How forecasting with Prolepsia works
Plain guides to the ideas behind Prolepsia’s forecasts. Each one explains a single question in full, with links to the pages where you can check it.
- How a forecast differs from asking an LLM with a search toolAn LLM with a search tool answers once, in prose, and is never graded. A Prolepsia forecast is a probability with its reasons and sources, kept current and scored once the outcome is known.
- How a historical forecast avoids information that did not exist yetTo test a forecasting method on past questions, the run must not read anything published after the date it pretends to be. How Prolepsia limits a historical run, and where that has limits.
- How forecasts are scored: the Brier score and log lossA forecast is scored once its outcome is established. How the Brier score and log loss work, with worked examples, and which questions cannot be scored.