Why Are Large Language Models so Terrible at Video Games?

Large language models LLMs have improved so quickly that the benchmarks themselves have evolved, adding more complex problems in an effort to challenge the latest models.