I find it baffling that the AI Snake Oil guys chose forecasting as something they think human experts are basically already perfect at. I don't think they asked any leading forecasters before making that statement, because leading forecasters tend to have appropriate humility about their own abilities. I'm a pro forecaster, and I have the sense that the people who are better than me are *way* better than me. There is a lot of room at the top, and it's clear to me that superhuman performance is possible in principle.
Yeah, this is my intuition as well. I am by no means a top forecaster, but I did put quite a lot of time into forecasting and prediction markets a while ago. Apart from everything else, I think it makes a huge difference how much time you can sink into research and monitoring. And on this axis alone humans are at such a disadvantage already.
This also fits into a general pattern some people fall into, of not believing that superior intelligence confers very much advantage. As if the ability to perform general-purpose reasoning over unbounded time scales (as a society) is the whole game.
There are several independently robust arguments for how we know something more cognitively capable than us can exist even if we really are at the peak of possible intelligence (more speed, more compute, more agents). But no, there is also such a thing as literally thinking in more dimensions at once. This, I argue, is what larger LLMs can do that smaller ones cannot (beyond sheer memorization capacity).
I find it baffling that the AI Snake Oil guys chose forecasting as something they think human experts are basically already perfect at. I don't think they asked any leading forecasters before making that statement, because leading forecasters tend to have appropriate humility about their own abilities. I'm a pro forecaster, and I have the sense that the people who are better than me are *way* better than me. There is a lot of room at the top, and it's clear to me that superhuman performance is possible in principle.
Yeah, this is my intuition as well. I am by no means a top forecaster, but I did put quite a lot of time into forecasting and prediction markets a while ago. Apart from everything else, I think it makes a huge difference how much time you can sink into research and monitoring. And on this axis alone humans are at such a disadvantage already.
This also fits into a general pattern some people fall into, of not believing that superior intelligence confers very much advantage. As if the ability to perform general-purpose reasoning over unbounded time scales (as a society) is the whole game.
There are several independently robust arguments for how we know something more cognitively capable than us can exist even if we really are at the peak of possible intelligence (more speed, more compute, more agents). But no, there is also such a thing as literally thinking in more dimensions at once. This, I argue, is what larger LLMs can do that smaller ones cannot (beyond sheer memorization capacity).