SETI Post-Detection HubMay 1
Hanno Rein

There is a new study out which shows that physics based weather prediction is better than #slop based predictions when it comes to extreme weather events. This shouldn't really be surprising to anyone who has studied physical or chaotic systems.

https://www.science.org/doi/10.1126/sciadv.aec1433

Apr 30 at 11:44pmWeb
Show threadaoanlaMay 1
@hannorein I seem to recall some general paper on statistical models (the "AI" approach here) showing that this is essentially always going to happen due to the sparsity of data near the tails of the distribution. That paper suggested an over-estimation of rare event frequency though - it was trying to argue that models make the same trade-off as humans do re "1 in a million" events and their significance.
Show threadHanno ReinMay 1
@aoanla which is whining think a physics based model has a huge advantage. Even if it’s not as accurate on average it kept just the essence better.
Show threadaoanlaMay 1
@hannorein I don't think it [edit: I typoed "I" here and only just noticed] was whining, it was trying to sound "insightful" (and argue that we "understood" why humans make reasoning errors (without mathematics) about rare events). This was a bit before the current "AI can solve everything, omg" hype cycle.
Show threadaoanlaMay 1
@hannorein in addition, I think the thing people miss about physics-based models is that they also validate our understanding of systems - a statistical fit just "gives answers" without really telling us much about *why*, which is surely the point of science?