This seems to imply that in the near future our weather models will become useless, but we won’t know when it has happened. Rohde’s Uncertainty Principle.
There is a broader pattern showing that generally most forecasts and predictive models are utterly useless or detrimental in the case of actual novelty. Look at just-in-time supply chains during the pandemic: in many sectors the system grinded to a halt. First, the run on toilet paper, then meat products, then wood (and many more). Did refrigerant suppliers know that the heat dome would cause millions of PNW denizens realize that AC is now an essential?
Most AI is trained on historical data. Adoption of such tools is going to need to rely further on wild simulation and adversarial networks if they are to be effective in the highly dynamic world we have entered into.
That thread also pointed out that the local (2 weeks out) models did predict the temperature but the forecasters didn't believe it.
So, that means that the localized processes are being monitored and predicted. So, this isn't completely outside the models.
However, as was pointed out, this may be a type of "hurricane" (used as a stand-in for a random-ish large scale variation) which may now become more common. That kind of large-scale change will require some updates to the models. (ie. we can model the fact that hurricanes occur seasonally, but we can't model individual hurricanes until they occur).
All right, so what this means is climatists can't even predict what weather will be like next month, but claim to be able to predict global climate 20 years from now.
5 comments
[ 2.9 ms ] story [ 24.8 ms ] threadMost AI is trained on historical data. Adoption of such tools is going to need to rely further on wild simulation and adversarial networks if they are to be effective in the highly dynamic world we have entered into.
That thread also pointed out that the local (2 weeks out) models did predict the temperature but the forecasters didn't believe it.
So, that means that the localized processes are being monitored and predicted. So, this isn't completely outside the models.
However, as was pointed out, this may be a type of "hurricane" (used as a stand-in for a random-ish large scale variation) which may now become more common. That kind of large-scale change will require some updates to the models. (ie. we can model the fact that hurricanes occur seasonally, but we can't model individual hurricanes until they occur).
It becomes invalid when the model becomes invalid.
Here, we had a new phenomena that was not included in the existing model.
It was not "statistically impossible". Rather, it was "not covered by existing statistical model".
Did I get that right?