So this excellent paper has good data on why AGI is unlikely by 2043.
But then very optimistic assumptions to conclude AGI is plausible by 2100.
My brother in tech, we're not getting 10^5 better chips; we're lucky if they don't shrinkflate 50% again.
https://twitter.com/sandersted/status/1667231392030396421
@sj I'd be willing to bet against their claim at 50:1 odds. Their methodology is likely broken, and I will try to figure out whether there is any scientific support for this kind of question decomposition
@niplav their 2043 claim or 2100 claim?
@sj their 2043 claim
@niplav read their analysis and let me know what step you disagree with
@sj I will probably not do that, since it doesn't feel like it passes my relevance threshold. I acknowledge that this is a bit of a dick move (to criticize a paper based on the abstract and then refusing to give a detailed critique).
Best thing I might do is to argue against the decomposition approach in forecasting and try to find evaluations of it, I have the intuition it's either not been investigated or shown to be biased downwards
