Forecasting Research Institute

AIRO is a regularly updated dashboard that tracks the probability of catastrophic risk events as forecast by an ensemble of frontier AI models, including AI-related catastrophes and other AI-related harms, and it runs weekly.

LEAP is a three-year project that tracks views of leading AI experts, superforecasters, and a representative general-public participant group, collecting thousands of monthly probabilistic forecasts about AI capabilities, adoption, and impacts on topics such as AI progress indicators and related consequences.

ForecastBench is a dynamic, continuously updated benchmark that measures ML forecasting accuracy on a changing set of probabilistic forecasting questions, combining dataset-style time-series questions and market questions with automated question generation and nightly leaderboard updates.

XPT was a long-run forecasting tournament that convened experts and superforecasters to generate multi-stage probabilistic forecasts, explanations, persuasion, and updates about existential and catastrophic risks over the next century.

The Forecasting Proficiency Test is an assessment intended to measure forecasting ability using probabilistic forecasting tasks, designed to support higher-stakes decision-making that depends on calibration and accuracy.

Project Improbable focuses on improving low-probability judgments, including methods for eliciting and assessing forecasts about rare outcomes where traditional inference is often challenging.