Implicit vs. Explicit Prompting Strategies for LVLMs in Referential Communication

This article compares implicit and explicit prompting strategies for Large Vision-Language Models (LVLMs) in referential communication, based on studies by Jones et al. and Zeng et al. It finds that while LVLMs can effectively coordinate efficient referring expressions when explicitly prompted, they struggle to do so with implicit prompts, indicating a significant gap in communication strategies between humans and AI systems. This research provides insights into the design of prompting strategies that could enhance model performance in communicative tasks.