How Smart are Smart Readers? LLMs and the Future of the No-Reading Problem

Abstract

Large Language Models (LLMs) can be used to summarize and simplify complex texts. In this study, we investigate the extent to which state-of-the-art models can reliably operate as ‘smart readers’: applications that empower consumers to tackle lengthy, difficult-to-read, and inaccessible standard form contracts and privacy policies.
Our analysis reveals that smart readers (1) reduce by 66.9% the length of contracts; (2) reduce reading time by 14:41 minutes (3) improve text readability by converting college-level texts to texts readable by fifth-grade students; and (4) do so without considerably compromising the essential information in the original contracts. However, smart readers are not flawless. They sometimes miscommunicate legal terminology and occasionally present information in a misleading or erroneous manner. Such issues prevent smart readers from replacing the advice of a qualified lawyer. However, for the large mass of daily transactions where consumers would not consider using a lawyer, current-generation smart readers could be an effective tool. In the context of these transactions, we conclude that current generation smart readers have arrived and that their arrival invites an academic and policy paradigm change.

Keywords: Consumer contracts, consumer protection, contractual complexity, GPT4, large language models, readab