• Hello, please take a minute to check out our awesome content, contributed by the wonderful members of our community. We hope you'll add your own thoughts and opinions by making a free account!

Misc 2023 Spring/Summer Whamby Thread

Status
Not open for further replies.
Would have thought you would make the trip for this festival! The lineup is insane. I’m going for Friday !



346608689_1307096040165336_7578099644156677895_n.png
You get to see sleep token.
 
Huawei's Pangu-Weather AI weather model's verification scores are impressive thus far.

F2fPcS-WQAAKr00
Interesting. It looks like one of its biggest selling points is the speed at which it produces forecasts, too. It is supposedly much quicker than NWP methods.
 
Yup, 1 min to run 240 hours. They could run this hourly. The AI allows this not to be run on a supercomputer too.
I’m curious if they could apply machine learning to NWP forecasts to get an even better forecast (albeit longer to produce)?
 
Interesting. It looks like one of its biggest selling points is the speed at which it produces forecasts, too. It is supposedly much quicker than NWP methods.
So the AI just "learned" what we already knew? Heavily weight your forecasts on the Euro! It's verification is right along the same line as the Euro. Does the AI ingest all of the weather forum data and then choose the available model output that is heavily favored by that polling?
 
So the AI just "learned" what we already knew? Heavily weight your forecasts on the Euro! It's verification is right along the same line as the Euro. Does the AI ingest all of the weather forum data and then choose the available model output that is heavily favored by that polling?
The AI modeling is running on desktop computers in minutes with comparable verification to the Euro already.

Considering it's still "learning", it's a huge deal just by sheer accessibility and ability for a collective of average humanity to make it better based on data the modeling will be trained on.

 
The AI modeling is running on desktop computers in minutes with comparable verification to the Euro already.

Considering it's still "learning", it's a huge deal just by sheer accessibility and ability for a collective of average humanity to make it better based on data the modeling will be trained on.

So basically we can clone the repository and just run the code/model on desktop?
 
So basically we can clone the repository and just run the code/model on desktop?

I haven't looked into all that. I'm not sure if it outputs graphics or raw, netcdf etc. You'll need the separate training files too.
 
Status
Not open for further replies.
Back
Top