In its growing AI war with Google, Facebook is open sourcing the specialised hardware it uses for Deep Learning AI, known as Big Sur, which is now twice as capable as its last setup.
Deep Learning essentially forces oodles of data through several “layers” of an AI engine, each running on a silly amount of graphics cards, to better understand what that information is. Feed it pictures of a horse and tell it when it’s right and wrong about what a horse it, and eventually it’ll be quite good at seeing horses. Deliberately feed it the wrong images after that, and it’ll be attempting to see horses everywhere, which is how we end up with things like this.
As Wired points out, it’s a crazy software world we live in, because for a few different reasons, both Facebook and Google are giving away their secrets in order to better compete. Deep Learning depends heavily on massive amounts of data being fed into it. The more it gets, the better it is all around. So even though Facebook and Google have access to massive amounts of data, it’s all the more massive if they become open about it. On top of that, the more people using it, the cheaper the hardware will become, and the more the code will be improved by interested parties.
Even though Deep Learning runs on GPUs and was born of image detection, it helps both Facebook and Google across the board. In addition to Facebook’s creepily accurate facial detection, it’ll get to know the way you communicate, and the stories you want to see. Google’s base function, search, will benefit greatly from it, as well as its targetted ads and maps software. That’s just the tip of the iceberg.
Both companies have a history of open sourcing, though a little differently. Google has been a little slower on the uptake, though it did recently open source its TensorFlow system recently. It’s a big step, though only part of the AI engine was made public, and details on the hardware it used to run it were kept secret. Facebook, on the other hand, has a history of being open about its hardware details, as it is right now.
Since Deep Learning specialists are not exactly a dime a dozen, Facebook and Google are also competing for staff expertise, and open source is something that software (and now hardware) engineers care about. So, part of this is about attracting the right people as well.
Image via Shutterstock