I'm personally excited about LLMs becoming more number-fluent for regression tasks, e.g. read this paper and give it a grade 0-100