
AI forecasters now match or beat top humans on traditional forecasting questions, so the question has moved on to where that skill holds and where it breaks. Scott Alexander walks through seven open problems raised in conversations with the Forecasting Research Institute: conditional and causal forecasts, forecasting AI progress itself, long horizons, questions that never resolve, finding the right questions to ask, and whether an AI can break from the respectable consensus. Then he explains why he is already using AI forecasts in real debates before any of this is settled.
* 00:00 - Introduction
* 02:02 - One: Can AI do conditional forecasts?
* 04:10 - Two: Can AI do truly causal forecasts?
* 07:34 - Three: Can AI forecast AI progress?
* 10:23 - Four: Can AI make good long-term forecasts?
* 11:35 - Five: Can AI answer questions that don’t “resolve” in a traditional forecasting sense?
* 14:06 - Six: Can AI think of the right questions?
* 15:53 - Seven: Can AI disagree with the respectable liberal consensus?
* 19:15 - Why I personally am erring on the side of using AI forecasters before these problems are proven solved
This is a public episode. If you'd like to discuss this with other subscribers or get access to bonus episodes, visit askwhocastsai.substack.com/subscribe
Weitere Episoden von „AI Article Readings“



Verpasse keine Episode von “AI Article Readings” und abonniere ihn in der kostenlosen GetPodcast App.








