I feel like this task might be testing if LLMs can reason about ambiguity more so than if LLMs model ambiguity. To test the latter I would give it natural langu

I feel like this task might be testing if LLMs can reason about ambiguity more so than if LLMs model ambiguity. To test the latter I would give it natural language tasks that are ambigious require the recognition and modelling of this ambiguity to solve them. It's another thing to make talking about this possibly internally modelled ambiguity the goal of the task.