What’s the point though? That LLMs tend to be constraint by constraints in their prompting? That seems unsurprising. Humans are visual animals. We can spot a c

What’s the point though? That LLMs tend to be constraint by constraints in their prompting? That seems unsurprising. Humans are visual animals. We can spot a chicken in a graph, but we’re unlikely to be able to tell that a different graph is using XY coordinates to encode a message against a one-time pad. But so what?