Why Are Large Language Models So Terrible at Video Games?

A new analysis explores why large language models struggle with video games, despite their text-based prowess. The author argues that LLMs lack the real-time perception and spatial reasoning required for dynamic game environments. These models process static text, not the continuous visual and auditory streams that games demand. They also fail to handle the iterative trial-and-error learning that human players use. The piece suggests that LLMs' inability to form persistent world models limits their gameplay. This highlights a fundamental gap between language understanding and embodied interaction in complex systems.