A London Startup Just Beat Human Forecas ...

A London Startup Just Beat Human Forecasters at Predicting the Future — And Raised $20M+ f

Sep 24, 2026

A London Startup Just Beat Human Forecasters at Predicting the Future — And Raised $20M+ for It

While most AI headlines are about writing code or generating images, one London startup just proved AI can do something much harder to fake: predict real-world events better than experienced humans can.

ManticAI has raised over $20 million, led by Balderton Capital and Radical Ventures, with Microsoft's own venture arm M12 also joining the round. The company is being valued at close to $90 million.

image

Here's what makes this one worth paying attention to:

→ Co-founded by Toby Shevlane, a former Google DeepMind researcher, and Ben Day
→ The product adapts frontier AI models to assign probabilities to future political, economic, business, and cultural events
→ Mantic's system performed strongly in the 2026 Metaculus forecasting competition — a contest where AI systems went head-to-head with experienced human forecasters and, in several cases, won
→ Unlike most AI products, forecasting has a built-in scoreboard: a system predicts a probability, the event happens or doesn't, and accuracy gets tracked over time — there's nowhere to hide behind confident-sounding language

That last point is what makes this interesting to me. Most AI tools get judged on how convincing they sound. A forecasting model gets judged on whether it was actually right — repeatedly, against a real outcome, over time. That's a much higher bar, and it's exactly the kind of test that separates genuine capability from good prompting.

If this holds up outside competition settings, forecasting could become a real enterprise-AI category alongside coding and data analysis — useful for markets, supply chains, risk planning, and intelligence work. The open question is whether a model that wins a structured competition can hold that same calibration in the far messier judgment calls real businesses actually have to make.

Would you trust an AI's probability estimate on something that actually affects your business decisions — or is "confidently calibrated" still a hard thing to verify from the outside?

#AI #Forecasting #StartupFunding #ManticAI #ArtificialIntelligence #VentureCapital

— 𝔖𝔞𝔫𝔡𝔢𝔢𝔭 ℜ𝔞𝔦𝔷𝔞

Подобається цей допис?

Купити для Sandeep Raiza каву

Більше від Sandeep Raiza

КонфіденційністьУмовиПоскаржитись