ARTIFICIAL INTELLIGENCE DEVICE FOR PERSONAL LARGE LANGUAGE MODEL AGENTS FOR COMPLEX TASK EXECUTION AND METHOD THEREOF

Smart overview of the Invention

The patent application discusses a method and device for enhancing the personalization of large language model (LLM) agents when executing complex tasks. This involves creating a structured user model that incorporates user preferences, history, and personal concepts to tailor task execution. The aim is to improve user satisfaction by providing outputs that are more closely aligned with individual user needs and contexts.

Structured User Model

A key component of the system is the structured user model synthesizer, which generates a model based on user-related information. This model serves as a basis for personalizing task execution. The personalized task execution engine then integrates this model with external data to generate a tailored response to user queries, offering a more customized experience than generic outputs.

Quality Assessment

The framework includes an automated quality assessor, configured as an LLM-as-a-Judge, which evaluates the personalized result against a baseline non-personalized result. This comparison helps identify the preferred output and provides a reasoned justification, ensuring that the chosen result aligns well with the user's profile. This process quantitatively assesses the effectiveness of the personalization.

Implementation

The system employs a multi-component framework where at least one component, such as the structured user model synthesizer or the task execution engine, is based on a large language model. The integration of user models can be achieved using non-parametric or parametric approaches, including fine-tuning and adapter layers, to adapt the task execution engine effectively.

Potential Applications

This method and device can significantly enhance the utility of AI systems in real-world applications by providing personalized recommendations and solutions. The inclusion of a structured user model and a robust quality assessment framework ensures that outputs not only meet task requirements but also satisfy individual user preferences, thereby improving the overall effectiveness and user experience of AI-driven services.