Growth without the mindset

Learning when you’re forced to

I basically have a fixed mindset about my own aptitudes and personality traits. To some extent, I was raised with it: the many Telugu aphorisms my dad would repeat included “You can’t straighten a dog’s tail” and “The fox was jealous of the tiger so it branded stripes onto itself.”¹ At Open Phil, where I grew up a second time, they did not tell you about your “growth areas” — they told you about your goddamn weaknesses.

Mostly though, I was born with it.² The can-do positivity associated with “growth mindset” is fundamentally pretty alien to me, and I can’t help but feel like it’s a bit twee or cringe. Flattering my biases, the dramatic pop psychology claims about the benefits of growth mindset mostly crumbled in the replication crisis. But I do think there’s something to it.

By growth mindset, I mean a tendency to consciously and deliberately invest effort into addressing weaknesses, practicing skills, and learning new things. Some people are really into this. They’ll make Anki decks to memorize facts about AI trends, or hire a coach to improve their management skills, or work through exercises in physics textbooks.

I am not like that. I can’t remember the last time I read a nonfiction book. I have basically no inclination to hire a coach or therapist, and it’s not for a lack of issues. I pretty much only learn stuff that I have to learn to do my job:³

• As part of writing an old report on when we might train a model as powerful as the human brain, I read up on the transformer architecture and empirical risk minimization theory and actor-critic policy gradients and efficient batch sizes and a number of other technical topics to understand what I believed and how I should construct my arguments.⁴ I never studied that stuff before, and never have since. Other projects led me to brief forays into hyper-specific corners of thermodynamics, statistics, economics, and ethology.

• I spent years at Open Phil working on a range of nebulous research and writing projects like this, vaguely feeling like I should get experience with grantmaking at some point but never getting around to it. Eventually, I developed a bunch of takes about AI safety research directions, which led me to want to fund certain kinds of research, which led me to finally make my first grant. From there I quickly escalated to take on our technical safety portfolio for a couple years.

• To this day, I haven’t properly fired up an AI agent and vibe coded something. Sure, I use models to help me edit tricky emails and review contracts and do a bit of desk research here and there for my reports, but…well, my job is largely sending tricky emails and negotiating contracts and writing long carefully-worded reports; AI agents aren’t amazing at this. I have some dim sense in the back of my mind that whole new vistas of possibility should somehow be opening up, but I am never inspired to do anything.⁵

I do think I’m leaving value on the table because it’s so unnatural for me to adopt a growth mindset. I recently found myself investigating a misalignment incident, and my skillset meant that I worked almost entirely on writing. If at some point in the past I had sat down and decided to invest in developing my ML research skills, I would have been more useful in that project. I could have become decent at it if it was crucial for my job, but it never was so I never did.⁶

Still, you grow whether or not you have a growth mindset. I’ve certainly grown a lot in the last decade. Yes, I very rarely put deliberate practice into working on my weaknesses — in fact, I tried to select projects that would be loaded on my strengths and forgiving of my weaknesses. But this work was still challenging, and forced me to push even harder in the areas where I was already strong. As a result, I have become more and more extremely myself with each passing year.⁷ As an approach to work and life, this has a fair amount to recommend it.

1. At one point as a kid, I was sad that I wasn’t very good at music. My dad told me that when he was a kid he was sad that he couldn’t sing well and his big sister was like, “It’s okay, the world needs people to appreciate good singers too.” I soon gave up on music lessons and to this day am not particularly musical.

2. When I first learned about Calvinism in AP Euro my sophomore year of high school, I was entranced by how right it felt in my bones. God decides when you are born whether you go to Heaven or Hell, and nothing you can do will ever change that. But those predestined for Heaven work piously their whole life anyway, because that is simply who they are.

3. When my job was to be a student, I was a great student — I was genuinely moved by the beauty of the subject matter and formed a deep understanding of it. When I got a real job, I pretty much never thought about any of that again.

4. I now think that I was overly thorough in how I approached that project, and if I were doing it again I would find elegant ways to sidestep the reasons I felt like I needed to nail down all the stuff in order to write it.

5. Just like I never set up the elaborate cool-kid keyboard shortcuts that make typing more efficient.

6. At one point when I was burned out I made the uncharacteristic move of attending an ML / AI safety boot camp on the vague hope that I could transition to more technical work, and took away very little from it.

7. This can lead you to double- and triple-down on very niche superpowers. For example, one of my main skills is talking to very smart and slightly crazy guys with intriguing and very big-if-true ideas, and translating the actually good ideas into impact by explaining them clearly. There are many sub-skills to this skill, including the taste for who has good things to say, the conceptual vocabulary to understand what the hell they’re talking about, a nose for when something won’t quite hold up, the social presence and courage to push on it, the taste for how to modify it into something that is a bit more right, the particular writing skill needed for producing this niche type of artifact, etc. You can’t learn how to do this in any systematic way because it’s a crazy and incredibly niche job.