JoLT leverages the in-context learning capabilities of Large Language Models to create powerful joint probability distributions for tabular data, enabling more accurate predictions with uncertainty quantification.

• Joint predictions across multiple target variables, handling heterogeneous data types
• Context-aware analysis that incorporates domain knowledge embedded in LLMs
• Flexible conditioning on user-specified side information
• Uncertainty quantification through probabilistic predictions

This engineering advancement matters because it provides a streamlined method for complex tabular data analysis without requiring specialized model architectures or extensive domain expertise.

JoLT: Joint Probabilistic Predictions on Tabular Data Using LLMs