News from the frontier of AI forecasting
FORAXION, the company behind The Forecasting Machine, pushes the limits of actionable foresight.
Results for “Brier score”
Four lessons from 25 years of crowd forecasting: Servan-Schreiber at the Lürssen Foundation workshop
At a workshop convened by MIT’s Dražen Prelec in Rijeka on 1–2 October 2026, Émile Servan-Schreiber was scheduled to present four lessons on money, Brier scores, the Bayesian Truth Serum and AI forecasting.
Play money, same ballpark: Hypermind’s 12-year accuracy benchmark against Polymarket and Kalshi
Hypermind published the data from 1,141 play-money markets resolved since 2014 and compared them with Polymarket and Kalshi. The report concludes the three are at parity, and an independent commentator agreed that parity, not victory, is the defensible reading.
Crowds versus language models: the more accurate the AI forecaster, the more it errs like humans
A Philosophical Transactions B paper co-authored by Émile Servan-Schreiber compares 76 language-model setups with human crowds on 580 ForecastBench questions and finds evidence for the accuracy–correlation effect.
What 15 months of crowd forecasting taught Johns Hopkins about disease outbreaks
A study by the Johns Hopkins Center for Health Security and Hypermind, published in November 2021, followed 562 forecasters through 61 questions on 19 diseases. The combined crowd forecast was well calibrated, beat chance by a wide margin and outscored every individual, including the public-health experts.
Hypermind’s 2016: ahead of the betting markets on Trump’s nomination, then one chance in four for Brexit and the White House
Hypermind scored its 2016 Republican nomination forecasts 35–40% better than Betfair, PredictIt and the Iowa Electronic Markets. Months later it gave Brexit and a Trump presidency about one chance in four. What a year of hits and misses says about probabilities.
