Ice-kun: A Virtual Pet on a Low Energy Display Controlled by a Large Language Model

Abstract

This paper presents "Ice-kun," a virtual pet controlled by a large language model (LLM) and displayed on addressable light-emitting diode (LED) boards. Unlike virtual pets with limited interactions, Ice-kun leverages an LLM to interpret natural language input and generate countless playful animations. Ice-kun is configured into several parts to generate the animations, and the LLM is tasked with generating animations for these parts. To achieve fast, correctly formatted responses from the LLM as a real-time control agent, two prompting strategies and six prompt engineering techniques are compared. Fast response time is achieved through a prompting strategy that uses multiple shorter prompts, each corresponding to a different part and executed asynchronously in parallel, while few-shot prompting and zero-shot CoT prompting are incorporated to ensure high-accuracy, correctly formatted responses. Code, experiment results, and demo videos are available in the authors' repository.