Are LLMs able to notice the “gorilla in the data”?