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.

翻译：本文探讨了在大语言模型（LLM）领域应用于在线广告系统的潜力。我们深入分析了此类系统必须满足的关键需求，包括隐私性、低延迟、可靠性以及用户和广告主的满意度。进一步地，我们提出了一个通用的大语言模型广告框架，该框架包含修改、竞价、预测和拍卖模块。针对每个模块，我们阐述了不同的设计考量，并深入检验了其实用性以及实现过程中固有的技术挑战。最后，我们探索了基于大语言模型的动态创意优化前景，作为一种显著提升广告对用户吸引力的手段，并讨论了其额外挑战。