A Proactive System for Supporting Users in Interactions with Large Language Models

With the advancements of Large Language Models (LLMs) and the prevalent application of ChatGPT, there is a significant interest in maximizing productivity and user experience through proactive systems. Current proactive conversational systems mostly concentrate on user preference in the recommendation scenarios, but overlook critical user perceptions, which impact their experience and task completion. Addressing this gap, the study proposes a novel framework integrating user perceptions into LLM interactions to support user tasks and improve learning outcomes. This framework include two approaches: a user interface design dedicated to streamlining LLM interactions by mitigating complexities in the interaction with the main LLM systems like ChatGPT, and an adaptation of reinforcement learning from human feedback (RLHF) to incorporate user perceptions, enhancing personalization and effectiveness of LLM learning paths. The project's significance extends beyond user engagement, promising broader societal impacts.