https://www.dwarkesh.com/p/sholto-douglas-trenton-bricken

i am mainly interested in how they got hired

It means that since you're getting up to speed on everything at the same time, rather than spending grad school going deep in one specific way of doing RL, you can actually take the global view and aren't totally bought in on one thing.

So not only is it something that's possible, but it has greater returns potentially than just hiring somebody at a grad school. Just like getting a GPT-8 and fine-tuning the model for one year.

Sholto Douglas

You come at everything with fresh eyes and you don't come in locked to any particular field. Now one caveat to that is that before, during my self-experimentation, I was reading everything I could. I was obsessively reading papers every night. Funnily enough, I read much less widely now that my day is occupied by working on things. And in some respect, I had this very broad perspective whereas in a PhD program, you'll just focus on a particular area. If you just read all the NLP work and all the computer vision work and like all the robotics work, you see all these patterns that start to emerge across subfields, in a way that foreshadowed some of the work that I would later do.

Dwarkesh Patel 01:50:26

That's super interesting. One of the reasons that you've been able to be agentic within Google is you're [pair programming](https://en.wikipedia.org/wiki/Pair_programming#:~:text=Pair programming is a software,as it is typed in.) half the days, or most of the days, with Sergey Brin, right? So it's really interesting that there's a person who's willing to just push ahead on this LLM stuff and get rid of the local blockers in place.

<aside> 💡

WHAT IN THE ACTUAL HELL??

</aside>

Sholto Douglas 01:50:46

It’s important to say it’s not like everyday or anything. There are particular projects that he's interested in, and then we'll work together on those.

But there's also been times when he's been focused on projects with other people.

But in general, yes, there's a surprising alpha to being one of the people who actually goes down to the office every day.

It shouldn't be, but that is surprisingly impactful. As a result, I've benefited a lot from basically being close friends with people in leadership who care, and from being able to really argue convincingly about why we should do X as opposed to Y, and having that vector. Google is a big organization and having those vectors helps a little bit. But also it's the kind of thing you don't want to ever abuse. You want to make the argument through the right channels and only sometimes do you need to.