Cassi places 2nd overall and 1st on “Dataset” questions on Forecastbench

Only Elon Musk’s XAI outperformed us overall.

And that bland ‘dataset’ title belies the import of the victory in this category.

What it means

Cassi is the best in the world: #1 for questions where there are no good odds out there already.

Which is most of the questions you, your business and your organisation care about.

On “Dataset” questions, Cassi was within 0.006 Brier score of the median Superforecaster—practically a dead heat.

When you can’t leverage the “wisdom of crowds” to base-rate your odds, the gap between AI and elite humans almost vanishes. 🎯

For questions where there are odds to get you started, (e.g. where you start figuring out the chances of your team winning the League, or of country X being invaded by country Y, by looking at the odds at Metaculus, Kalshi, Polymarket (or outside the competition Betfair for that matter)) human superforecasters still have the edge… …but vs AI forecasting we are second only to Grok.

Why did we excel?

Cassi has focused intensely on real-world performance – accuracy in prediction in areas where there are no good odds – as part of our wider work with clients and customers.

Ironically maybe, we won because we are not ‘bench-maxing’ by tailoring our model to the scoring systems of other forecasting competitions.

Why does it matter?

It’s highly unlikely your organisation has access to a team of proven superforecasters, and even if you did, forecasting takes time.

Cassi’s forecasts are with you at the touch of a button.

Moreover, prediction is about more than forecasting.

‘Everything is Prediction’.