Google's DeepMind finds 2.2M crystal structures in materials science win | Hacker News Reader

Google's DeepMind finds 2.2M crystal structures in materials science win

arstechnica.com

51 pointsPaulHoule11 comments

"Finds" is far too generous. There was no experimental research done. It was structure prediction using AI.

Not a material science expert, however per their paper, they use DFT to verify the stability, then use the verification status to improve the model.

>candidate structures filtered using GNoME are evaluated using DFT calculations with standardized settings from the Materials Project. Resulting energies of relaxed structures not only verify the stability of crystal structures but are also incorporated into the iterative active-learning workflow as further training data and structures for candidate generation

There was a follow up paper that coupled this with an autonomous lab and was able to reproduce a lot of the structures.

Then, as science goes, an enlightening thread on Twitter about what Science they actually did: https://x.com/Robert_Palgrave/status/1730358675523424344?s=2...

Nitter'd:

https://nitter.net/Robert_Palgrave/status/173035867552342434...

Thanks.

As a self proclaimed expert on sort implementations, the work DeepMind published in that field was also deeply lacking. Essentially presenting tiny 50 year old sorting networks as novel result. And adding them into libc++ in a way that exclusively looks good in hot microbenchmarks. But degrades performance if your program does more than only spin a hot loop that sorts fixed size inputs of 3 or 4 elements.

I'm starting to see a pattern with DeepMind and the quality of their papers.

Enlightening AND highly critical overview. Read this critique before getting too excited.

Should be the response to this tweet:

https://www.linkedin.com/pulse/regarding-our-recent-a-lab-ar...

Paper: https://www.nature.com/articles/s41586-023-06735-9

aaand they’re all under patent now.

Reply on news.ycombinator.com