AutoElicit harnesses LLMs to generate expert priors for specialized predictive models, reducing dependency on human experts and limited datasets.

• Creates domain-specific prior distributions for interpretable models
• Provides comparable performance to human expert priors
• Demonstrates effectiveness in healthcare applications including UTI prediction
• Maintains model interpretability while leveraging LLM knowledge

This approach is particularly valuable in medical settings where data privacy is critical, labeled data is scarce, and model interpretability is essential for clinical adoption and trust.

AutoElicit: Using Large Language Models for Expert Prior Elicitation in Predictive Modelling