Prompted LLMs as Chatbot Modules for Long Open-domain Conversation
Gibbeum Lee, Volker Hartmann, Jong-Ho Park, Dimitris Papailiopoulos, Kangwook Lee · 2023
In this paper, we propose MPC (Modular Prompted Chatbot), a new approach for creating high-quality conversational agents without the need for fine-tuning.Our method utilizes pre-trained large language models (LLMs) as individual modules for long-term consistency and flexibility, by using techniques such as fewshot prompting, chain-of-thought (CoT), and external memory.Our human evaluation results show that MPC is on par with fine-tuned chatbot models in open-domain conversations, making it an effective solution for creating consistent and engaging chatbots.1 * Equal contributions; randomized order 1 Code is available in https://github.com/krafton-ai/MPC.I'm 27 years old.How old are you?How old is Sarah?-Sarah is 25.-Sarah is a student.Clarifier (few-shot) Rephrase User's question in third-person.User: I'm 27 years old.How old are you?Specifically, Memory Processor (few-shot) (CoT) This is the list of Sarah's knowledge.(1)Sarah is 25.(2) Sarah is a student.Q: How old is Sarah?A: Let's think step by step.… Answer: Sarah thinks Utterance Generator (zero-shot) This is the list of Sarah's persona.Sarah thinks Sarah is 25 and a student.This is the conversation between Sarah and User.{Dialogue history} User: I'm 27 years old.How old are you?Sarah: Summarizer (few-shot) {Dialogue History} Summary -DPR (bi-encoder) Sarah is 25 and a student.I'm 25 and a student.