[0:00:20] So I think my definition of AI is the same generally and related to work. And I guess I think my definition of intelligence is basically a functional definition. So intelligence is the ability of an agent to achieve its goals efficiently in a wide variety of circumstances. So a thermostat interacts with its environment.
[0:00:49] But it's not intelligent. If you take a Google supercomputing cluster, it might be far more capable than a human being at answering search queries. But if you drop it off in the middle of the woods and ask it to get power, it's not going to succeed. Whereas a human, basically, if you give them any sort of goal, they'll be able to figure out a way to efficiently achieve that goal. So that's intelligence. And then artificial is just not from.
[0:01:19] Biological evolution. Yeah, so I think basically supervised learning is solved. So anything where you can collect maybe 100x more hours of experience doing a task than a human would, than you would expect a human to need to become an expert in that task. If you can get 100 times more data, like labeled data.
[0:01:51] Then you can probably train an agent that will, or you can train a system that can complete that task to the same quality level. So I think that that right now is where AI is already successful. And so if you imagine like a self-driving car, we can collect, a human might drive for a thousand hours before being an expert. We can collect a hundred thousand hours of labeled data. And so driving will be solved.
[0:02:21] I think that sometimes it's very expensive to do that. So if you imagine trying to collect that amount of data and labeling it for engineering a rocket or something like that, it's very expensive and hard to do that. So you wouldn't be able to. So I think figuring out ways to get systems to transfer their knowledge from something to a new domain.
[0:02:50] is a big unsolved problem that I think will push the field forward a lot. I basically think that we'll be more successful. I think we're going to make human level AI within our lifetimes. And I think it'll be the most important invention humans ever make. So I think that's important. That keeps me up at night. I think one way that we can make sure that
[0:03:26] When AI does happen, it's just a good thing for everybody is to make sure that we have a global research community that is unified and well-informed. So doing things that help make there be a global research community and help make that global research community capable.
[0:03:53] Well-informed and unified are things that I care about. My role is mostly building tools that help those three things happen. We are building tools just to make people more capable and making tools that help researchers collaborate more effectively on the most important problems in the field. I am excited just to see all the different things that people are building out.
[0:04:28] I think that it'll be fun to actually like get my hands dirty just like pulling out a laptop and doing some coding with people too. So that's what I'm most excited for.