SAGE 0.2

LLMs for DOM Informed Internet Guidance

Abstract

The grey divide affects many older adults, leaving them vulnerable to digital exclusion and fraud. Education has largely
failed to be effective, as the digital ecosystem is constantly changing. Previous work has proposed that a system providing just-in-time support through a text-based large language model (LLM) assistant designed to provide patient and context-aware support may be able to dynamically augment the user’s capabilities in place of proactive education alone. This paper describes work-in-progress toward such a practical system, called SAGE 0.2. SAGE is an API-based agent that is made available to the user through a browser extension. By injecting a lightweight content script, the document object model (DOM) is parsed and provided to the LLM as a context. Responses to user queries are then informed by the current webpage, allowing SAGE to answer questions and provide simple on-screen guidance. This early prototype uses free-tier models to show the feasibility of such a practical and impactful application of LLMs, however it also demonstrates a number of critical issues that will need to be addressed to apply such a system at scale.