This is why I really like karapathy's idea of llms having spiky intelligence.

We would assume that if tasks A and B are closely related. Mastery in A would mean mastery in B but that doesn't always work with an LLM