A General Approach to Website Question Answering with Large Language Models
Yilang Ding, Jiawei Nie, Di Wu, Chang Liu · 2024
Language Models (LMs), in their most basic form, perform just like any other machine learning model - they produce interpolations and extrapolations based on their training distribution. Although recent models such as OpenAI's GPT-4 have demonstrated unprecedented capabilities in absorbing the copious volumes of information in their training data, their ability to consistently reproduce factual information still remains unproven. Additionally, LMs on their own lack the ability to keep up to date with real life data without frequent fine-tuning. These drawbacks effectively render base LMs unserviceable in Question Answering scenarios where they must respond to queries regarding volatile information. Retrieval Augmented Generation (RAG) and Tool Learning [1] were proposed as solutions to these problems, and with the development and usage of associated libraries, the aforementioned problems can be greatly mitigated. In this paper, we ponder a general approach to website Question Answering that integrates the zero-shot decision-making capabilities of LMs with the RAG capabilities of LangChain and is able to be kept up to date with dynamic information without the need for constant fine-tuning.