The answer Bard gave is not even very coherent. Very similar results with GPT-4V as well. This makes me very cusrious how exactly do these models "see". Are the

The answer Bard gave is not even very coherent. Very similar results with GPT-4V as well. This makes me very cusrious how exactly do these models "see". Are they intelligently following the route starting from one point all along, or are they just tracing it top-to-bottom-left-to-right? Seemingly, latter is the case. I expected that the AI would be able to understand that say taking a right turn from a straight road to another sub-road definitely involves crossing (since I specified that one is running on the left of the road). And try answering along those lines.