Large Language Models Struggle with AI System Design

Despite their proficiency in writing regular programs, large language models (LLMs) are poor at designing AI systems. They tend to create overly complex symbolic systems instead of leveraging other LLMs, leading to inefficient solutions.