people think that's sufficient. Then there were a series of breakthroughs, something called reinforcement learning, of which DeepMind is one of the most successful. We use an example in the book around playing Go and the Game of Chess, which are using reinforcement learning. There, what happens is the computer uses a clever algorithm to look at an infinite number of future choices and figures out which one is the best. My own opinion is that there will be another breakthrough or two before we get to AGI. And I think Dan agrees on that. And the problem is that you probably can get to fantastic pattern matching, fantastic text generation, fantastic human-like insights in text and so forth. But as Dan likes to say, it's all about the objective function at the moment. And how would you do an objective function which consists of go think about interesting things and tell me what you want to think about, which is what you can do with a human. We're probably a breakthrough or through a way (12/45)
You are viewing a single comment's thread from: