consists of two phases: (1) an offline phase where we train a single policy model using a history-dependent loss function, and (2) an online phase where the model adapts to user preferences through in-context learning. We demonstrate PPT's effectiveness in a contextual bandit setting, showing that it achieves personalized adaptation superior to existing methods while significantly reducing the computational costs. Our results suggest the potential of in-context learning for scalable and efficient personalization in large language models.