TopBench: A Benchmark for Implicit Predictive Reasoning in Tabular Question Answering

TopBench is a newly introduced benchmark specifically designed for implicit predictive reasoning in Tabular Question Answering, consisting of 779 samples across four sub-tasks, including single-point prediction and treatment effect analysis. The benchmark evaluates various large language models (LLMs) on their ability to recognize latent intent and perform reliable predictive reasoning, revealing that current models often default to simple lookups rather than effective intent recognition. This work emphasizes the need for enhanced modeling capabilities to improve prediction precision, which is crucial for practitioners developing AI systems that require advanced reasoning over tabular data.