Online Advertisements with LLMs: Opportunities and Challenges

This paper explores the potential for leveraging Large Language Models (LLM) in the realm of online advertising systems. We delve into essential requirements including privacy, latency, reliability as well as the satisfaction of users and advertisers which such a system must fulfill. We further introduce a general framework for LLM advertisement, consisting of modification, bidding, prediction, and auction modules. Different design considerations for each module is presented, with an in-depth examination of their practicality and the technical challenges inherent to their implementation. Finally, we explore the prospect of LLM-based dynamic creative optimization as a means to significantly enhance the appeal of advertisements to users and discuss its additional challenges.