That's a really neat experiment. I did something similar with an ASCII maze and had Claude work through it. Seems like you have the same concept, but a tougher "maze". Soon models will be able to reason outside of their context window and directly in our world. I think there is room for us to build that bridge, and I'm really excited to see that happen. Models mostly have the tools already to reason about things, but there currently isn't a way to bridge the divide.
That's a really neat experiment. I did something similar with an ASCII maze and had Claude work through it. Seems like you have the same concept, but a tougher "maze". Soon models will be able to reason outside of their context window and directly in our world. I think there is room for us to build that bridge, and I'm really excited to see that happen. Models mostly have the tools already to reason about things, but there currently isn't a way to bridge the divide.