https://arstechnica.com/ai/2023/11/googles-deepmind-finds-2-2m-crystal-structures-in-materials-science-win/
Trove of combos is >45 times larger than number unearthed in entire history of science.
Headline should read, DeepMind calculated 2.2 million hypothetically stable crystal structures with the possibility to advance everything from materials science to biology (emphasis mine)
There's a large difference between a hypothetical crystal structure and being able to produce that crystal structure in the real world. And an even larger difference between the hypothetical and the ability to produce the crystal structure in an efficient and cost-effective enough manner to be usable. There are already multiple materials that are known that would make better batteries than lithium-ion batteries, but either the manufacturing method is too hard or expensive, or the materials are extremely rare. So it's nice DeepMind was able to do this, but it's years from being usable still.
Didn't they synthesize like 50 of them and had a success rate of like 70%? Continuing that process, they can continue to refine the model.
Success rate of 70%. So they didn’t try all 2.2 million yet, obviously. 70% of 2.2 million is 1.54 million btw.
So essentially the news is that a computer got baked and went “dude, what if there was a stable crystal that looked like this” 2.2 million times
This is insane. Right now I’m reading The Coming Wave by Mustafa Suleyman (formerly of DeepMind) and I’d been a bit concerned that he was all hype, but giving humanity a 45-fold increase in materials we know about? That’s enormous. The future is going to be a crazy place.