LEAP framework improves LLM probabilistic forecasting accuracy

LEAP replaces monolithic prediction with a method that elicits individual likelihood parameters for each piece of evidence. These parameters are combined via a deterministic probabilistic model to form a posterior distribution, preventing the collapse of uncertainty during forecasting.