"Rick likes books from Tufte. Tufte is known for his work on data visualization. Is Rick interested in data visualizations?" (all frontier reasoning models get

"Rick likes books from Tufte. Tufte is known for his work on data visualization. Is Rick interested in data visualizations?" (all frontier reasoning models get that right). -> This qualifies for me as a super simple reasoning task (one reasoning step). From that you can construct arbitrarily more complex context + task definitions (prompts). Is that "just" statistical pattern matching? I think so. Not sure what humans do, but probably you can implement the same capability in different ways.