Computers managed to beat the world's best humans at chess long before they managed how learn how to walk. What's easy and hard for humans isn't necessarily a good guide to what's easy or hard for a computer.
(And I'm fairly sure it would be pretty easy to build a system that uses an LLM and a few other small components to beat Pokemon Red. The experiment you are talking about is deliberately hobbled by using a stock LLM without any such tools to make the whole thing entertaining. But when you are trading, you'd want to give your LLM as much help as possible.)
(And I'm fairly sure it would be pretty easy to build a system that uses an LLM and a few other small components to beat Pokemon Red. The experiment you are talking about is deliberately hobbled by using a stock LLM without any such tools to make the whole thing entertaining. But when you are trading, you'd want to give your LLM as much help as possible.)