Scott Alexander

My perspective is that the few times I've talked to Polymarket people, I've begged them to implement various cool features, and they've always said nope, sorry, too busy figuring out ways to make rules clearer.

Prediction market people obsess over maximally finicky resolution criteria, but somehow it's never enough.

You just can't specify every possible state of the world beforehand.

The most interesting proposal I've seen in this space is to make LLMs do it.

You can train them on good rule sets, and they're tolerant enough of tedium to print out pages and pages of every possible edge case without going crazy.

It'll be fun the first time one of them hallucinates, though.

I include this section under protest.

The media likes engaging with prediction markets through dramatic stories about insider trading and market manipulation.

This is as useful as engaging with Waymo through stories about cats being run over, link in post.

It doesn't matter whether you can find one lurid example of something going wrong.

What matters is the base rates, the consequences, and the alternatives.

Polymarket resolves about 1,000 markets a month, and Kaoshi closer to 5,000.

It's no surprise that a few go wrong.

It's even less surprise that there are false accusations of a few going wrong.

Still, I would be remiss not to mention this at all.

So here are some of the more interesting stories.

Phantom bets.

Who will win the 2025 Nobel Peace Prize?

Twelve hours before the announcement, someone placed a large polymarket bet on Venezuelan opposition leader Maria Corina Machado.

bringing her probability from 4% to 73%.