On: The future of AI according to thousands of forecasters

[saulmunn]

this isn't quite true -- on metaculus, accounts that have a history of forecasting things well are weighted more heavily

[zgluck → saulmunn]

SISO (Shit In, Shit Out) still applies. You guys need a high quality user base with domain knowledge, at least as a seed. There is no proof that you have that at the moment.

Edit:

Okay, that track record page avionical posted in a separate comment is actually a bit convincing now that I dig deeper into it. :-)

I suppose that for e.g. AI/AGI a weakness could be that the estimates for most of the users have been short term (a few years at most) but the AGI estimates are 8-17 years away. Those estimates are a lot harder to do make and hence surely a lot less accurate.

[avionical → zgluck]

What would be good evidence of a high-quality user base with the relevant skills? A transparent, well-calibrated track record?