Noam Shazeer: belief

Transcript context

…One thing that I like to do is to find out about a new and interesting area, and one of the best ways to do that is to pay attention to what's going on, talk to colleagues, pay attention to research papers that are being published, and look at the kind of research landscape as it's evolving. Be willing to say, "Oh, chip design. I wonder if we could use reinforcement learning for some aspect of that." Be able to dive into a new area, work with people who know a lot about a different domain or AI for healthcare or something. I've done a bit of working with clinicians about what are the real problems, how could AI help? It wouldn't be that useful for this thing, but it would be super useful for this. Getting those insights, and often working with a set of five or six colleagues who have different expertise than you do. It enables you to collectively do something that none of you could do individually. Then some of their expertise rubs off on you and some of your expertise rubs off on them, and now you have this bigger set of tools in your tool belt as an engineering researcher to go tackle the next thing. I think that's one of the beauties of continuing to learn on the job. It's something I treasure. I really enjoy diving into new things and seeing what we can do. I'd say probably a big thing is humility, like I’d say I’m the most humble. But seriously, to say what I just did is nothing compared to what I can do or what can be done. And to be able to drop an idea as soon as you see something better, like you or somebody with some better idea, and you see how maybe what you're thinking about, what they're thinking about or something totally different can conceivably work better. I think there is a drive in some sense to say, "Hey, the thing I just invented is awesome, give me more chips. " Particularly if there's a lot of top-down resource assignment. But I think we also need to incentivize people to say, "Hey, this thing I am doing is not working at all. Let me just drop it completely and try something else." Which I think Google Brain did quite well. We had the very kind of bottoms-up UBI kind of chip allocation. You had a UBI?…

Stored transcript either side of the excerpt. The highlighted words are the published quote; the surrounding text is unedited source, never generated.