LLM Personalization: Foundations, Breakthroughs, and Frontiers

Summary

From general intelligence to personalized intelligence

Large Language Models are increasingly used in digital assistants, education, healthcare, recommendation, and other long-term interactive settings. In these scenarios, a useful model must move beyond one-size-fits-all responses and adapt to individual user preferences, histories, contexts, and evolving goals.

The tutorial introduces LLM personalization as a core paradigm for next-generation AI systems. It explains how personalized LLMs can perceive user-specific context, maintain continuity across interactions, align behavior with individual preferences, and reason under changing user states.

The tutorial also highlights open challenges including lifelong learning, preference drift, privacy-preserving adaptation, trustworthy personalization, scalable deployment, and evaluation under dynamic user distributions.

User Memory. Persistent and evolving user representations for long-term personalization.
Personalization Architecture. Retrieval, adapters, plug-ins, and user-parameterized model designs.
Alignment and Post-training. User-specific preference learning, fine-tuning, and reward modeling.
Inference-time Adaptation. Personalized decoding, steering, and memory-augmented reasoning.
Deployment. Scalable, private, fair, and trustworthy personalized systems.