Scott Alexander

The y-axis shows a difficulty-adjusted Briar score.

And we see horizontal lines representing public performance and superforecaster performance.

And we see that by the time Gemini 2.5 Flash is released in April 2025, we're beginning to approach public performance.

with the line extrapolated to cross that line by around July or August 2025.

Once superforecaster bots can consider questions for pennies, we can create play money prediction markets for them and trust that the consensus answers will be as canonical as perfectly designed real money prediction markets would be for humans.

Expecting this to happen in 2027, what will that look like, and who should we invest in?

Maybe this benefits Manifold.

All of a sudden, play money markets become much more important, and quantity becomes more important at the expense of quality.

But branding and perception are important, so the victory could also go to someone who designs around super-forecasting bots from the ground up.

The Trump administration has signalled willingness to allow innovation in this space, so we have at least another three years of friendly regulators.

Three years when auto-forecasters will be improving quickly, and AI will be lowering the barrier to starting new businesses.

A lot can happen during that time.

This month in the markets.

Will Viktor Orban remain Hungary's Prime Minister after the 2026 elections?

Showing a 46% chance.

It seems to have been fairly high at 70 or 80% until July 2024.

Then it stepped down.

In around April 2025, it started to decline rapidly, reaching a low of about 20 or 30%.

It's been climbing up since and is currently hovering just below 50%.

Scott writes, I've previously written about Orban under the assumption that he's a dictator-adjacent figure who's hacked Hungary's election system so he can't possibly lose.