Personalized Large Language Model Assistant with Evolving Conditional Memory

TL;DR

Long-running LLM assistants lack cross-session memory, limiting personalization. The paper introduces a plug-and-play evolving conditional memory framework that externalizes memory, with history-based, summary-based, and novel conditional memory modules, plus a self-reflection mechanism to regulate retrieval. Empirical results across three synthetic datasets show conditional memory generally yields the best performance, especially for learning new knowledge and user feedback, while self-reflection improves memory usefulness. The approach enables personalized responses without fine-tuning and highlights practical benefits and avenues for future work including testing other LLMs and memory dynamics such as timing and forgetting.

Abstract

With the rapid development of large language models, AI assistants like ChatGPT have become increasingly integrated into people's works and lives but are limited in personalized services. In this paper, we present a plug-and-play framework that could facilitate personalized large language model assistants with evolving conditional memory. The personalized assistant focuses on intelligently preserving the knowledge and experience from the history dialogue with the user, which can be applied to future tailored responses that better align with the user's preferences. Generally, the assistant generates a set of records from the dialogue dialogue, stores them in a memory bank, and retrieves related memory to improve the quality of the response. For the crucial memory design, we explore different ways of constructing the memory and propose a new memorizing mechanism named conditional memory. We also investigate the retrieval and usage of memory in the generation process. We build the first benchmark to evaluate personalized assistants' ability from three aspects. The experimental results illustrate the effectiveness of our method.

Figures (6)

• Figure 1: An example of what a personalized LLM assistant with evolving memory can do.
• Figure 2: The framework of personalized assistant with evolving memory.
• Figure 3: The LLM assistant using conditional memory with self-reflection retrieval.
• Figure 4: The examples of the three constructed test datasets.
• Figure 5: The multi-choice result of different types of memory in learning from feedback and continuing previous dialogue.