You know... I genuinely don't get why this post is getting downvoted in a technology community. Wtf is up with lemmy's hatred of anything AI related? This isn't even related to the LLM or image generation stuff. This is WEATHER FORECASTING. The cool part is the resources it requires to forecast weather correctly. Compared to supercomuters, it can forecast stuff while running on relatively low end hardware within a minute's time. That's the cool part you jackasses. But no, lemmings are like "AI bad, capitalism something something..."
My only problem is that his videos have become increasingly clickbaity over the past couple of years. AI Explained does a much better job.
You ain't wrong, but also, I think a lot of people here just don't want this kind of information in the form of a YouTube video. It would probably get a better response if the link was to the GitHub page, or if there were a meaningful text summary in the main post.
Props to the YouTube channel for including good links in the video description, though.
GitHub: https://github.com/google-deepmind/graphcast
Full paper: https://arxiv.org/abs/2212.12794
interesting, I already had read, that lambda was used and was at least partially more effective than classical models. Question is how it will fare when the data it was trained on becomes less relevant as weather patterns change in the future due to climate change.
It is to be seen experimentally I guess. But from what I understood, I don't really think this would be that big of an issue due to two factors:
- Climate change has been quite predictable in the past few decades. Therefore, assuming that it is trained on data from the past few decades, wouldn't it know how to anticipate it?
- In the beginning, it is kinda seen to be doing cloth sims which use physics. Here, it seems to know which calculations to prioritize and which to approximate. This results in an approximation of a physics sim while using a fraction of the resources. The most computationally expensive part of the weather models seem to be the fluid sims. The AI here seems to be able to essentially come up with a very efficient fluid sim model. I can't see how climate change can affect the physics of fluid sims and stuff. Only factors like temperature distribution and stuff would change, no?
Again, I don't really know as I'm not a meteorologist nor an expert in how this thing works (although I would like to be hehe).
Horrible video. God awful voice.
TwoMinutePapers is a well-established YouTube channel that does a great job of explaining new scientific techniques/advancements (usually in the areas of computer science and graphics). Do you actually have anything to say about the content of the video?
I can understand why some viewers would have a negative reaction: it’s like listening to an overexcited sports commentator paraphrase the paper’s abstract and speculate about its impact, without really trying to explain the content. The value for those who follow the channel would presumably come from its discrimination in selecting significant papers and accuracy in summarizing them, but that’s not apparent to a first-time viewer.
Are you a troll or genuinely just an absolutely miserable person? Since when did Lemmy's comments turn like that of Instagram?
You're right. we shouldn't resort to ad hominems here.
If that video is yours, i must admit though, that you might want to work on your cadence and inflection a bit, otherwise it is difficult to listen for the whole video. The video itself was interesting, I already had read, that lambda was used and was at least partially more effective than classical models. Question is how it will fare when the data it was trained on becomes less relevant as weather patterns change in the future due to climate change.
No need to get personal here
