 Manuel Orbegozo/ReutersMeta announced this morning it’s releasing a new AI model and data set aimed at helping scientists make new raw materials. If that doesn’t sound exciting, think of it this way: Of all the advances in technology, humanity’s survival is still largely dependent on things we’ve dug up from the ground, oil, minerals, plants, etc. All of those things have served us incredibly well for thousands of years, but they may not be enough to get us through the next thousand. We need to find new materials that can make better and cheaper batteries, more efficient solar panels, and faster computers. Until recently, the way scientists have looked for these new materials is by essentially running lab experiments based on educated guesses. AI researchers at Meta hope they can speed up that process by providing a massive new data set that can be used to train AI models, as well as a new AI model of its own, that can predict new inorganic materials, significantly cutting research time. To get the data, Meta used something called Density Functional Theory, which is basically a way to express the essential makeup of atoms on paper, in computational form. Running a Density Functional Theory calculation takes a lot of compute power, so Meta used its own servers when CPUs weren’t being used by Facebook and Instagram members. “When the sun was over the Pacific Ocean and we didn’t have to surface as many cat videos, we just ran atomic simulations,” said Larry Zitnick, a research scientist at Meta’s Fundamental AI Research (FAIR) team. The team ended up with more than 100 million of these calculations, releasing them under the name Open Materials 2024. An AI model trained on the data set can then make guesses about possible new atomic structures for materials without doing all of the complex calculations. Meta’s hope is that the data will lead to new materials that will help solve problems like climate change, but also potentially help Meta. The lenses of its prototype augmented reality glasses called Orion, for instance, are made of silicon carbide wafers, an advanced semiconductor material that allows Meta to push the envelope on “field of view,” creating an image that appears to be much larger than previous attempts at the technology. But it’s extremely expensive to make. It’s possible that another material that doesn’t yet exist could one day replace silicon carbide. “When it comes to materials discovery, this is a huge, huge area that has potential to completely revolutionize how we will handle the energy crisis, how we will handle some of the current crunch for compute of AI models, and how we will potentially build some of the AR/VR devices in the future,” said Joelle Pineau, who leads the FAIR team at Meta. | 
