Do You Even Need More Data? Rethinking LLM Deployment for Label-Scarce Tabular Classification with TabLens
The LLMTabBench evaluates how large language models perform tabular classification with little data by testing various factors like examples and data complexity. Findings show zero-shot LLMs can be competitive, but performance saturates quickly with few-shot examples. Data complexity acts as a hard ceiling on performance, and combining in-context information, few-shot examples, and external instructions provides the best results.