Keeping Humans in the Loop: The Role of Human Skills in Large Language Model–Based Natural Language Applications

Human skills in the form of data annotation and generation have long been the cornerstone of building and testing machine learning algorithms. With the advent of large language models (LLMs) and their unsupervised learning on large quantities of data, the role of human data experts has been called into question. This article examines the changing role of human data annotation and linguistic analysis and provides an overview of past, current, and possible future paths. We discuss the targeted use of annotated data for fine-tuning and for feedback and reward models to supplement and improve LLMs, as well as the use of human linguistic skills for low-resource languages and for creating multimodal data sets, such as those required for human–robot interaction.