Alternative Paths to Becoming a Research Scientist: Exploring the OpenAI Residency and Scholar Program
![avatar](https://cdn.gradual.com/images/https://d2xo500swnpgl1.cloudfront.net/uploads/openai/6632692-b87ecc21-dfb2-4676-8b19-774135047d39-1713292154770.jpeg?fit=scale-down&width=320)
Christina Kim is a Research Engineer at OpenAI. She has contributed to WebGPT, ChatGPT, ChatGPT with Browsing, and GPT-4. Before joining OpenAI, Christina was the founding engineer at Sourceress.
![avatar](https://cdn.gradual.com/images/https://d2xo500swnpgl1.cloudfront.net/uploads/openai/d60550f3-a3c9-4036-8075-9a9618ccdb20-17dfd4cf-95a3-4de1-89d9-c3244be69787-1713291856052.jpeg?fit=scale-down&width=320)
Brydon Eastman is an Applied Mathematician and Research Scientist at OpenAI. Before joining OpenAI, Brydon worked for a fintech company called MinervaAI and held a post-doc at the University of Waterloo where he taught differential equations courses and helped supervise and conduct research in mathematical oncology.
In this event, we delved into the crucial aspects of a successful career in AI research, emphasizing the importance of maintaining curiosity, proactively seeking feedback, and continuously enhancing both technical and interpersonal skills. Attendees learned how these elements could shape their journey into the AI field, helping them to adapt and thrive in a rapidly evolving environment. The interactive live Q&A session with Brydon and Christina offered a unique opportunity for community members to ask questions, gain insights, and receive personalized advice from professionals who had navigated non-traditional paths to success in AI. This engaging dialogue provided practical strategies and inspiration for anyone looking to break into or advance within the AI research community.
Welcome to another exciting event with the OpenAI Forum. I'm really excited about what we are going to be talking about today. We'll do some general housekeeping, and then I'll do an introduction of our speakers. And then we'll just have a conversation, and we'll invite and encourage you all to participate in that conversation. So that way you feel like you can just have a casual conversation about a non-traditional career path into AI.
Let's go ahead and get started, everyone. I will, before we begin, just want to make sure that you all know that the event is being recorded and will be published on the OpenAI Forum after the event. So I just wanted to flag that for everyone.
And then a few notes on navigating on the platform, if this is your first time on the platform, at the bottom center of your screen, there's a button that you can use to raise a virtual hand. We ask that you please stay on mute as the guests do their conversation. And then during the Q&A, if you're going to be asking a question, unmute yourself then and raise your virtual hand, and we'll have folks on the call who will help with getting your question spotlighted and answered.
On the right side of your screen, you'll see a panel with all the community members who are here. You can feel free to connect with folks one-on-one. You can hover over their LinkedIn icon embedded in each of their profiles. So we really want, you know, as we, this is your community, and we really encourage you to continue the conversations beyond even this event that we're hosting tonight.
And so with that, for those of you who don't know me, I'm Artem Trotsiuk. I'm your OpenAI Forum Fellow. I wanted to start off by reminding us of OpenAI's mission. OpenAI's mission is to ensure that artificial general intelligence, by which we mean highly autonomous systems that outperform humans, are most economically valuable for work, benefit all of humanity. So with that, let's get started. Today, as I mentioned, we're going to have a conversation about non-traditional career paths into AI research. And we're honored to have two distinguished speakers from OpenAI who will share their insights and expertise with us tonight. Let me tell you a little bit more about our speakers.
So Brydon Eastman, he is an applied mathematician and a research scientist at OpenAI. Before joining OpenAI, Brydon worked for a fintech company called Minerva AI and held a postdoctoral position at the University of Waterloo. I saw some folks from Canada in the chat, so some connection there, where he taught differential equation courses and helped supervise and conduct research on mathematical oncology. We'll have to dissect a little bit more about that, Brydon, in our conversation, what that meant to just help me understand that a little bit more.
And then Christina Kim, she is a research engineer at OpenAI. She has contributed to WebGBT, ChatGBT, ChatGBT with Browsing, and GBT4. Before joining OpenAI, Christina was a founding engineer at SourceSource.
So with that, the way we're going to facilitate our discussion is just to have a conversation with Brydon and Christina. So I'll kind of kick it off to allow for them to tell us a little bit more about themselves, tell us a little bit more about just like your background, to help paint a picture about you as a person and kind of pre-OpenAI, what you did, and we can kind of take it from there. I don't, how about Christina, you go first, and then we'll go from there. Cool.
Yeah. So I studied computer science in undergraduate, and right afterwards, I moved out to San Francisco, where I worked as a software engineer. I worked for a few startups before I started doing a bit of machine learning at my last startup, SourceSource, and that's where I really got a lot more interested in machine learning and deep learning. It also helped that actually the few of my roommates at the time were researchers at OpenAI, and they had just published, I think, GBT2, and they were telling me, hey, if you are going to be interested in startups, maybe you should start looking at large language models at something for a startup. But instead, I think I just became way more interested in the idea of AGI, and I realized that I wanted to figure out how I could contribute, and I thought this is maybe the coolest technology that anyone could contribute to in their life. And so I started self-studying, and then OpenAI at the time had this program called the Scholars Program, and that's how I ended up joining OpenAI. It was like a six-month program. We don't have this anymore, like Artem said. We now have the Residency, which I think is a much better program, but basically, it's a way for getting folks who, without a deep learning or research background, to transition into these research roles. Talk a little bit more, I think, about the Residency Program with Brydon.
Brydon, maybe you can take it from here and kind of tell us a little bit more about yourself. I'm seeing in the chat, people are talking about you TA-ing a class or something like that. It's a small world. I think you're on mute, so if you can unmute yourself, and then...
Can I mute it now?
Yeah, you're good. You're good to go.
All right. Yeah, so I also, in undergrad, I did computer science. I entered thinking I was only going to do CS, and then during my first summer there, one of the math profs approached me and asked me to join him for doing a research thing. So I ended up getting into math research at the time, and so I ended up double majoring math and computer science, and then went off and did grad school in math. I think, in general, if your goal is to break into AI and AI research, that doing a master's and PhD in mathematical biology is probably the worst way to get into it. But I was very, very interested in it and very glad that I went that way. I think what really drew me to that direction was that mathematical biology was a very young field at the time, and physics was cool and everything, but it was already all done. To come up with new stuff there was really hard, whereas math bio was super young and felt like there was lots of areas to make contributions. During my master's degree and my PhD, I started using machine learning a lot and my math bio research. I started following OpenAI after Andrej Karpathy had this blog about the unreasonable effectiveness of RNNs. So I trained my own and used it to write a song with my friends. And then after that, I started following OpenAI more closely, and then after my PhD, I started looking into different places I was going to work. I started at this fintech startup and then started looking at different OpenAI postings and ended up here.
What's math bio, mathematical biology? I don't know if anyone else in the audience is curious about that, if he has to do a plus one or something or thumbs up in the middle icon, but I'm genuinely curious, what's mathematical biology?
So it's using math to model and describe different biological things. During my master's, I was mostly focused on predator-prey systems, which are used for ecologists and they also have different applications in chemistry, especially in the use of bioreactors for green energy. And that was the particular application we dealt with in my master's. Then in my PhD, I was mostly focused on health stuff. So oncology was the main thing and we had a bunch of different imaging data from 12 women with breast cancer at St. Mike's Hospital. And we're just using that and differential equations and stuff to model the spread of cancer and its response to different treatments.
Got it. Okay. Christina, when you went through the program, the program that used to, the one that you were a part of, what was it like? How do you think your educational training helped prepare you for that program and kind of the path into an AI career? I'm curious about that.
Yeah, I think it's actually pretty common for a lot of folks at OpenAI on the research side to actually have a background that is non-traditional. And so they're not coming from like PhD programs or a lot of people actually have quite similar backgrounds where they were like Bryden were doing a PhD in like a slightly different field or maybe even like totally different. And in my case, I think there are a lot of folks at OpenAI whose backgrounds were also software engineers.
I think some skills that really help with research is writing code. So I think if you enjoy writing code, I think it's definitely learning the skills to become a researcher. I think they're definitely like learnable. And I think like having a headstart as like knowing or already had a program was definitely extremely useful.
Bryden, what do you think for when you did the residency program, how much of what you did in your PhD world helped with any of that? Like the transition into it, you know, for, I know we have a lot of audience members who are students in the room and PhDs and postdocs, what's a good takeaway? What was the residency program like for you and kind of what helped you even like with that?
I think like transitioning into the residency program from a PhD program felt very natural. It's like a high pressure, high stress environment, but there are people dedicated to ensure that that pressure makes diamonds and it doesn't make you crumble. And so it actually felt like, like grad school, if grad school was supportive, it was, yeah. And I think like the research mechanism that was very like similar, so it felt like very familiar in that regard. But yeah, like to Christina's point that the software engineering stuff was, it's just so key, like how quickly you can iterate is usually the most important thing. And so like being able to not only write code, but write good code and write good code that will make sense in six months to a stranger is like a super important skill. And yeah, it was, I think like the one thing.
I found myself going back on my old undergrad stuff and trying to remember the things I learned back then was trying to go over all this different software engineering principles.
Got it. For folks who want to start in this space, say they're working in biology and they're interested in transitioning and taking some of their computational work and then transitioning, what would you advise someone who wants to start exploring a career in AI and just like machine learning? What would be a good place that you usually recommend for folks to start with?
I see some discussion in the chat of just like, what are the cold start parts of like when you want to get involved, what does one do? Maybe Christina, you can start with this one.
Yeah. I think now that AI is such a hot field, luckily there's actually like a lot of material out there. I think even when I was trying to jumpstart and learn as much as I could, one thing I really liked was, I think they might call it something different now, but it's called fast AI. It's like a course and there's a lot of different courses. And I think most universities also make a lot of their courses public, but I found them a little bit unapproachable from my point of view. But I think if that's something of interest, I think another thing that's really useful is that there's a lot of people that really make these beautiful blogs. So if there's any research idea that you see a lot of people talking about on Twitter, and that's actually one of the ways I learn a lot about new papers is actually Twitter. So I'd actually recommend following a bunch of researchers on Twitter, and then you'll kind of just buy Osmosis through Twitter and get up to date on a bunch of things.
But yeah, they'll make these really beautiful blogs. So you don't actually have to spend all your time reading these papers. You're going to read these beautiful blogs instead that concisely and nicely diagram what the major points are from papers. So I think one way is I definitely really like Twitter for discovering and trying to figure out how to be up to date for things. And then there's also just a bunch of courses I would recommend taking a look at. I think because of the interest in AI now, I think you'd probably be able to find something that's at the right level of introduction for you.
Cool. Brandon, what do you think? What advice do you give your younger self and or someone that just is curious about transitioning into the role like yours?
Yeah, totally. I would echo what Christine is saying. Like I said, when I first was getting into this stuff, I found Andrej Karpathy's blogs really helpful, and he's still making great blogs. So that's still a good way to get your feet wet. And just in general, I think, yeah, like Tina said, there's lots of really good resources on the internet for learning the basics of it. And then just having... There's projects you can be doing. You can jump on Kaggle or something, join a Kaggle contest. There are well-scoped out, well-defined projects on the internet that are a great way to learn how this stuff works in practice. Go train MNIST or something. Do the classics and just train models, build up a small portfolio of doing that. And that'll supercharge you way more than studying some book.
Got it. Do you have a few favorite people you follow on Twitter, both of y'all? I know that was like an osmosis mention I was seeing in the chat, like folks are asking or just talking about that. Do you have favorite folks that you recommend people to start kind of listening, not listening, I guess, but reading up on or anything like that?
Yeah, I mean, I think Brighton said Andrej is definitely very good. I can try to drop some links in the chat here for some folks I like on Twitter.
Cool. Brighton, how was the residency program process for you, kind of going through the interview process and just in general, the support network? You mentioned the support network. What's that like for folks to kind of get an idea of what you meant by being a diamond?
Yeah. So I think when you join the residency program, there's the people who are coordinating the program. And it's usually quite small, like the amount of residents that come on at any one point in time is like a fairly manageable numbers. So these like residency coordinators, they know who everyone is. When I was going through the program, we used to come in in cohorts. So we all started on the same day. They're no longer doing that. It's more organic to just kind of take people on when the need arises.
But yeah, so there's these people I'm no longer sure who's managing the program because the person who did it when I came in has since moved on to other things. But there will be a person who's like the residency coordinator who is a very useful point of contact. And then additionally, like throughout the residency program, you're matched with a mentor and a manager. And so like the nice thing about that split is that the manager is the person who like at the end of the day is going to make the decision on if you get a full time offer or not. Right?
But the mentor is then like a bit removed. So you can you can ask you can look a little dumber, like you can feel more free to look a little dumber in front of the mentor and ask them questions, like get an honest opinion on how far along you are, like how close you are to getting hired and stuff like that from someone who's it's a bit more approachable to ask that person because they're not going to be making that choice. That makes sense.
Christina, when you did your program, did you have a similar experience as what Brighton was just discussing around? And also, thank you for dropping the link for Fast.ai into the chat for folks to see that right now. But when you went through the program, what was the mentorship like for you and kind of like your general experience through the program that you were participating?
Yeah, so the Scholars Program was structured a bit differently. So I think for the residency now, you're like fully like part of a team and you're working on like a team, a project that's usually with other people on that team. But the Scholars Program was structured differently. So we were actually basically doing our own research project.
So it's a little bit more isolated. I think that's one of the reasons why we have the residency now instead and deprecated the Scholars Program. But I think one of the things was just like the cohort for the scholars. We had like a the I think for the residency now, it's on a rolling basis. So you can actually start whenever. But for the Scholars Program, we had a cohort. So it was really nice to have like eight other people that I was like learning with and talking to and like the same struggles of like trying to ramp up quickly and like issues with like code. So I think that was really nice to have like a group of people to like talk to as we're like all going through the same thing.
Cool. What do you, how's and now that you've kind of gone through that, what's life now like in terms of kind of just the work that you're working on? What excites you about the general future direction that you're involved with? Maybe Brighton, you can start with that.
Yeah, I think like similar to what I was saying about why I liked Math Bio with it being so young and there being so many like low hanging fruit and obvious places to research. It sort of feels like the same here. Christina and I both work in post-training research and in post-training, there's like, there's a lot I think that's still, it's a young discipline in an already younger discipline or an already young discipline.
So there's a lot of new things and it's, it feels like super exciting. Like we're discovering new ways to do things every couple of months and it's yeah, just like that kind of research muscle of learning a new thing and then seeing it kind of like seeing that idea or that pattern kind of copy into other places. It's just like a really rewarding, really exciting thing.
Yeah. Fun to be part of.
Cool. Christina, what are your, what are your thoughts on the question?
Yeah, I think similar. Like I think there's just a lot of new stuff and I think it's just it's also like interesting to like explore just because it is like everything that we're doing is a bit very new.
And so I think there's like a lot to explore. And so I think it's like hard to feel like bored in any sense.
Fair enough. Fair enough. What, while, while folks in the audience, if anyone does have a question, I've been kind of reading the chat as we're going, but if folks do have questions, we can we can have those conversations and you can ask Christina and Brighton directly. But I do have another question in terms of just kind of thinking about challenges. I know we're talking about all the things that are good and things that like people are excited about and how you transition to this role, but what are the types of challenges that like, you know, like how does, how does one overcome challenges of getting into the space or maybe challenges that you, people in the audience might understand and relate to?
I'm going to keep it a little bit open-ended, but genuinely curious about like, what do you, what are your biggest challenges currently in your role or maybe in the way in transitioning into a role like this?
I think there's like a common failure mode that I saw in a lot of the different residents. My cohort, I think we had like a 50% survival rate. It was like a snap. And I think like the thing that was pretty clearly different between the people who got full-time offers and the people who weren't able to was typically like how execution-oriented people were.
Like I found that some of the other residents, especially some of them who came from, also came from academic backgrounds, got too nerd sniped by trying to make like the most mathematically perfect idea or trying to like, yeah, like really trying to over-optimize before they even get like a good result at the door. And I think that was, yeah, just kind of the difference of just getting things out there, failing forward was a pretty important, at least in my cohort.
Justina, what do you think?
Yeah. I think one thing that's definitely different about research, if you're coming from industry, is that it's a lot of things, things might not work out or like, it's also hard to like plan for how long something should take.
Like there's already a joke that like for engineering, like for engineers that are like really bad at giving time estimates, but it's actually, it's way worse for research. So I think being able to like manage your time well and then also like know when you should keep pushing on something or not pushing on something is I think a skill to have that's quite different than I think maybe coming from someone who's just worked as a software engineer and didn't have any research background.
Cool. Cool. Let me...
some folks from the audience. Now I've been kind of asking questions just to understand a little bit more about the both of you on the program, but perhaps we will open it up to a few members of the audience to kind of talk a little bit more about that.
I'm sorry, I'm going to mispronounce your name, Hudhaifa, if you want to unmute yourself and ask your question. I think you're muted. I think you're still muted. I don't think I can hear you. There we go. You want to give it a shot? Okay. Perhaps we can come, maybe you can type your question in the chat and we'll go ahead and try to answer it that way. I don't know why.
Can you hear me now? The internet connection seems a little bit unstable, so we can't fully hear you. Maybe try turning off your video. Yeah, try it that way. Okay. This is better. Better. Yes. Okay. Sweet. Okay. Sorry. And it's okay for butchering my name. It happens more often. So my question was kind of on like the domains of research that the residency program kind of offers. I was curious if you guys have like something specific, like synthetic data and stuff as it's becoming like more relevant and more important in how we're training models now. I think like the residency program, it'll match you with a team. And so there's just the scope of it is basically the scope of research at the company, which certainly includes people looking into these sorts of directions.
Are there any like specific fixed things that you guys are working on right now? Or is it just kind of just like anything and everything?
Yeah, I think it really depends. Oh, sorry. I think it depends on the mentor you get matched with. Sometimes the mentors will have a project already in mind, so it will be less like project selection for you. And I've heard of some residents where they're kind of coming up with a project as they go along.
Okay. And how does that matching process kind of work? If you can elaborate a bit more on it?
The interview process, they'll be like, they'll put your portfolio up. And then whoever's looking for a resident can see your portfolio. And then they can like, if they're interested, they'll schedule a mentor match interview. And yeah.
Okay, that's good. Thanks a lot. And we did drop a link for folks. We pinned to the top, the link to the OpenAI Residency Program website, which has a lot more information about kind of what we look for and the general residency program and some FAQs. So that also is a good resource for folks to have some general questions about that.
Let's go ahead and Mohsen, I will call on you next. If you can unmute yourself.
Hi, I cannot. Okay. So, oh, yeah. Can you hear me? Yes, I can hear you. Yeah. Thanks, Artem. Thanks, everyone. I don't know, Christina for insight. Yeah, I do have a question regarding the requirements because as I'm checking the link for residency program, I do not see the requirements. You know, anything could be specialty, could be, you know, age. I don't know the background. Is there anything because I cannot find in the FAQ page?
So if your question is just like general requirements, the program is a means of getting folks who are coming from non-traditional backgrounds into a path to AI. And maybe right in, what is your general sense of what helped you prepare to be a good candidate? Like what kind of skills as you kind of moved forward into the residency did you think were valuable for you to have as part of like just that process? I think that will help answer this question here.
Yeah. Yeah. I think like the people that they're trying to recruit through the program are very varied. Like there's not like one cookie cutter mold. And that's why I don't think there actually are like hard and fast requirements for the role necessarily. Like there's obviously an interview process and everything, but I don't know necessarily that it's like, oh, you have to hold X degree or something like that. And in fact, I think it's specifically not that.
During the interview process though, like they will be testing for certain skills and they have a particularly high expectation for coding skills. You know, of course, just kind of like math and problem solving skills will always be helpful in those sorts of interviews too.
Okay, cool. Thank you. And is there any open time or as you said, is more rolling bases because again, there should be any waiting lists for that, how it works with respect to the residency program.
Yeah. When the applications are open, they're on the page. The best thing to do is you check back on the page and as soon as there's applications open, then one can apply. And then yeah, I think I covered that right in, but if I didn't, please add anything else to that. Cool.
All right. Next question, Omar.
Yeah. Can you hear me?
Yes, I can hear you. Hello.
Hello. Yeah, thank you for your time.
Just so I think. So, Bryden, you mentioned earlier that, I think you said pressure makes diamonds, right? And I think just out of curiosity, the thing I was wondering was, you know, OpenAI being OpenAI and being at the top of its field, right? I would imagine that, you know, pressure obviously is there for everyone who works there as far as getting results, shipping, all that kind of stuff. So like, what's an example of that? Like how they have a deadline or of a type of deliverables that you faced or that you had to meet during your residency?
Yeah, like it could be all of those things. It could be tight deadlines. During my residency program that happened quite frequently where there would just like deadlines would move up on things kind of unexpectedly due to kind of external factors. Things like pivots are very common. I think like hard pivoting, like you're pushing on some research project and it's looking like incredibly promising and then someone else somewhere else at the company like all of a sudden developed something that just means, you know, we're all pivoting that way. And this area doesn't matter anymore or isn't as important as we thought it was or something like things like that happen very frequently. It's also just, I think we have very ambitious goals. So sometimes you'll be taking on a project and the pressure is not necessarily like you need to have solved this by X date. It's more just like the magnitude of the thing you're working on can be quite a bit of pressure sometimes.
Cristina, what are your thoughts on that? Like generally, what's your advice for someone to like how to cope with deliverables and pressure and anything like that when it comes to kind of working in an environment where they're transitioning into a new space? How does one cope with that? How does one work through?
I think I wanted to build on that question a little bit more.
Yeah, I think a big part is like communication. I actually just had a resident who did convert full-time. And I think one of the things was just like really, and I noticed it from the other side now as a mentor was like being able to communicate a lot and like being up to date. I think it's hard if you're kind of just like going off in a corner and just working on things and people don't really know what you're doing and like where the progress is at. So I think one thing is like being able to communicate like your negative results and positive results. And I think all of those things like, because the whole purpose of the residency too is for you to be learning and seeing like how good of a fit it is or the research culture is a fit for you. So I think one thing for research is it's really important to share all results negative or positive. So I think just being like communicative and like over communicating and like being able to have decent good write-ups to point to and being able to share code so other folks can work on it or like help you out as well. Cool. Omar, thank you for the good question. And actually while I'm here, can I ask another one?
Sure. Yeah. Yeah. I think most people here are students and especially grad students and the classic thing that, you know, a huge chunk of time is spent on is learning, keeping up with literature, new papers, all that. So I'm also just curious, like during your residency and then even now, how much of your time is spent on just staying up to date with the latest literature or just learning in general?
During the residency program, I think quite a bit more, like it's pretty natural, I think. And like at least for me, like when I first joined with my manager, the first thing he did was give me like a huge reading list. So there's in the residency program, I think an expectation that you're learning quite a bit. Now, like keeping up to date with literature and research is probably where I spend more of my time rather than like trying to learn a new technique necessarily. Though every once in a while you do try to learn like a new programming technique or something. But yeah, I have like specific times on my calendar carved out for just like not being on Slack, just reading papers and stuff.
Cool. Christina, what about you?
I think it's a very good question, Omar. Thank you. That's another pretty good question. Yeah, I think mostly, like I said, Twitter, I think it's a pretty good resource for staying up to date. I think that's, yeah, the best way I get up to date on like what other labs are doing.
And then also when people are sharing things like internally like on Slack for us
But yeah, I definitely read less papers than I did when I was in my scholars program unfortunately
Awesome one more. Thank you, Angela. I have a question I Was just wondering about your interview processes so like you mentioned a couple of times like holding and having of course the base for research, but we'd love to hear your experience like from like start to your Your final round and also if our question if this interview is like common for all residents or is different per group
Yeah, so for the residency I think there is like a set Rounds I think I don't I don't remember exactly what the rounds are but the interviews should be the same ish like in terms of like Hardness of the question and so a lot of the interviews are like yeah around coding. I think there might be Some that might be a little bit more Math ish focus, but most of the interviews technical interviews you'll do will involve you writing code Enjoy thank you Roy next
Thanks Hey guys, this is awesome, so I'm non-technical but back myself and say I can learn pretty fast if I need to question is what's the minimum level of technical abilities that you think would be required for the residency and Has anyone come through and then Say Just interesting people I think like it's not the not the only way that open AI hires but it's yeah one way by which the Sort of recruiting new talent into the field especially people with like fresh ideas and thinking about problems from different directions And The I think like the intention is that for the residency program? That the selection process would be good enough that every resident that gets picked is someone who could Become a full-time employee and like the goal is to make sure that that happens And that of course doesn't happen for everyone, but the the rate of conversion is like quite high Roy thank you Yoko, you're next
Yeah Yeah, I can't hear you, okay I'm mute. Okay. How about this? Good. Good. We can hear you. Okay. Thank you Sorry, I'm concerned. I think my question is a little bit out of focus for this particular workshop But yeah, I am a program manager in high-tech industry and I'm trying to get into AI and I have been working with software engineers Software engineers and PhDs a lot, but not so much with researchers but based on the current situation and landscape of AI's I What do you? think that difference to work with a Researchers as a program manager and what do you expect to program managers? So maybe the way we take that question is kind of more of The difference between software engineering Yeah, yeah Christina or Brighton, what do you all think about the differences and similarities between the fields and kind of the translatable skill sets between software versus machine learning engineering I Think that there are like quite a lot of similarities like there's not no real obvious differences to me I think like some of the things Christina pointed out about Timelines, that's that's one thing I remember when I first joined there was like some research project that by its like very definition was incredibly open-ended and One of the engineers came to me and was like, let's get this thing shipped. I was like, that's not really how it works It's just It'll work or it won't who knows right? And so there's I think yeah, like timelines and stuff is probably the biggest difference. But other than that, I think like Nerds are nerds. It's just it feels very similar similar kinds of people like highly motivated intrinsically motivated people working on things they like Okay, so Something like a ball being requirement and you but you have a time requirement So you have to meet that time even though your requirement constantly changing. Is that something?
Yeah, yeah, I think so I think that's part of it yeah Yeah, and then I have one more quick very quick question Um, so I'm trying to take a class from a local school and to get into AI and what would you guys? Recommend what class should I take except, you know, AI JNL and also machine language That's probably I need to take other than that. What kind of classes do you guys recommend to get into AI industry? I Think Christina you dropped a link in the chat for a fast AI right? Tell us. Yeah high level and what? What that course just another recap on what that courses? Yeah, so it's made for there's a couple different courses now that I'm seeing it but the one that I did was is basically like deep learning for coders and so it's a kind of Expects you to have some programming background, but then it quickly kind of just ramps you through like actually like Coding like a transformer and things like that and so it's like very practical and so it kind of gets you like hands dirty you really quick and so you'll kind of be Like programming things without like actually fully understanding what it's doing yet And then as the course goes through like actually explain to you like the science and the math of it And so I think that course is really good for like a quick Deep dive into things. Um, but then I think if you're looking for other courses, too I think like linear algebra is quite good And having a basis in that And maybe folks in audience in the chat if you all have any other suggestions on courses just drop them in the chat and That way everyone can we can do a group think about good courses out there Yoko, thank you. Thank you our next question by Severin I Don't think we can hear you you're on your there we go We cannot hear you now Maybe while we can work on the technical component on the back end. I'll go for Ryan. I'll go next file Will you figure out your audio? Hello So, I guess my big question was like There's a lot of students here, but what about people who have not gone to like an Ivy League school do you guys know residents who have not had like, you know have not gone to Stanford or Oxford or something like that and still have had success getting into these programs Because I feel like you know even when you come from non-traditional backgrounds you can still Lots of people still prefer from that area and when you don't even though you might have the skills to do So it can be a little bit more kind of difficult to break through and be like, yes I have that kind of technical skill. So yeah, I Think like during my program release like in my cohort there was people were from like a bunch of different Educational institutions, there's at least one person who had dropped out So they were they had dropped they were at McGill, which is like a very nice school in Canada they had dropped out of it though and and had done like their own thing had kind of like just Had a great like portfolio of projects to kind of show that they they were a very capable person I think yeah, like more for the residency program more so than necessarily like having the right pedigree or necessary or anything like that. It's more about Being able to prove that that you're someone who is Capable of doing AI research and degrees are one way to prove that but they're not the only way and I think I would say like a good portfolio would be quite a bit stronger Thanks Ryan Seeing a lot of good discussion in the chat and some examples of different courses, especially folks who are trying to foray into that so Good we do want to try Severin again and see if his his microphone works Or Severin you can drop your question in the chat if you'd like if that would be easier for you as well Nope I don't think we can hear you Just go ahead and drop in the chat and then we'll try to answer it that way That'd be the best way to go about it Sounds good Gotham do you want to go next? Hey, can you hear me? Yeah, we can hear you Yeah, so I was curious about like, you know what a machine learning engineer versus like a researcher role at OpenAI looks like and I'm particularly interested in safety and if like research is the right direction to look for that or Would that be more like an engineering thing? I think both like there's definitely both research
and research engineers on the different safety teams and alignment teams and stuff, and always looking for more talent in there. I think people get really hung up on the difference between software engineer, research engineer, and research scientist, which, as far as I know, are the three different tracks of the residency program.
It's a very osmotic barrier. One of the guys I know who went through the software engineering thing is now a researcher on a completely different team than the team he did his software engineering residency on. Yeah, people just kind of move to wherever they fit the best, and the labels are, OpenAI is a fairly flat organisation, and it's quite good about not necessarily being too dogmatic about labels like that. Got it.
And in terms of interviews for the residency program, what does the interview look like for each of these tracks? Are they pretty similar, general set of software skills and machine learning knowledge, or do they differ across the tracks?
Yeah, I think they're very similar interviews between the three different tracks. I think the only difference is that the software engineering interview, there will be more software engineering-style questions, and then the research-tracked interviews will have more theory questions, but in general, just, yeah, software engineering stuff, machine learning theory stuff is all good stuff to brush up on. Got it. Thank you.
Sweet. Maybe for Christina and Brighton, I have more of a general question, I think. How much of the work that you were doing as a resident or in the scholars program translated into the work that you're doing post that program? So I know that with the residency, there's an aspect of working on a project. How much of that got translated into your full-time role and what you're doing now, and what have you seen with other residents, perhaps, that have gone through the program of the project they're working on, and how does that translate into their current work?
Yeah, I think what's nice about the residency is that you are embedded on a team, and so you're working with that team. So oftentimes, the residency project is just the start, and so they normally will actually continue that project if you are joining full-time, and so you get to continue that work and expand it. But like Brighton said, there are some people who, after their residency project, and they want to join full-time, but they decide to do something completely different, and I think that's fine, too. I think the point of the residency is also to expose you to all the different types of research that's happening at OpenAI, and so I think people are also very open for residents to switch teams after the residency is over.
Cool. Brighton, what's research culture when we talk about the idea of doing research versus the idea of at OpenAI, research culture? What's that generally like? Describe that for us a little bit more.
Yeah, I think a lot of what Christine was saying just about sharing write-ups a lot, I think what makes the research Slack channels slightly different from the engineering Slack channels is just, in the research Slack channels, you'll always see people posting updates of plots of failed experiments more frequently than plots of successful experiments. People will just be posting about an interesting paper. You'll be working on some direction, and then some random other lab will publish actually the solution to your problem, and then it'll start this huge really active firestorm on Slack of people chiming in and trying to see if everything's been completely scooped, and it's always, yeah, just kind of a very active place of just being really, I think, curious probably more than anything.
Awesome. Another question I've kind of seen come through the chat. For someone who's considering going into like a residency or a research fellowship or kind of in the beginning of their career deciding deep learning and kind of directly applying for an engineering role or going through residency, how beneficial do you think the residency program was for you going from in the non-traditional path into this? Do you think it helped you in terms of understanding a little bit more about what the work that you're now doing would entail, or how did that help you in terms of more success for where you are now?
Personally, I found like the residency program was, yeah, really like one-to-one basically with the day-to-day. It really ramps you up quite quickly to what the full-time employee kind of expectation will be like, and yeah, I found the transition to be like super smooth, and it feels very similar to what it felt like back then, and I see that a lot with like the other residents, too. People are already treating them like full-time employees usually by like month three. Super cool.
Mike, I think you had your hand up if you want to go ahead and ask your question as well. Thank you for your contribution to the chat about the different courses available for for folks who are interested.
Two-part question. The first part is having worked on research teams and on development teams in the past, yes, they are different in terms of day-to-day operation, but there always is a transition in terms of how does the research make it into the development path? I'm sort of curious as to the culture there in terms of bringing what has been a greenlit from research to become actually on the roadmap.
The second part is for this residency, in establishing a goal for six months, how is that formulated, and how do you work with the residents in terms of how those goals are set and achieved? Maybe we'll start with the second question since that builds on what we just talked about with the residency program. You can talk a little bit more about broad scale, how the goals were set, and how that transitioned into kind of the work that you were doing. Then we can go back to the first one.
I think for the residency program, far more than goals of a particular deliverable, it's more like the deliverable is the resident in a way. There's just an expectation of the level that the person will be performing at by the end of the residency program, and that's made very clear. I actually think it's one of my favourite things about OpenAI is how objective it feels like performance stuff is. It's spelled out very plainly. For the residency program, I sort of felt like that was the main thing, and less so about the particular project, and because often, like Christine was saying earlier, often the project is picked before the resident is matched to the project. It's usually the case that it's an ongoing thing, and the resident's maybe joining something that's already been being worked on for a couple of months.
Got it. Thank you. Maybe what's been your experience like, Brendan and Christina, around the collaboration and collaborative nature of working at OpenAI? Maybe going back to that first question, I would like to kind of ask about that. What's collaboration like, and just the overall kind of thought experience more broadly?
Yeah, I think post-training, the team Bryden and I are both on, is interesting because it's a team that gets to interact with a lot of different teams. We're also kind of like the last research team right before maybe something will go on to become a product, so we interact a lot more with the applied teams, and so a lot of the things that you'll see, like model updates are coming from our post-training team into like chat GPT. So there's a lot of collaboration between the research and applied orgs as well there, and so I think it's, like Bryden said earlier also, it's like a pretty flat organization that it doesn't really feel like everyone is working together, and it's easy to just message someone, and there's not really, I don't feel hesitant ever to just message someone out of the blue, even if I've never worked with them, to ask them a question about something I've seen or something they may have posted, and so I think it's a really easy place to ask questions, and it's a really easy place to like just start collaborating. It's also like if you have an idea, I think another thing about OpenAI is like regardless of like maybe your level or like how long you've been at the company, I think people are very eager for like if you have an idea and you're excited about it, for you to pursue that and like work with other people who are also excited about that, so I think yeah, I think I find OpenAI to be like extremely collaborative and like definitely the best place I've worked in terms of that.
Cool, thank you. Thank you, Mike. Thanks for your contribution. Last question by Stephen, and then we'll work towards wrapping up tonight.
Hello? Yes, we can hear you, Stephen. Okay, so I was curious about if there's any interest in research on analog systems, spiking neural networks, you know, other sort of like non-traditional methods of building neural networks, rather than, you know, like given the large computation cost of traditional back-propagating neural networks.
Yeah, I think there are people looking in all kinds of different directions at like different architectures, but yeah, I think a lot of the times when you're dealing with things at the scale that we're kind of dealing with, you tend to work with things that are amenable to the hardware. Stephen, if you aren't already following, we actually post blogs on the OpenAI website about the ongoing research work that we have going on, and that's honestly the best way to keep in touch with what's happening on the research front with the stuff that the teams are working on. They're very interactive. It's much better than reading a paper.
Yes, yes. I agree with Christina's mention earlier about reading blogs as a means of digesting information.
particularly when it's a very dense paper and you look at the paper you're like what are you trying to get out here? I do not understand and then through a blog it's much more manageable.
Well thank you everyone. I know we've had a lot of good questions so before we break, Stephen thank you for your good question as well.
Before we break I do have some final closing, a few things we're going to drop into the chat. One thing that I wanted to let everyone know here is we do have a program called the OpenAI AI Trainers Program which is a program in which folks who are interested in helping us evaluate, helping us with our AI training efforts and being domain experts where they can contribute to the ongoing model evaluations that we have. We just dropped a link into the chat and we'll pin it at the top for those of you who are interested.
We have an ongoing project right now and many research projects in the pipeline so we're excited that if any of you are interested across different domains, across different areas of expertise, you can contribute in that capacity. I'll complete the form there and then we do have an upcoming event coming up on the 30th which will be a roundtable as part of our community workshop series and it's going to be focusing on lowering barriers to AI adoption.
This will be with Conor Grennan who is also a forum member here and he'll be talking about just the psychology around adoption of different technologies and then we'll dissect and jump right into different use cases. We'll kind of talk about how to build a custom GPT, how is AI being used in a non-traditional sense like social science work and how can you apply that domain knowledge into that field as well.
So definitely if you're interested, we just dropped a link in the chat for you as well there to sign up for that virtual event that's coming up and then for those of you who missed any of our previous events, we had an event on April 25th, Practices for Governing Agentic Systems as well as April 3rd. We had a talk by one of our forum members who's also a Turing Award recipient on Backdoor Vulnerabilities and Mitigation Strategies there.
So those are available for you to watch in the platform and as a reminder, the chat here, this will continue to be active as a chat in the platform. If you go into your messages and everyone who was here at the event, you'll be able to see the chat transcript and you can interact with folks there as well.
But really appreciate Christina and Bryden, you coming and talking about your experience and helping us see what options there are for people who are interested in going through a similar career path as what you guys are doing. So everyone, please give us, give a virtual round of applause, press the buttons, press the likes at the bottom just to kind of really thank our speakers for their time and I really look forward to seeing everyone in our upcoming community sessions as well.
![](https://cdn.gradual.com/images/https://d2xo500swnpgl1.cloudfront.net/uploads/openai/small-square-logo-on-dark-background-80758a52-cfe8-4676-b9b5-a3cdd8c7d230-1693350163347.png?fit=scale-down&width=100)