Aligning LLMs with Human Uncertainty: A Beta-Bernoulli Calibrator for LLM Forecasting

In three linesBBC (Beta-Bernoulli Calibrator) converts point forecasts from any LLM into probability distributions using supervision from binary outcomes and aggregated human forecasts. The model captures epistemic uncertainty through variance, outperforming post-hoc calibration and specialized fine-tuning approaches.

Summary generated by Claude — human-verified