Hand-Crafted Features and Templates as Early Prompts

Consider an older text classification system designed to determine if a movie review is positive or negative. The system works by providing the model with a structured input like: 'Review: [movie review text]. The sentiment of this review is __.' The model is then trained to fill in the blank with either 'positive' or 'negative'. Which of the following statements best analyzes the connection between this older technique and the methods used to guide today's large-scale language models?

Connecting Past and Present NLP Interaction Methods

Analyzing the Evolution of Model Interaction