Google DeepMind touts AI model for 'better' global weather forecasting Google DeepMind researchers claim they’ve used machine learning to devise a model that can deliver better 15-day weather forecasts and requires only modest quantities of compute resources to produce its predictions. The model, dubbed GenCast, provides a probabilistic ensemble weather forecast – a distribution of probable …
The reason they do not give you a confidence interval is probably that they are not doing statistical modelling and therefore generally don't have much quantitative idea how good their forecast is. However I'm 70% confident that AI will be happy to make up a number and report it to you. e,g, "There is an 83% chance that a tornado will carry you off to Oz between 1400 an 1600 this afternoon" (Which is why it might be a good idea to keep the works of L. Frank Baum out of your AI's training materials).
This post has been deleted by its author
Or in high buildings in Russia.
But really I don't trust this and it's likely to be dangerous sometimes.
See also IBM claims for Medical AI for "Watson" after "winning" Jeopardy. Turned out mostly only shared branding and didn't work.
A lot of these so called AI systems are doing pattern matching based on past human selection or results. This is ultimately doomed, especially if after a while there are no human experts to scrape the data from.
All the "AI" purveyors have been proven to mislead, hype or even lie. Even the terms used are a lie.
only works IF you are looking in the direction of where the weather is coming from or will be coming from in the next 2-3 hours. Looking the other way only tells you what has been.
Personally, I'd rather go with the European Medium Range forecast. The US one forecasting a deluge over my home right now. It might be dull and overcast but it isn't raining.
These AI models that use less input data are more easily swayed by garbage data. GIGO.
@sabroni
"Hopefully it succeeds. Could be the first and so far only good product from the AI boom."
Maybe. We will see. The joy of markets and competition is to try out new things. Some of the features on cars for self parking and self driving also seem pretty good. Then there are translation programmes which I can attest are helpful.
"Dunno why you got downvoted. You said exactly what I was thinking."
Downvoter here. Yes the climate changes and that is normal. Crying against a new method that looks very promising because of the latest fad seems daft to me.
But the bit that jumped out at me is the complaint of using electricity to train the AI when it takes less time and energy to produce a forecast which is more accurate (according to the article) and would benefit the cult of the wind god for placing their monuments.
So I downvoted for 2 reasons, cult beliefs in madness and being such a believer but against solutions to that cults believed problems.
@Will Godfrey
"I can't make head nor tail of your comment, so no point in carrying"
I am really not used to people admitting to stupidity but it is refreshingly different.
I will try to dumb it down for you.
First it assumes the MMCC co2 we are all doomed cult is correct.
Second if you believe in the cult (first point above) then if this ML method uses less time and energy to make the predictions (in the article) then Irongut should be all for it. Instead Irongut complains about the energy to train the model.
If this is still too complicated to understand then yeah you might not want to reply. But I hope this helps.
@Headley_Grange
"Top Tip: if you think everyone else is in a cult it's probably you that's in a cult."
Very possibly. But since I dont think everyone else is in a cult, only those in the current MMCC cult I think I am on safe ground.
But I do notice you dont have an answer for the cult problem which is to oppose solutions to the problems they claim are real. This is where the cult makes no sense as if they truly believed in the theory they would be for solutions and not against them. My favourite is how nuclear is only now a consideration after over 2 decades of knowing it was the solution to the imagined problem. Instead we still suffer the insanity of trying to apply technology that doesnt work which relies on some magic technology we currently dont have and always requires a bit more investment to make it work, honest.
@sabroni
"Could you contemplate getting over yourself for a second?"
Try reading the comments. Had Will Godfrey asked a question for what he didnt understand then I would have been more gracious in my reply. Instead his argument was that he doesnt understand so not worth replying (this was his reply ffs) and so will flounce off.
If he wants to act childish then I will treat him as such. I made it simpler but agreed he probably shouldnt respond if the conversation is too complicated.
But anyway did you have something of value you wanted to add to the conversation?
It reminds me of a theory called "six day weather types" that was propounded in the early to middle 1960s by a teevee weathercaster named Harry Geise, who did forecasting for a station in Sacramento, California.
This was in the days before we had the GOES satellites and supercomputing number crunching systems and it had a pretty decent predicitive accuracy. It was taken seriously enough by the professional community that Geise gave a paper on the subject at an American Meteorological Society conference in 1965.
How well that would work in an increasingly unstable climate is anyone's guess (or Geise. . .).
I remember in a previous life attending a talk on financial modelling, stochastic differential equations and all that stuff. Not my thing at all, but I do remember one interesting observation which was along the lines of financial modelling is unlike climate modelling and weather forecasting because the outputs of the model can directly and measurably affect the thing that is being modelled. Looks like that's no longer true.
In rural China it was/is fairly common to keep a ‘weather loach’ (a bit like a 6inch fat eel) in a large jar. When it’s calm - the weather would be fine, if it was particularly wiggly then bad weather was on the way.
I kept a couple and they were great fun - until they escaped and died on the floor. Can remember what weather that was supposed to foretell.
To allow for fair evaluation, the prediction must be projected onto the same scale as is currently used - e.g. 50% chance of 1/10 inch of rain, etc. Then the accuracy can be objectively evaluated.
In fact a range of predictions is not practical for daily human use (although it might be better for some special cases like farm irrigation).
The Register 
Biting the hand that feeds IT
Copyright. All rights reserved © 1998–2024
