News from the frontier of AI forecasting

FORAXION, the company behind The Forecasting Machine, pushes the limits of actionable foresight.

2025-10-03The Forecasting Machine

Three early questions for The Forecasting Machine, a year on: the shutdown, the Trump coin and late-night TV

In its first weeks, The Forecasting Machine was asked how long the 2025 US shutdown would last, whether a Trump dollar coin would be minted and which talk-show host would be cancelled next. Here is how its reasoning held up.

2025-09-30The Forecasting Machine

Superforecasting at ten, and the “proven forecasters” NewsFutures spotted in 2006

Ten years after Tetlock and Gardner’s Superforecasting, Émile Servan-Schreiber dug out a 2006 NewsFutures market in which the previous quarter’s best forecasters beat the crowd. Why the idea took a decade to be recognised, and where AI fits now.

2025-09-27The Forecasting Machine

Who controls collective intelligence? Servan-Schreiber’s case for public forecasting

Two posts in 2025 connect a criticism of concentrated platform power with a proposal to make evidence-based forecasts available to anyone.

2025-09-16The Forecasting Machine

The Forecasting Machine opens to the public after 18 months of research

On 16 September 2025, The Forecasting Machine came out of closed beta: an AI platform that reads global news, runs an ensemble of language models and turns open questions into updated probabilities.

2025-05-20The Forecasting Machine

AI gathers, crowds guess: Servan-Schreiber’s hybrid forecasting model in Prague

At the Europe as a Task conference in Prague in May 2025, Émile Servan-Schreiber split forecasting into information gathering, where AI now leads, and guessing, where crowds still take better-calculated risks.

2025-03-19The Forecasting Machine

How humans and an early forecasting AI saw 2030: the OECD's Hypermind pilot

An OECD foresight exercise compared aggregate human forecasts with an early LLM-based system across 25 possible AI benefits and risks to 2030—and emphasized the limits of both.

2025-02-22The Forecasting Machine

Human diversity and AI model collapse: a question from UM6P’s 2025 debate

A February 2025 essay turned a debate about the future of human intelligence around: could AI sustain its performance without fresh, diverse human observations? Research on recursive training gives the question a concrete basis.

2024-09-18The Forecasting Machine

AI versus the crowd: Hypermind’s hybrid forecast map of the 2024 US presidential election

In September 2024 Hypermind set its new retrieval-augmented AI Forecasting Machine against its human prediction market on a state-by-state election map. How the experiment worked, what it forecast, and what the result can and cannot tell us.

2023-10-06The Forecasting Machine

At the OECD’s 2023 foresight meeting, Hypermind shares crowd-forecasting use cases

Servan-Schreiber discussed geopolitics, public health and AI forecasting at the Government Foresight Community’s October 2023 meeting in Paris.

2023-02-23The Forecasting Machine

When ChatGPT invented a biography: an early argument for accountable forecasts

In January and February 2023, Émile Servan-Schreiber contrasted a fabricated ChatGPT biography with Hypermind’s scored probability record. The comparison was about evidence as much as intelligence.

2022-07-11The Forecasting Machine

“Arising Intelligence”: what Hypermind’s crowd got right and wrong about AI progress

From 2021, Hypermind ran contests asking forecasters to predict AI benchmarks with Jacob Steinhardt of UC Berkeley. Progress on maths and language beat the crowd’s expectations; robustness lagged them.

2021-09-23The Forecasting Machine

The US–China compute race in Hypermind’s September 2021 AI contest

Bilingual invitations asked forecasters to move beyond AI-supremacy rhetoric and predict documented training compute. Their six-to-one baseline was a historical measurement, not a probability of winning.