abstract

https://arxiv.org/abs/2308.01542 How can we reveal a chatbot’s memory state and allow a user to manipulate its memory representation? The recent advent of large language models (LLM) has resulted in high-performing conversational agents such as ChatGPT. These agents must remember key information from an ongoing conversation to provide responses that are contextually relevant to the user. However, these agents have limited memory and can be distracted by irrelevant parts of the conversation. While many strategies exist to manage conversational memory, users currently lack affordances for viewing and controlling what the agent remembers, resulting in a poor mental model and conversational breakdowns. In this paper, we present Memory Sandbox, an interactive system and design probe that allows users to manage the conversational memory of LLM-powered agents. By treating memories as data objects that can be viewed, manipulated, recorded, summarized, and shared across conversations, Memory Sandbox provides interaction affordances for users to manage how the agent should `see' the conversation.

quickstart

setup the application

1. install node.js.
2. run git clone https://github.com/hzhfred/chatbot-memory-hci.git.
3. run npm install to get the dependencies.

run the development server

1. run npm run dev to start the development server.
2. visit http://localhost:3000 to view application.
3. edit app/layout.js or app/page.js and save to see the updated result in browser.
4. get the .env.local file to access OpenAI.

concept

Memory Sandbox is a system that enables users to see and manage the memory of conversational agents. Memory Sandbox provides the following interaction affordances: 1) toggle memory visibility, 2) add memory, 3) edit memory, 4) delete memory, 5) summarize memory, 6) create a new conversation, and 7) share memory.