Scott Alexander

Scott writes, here's a pretty crazy, metaculous question.

The resolution criteria specify it's not about scammers using AIs to blackmail their victims, it's about an AI independently developing and executing a blackmail plan without human prompting or support.

Something like this has already happened in toy experiments conducted by safety teams when all the conditions were exactly right, but forecasters seem confident it will happen in real life sometime in the next three years.

I don't understand what's going on here, and I'm going to recheck this question after signal boosting it to see if it changes.

Which company has the best AI model end of March?

A polymarket question?

Google is sitting on 79%, XAI on 10%, and all the others below that.

Polymarket has a few of these who-has-the-best-AI-when markets.

Resolution is usually positioned on the LM Arena leaderboard, which usually but not always mirrors common sense consensus.

I get more interested in these the further out they go.

But the Dune version is bizarre.

It doesn't even list Google as an option.

And there's nothing past mid-year.

Other implied claims from Polymarket's tech section?

Only 44% chance Anthropic will still dominate coding by late March.

Anthropic, and especially OpenAI, probably won't IPO this year.

XAI will call their next model Grok 4.20, of course.

And Nathan Young has put together an AGI Timelines Forecasting Dashboard.

AGI timeline forecasts.

So we see a timeline on the x-axis, and then years when AGI arrives on the y-axis, and we see lots of different lines plotted.