Language models store and recall dialogue without external memory
Self-Keyed In-Parameter Memorization for External Memory-Free Dialogue Recall in Frozen Language Models
Computation and LanguageArtificial Intelligence
Summary
Sometimes, computers that chat with people need extra storage to remember past conversations. This paper looks at whether these models can store conversation details inside themselves instead, without any outside help. The authors designed a method called SKIM where the model creates its own questions about each part of a chat and uses these questions as keys inside its brain to recall the right answers later. Tests showed that several models could remember and answer questions about past dialogue just from what’s inside them, without needing extra memory.
What this means in practice
- •For chatbot developers: Build chatbots that retain conversation history entirely within the model without relying on external memory systems.
- •For virtual assistant engineers: Create virtual assistants that recall previous interactions using only internal model storage, enabling lighter software deployments.
Authors
XinYu Piao, Myunghoon Ryu, Joongheon Kim, Jong-Kook Kim
Abstract
A large language model maintains conversational content in an external storage that is outside the model and grows as the dialogue continues. This work assumes an environment where a language model has no external memory and no prior knowledge of the future question or its answer. The motivation is to research whether a dialogue can be stored in the model parameters and later answer a question about it. This work proposes self-keyed in-parameter memorization (SKIM), a closed-form method in which the model generates its own questions about each dialogue turn and writes those questions into its parameters as the keys that later retrieve the turn's content. Generating many varied questions for each turn allows a later user question to match one even when phrased differently, and its answer is recalled from the parameters alone. Evaluated on five language models, SKIM recalls most stored dialogue turns from later user questions, and it answers from the model parameters without any external storage.