Evaluating and enhancing probabilistic reasoning in language models
Language models are capable of remarkably complex linguistic tasks. However, numerical reasoning is an area in which they frequently struggle. We systematically evaluate the probabilistic reasoning capabilities of LLMs and show that they can make more accurate inferences about distributions aided by the incorporation of real-world context and simplified assumptions.
https://research.google/blog/evaluating-and-enhancing-probabilistic-reasoning-in-language-models/