p19 / Conference
Large Language Models as Universal Predictors? An Empirical Study on Small Tabular Datasets
Model and Data Engineering, MEDI 2025, pp. 109-116 · 2026
In plain language
An empirical investigation of prompted LLMs as predictive models for small structured datasets.
Research context
The results help distinguish useful low-overhead data exploration from tasks that still need dedicated predictive models.
Approach
Few-shot GPT, Gemini, and DeepSeek predictions are compared with conventional ML and tabular foundation models across classification, regression, and clustering.
Findings
Classification can provide strong baselines without task-specific training. Regression and clustering show substantial limitations relative to established ML approaches.
Summary based on abstract. Springer citation year and first online publication are 2026. Conference and original preprint are 2025.