This AI Paper from Microsoft Proposes a Machine Learning Benchmark to Compare Various Input Designs and Study the Structural Understanding Capabilities of LLMs on Tables

A team of researchers from Microsoft has presented a benchmark to assess the Structural Understanding Capabilities (SUC) of Large Language Models (LLMs) in understanding structured data like tables. The study found that input options, such as partition markers and content order, affect LLM performance. Self-augmentation techniques, utilizing LLMs' internal knowledge, have shown significant improvements in LLM performance on tabular tasks.