Rechenengpässe führen dazu, dass 94 % der Projekte in der US-amerikanischen Forschung und Entwicklung abgebrochen werden
https://www.azom.com/article.aspx?ArticleID=24821
Rechenengpässe führen dazu, dass 94 % der Projekte in der US-amerikanischen Forschung und Entwicklung abgebrochen werden
https://www.azom.com/article.aspx?ArticleID=24821
6 Kommentare
[https://matlantis.com/en/resources/useful/accelerating-discovery-ai-trends-in-materials-rd/](https://matlantis.com/en/resources/useful/accelerating-discovery-ai-trends-in-materials-rd/) – white paper this is based on!
So real issues that can be resolved by AI can’t get compute power because LLMs driving sex chats and search summaries are eating up the available compute power.
I don’t really see how this is different from any other research in the past. Teams of researchers are constrained by time, money, and available tools? Ya don’t say!
This is a clickbait headline that is not from a peer-reviewed publication. It reflects only the opinions of so-called „[simulation leaders](https://matlantis.com/en/resources/useful/accelerating-discovery-ai-trends-in-materials-rd/)“.
There are a few web3 companies whose focus is decentralized compute/storage and who are looking for customers.
I’ll accept that as true but it’s not new.
Even 30 years ago we had to shape models / simulations for efficiency and run-time. I myself had plenty of finite element runs that would have been better with more nodes for greater resolution but they’d take far too long to run (and thus more likely to fail). There’s a definite art in make efficient models, and that art takes ample time, and many mistakes, to learn.