The machine-learning model takes less than a minute to predict future weather worldwide more precisely than other approaches.

That's saying something because the super-accurate and processing intensive ECMWF takes 12 hours per iteration. But then again, GraphCast was trained on the ECMWF predictions, among others.

The snag:

"In the meantime, problems with machine-learning approaches must be ironed out. Unlike NWP models, researchers cannot fully understand how AIs such as GraphCast work because the decision-making processes happen in AI’s ‘black box’, says Grover. “This calls into question their reliability,” he says."

How could they possibly not understand how it works???

https://www.nature.com/articles/d41586-023-03552-y

On Wed, 22 Nov 2023 13:19:55 -0800 (PST), Fred Bloggs
<bloggs.fredbloggs.fred@gmail.com> wrote:

>The machine-learning model takes less than a minute to predict future weather worldwide more precisely than other approaches.
>
>That's saying something because the super-accurate and processing intensive ECMWF takes 12 hours per iteration. But then again, GraphCast was trained on the ECMWF predictions, among others.
>
>The snag:
>
>"In the meantime, problems with machine-learning approaches must be ironed out. Unlike NWP models, researchers cannot fully understand how AIs such as GraphCast work because the decision-making processes happen in AI’s ‘black box’, says Grover. “This calls into question their reliability,” he says."
>
>How could they possibly not understand how it works???
>
>https://www.nature.com/articles/d41586-023-03552-y

My wife is much better than any of the supercomputer models at
predicting if it will rain tomorrow.

Nothing artificial about her!

But no amount of intelligence can predict the future state of a wildly
chaotic system.

On Thursday, November 23, 2023 at 8:33:38 AM UTC+11, john larkin wrote:
> On Wed, 22 Nov 2023 13:19:55 -0800 (PST), Fred Bloggs
> <bloggs.fred...@gmail.com> wrote:
>
> >The machine-learning model takes less than a minute to predict future weather worldwide more precisely than other approaches.
> >
> >That's saying something because the super-accurate and processing intensive ECMWF takes 12 hours per iteration. But then again, GraphCast was trained on the ECMWF predictions, among others.
> >
> >The snag:
> >
> >"In the meantime, problems with machine-learning approaches must be ironed out. Unlike NWP models, researchers cannot fully understand how AIs such as GraphCast work because the decision-making processes happen in AI’s ‘black box’, says Grover. “This calls into question their reliability,” he says."
> >
> >How could they possibly not understand how it works???
> >
> >https://www.nature.com/articles/d41586-023-03552-y
> My wife is much better than any of the supercomputer models at
> predicting if it will rain tomorrow.
>
> Nothing artificial about her!
>
> But no amount of intelligence can predict the future state of a wildly chaotic system.

Not moment to moment. But climate change is predictable in terms of short term averages.

--
Bill Sloman, Sydney