If AI tells you:
"There is a 40% chance this will happen."
Would you trust AI more than a professional analyst?
A news story tonight is pushing this question from the lab into reality.
London-based AI startup Mantic announced it has completed a $25 million seed funding round.
But what attracted investors wasn’t just another AI tool.
It’s this:
The company is training AI specialized in one thing—predicting the future before it happens.
What has Mantic just achieved?
Reuters reported on the recently concluded Summer 2026 Metaculus Cup, where participants must assign probability predictions to upcoming political, economic, and cultural events.
For example, not just answering:
"Will this happen or not?"
But rather:
"What is the probability that this will happen?"
Once the events occur, participants are scored based on how their predictions matched reality.
According to Reuters, Mantic ultimately:
outperformed all human participants.
In overall ranking, it was only behind another AI bot named "laertes."
Metaculus itself posed this question:
"Will any bot beat all humans in the Summer 2026 Metaculus Cup?"
The official answer is now:
Yes.
This is a clear milestone for AI forecasting.
Mantic Didn’t Train a Large Model from Scratch
This is one interesting aspect of the company.
Mantic CEO Toby Shevlane, a former research scientist at Google DeepMind, co-founded Mantic with Ben Day in 2024.
Mantic’s approach is not to spend billions training its own frontier model.
Instead, it leverages frontier models from other AI labs and specializes them specifically for forecasting.
The general process can be understood as:
Make a prediction
→ Test it against historical data
→ Wait for the event outcome
→ Score the prediction performance
→ Identify errors
→ Improve the system for the next round.
This approach has a huge advantage:
Forecasting is one of the few AI tasks where the "correct answer" eventually appears.
A beautifully written summary can be hard to score absolutely.
But if you predict:
An event has a 20% chance of happening,
Eventually, you can check if it actually happened.
Thus, forecasting suits AI well for continuous, measurable feedback.
Why Might AI Be Especially Suited for Forecasting?
Humans face one practical limit in predicting the future:
Humans get tired.
No one can spend 24/7 reading thousands of news articles.
Nor constantly update probabilities for hundreds of events with every new piece of information.
AI, on the other hand, can:
- Continuously search for new information
- Track numerous events simultaneously
- Quickly review historical cases
- Recalculate base rates
- Constantly update probabilities
- Run the same process repeatedly across many questions
Mantic’s CEO also mentioned another possible advantage AI had in the Metaculus Cup:
It’s less prone to herd behavior, following the consensus blindly.
If the market or human forecasters heavily favor one outcome, AI can recalibrate based on data instead of simply following the crowd.
This doesn’t mean AI has no bias.
But its biases may differ from those of humans.
But There Is a Very Important "But"
Seeing "AI beats all human forecasters"
It’s easy to jump to the conclusion that:
AI already predicts the future better than humans.
The current evidence does not support such a sweeping claim.
The reason is Metaculus also runs another test called FutureEval, directly comparing AI with top human forecasters.
In the Spring 2026 test:
10 Metaculus Pro Forecasters
were matched against
the top 10 forecasting bots.
Both sides answered the same set of 99 questions.
The result was:
Human pro forecasters still averaged 1.25 head-to-head points per question ahead of bots.
This gap was not statistically significant.
In other words:
This data doesn’t prove humans are stronger or that AI has surpassed top humans.
The more accurate statement is:
The gap has become very small.
Even more notable is the speed of improvement.
In a previous Q2 2025 FutureEval, the same bot team trailed human pros by about:
20.03 points per question.
By Spring 2026, it had shrunk to:
1.25 points.
This is the number worth paying attention to tonight.
So What Does Mantic’s Victory Mean?
It does not mean:
"AI already knows the future."
Instead, it means:
"AI has reached the stage where it can directly compete with professional human forecasters."
There’s a big difference.
Forecasting is never about prophecy.
For example:
If AI says there’s a 70% chance of an event happening,
and the event ultimately doesn’t occur,
that doesn’t necessarily mean the prediction was wrong.
If you group many "70%" predictions, approximately 70% should occur—that's good calibration.
So valuable AI forecasting shouldn’t claim:
"The future will definitely happen like this."
Instead, it should tell you:
"Based on current information, here are the probabilities of each possible outcome."
Why Are Businesses Beginning to Take Interest?
Reuters quotes investors saying Mantic has already attracted attention from global corporations and governments, with some clients integrating its AI—though the company hasn’t disclosed client names.
One particularly active group is:
Hedge Funds and Trading Firms.
The reason is clear.
If you can estimate
- economic events
- policy changes
- industry demand
- commodity supply
- company events
earlier and more accurately than the market, even a small forecasting edge can be very valuable.
But caution is needed here.
Just being slightly more accurate in estimating event probabilities than others doesn’t guarantee consistent investment returns.
Once entering financial markets, factors such as:
whether prices already reflect information,
transaction costs,
timing of entry and exit,
position sizing,
liquidity,
risk management
and many other variables come into play.
Forecasting benchmark scores can’t be directly equated to trading success rates.
What Are Investors Really Betting On With $25M?
The current seed round was led by Radical Ventures.
Other investors include:
- Microsoft’s M12
- Thinking Machines Lab
- Balderton Capital
- Other investors
The valuation was not disclosed.
What investors might be really betting on is not a "chatbot that guesses answers better."
But a bigger vision:
If AI can continuously turn uncertain futures into more reliable probabilities, forecasting may become a new foundational layer of enterprise infrastructure.
Companies strategize,
governments assess risks,
supply chains plan demand,
financial institutions run scenario analyses,
all often relying on humans to judge incomplete information.
AI might not replace final decision-making.
But it could first transform:
"I think this will happen."
into:
"Given current data, I estimate a 37% chance it will happen, up from 29% yesterday."
This step alone can be very valuable.
The Real Takeaway Tonight Isn’t "AI Can Predict the Future"
Mantic’s result is definitely worth noting.
But even more important is this:
AI is moving from answering "What is happening now?" to quantifying "What is likely to happen next?"
Until now, we mostly tested AI on:
Can it write?
Can it calculate?
Can it code?
Increasingly, we’ll test:
Can it judge future probabilities more accurately than humans, despite incomplete information?
The Summer 2026 Metaculus Cup sends a strong signal.
But FutureEval also reminds us:
Beating all humans in a competition doesn’t mean AI has fully surpassed top human forecasters.
The real change is:
This contest, once seeming far away,
is now very close.
Today, grow with AI a little.
Learn an AI skill daily.
Save a bit of time every day.
Improve your abilities little by little.
SasaDaily grows with you.
