Joshua Achiam, AI researcher and Chief Futurist at OpenAI at the time of recording, on curiosity, craftsmanship, and finding your own voice.
Joshua Achiam’s account of learning is full of curiosity, difficult books, generous teachers, and projects that forced him to turn ideas into something real. In conversation with Siya Purohit, he describes a standard of excellence built through practice, collaboration, and attention to the questions a piece of work leaves unanswered.
For Josh, that standard also shapes how he uses AI. He starts with a detailed explanation of what he wants to do and why. He pushes vague recommendations toward concrete answers, asks the system to work through real constraints, and checks the original sources. AI helps inform his research and thinking; when he writes something for others to read, he wants the words to be his own.
The conversation explores the teachers and peers who encouraged his curiosity, the discipline of building and testing ideas, and the patient listening that makes collaboration work. It closes with a case for continued learning: developing enough depth to judge the work and enough breadth to keep growing as our roles change.
The Interview
Siya: So, Josh, growing up, what were you like as a student?
Josh: So when I was a little kid, I was not particularly attentive. I wasn’t a great student. I would run around. I was kind of a little unfocused. In elementary school, I would, there’s a story that my mom loves to tell where the superintendent came to visit my school when I was in like third grade. And when the superintendent and the principal came into my classroom, I was ignoring their presentation completely so I could just read Captain Underpants and you know that’s like so when I was little that was the kind of student that I was. I got a bit more serious about things when I was in high school and that was when I really started taking academics quite seriously and thinking like okay this is important I’ve got to prep for college and all that. Is there
Siya: a teacher that you still think about who may have like seen something in you or influenced you in some way?
Josh: Quite a few of them actually. There was a teacher who in middle school English class recognized that even though I was one of the much quieter kids in the class and I didn’t kind of naturally like pop socially the way that other kids did that I was thinking of interesting things and he wanted to encourage that in me and he like really took time and helped engage me to think about creative writing to think about analytical work that I could do and gave me a lot of confidence to go forward. So, you know, Mr. Gordon, if you’re ever watching this, thank you. And in high school, I had a couple of teachers who really nurtured my interest in science and who helped me develop mathematically. One of them took a chance on having me in an AP physics class, even though I hadn’t yet taken AP calculus. And he just kind of gambled that I would figure out the math as I went. And he was right. And in the first week of his class, he did a review of all the calculus material for everyone. And that was how I learned calculus. It was great.
Siya: Wow. Calculus is a hard subject. So how do you think you learned it in a week?
Josh: I think he broke things down in this really nice, systematic, thoughtful way that motivated everything kind of intuitively. Part of mathematical instruction sometimes is intimidating or difficult to approach because people throw a lot at you at once and don’t really tell you what the important part is and don’t necessarily get to the intuition. But he was a teacher who really knew how to show you where the intuitive part of mathematics was. And I appreciated that a lot. It made it really, really like clear and simple and straightforward, which is not usually what you hear about for math, but that he was able to do it. And one other teacher who I just, I have to kind of, while I’m on a camera and talking but I’m thank. I had a teacher for Latin classes for four years in a row when I was in middle school and high school who was close to retirement, Mr. Goufar, and he was kind of like a... he was like a little curmudgeonly but in sort of the delightful and loving way. He felt like a class grandpa and he nurtured all of our interests in the classics by making it feel very accessible. I mean, there are not very many students who otherwise go into technical fields who come away with like an appreciation for the Aeneid, but I totally did. And it was because I had a teacher like that who was warm and who made it accessible and made us feel like we could kind of have it become a part of ourselves in the process of studying it. And it was just great.
Siya: That’s beautifully put. You said you became more serious in high school. What prompted that? I
Josh: think my parents cared a lot about, you know, me and my sister being academically successful, successful in life, and they really encouraged us to take it seriously because we kind of had this implicit cultural pressure. Going to college is important. Getting a degree is important. Going to the best college you possibly can go to is important. And they weren’t super hard on us about it or anything like that. They didn’t really like push us in a in like a bad way, but it was it was like encouraged to take it seriously and I think at that point I also started getting like a little more competitive. I wanted to do well. There was a lesson I got from my dad which was super important though. He would always tell us don’t compete against other people, compete against yourself. And I really took that to heart. I wanted to I wanted to constantly like outperform myself. I wanted to do better than I could before and the material also got more interesting. So I wound up taking like a lot of AP classes in high school and then when there was material that wasn’t covered in in classes, one of the things I was also very lucky about was that I had a number of friends in high school who were super curious about advanced topics in mathematics and science that wouldn’t have come up in a high school class. And they were very autodidactic. They liked to go out and get their own books and study for themselves. And we had a nerdy group of friends that had jokes about advanced mathematical things. And so I felt very influenced by that. And I also wanted to go and get advanced math books and study them. And I did that. And I, oh gosh, the first one that I picked out, I went to Barnes & Noble and I was like, all right, let me find something kind of approachable. Let me find something that looks, all right, the book that’s about this thick and it’s got like a nice friendly cover, it can’t be that bad. It was vectors, tensors, and the basic equations of fluid mechanics. And it was, whew! I spent like months staring at just like the first three pages of this book trying to understand the notation. It was a challenge. But because I had this like extremely hard thing in front of me that I didn’t know how to interpret at all. I was like frustrated, like, okay, I have to do something that I can read this and get it. And
Siya: so I pushed myself to do a bit more. So this sounds like way beyond the curriculum of your high school. Yeah,
Josh: that was not a high school curriculum textbook.
Siya: That was... It was just your peer group that was encouraging you to think like that.
Josh: Yeah, it’s very helpful to have a peer group that is excited about learning and is excited about academics and studying interesting topics.
Siya: What do you think that school taught you well? and what do you think you had to learn outside of school? Like more in terms of the skills that you developed for life.
Josh: That’s a good question. I think my experience of school taught me that being able to engage in a really wide range of intellectual areas was going to lead to having deeper connections with people. Like the more that I learned, the better I was able to connect with peers and with teachers and with folks who had something really interesting to offer. I was pretty lucky. The type of school that I went to had this very kind of academic and intellectual environment, and it was really stimulating. I think I didn’t quite get disciplined, which maybe is like an odd thing to say, but part of pursuing something effectively in the real world is the discipline that comes with every day making progress on a goal and especially pushing to make something, to make something tangible, to make something real, to build something, and to build something in collaboration with others. And in school, I don’t know that... Sometimes the goals were so academic that this process of actually making something wasn’t really necessary, and so the discipline that you kind of have to have to make something wasn’t always required. And so I could spend my time studying really interesting topics or writing about really interesting topics or deriving something, but that was different from having to put together a completed project end-to-end that then I could test or physically build or run an experiment on. And so I wound up with a much more theoretical bent and not as much of an experimental bent. That I think kind of came later when I went to grad school eventually. Because in grad school you really have to put things together. You have to test your hypotheses. You have to prove that something works. You have to become very rigorous about the experiments. And I got quite lucky when I went to grad school that I had a number of mentors who really emphasized this kind of experimental and building side of things.
Siya: Was that a hard skill to pick up in grad school?
Josh: Totally. Because
Siya: you must have been, what, 22, 23 by then. So it was a very different way of approaching. How did you go about, like, changing your mindset to be more experimental?
Josh: So what I started to do in early grad school, I wanted to pursue studies in AI. And I had these kind of theoretical ideas of what to do with AI. In order to make them tangible and actually publish something on them and produce a project about them, I would have to get good at writing a lot of code and running a lot of experiments. And I didn’t have that skill set coming in. That wasn’t supernatural. So, I had programmed since I was in middle school in settings that were not rigorous and scientific and where you were trying to make something consequential, but the extent of programming and building a codebase from scratch I hadn’t really done. So I set about doing a lot of re-implementation projects where you take a thing that already exists and you just try to figure out how do I build a copy of this thing? How do I build a replica of this thing? How do I make it from the inside out and thereby learn the contours of how it works and also learn how to make other things down the road. So I re-implemented a number of deep learning and AI papers at the time and kind of struggled through the process of doing it and learning to write a moderately sized code base and maintain it and checking the experimental results and kind of bearing it out the hard way. I also got very lucky that the research group that I was in in grad school published certain types of starter kits where they made reference implementations for a number of different algorithms and they made logging tools that let you run experiments very, very easily. And from studying that work, I’m thinking in particular of a piece of software called RLLAB, which was published by Rocky Duan and Peter Chen and a couple of other folks circa 2016. I don’t know if it was an early open AI paper. It might have been an early OpenAI. It was kind of overlapped. Those people were half at OpenAI and half here in that time frame. But I studied this code base, and I tried implementing some ideas that I had in this code base, and that was how I started building a first research project that became a publication. So I think kind of finding your tribe that’s able to do the things and has the culture of knowledge and expertise that you want to learn is a really important part of education and kind of cultivating talent in the direction of, you know, being able to do those things.
Siya: I think one thing that really stands out in your story is how deep you go into these different topics. Sometimes, like, even if the topic seems so cumbersome, like they did in high school, possibly, what motivates you to actually go that deep?
Josh: The motivation usually comes from curiosity and stubbornness. I don’t know exactly how to put it, but there’s a feeling when you sink your teeth into a problem because you want to solve it, or you want to know what the solution is, and you just keep going at it over and over and over again. It fills your spare time thoughts. You wind up postponing other things because you’re more interested in that problem than you are in something else. And that intrinsic drive is kind of the important thing. If I were to think about, you know, how to cultivate talent in others, nurturing that intrinsic drive, finding where they have it for, where someone has it for something, and encouraging them over and over to pursue it, to spend the time doing the thing that they’re really excited about, that they’re really passionate about, that they can’t get out of their head, that will cause them to then actually spend the time, do the thing, get the skill level up that comes with pushing themselves through it. And I think it leads to the outcome in the end.
Siya: Something a lot of educators ask us is about like, how do you even identify that in a student? Like, how do you identify like what they’re actually motivated by? But especially when they’re like using AI solutions to get answers so much more quickly than ever before. So they’re not going through that process of discovery as they used to in the past.
Josh: Yeah, I think that’s a good question. It depends a lot on what age range you’re talking about, right? Because figuring out where someone’s interests are and where someone’s passions are and when to encourage it and how to encourage it, it varies a lot by age. I think for younger kids, look for the thing where they can’t sit still and they’re distracted by something. And the thing that they’re distracted by may actually be the beginning of a superpower. Like it might not be bad behavior that you have to kind of push them away from. It might be something where you should figure out, all right, how do I get them to lean in on this? know, like they can’t stop reading Captain Underpants in class. Maybe they’re going to be a big reader. Maybe I should just like put, you know, more and more interesting books in front of them. And even if they haven’t figured out exactly how to, you know, navigate the social niceties or doing the thing that they have to do in addition to that, pushing them more in that direction will give them a big unlock later in life. for folks who are maybe more at the high school or college level, there are things that you see people have in their individual personalities. The part of them that’s distinct from the people around them, the thing that they’re interested in, even if their peers kind of are all interested in something else, or the thing that they kind of, that maybe they like get shy about because it’s sort of too important to them to want to talk about. And I think looking for those kinds of unique features of those people and giving them examples of things that they could do in those spaces where it would excite them to kind of do it. Like, that’s the thing to do. Like, there’s no getting away from the fact that young folks are going to use AI to do half of their schoolwork or half of their projects. If you give them busy work to do, they’re going to short circuit the busy work. There’s something else they really want to be doing. and you kind of have to find that other thing because they’re not going to go and use the AI for that unless the AI is actually an important piece of doing that thing. If AI is helping them build software and the outcome that they really care about is having built an important piece of software that does something exciting for them, that’s great. Help them do it. Push them a little more in that direction. Otherwise, look for the thing that they’re not using the AI for that they don’t think of as busy work, that they think of as the fulfilling thing and help them double down on that.
Siya: Amazing. Let’s talk about OpenAI. So you once said that you kind of like grew up at OpenAI, given how early you joined and like how many different roles you’ve had at the company. So we’d love to hear, like, when do you think your standards for work became much higher at OpenAI? It’s
Josh: a good question. For what it’s worth, I think OpenAI is a place where everyone comes in with a relatively high standard of work. It’s a place that attracts top talent and it’s a place that prides itself on getting to the frontier. So I think even when I came in, my standard for work was quite high. Maybe the types of things that changed over time as I grew up in here were my expectations and norms and standards inside of collaboration and what I expected good collaboration to look like and good leadership to look like. Because those are qualities that were not really things I had to hone or learn in the type of schooling that I was doing before being at OpenAI. I came to really appreciate when folks would be able to take their ego out of a hard question and a hard strategy question or a hard question about what research direction was going to be productive and in like a low temperature methodical thoughtful way elicit the pros and cons from everyone around them, just elicit the different points of view and listening very effectively. And I came in as almost like a brash hothead in some respects. Like I was good at doing technical research, but I was really annoying when I would try to like impose my own opinions on different things. And what I learned from watching, you know, who was effective and who was less effective was that the people who were patient and thoughtful and really able to take themselves out of it for a little while to just listen and see where a consensus might exist and then nurture that kernel of consensus between others and help it grow, that that was very successful. And so that kind of gradually became my standard for collaboration and leadership within the company.
Siya: When you see a piece of work, what makes you think it’s, like, truly exceptional versus good enough?
Josh: I love this question. I was thinking about it recently. one quality that I look for that tells me that this is really like a work of great craftsmanship that this was something that was put together by someone who cared a lot and was extremely talented is if for all of the questions that I naturally have as I go through the work do I find that the author understood the question internalized it has already prepared some kind of an answer or response? Do I find that they’ve covered all the bases for the things that one would have to do? Do I find that they’ve covered all the bases for the natural set of objections that one might have to the thesis of the work or the basic thing that you would do in here? Have they prepared an answer for why they made a trade-off a certain way? Great craftsmanship involves trade-offs. There are trade-offs everywhere. You have to decide between quality, speed. In software, have to decide latency or what does the user really need. In policy, you have to have trade-offs about what do different stakeholders need. And just when you see a work where the author of the work or the authors of the work have predicted in advance all of the objections to the trade-offs that they’ve made, and they have a defensible stance on why that trade-off was chosen. and it reflects something deep in the craft, either about the technical side of things or about the cultural or societal implications of the things. That’s how you know that it’s like a genuinely exceptional thing, this detailed account of the trade-offs and the depth of the craft. What
Siya: I’m hearing is basically someone who’s very comprehensive in their thought process. Yeah. And that is like, I spend a lot of time with like Gen Z and it seems like that’s a skill that they’re really struggling to develop because they have such strong points of view sometimes on the things that matter to them that they almost fail to look at the other sides of it. And that’s something educators keep bringing up to us too. How does one develop comprehensiveness of thought like that or like me see as consultancy?
Josh: Oh, that’s a great question. So it’s hard for me to say one way or another what Gen Z really is like or thinks because everyone in a cohort is so unique and individual and different from each other. And I haven’t spent enough time around the Gen Z kids to know exactly what their cohort behaviors are. And I always want to be a little bit cautious and humble in how I judge the younger cohort because it’s very easy for me as now a 30-something to want to be like an old complainer and say, like, ah, you know, kids these days. But I’ve also heard these kinds of things, and I’ve seen some of it. And my impression is that there are maybe some cultural points of view right now that are common in that generation. And there’s like a lot of social pressure to adopt this point of view or that point of view. I think something about the prevalence of social media, Instagram, TikTok, etc., made it much easier to apply social pressure among young people. And if we can encourage people, if we can encourage these young people to not pay so much attention to what everyone else on social media wants them to think, to not feel like, you seeing someone else having that thing means they’ve got to have it to to not worry about where they are in life relative to other people who are painting kind of a sort of false picture of how good their lives are and not feel obliged or pressured to take a stance on something before they themselves have deeply studied it, then maybe we can make some progress. And I think that’s got to come from parents. I think that’s got to come from educators. I think that’s got to come from trusted people in kids’ lives. And, you know, it’s kind of difficult. It really takes a village to help get people out of a place of receiving points of view and willing to start considering them from scratch. You know, you often hear or see complaints these days that universities are not doing enough to encourage diversity of thought. And it’s possible that there is some truth to this. I think we I think we can... Oh gosh, you know, there’s another thing that comes to mind thinking about universities. I was reading recently about universities that are considering analysis of student personalities as a part of the admission process and getting peers to maybe like rate each other to figure out, you know, does this person, you know, win the approval of their peers in these respects as maybe like a component of a comprehensive evaluation. and I hope that the way that we evaluate students, the way we evaluate their personal essays, can update a little bit so we’re not incentivizing them to homogenize too much, right? Because once you have an evaluation process that you’re telling students is an important gate for them to make progress in life, they start trying to figure out how do I, you know, succeed in that evaluation process. And in AI, you we think about that as the AI trying to figure out how to game the benchmark. Well, students try to do that too, you know. We tell them, like, you’ve got to write this personal essay that must make you look like either a saint or someone who comes from a very specific type of background and students are very responsive to that. They feel like, okay, if I want to get a degree in life and get to participate in the economy and have economic opportunity, I have to play this game. We should stop making kids play games like this. We should stop making kids feel like their life outcome is dependent on whether they conform to a very specific standard from early on. And that might be another piece of how we help encourage some of the you know diversity
Siya: of thinking. Because when I think back to college the most fun parts were like the funky people you met. Yeah, yeah, weirdos. than what you had grown up with. Totally. And it like really expands your way of thinking but also like how you socialize with them and how you learn how to talk to people. Yep. That really matters. What separates someone who is very smart from someone whose work you deeply respect?
Josh: Having done the work. And the subtlety here is that there are lots of people who are incredibly smart who are capable of extraordinary work. Not everyone winds up doing the best work they can possibly do. And there are lots of reasons that they don’t necessarily do it. One might be that they talk themselves out of it. Another might be that they don’t wind up in the right place to foster their talent and their creativity. I think everyone should try to go to work in a place that’s going to encourage them to do their best work and where it’s easy or possible for them to do it. Sometimes people get discouraged by something going wrong and then they wait too long and they don’t go after doing the thing. But the real differentiator between people who are just possibly capable and people who have accomplished something that I look at and I respect the work quite a lot is the process of doing the work. The point where they sit down and say, I’m going to carve out time, I’m going to find a specific type of job, I’m going to find a specific type of role, I’m going to put myself in the place where I can do the best work of my life and then I’m going to just grind really hard until it’s done. That I think is the thing. And something I like to believe is that almost everyone I meet is unique in some way and It has the capacity to be exceptional in some way. Whether that’s in a local way, they can be exceptional in loving and taking care of the people in their lives, or in a global way, they can be exceptional by solving some frontier problem in science and technology that winds up having a lot of downstream impacts. I think everyone has some unique capacity to be exceptional, but it’s the thing where they resolve to themselves that they want to do it and then choose consciously day in, day out to do it that makes the difference. When
Siya: I talk to students I talk about how important reps are basically have to go through the motions over and over to be able to master anything whether it’s talking to a recruiter or getting even a document out. Something that like I get a lot of pushback on from campuses is that like students are like okay I used AI am I allowed to use in the classroom and I submitted the first like submission I got back from AI and like the entire conversation we’re having is like no you need to push AI to Think more, like, think bigger. Think like, I mean, capture your own voice within that also. And that’s also like going through different kinds of reps, right?
Josh: Totally.
Siya: When you use AI, like, how often do you push back? Like, what is your thought process on, like, getting really good output from AI?
Josh: Oh, that’s a very good question. Yeah, I do. I tend to push back a lot. When I am working on something where I know that the AI is going to be helpful as a piece of the loop, I put a lot of effort into trying to specify what it is that I’m trying to do. So I’ll write very, very long multi-paragraph prompts to explain what I’m going for and why I’m going for it and what the motivation is. And then the AI will generally be doing some kind of research on my behalf or making recommendations in some set of areas. And within the recommendations that it makes, I’ll push back and I’ll say I want to drill in further on that. I don’t just want this very high level vague description of something that could be, I want you to make this progressively more tangible and refined. I want you to grapple with the hard constraints in the reality of this thing. And I want to learn about those hard constraints as we’re going. So a focus on the depth and the detail, kind of the resolution of answer is, in my experience, very helpful for getting useful things out of AI. And I would encourage students to do the same.
Siya: And how do you make sure that AI sounds different for you than it does for everyone else?
Josh: You know, it is like a genuinely hard technical challenge to get the AI to have a different voice in different things. So for the type of writing that I put into the world, I don’t actually use it to produce that writing. I use it to inform my research process and my thinking and what things I know that can then be reflected in writing. Where I also really hold it to a very high bar for correctness and for citation to original sources because I want to make sure that the things that I go and say and advocate for are grounded in facts. But when I write, I try to write in my own voice. And I would encourage students to do the same thing, honestly. I think that learning to speak in your own voice is just such an important part of becoming an expert yourself or becoming the best and most capable version of yourself. So even if you can use AI, and maybe using it to do a whole lot of busy work is fine, but when you’re trying to say something that you think is important to say, get the AI to help you, but say it in your own voice in the end.
Siya: As AI makes more people capable of producing good work, what becomes more important?
Josh: There are a few things that I think will always be important. One, being able to listen to and engage productively with your colleagues, with your friends, with your family, with your community. Communication skills, respect, and collaboration skills are always, always going to be useful. No matter what the AI can do, we’re still going to be people who need to get along with other people. Two, I think... Coming back to depth, there are a lot of things that AI will be able to do and that we might be able to get without too much effort or too much thinking. But pushing the AI for depth and being able to yourself understand the domain in which you’re asking the AI to do work to a place where you will elicit from it better work, where you can push it, where you can be a part of composing its outputs productively. It requires people to have a relatively deep understanding and a relatively broad understanding of a lot of different things. So I think it becomes important for people to know more and study more maybe than they have and to have more theoretical knowledge, to have more ideas than they would have necessarily had to in the past. It becomes important for people to avoid over-specializing in a specific narrow area and to remain kind of capable of breadth and to maybe see themselves as lifelong learners, that there’s never a point where their education is done and they’ve figured out enough to do the job that they have to do. What their job is, what their role is, it’s always going to be changing and there will always be new opportunities, and they should embrace those and be willing to just learn a lot about a lot of fields so they can productively engage with the AI to address something that’s important either societally or produce something that they feel fulfilled by creatively.
