Reddit is known for many things: lively communities, a dedicated user base, cum boxes, incest. Now, the Elon Musk-and-Peter Thiel-backed nonprofit OpenAI wants to use Reddit’s vast array of content as a guide for its new machine learning programs.
MIT Technology Review reports that OpenAI has partnered with NVIDIA to use the latter company’s new DGX-1 supercomputer to train its deep learning systems both more rapidly and with more data. One way they’re going about that, apparently, is by using Reddit, so cross your fingers that the robots don’t start spouting abuse and garbage!
“One very easy way of always getting our models to work better is to just scale the amount of compute,” OpenAI research scientist Andrej Karpathy said in a press release. “So right now, if we’re training on, say, a month of conversations on Reddit, we can, instead, train on entire years of conversations of people talking to each other on all of Reddit.”
“Eventually,” he added, “we’ll use that to talk to computers, just like we talk to each other.”
MIT Tech Review notes that this kind of research is still in its infancy, but that using Reddit, with its almost infinite source material, is one way to go about training the machine learning programs. Besides language, OpenAI is also reportedly looking into “training robots do the dishes through experimentation” and “building algorithms capable of learning to play a wide variety of different computer games.”
But back to Reddit. In a way, using the social platform makes sense: If it’s a conversational tone the researchers are after, Reddit has that in spades. The communities are also wide-ranging enough that there’s a wealth of diverse source material, though that admittedly can’t be said for the users themselves. But Reddit has a tendency to capture the weird, the spontaneous, and the organic, in a way that, say, Facebook doesn’t. (Then again, it can also be a complete and utter cesspool, so there’s also that.)
Will the robots adopt Godwin’s law? Will they randomly blurt out “u wot m8” and “moms spaghetti” mid-conversation? Will they go after the wrong terror suspect? Will they take after Microsoft’s racist AI chatbot and turn into Hitler clones? Only time will tell.