I met some people who needed my help with something called deep learning. At that time, I didn't even know what deep learning was, but they asked us to create a domain-specific language so that all their algorithms could be easily expressed on our processors.
We created something called CUDA, which is essentially what SQL is for storage computing, but for neural network computing. We made a domain-specific language for them, kind of like the OpenGL of deep learning. They needed us to do this so they could express their mathematics. They didn't understand CUDA, but they understood deep learning, so we built something in the middle for them.
The main reason we did this was because these researchers had no money. One of the great skills of our company is being willing to do things even when the financial returns are completely non-existent, or maybe very far away.
We always ask ourselves: Is this worthy work? Does this move a field of science forward in a meaningful way? I've been talking about this idea since the very beginning. Our inspiration doesn't come from the size of a market, but from the importance of the work itself, because truly important work is often an early sign of a future market.
