Patri Friedman (commenting on the Robin Hanson gadfly’s post):
Ah, Mencius. Combines a brilliant insight on prediction markets that I hadn’t seen before and increases my understanding (they require training) with ridiculously stupid points like “They can’t predict things that no one knows”.
Duh, they are information aggregators which give us a better estimate of the consensus probability than other methods (b/c if anything else works better than the current market you can use it and make money). This is an awesome property – that they automatically incorporate other prediction methods. But only over time through the training process.
I agree with that overall, but I disagree with the choice of words for “better” and “awesome”.
