Tag
This paper evaluates nine LLMs on their ability to accurately communicate probabilistic predictions in natural language, finding that models are consistent but miscalibrated, particularly for uncertainty tasks.