I'm not sure how I feel about the details of the benchmark, but I think this is an important direction in which it would be nice if LLMs were improved. At pres

I'm not sure how I feel about the details of the benchmark, but I think this is an important direction in which it would be nice if LLMs were improved. At present they don't really understand either stories or even short character interactions. Microsoft Copilot can generate dialogue where characters who have never met before are suddenly addressing each other by name, so there's great room for improvement.