Umarbek

When I was building an early pizza ordering website, I put a Coke ad at the top as a joke because I thought it was so funny that there would be internet ads. Obviously, it turned out not to be that funny.

My pizza ordering website failed profoundly because of how it worked: you'd put in your order to the website, and since pizza places weren't online, it would automatically send them a fax with the order. Then I realized they don't actually check their faxes very often.

Larry Page is so ambitious that there's almost no plan you can suggest to him that he won't say is not ambitious enough. You need not just the solar system, but the galaxy.

We tried to license Google's technology to Excite. Vinod Khosla thought it was great and said they should buy it. We emailed saying we'd license it for 1.6 million dollars. Fifteen minutes later we got a reply saying "that's a lot of moolah" but okay. We were excited. Then our friend Scott came in laughing hysterically; he had faked the reply. Back then you could send an email from anybody.

When I was leaving my PhD program to start Google, my advisor Jeff Ullman said, "Why don't you just give it a try and if it doesn't work out, you can come back?" So I'm still technically on leave of absence from Stanford. Might still come back. We'll see how it goes.

We started a fairly academic-minded company. Larry and I both came out of a PhD program, unlike a lot of startups at the time that came out of college. I think that shifts how you think about things. The investment in foundational R&D was part of our culture quite early on.

We messed up with AI in that we underinvested and didn't take it seriously enough eight years ago when we published the transformer paper. We didn't invest in scaling the compute, and we were too scared to bring it to people because chatbots say dumb things. OpenAI ran with it instead, and good for them.

We were lucky to hire Jeff Dean, but we were in the mindset that deep technical things mattered. He was passionate about neural networks from his college days. He built up a whole effort. In my division at Google X, I let him do whatever he wanted. He'd say, "We can tell cats from dogs." I'd say, "Oh, okay, cool." But you trust your technical people.

Google developed its own AI chips, the TPUs, going back about 12 years. Initially we used GPUs and were probably among the earliest users. Then we tried FPGAs, then developed our own chips which have evolved through many generations. That trust in going after deep tech and getting more computation has paid off.

Today's AI systems are periodically dumb enough that you're always supervising them anyway. But occasionally they're brilliant and give you a great idea.

I wouldn't switch from computer science to comparative literature just because AI is good at coding. The AI is probably even better at comparative literature, just to be perfectly honest.

When AI writes code and makes a mistake, it can be pretty significant. But getting a sentence wrong in an essay about comparative literature doesn't have the same consequence. So it's actually easier for AI to do some of the creative things than the technical things.

If you carefully track progress in AI, you'll see that algorithmic improvements have actually outpaced the scaling of compute over the last decade. The algorithms matter more than the raw computing power.

Companies like ours will never turn down being at the frontier of compute. But compute is the dessert after your main course and vegetables of actually having done your algorithmic work.

When you have a cool new product idea, really fully bake it before you have a cool stunt to launch it. With Google Glass, I tried to commercialize it too quickly before we could make it cost-effective and polished enough for consumers.

Everybody thinks they're the next Steve Jobs. I've definitely made that mistake. But he was a pretty unique kind of guy.

There's a treadmill you get onto as a founder where outside expectations increase, expenses increase, and you're obligated to deliver by a certain time. You get this snowball of expectations and don't give yourself the time you need to properly develop your product.

Technically speaking, we don't even know if P is not equal to NP. On the computation front, there are so many unanswered questions. Quantum algorithms are specific to very particular structured problems. It's hard to know what quantum computing will ultimately bring.

The spotlight of attention moves around in technology. Right now it's very large on AI, but it was shining on biology and it shouldn't stop. There are exciting things happening in synthetic biology. We need to broaden that spotlight.

I retired about a month before COVID hit, and it was the worst decision. I had this vision of sitting in cafes and studying physics. That didn't work because there were no more cafes. I was stewing and felt myself spiraling, not staying sharp. I realized I had to get back to work.

Being able to have a technical creative outlet is very rewarding. If I had stayed retired, I think that would have been a big mistake.

AI empowers individuals because generally speaking, you don't have experts in various fields around you all the time. As a non-expert, I can pull out my phone, talk to an AI, and get an 80 to 90 percent decent overview of almost any topic.

I use AI constantly now, whether choosing a gift for friends or family, brainstorming new product ideas, or thinking about art. It doesn't do it for me; I typically ask for five ideas, and probably three will be junk in some way I can tell. But two will have some grain of brilliance or put things in perspective that I can refine.

Is there a ceiling to intelligence? We've had hundreds of thousands of years of human evolution with millions of primates, but that's a pretty slow process compared to what's going on with AI. We just don't know how smart a thing can be.

If you skip the news in AI for a month, you're way behind. The rate of innovation is absolutely amazing and hugely competitive between the top US companies and the top Chinese companies.

I don't know that for the coming century the idea of a school of engineering or a university is going to mean the same thing as it used to. Information spreads quickly now. Anyone can learn online, talk to an AI, or watch lectures. The geographically concentrated model may not be the format for the next hundred years.

At some level, if you have a hundred people together, it's kind of fine. They don't have to be at the same place as another hundred people. Increasingly, I see individuals who create new things regardless of degree. We've hired tons of people who don't have bachelor's degrees. They just figure things out on their own in some weird corner.

When I was a grad student at Stanford, the time from a new idea to it being commercially valuable was many decades. If that timeline compresses dramatically, it no longer makes as much sense to let ideas marinate in academia. The question is whether that long incubation period still matters when industry can scale new ideas so quickly.

I was born in Soviet Moscow in a 400 square foot apartment with my parents and grandmother. We were very poor. My father went to a conference in Poland where they told him what the western world was like, and he decided to move us. It was very controversial in the family at the time.

Moving to America meant learning a new language and making all new friends. Those transitions seemed painful at the time but later paid off. I had the experience of expanding my world in ways that seemed very painful but eventually created opportunity. Challenging transitions can pay off.

Something about California was very freeing and liberating in thought, given the tradition of the state. One that we're getting away from in California, if I'm being honest.

I talk to Gemini in the car often. I'll ask things like: I want to develop a data center, I need how many hundreds of megawatts of this kind of power, how much it's going to cost. I just talk to it about stuff on my drive. I prefer to have an interactive discussion rather than just listening to podcasts.