Episode 90

Hercules Konstantopoulos on Journey from Astrophysics to Data Science and Hiring in the Tech Age

TIM: We're live on the Objective Hiring Show with Hercules. Hercules, thank you so much for joining us today.

HERCULES: It's great to be here, Tim. Thanks for having me.

TIM: It's our pleasure. And I believe you're the first person in New Zealand that I've spoken to on this podcast. It's good to get a cross-Tasman conversation going, so thank you for joining on that front. And it'd be great for our audience just to hear a little bit about yourself just so they can start to contextualize our conversation today.

HERCULES: Yeah. Prepare for a ride around the world. The background is a little bit silly, so I'll try to keep it brief. because I've worked in a whole bunch of sectors and industries in a whole bunch of countries. I'll, yeah, let me try to keep it punchy here. So my background way back in the day is in research. So I started as a researcher in astrophysics. That's because I decided what I wanted to do when I was eight years old. And maybe I didn't go past the as, or I don't know what it was. But I basically dedicated my little boy childhood to becoming a scientist. So then I, and I did, eventually I did. It was great fun. took me to all kinds of places. I studied in the UK and in Chile, where there are big telescopes. Then went to the US and then to Australia to do postdoctoral research. So that's basically three years at a time, placement at a lab somewhere. So that was fantastic. I got to work at Penn State University, which is a fantastic and big American university, kind of like the things in the movies. So it was really interesting to have the real experience. And then that eventually brought me to Australia, where I worked in a national lab. It was the Australian astronomical observatory. And that oriented me toward more technical things, which I found out I enjoyed more than prestige, which was a very interesting realization. Then I started building data pipelines, which I had always been as an astrophysicist, but I started getting more serious about people, other people's problems rather than my own research problems. And I got quite addicted to that. It was pretty, pretty sweet. So when I got a little bit tired of academia and the grant writing and the boom-and-bust funding cycle, this was about 10 years ago. I'd spent 10 years in academia. And then I made what ended up being a very much lateral move. It wasn't much of a move to data science because he'd do the same things day in, day out, the technical part. And I worked in sustainability to begin with, then big tech. I worked at Atlassian for a few years, which was fantastic. And then. I moved to New Zealand five years ago, and I got to come back to renewables and sustainability. I worked on renewable energy, geothermal energy, and tried to get a geothermal AI startup going, which was a lot of fun. Then most recently, about one year ago, I became the head of data science at a medical research lab, the Malican Institute of Medical Search, which is a total trip for me because I'm not a biologist, but I am a scientist. But they hired me as a techie, so it's everything put together in a pretty sweet package.

TIM: What a wonderful summary. And yeah, as you said, a trip right around the world. And I'm not sure what the appropriate segues are from astrophysics to hiring, but I'm going to try to make it right now. `Hiring is one of those things that's so inaccurate, even at the best of times. Like, I was just listening to a podcast yesterday. About football, like soccer, football, and they were talking about the manager was on there talking about even when they've done all their due diligence, they've done all their research, they've just spent 50 million pounds on bringing a player to their squad. Even when there's that amount of money on the line and that amount of money spent, hiring at the best of times is so inaccurate, which I'm assuming is not the same as astrophysics. You've come from, like, a hard science background where you can actually measure things as, like, universal truths and whatnot. Yeah, putting you on the spot a little bit, is there anything from your astrophysics and science background that you were able to think of and apply? When hiring people and thinking about your people in general,

HERCULES: Wow. That's actually a really good question because astrophysics is an exact science, even though the joke among astrophysicists is that you're a million kilometers out. It's no big deal, but then the skills part is a little bit nebulous in that it is assumed that anybody can acquire any skills, which I say not derogatorily. It's true. That's the best part of the training of specifically astronomy, where you have to do all the things. There isn't a technical group next door you can hand over to do your coding for you, to do your stats, or anything like that. You have to learn absolutely everything. I remember when I was doing my postdoc in Australia, I, yeah, I needed to break into a new area; I needed to learn a new thing. So I talked to a professor, and he goes, Yeah, great, you've got a dark matter halo problem. You need to go learn about dark matter halos. And that was the end of that mentorship session, because what else is it going to do? Teach me about dark matter halos. No, I've got to spend two years figuring that out now. So that part is really interesting, because it's a hard science, but the skills are assumed to be attainable. Once you've reached that level, you'll learn the things you want to learn. Then you hire people based on their record in terms of delivery, typically. And the delivery is something we do quantify in a very flawed way, but we do quantify it through publications. Further quantify it through citations. How many people thought your research was interesting or useful? Although on the flip side, if somebody thought your research was awful, they still cited your research. Not a great metric. So you hire based on those attributes that go on your CV. The biggest problem there, and this was a while back; I haven't been involved in 10 years; maybe it's changed, but the biggest thing was that there are way more people than jobs. So you solve your problem by having a flood of applications. If you're applying for a professorship, as I did once, back in the day, I would have been one of 200 people applying for one job. All of those people have PhDs in astrophysics, which is the requisite. And, probably half of them have a lot of experience and would be pretty good as professors. and maybe half of those would be very good as professors. And maybe half of those would be very good as professors at a specific institute. So that's still, you know, 25 people. How do you choose among all those people? So there's a call for choice, which is a good problem to have. It's not good when you're the job seeker, but for the employer, it's pretty easy because you just ask around essentially. That's the interesting part. When you can't distinguish between your list of 20 people, you just go ask around because everybody knows everybody, which is where the biggest flaw in the process comes in. It creeps in because you get biased opinions about. individual humans.

TIM: A bit of cronyism, even in the world of academia and astrophysicists still, at the end of the day, it's a little bit about who, at a certain level, maybe at that dividing line, reminded me of. One of my favorite books that I'm reading at the moment is the original gangster of self-help books, How to Win Friends and Influence People. This is the Spanish copy here. Just trying to

HERCULES: Yeah.

TIM: Prove my Spanish skills a bit. And yeah, it would certainly advocate a job-seeking process that was focused on who and trying to unlock doors that way. I wonder whether the market's going even more now down that direction. Because I keep hearing about companies putting up job ads, particularly in the United States and Europe, and they're getting inundated with applicants, like 200; they'd be like, That would be in an hour. They're getting maybe a thousand for an average data scientist or analyst role. And so from a job seeker's perspective, I imagine going through the back channel, trying to navigate through who, rather than fighting it out with everyone else. And an online job portal might be the way to go. If you were searching for a job now, how would you approach it?

HERCULES: It's really interesting for me. Been pretty lucky in. That have known the right people when I needed to know the right people, but also I've just applied cold to various places, especially when we moved to New Zealand; I didn't know anybody. I had friends, but I didn't have professional connections. I didn't end up getting a really nice job when we moved here, sight unseen; I just applied, or I was one of however many applicants, but it is a smaller pool here, so it's. The whole country, the size of a large-ish city. It is different; I'm in Wellington. There's 200,000 people in Wellington proper, and then under half a million, I think, in the metro area. A company of 500 people is huge. It's a different context. So you're not going to get that many applications, and you can sort through the old-fashioned way, and it's fine. So that's how I got my first job in New Zealand. And in fact, the job I'm in right now, I also didn't know anybody at the Institute, and I got into the regular sort of method. When I was in Sydney, I can tell you it was the context you're describing. When I was at Alaskan, we would get, I don't know, 600 applications for an internship easily. And this was 10 years ago when it was, I don't know, there was a company that was a thousand people, maybe it wasn't. It's probably about 5,600 people today, probably bigger; I'm not even sure. We were getting heaps and heaps, and then it was still, our recruitment team didn't want to use very niche keyword searches because we would miss out on people. So they had to read all these CVs; they had to at least pause all these CVs. And so when I first got my job there, I didn't get through just applying; whatever the reason, I don't know, maybe my CV wasn't written the right way. I don't know. I really didn't make it through. I made it through first pass once and I never got the call to schedule the interview. And then one of my friends got a job there, and then a week later, I had a job there. So for sure networking, all the indications I have are that it's still extremely important.

TIM: Let's say my cohort. And I sometimes also feel like the perception of networking is a bit too transactional. Maybe it's, Oh great, I need a job at this company. I'm going to bomb a bunch of people at this company with a generic LinkedIn message and a generic CV. That's not the way to do it. How would you think about networking? What do you think is the right way to approach it?

HERCULES: The science meetups are back. Obviously that's a really important one. There were, as you say, a few nobody was meeting in person. so that was, that must have been pretty rough. I was safely ensconced in a job that I liked back in the day. So I was fine. I didn't need to meet anybody. Also in Wellington we barely had a lockdown. We were lacking that respect. I'd say I'd say in tech, and I definitely include myself in that group. It's always extremely awkward. Nobody really is comfortable with the whole in-person go-to-the-meetup. Hi, I'm Hercules. Ah.

TIM: Not a lot of pure extroverts.

HERCULES: No. I'm one of those heavily socialized. Introverts. So I learned how to do it, not consciously, but because it's a cultural thing where I grew up in Greece. There is no such thing as an introvert. It's not in the eighties.

TIM: Oh, really?

HERCULES: No, there was no such thing as, Oh, little Johnny likes to read books. He doesn't have to play with the other kids. No, just go. There was no such thing. So I had to do it. And then I learned how to cope with it, and it was fine. And that was a very, it's a very different context from how most people grow up.

TIM: Because

HERCULES: We acknowledge that now this one is bookish. It's fine. It doesn't have to play soccer. So that's, I obviously, it was hard, but as a kid, it taught me that thing that it's just. You get yourself into that, and you, it's very uncomfortable. And it took me a very long time to get at it too. What I always told myself was, and I tell this to everybody, you just need to meet one person. Just go to the meetup and do one intro. And if you do that, if you go to one meetup every month, it's not going to seem horrible the next time you have to do that cold intro, and then it'll be easy. Then, well, then you'll know how, what it is. It's just an experience. It's just gaining experience in something that looks icky and weird. And the transactional part, you're right. That definitely, I felt that's just a weird thing, but everybody's in on that transaction. That's why the people who are hiring are there as well. It's quite explicit, in fact, at meetups. There's a slot for networking. That's what that is. It's not a weird thing. It's not. I've heard people think that it's even to the point that, oh, it's an unfair advantage with those who aren't in the meetup. Yeah, it is, but they didn't come to the meetup.

TIM: I'm reminded of another one of my favorite books, actually, which I can see on my bookshelf over there, Atomic Habits. I'm not sure if you've read that one. now. The premise of the book is basically about how to improve your life by focusing on your good and bad habits and making these kinds of small micro changes that, through time, would compound. profound impact. And yeah, the way you were describing your philosophy on the networking reminded me of that book because you weren't saying, Oh my god, I have to go to this function, I have to speak to 50 people, and I have to do that. You set yourself a reasonable, achievable, and unintimidating goal, which then probably helped you do it in the first place, as opposed to overwhelming yourself with too high a goal. And so it reminded me of, yeah, this book and one of its principles around making it easy. That's the way to make a good habit stick and a bad habit. Get rid of that by making it hard.

HERCULES: And you'll get good at it over time. It's not something you can get frustrated at yourself for. Even failing, like failing, I went to the meetup. I was too uncomfortable to talk to anybody; I didn't talk to anybody. Fine. You'll do it next time. It doesn't have to happen immediately.

TIM: Yes. Don't beat yourself up too much. Yeah, for sure. Just to change tracks a little bit away from networking and landing roles and those kinds of things. What about on the hiring side? Of things as a hiring manager. Have you had a chance yet to dabble with AI tools in hiring? Have you seen perhaps candidates on their side use tooling?

HERCULES: I haven't used I haven't used ai tools as a hiring manager and that's because I'm hiring. I have been hiring for the past year. We've been hiring at the Malaghan; my role was to, for the first year, build a team. So there's four of us now, including myself. So there was a whole bunch of time spent on hiring, but again, we're in this interesting niche where. We didn't need it; we weren't overwhelmed with applications, although we were delighted to get it. I don't know; we hired a senior data scientist recently, and we had 50 applicants, which, again, for a country of 5 million, is huge. And in a very niche field, it is quite intimidating. For me, it was as well to walk into a biology lab and immunology lab and tell people, sure, I can do data science here; it's fine. It's really quite a bit; it seems really daunting because it's real science, and there are people literally curing cancer in that building. So it's quite daunting. So getting about 50 applications was phenomenal. And it was, so I used all the standard tricks. I made sure that the application was clearly written by a human. That's not to say that I wouldn't have used some Gen AI tool to do filler, and I used Gen AI for just the first pass at questions and reformatting the phrasing of questions, things like that. So I guess I have used AI tools now that I think of it, but not for making the application. The application was for me. I just bulleted it. I fleshed it out. I wrote it. I made sure it sounds like me, actually, but slightly more professional, if you match the Institute. You're clearly reading something that a person wrote who understands what the role is going to be, who understands what your hesitations might be, who understands why it's daunting, who understands all of these things, and who's giving you a really good reason to apply. I made sure to say, Hey, look, if you don't know biology, that's fine. We'll get you up to speed with all these things. If you know all the data science, come, or if you don't know all the data science, the biology, great, this is a mix of skills. making it appear. Right? So a lot of people can apply and try their luck in that process, and then because we got a lot of applicants, it wasn't overwhelming. We did read the whole panel. There was of us. Yeah. Reading every single one and grading every single one, and not in absolute terms, writing pros and cons, having conversations. These are advantages we have because we are very selective because we're a small institute. It's 135, 140 people or something, which again, for me, was quite large, but it's in real terms, quite manageable. And because from my perspective, I've been tasked with getting a team of four people. Fine. I can handle that over the course of a year. It's not too much. If I were in a place where there's natural turnover and we're hiring every quarter, we're hiring a couple of people. I couldn't have done that.

TIM: Do you find in New Zealand that the candidates applying are of any different quality? So if those 50 CVs, did it appear as though there was almost like a higher relevance than what there had been in other markets you'd hired in?

HERCULES: It's a great question. We had, so let me think about this one sec. We had a few people from overseas applying. And of course, if you're applying for a job in another country, you're really serious about it. So it's not going to be frivolous. So those people were top-notch. And we had then We had some junior people who were applying who were extremely good at their level, but they just weren't a senior enough level to be considered for this. Yeah, overall it was really high quality. And I don't know that I've thought about interpreting that, but there were basically no throwaway applications there.

TIM: Did you source them through the usual channels, like through LinkedIn or Seek or the equivalent?

HERCULES: Yes. So it was we posted the ad on Seek. So I posted a lot on LinkedIn and did the standard things, tagged my friends who might have friends, and again, networks, but online. because this one in particular was quite, there are a lot of different profiles that would fit the role. This is a data scientist in a biology lab. It's pretty. It's not a niche that exists, really, so I was very happy to adapt to the right person. The right person would manifest themselves, is what I thought, and they did. So then I tagged in my LinkedIn posts. I tagged people from biology, people from data science, people from data engineering, and people from startups, because that's very relevant in the sort of thing that we're doing. Then, you get a good enough reach that way, and you get really good candidates that way, and you get referrals that way.

TIM: It's interesting in the way that you've framed the role, which is that it's not necessarily A, B, C, X, Y, or Z. This is what we need. Which, under normal circumstances, I feel is the right way to go about it. Thinking in a lot of detail about what exactly you want, clear criteria. That way you can set up the process to then measure those things and not get derailed. But, it's almost like this role is so un, so atypical, it doesn't really fit a mold. And so you have to take this slightly more flexible approach where things developed as you were hiring, I imagine?

HERCULES: Yeah, quite a bit. So we wanted, we had a few things in mind that, so there's two of us, two science people, and if I could count myself as a science person anyway, and HR involved in this. So we're each bringing a different thing. So I'm thinking as a service org leader. or manager. I'm thinking, are the things we need? Then the scientist who is the client who's on the panel is thinking, this is what I need from this person. And the HR person is thinking, how did these skills tie up together? What is, let's think about this person's communication and all these kinds of really basic things that might be missed sometimes. So as we got these applicants through, yeah, we definitely were appreciating different things that they could do, but we had some real touchstones that we couldn't do without. So even though we were quite flexible, we needed the person coming in to code in Python. That was a real requirement because of various reasons. But being a, like, a proper coder, I put this in quotes because there is no such thing, but somebody who's been in production environments who understands how these things work, that was one touchstone. And then everything else was quite flexible. We wanted them to; we had a long wishlist of things. Hey, do you know how client work works? Hey, do you know how DevOps works? Hey, what kind of large have you delivered? Have you done any project management? So there were a lot of things that were important, and we were just going with essentially what wasn't explicit, but essentially a matrix where a person can code the way we need them to code and teach other people and be a set example. And then how many of these other things do they take? Okay, let's see how these people shape up. Let's invite. a few of them for an interview and see how we work together. And we also gave them not a coding exercise, but we gave them an assignment that we could talk about.

TIM: You mentioned you had a matrix, but it wasn't explicit in the sense that you weren't measuring them against each of these criteria? Or was it more like an on-off binary kind of classification?

HERCULES: Yeah, we—I was—it wasn't such a strict points ranking. It was a little bit, but it was basically, I'm blanking on what, though; I remember there were three basic things, and it was the Python coding was one of them, and there were two others that were not quite as important as that, but they were also basic. So essentially, to consider somebody to bring them for an interview, they'd have to fulfill all those things. And then we were a little more loose, but yeah, we were able to get. I think we've got five people in front for an interview in the room and figure out how they work and get a little bit of a vibe, obviously.

TIM: I'm interested in your thoughts on this, given how rapidly LLMs have advanced and their seeming ability to do coding pretty well. If you were to hire for this role again now, would you still have the hard Python requirement?

HERCULES: That's such a great question. This is why we talk about this all the time. And I talk about this with data scientists all the time and with people who teach Python or teach programming in general, where I think we're in this job, yes, I'd still want them to understand how coding works. So looking forward to the very near future where all code is just have to be testers and you don't actually need to do the coding. You just set some assertions, find niches, find exceptions, and all these kinds of things, and the machine codes for you. That's great because this role is also supposed to be somebody who trains people on how to think about coding. I would still want to hire somebody who can code; this has happened in every automation. You don't need to be able to do the process because the machine does the process, but you need to be able to fix the machine when it breaks, and to fix the machine, you need to understand the exact steps the machine has to take to make the widget. So we're still in that phase. We need one person like that in the team. and so I would hire the same way. But maybe what I might do differently today is to actually give them a coding assignment because they would do it with a co-pilot. I didn't want to give a coding assignment even though, yeah, co-pilots were still, were already pretty advanced to do that. But I wasn't interested in that knowing that a co-pilot will be available and they'll be getting better every day. I didn't really want to give somebody a coding assignment and say, Okay, go solve this problem, because it's not about the coding part. It's about the logic part, and it's about understanding the domain. So we gave him something that's a combined practices and quickly getting spooled up on the domain. That was the thing that's really crucial for this role. So that's a lucky thing that we were looking for somebody who would be a data scientist in a very niche domain. So we needed to see how quickly somebody could get spooled up with a domain enough to be useful. There isn't the expectation that you'll go learn genetics. In a week deliver a project, but you go out there and you read enough docs and you do enough co pilot coding to deliver something on the technical side of genomics, here's an algorithm that does the thing you asked for, is this good enough? Okay, great. So that's how we treated this.

TIM: I'm interested in those. I think you mentioned five candidates had gotten to that. Stage. Were they on a similar playing field in terms of that domain knowledge in that they all had to learn it from scratch, or did some already have some experience there?

HERCULES: We had a couple of people who were biologists. I think there were, yeah, I think there were two people on the shortlist who either had worked in. What was called bioinformatics, which is a word that kind of means data science, but it's got its own word because it's specific to biological domains, but it's also much more limited than data science because it is, essentially, domain expertise, I would say, in the bioinformatics domain somewhat. There are exceptions. But yeah, there were two people who had, who were, I think there was one biologist, one bioinformatician, and then other people who had some exposure to biology projects through work as data scientists, as consulting data scientists. I think two of them had that. yeah, so it was roughly half the people had some domain expertise. And yeah, the other half had none, essentially. Yeah.

TIM: all, and then somehow produce something of value in a week, thinking, Wow, that's pretty amazing. I can see their growth potential is high, or, even if their end result was not maybe as high quality as the ones who already had the domain knowledge, like, how did you think about that comparison?

HERCULES: That's where the assignment came in, because we didn't give them any coding. So we were trying to get it right? Okay, when I worked at Atlassian, there was a thing I thought about all the time. There was a long corridor that joined the Geo and Confluence teams at the time. And it was probably like 10 meters long or something, just a long corridor connecting two buildings. And over the whole 10 meters of it, it had progress over perfection on it, progress much greater than perfection. So I think about this all the time. I think about just delivery and iteration all the time. So that was one of the principles that kind of made me choose this assignment method, which was not a coding assignment. It was essentially a logic assignment. So it was something you could whiteboard and something that allowed us to really get into somebody's immediate problem-solving and comfort with domains. I got this from them; it's not my idea. I got this from somebody else, but I don't know if they want to be named. So I'll say. Thanks. But it was basically the niche that I'm in, which is a scientific domain, a research domain, but one that you might attract somebody who doesn't already know the underlying scientific domain. We gave them a paper that was a very technical paper, which was a techniques paper essentially, but specifically in biology. The paper is MetaCell, which is clustering but in a biologically specific way. So when you do. genomics, gene sequencing, and you identify genes, and you identify, and you have cells that have all these genes expressing or not expressing. And there are specific ways to cluster them and to get interesting results out of that. And MetaCell is an emerging technique in that area. And it's something we were interested in implementing in the lab. So I thought, great, let's get these people to read a summary paper for all the implementations of MetaCell. And I will make it really clear that, hey, don't go crazy. Spend a couple of hours reading this, and we're going to talk about these two things. So it was really basic stuff. How would you implement a pipeline, which you had to just read the paper? And if you could ask a question of the author of the paper, what would it be? That was the formula. That's not my formula. I took that from a friend. The great thing that happened is that then you don't actually need to spend your whole week on that, which is something I really don't like. I You know, I've never been in the position where I couldn't really do it as a job seeker, but every time I would do it, I would spend eight to 16 hours or a whole weekend on a coding assignment. I was thinking, what do I have? What do you have, kids? Do I have, or did I have, other responsibilities? No way. I now have a huge advantage over people in different circumstances. And I thought, what does this say about a place that. Is okay with that. What is it? What does it say? And okay. Yeah, anyway, because you can't control for people being in different circumstances, you just disqualify a whole bunch of people, essentially. And again, because we're doing this a year ago, whenever you can get a gen AI to write your whole assignment, that doesn't really matter. So what we really wanted to see was how they rationalized everything, how they, all the decisions they made, why they made them. And then we had people, we had two people come in with quite astonishing solutions to the thing. It was really impressive, really impressive. So we were even spoiled for choice. We had one person. How would you implement the thing? Obviously I would do the thing they say in the paper, except how about steps five and six? Why would they recommend that? Because I would do this. And it blew my mind when this person suggested it was not something I'd thought of. That's brilliant. Yeah, actually, that's a great point. That's a really good thing to implement. And then the other person who came in said, Look, this is how it is implemented; these are the steps. I actually wonder about this other thing and then pull out another paper from his satchel. Actually, I was reading one of the sources, and I was wondering about this. What does this mean? Does this mean what I think it means? I just look at the scientist next to me, who was on the panel. Oh my god, this is a little bit impressive. That's not to say that this person didn't spend endless hours on this, but that's what we were trying to diagnose, right? How quickly can you spool yourself up with the technical aspects of this? Don't get bogged down in the science, which you will not know. It's not the requirement for your role to understand what these genes do. It's the technical part. And that showed us, these two people showed us really clearly, they can focus on the technical bit, which is our role as a support team, as a science support team. And then they had great ideas, and then they had really good ways of communicating their ideas. So that's all we needed to know, because we're in an environment where the majority of people who applied clearly had and demonstrated the ability to learn new things. But that quickness, that, I read the thing you gave me, and I'm spooled up on the things I need to be spooled up on, which you can do in a couple of hours, even with a complicated paper.

TIM: Yeah, it shows that real laser focus and not being overwhelmed. a massively complicated paper, which I'm just trying to put myself in their shoes. What would my approach to that be? I would definitely overthink it. My starting point would be like, Oh my God, I'm going to have to go and research all these X, Y, Z things. If I were given that assignment right now, I'd probably go to Claude and Chachipiti and say

HERCULES: Yeah.

TIM: me some summary, teach me these concepts in like kindergarten terms or something. I'd be trying to. Go down that route. Did any candidates use, can you remember, an LLM to help them along the way to understand the paper?

HERCULES: As far as I know, they didn't. The people who really impressed us, I don't think they did. But I totally agree with you. And also just to put aside by that, made it clear that you don't need to understand the underlying science. So we did give him the instruction, but again, some people will get overwhelmed when we go overboard and try

TIM: Yep.

HERCULES: too much. So that's part of the test. But then also to just, to confirm, to just validate that I do that all the time. So if they had done that for the assignment, great. I do that on the job all the time because I don't know genetics. I don't know biology. I don't know immunology. When I'm in this context and I'm trying to manage a project I'm on, GVT 40 did a pretty good job of teaching me genomics, including the algorithms from the nineties that people use, including the specific packages that people use. It did a better job than courses I've taken. been really interesting. Somebody spent the better part of 10 years on sustainability. That's the carbon that I spent learning genomics was a little bit unacceptable, but if we can do that stuff in a more productive way and in a cleaner way. It's astonishing how much you can get and how, technically, people, how much of a shortcut you can get. Because if I'm working on a project that has deep domain, I just need somebody to get me unstuck for a second so I can do the technical bit, the job delivered, and frankly, do I need to remember this specific thing? in six months time? No, I don't. I just need to solve this problem. If it comes up again, then my memory will be rejigged, and I'll remember it the third time. If you need to remember a third time, it's important. But that's been the main way I've been using LLM foundation models and agents over the past year or so.

TIM: That's a good perspective you've given that you don't necessarily need to learn this. You don't need to put it in your hard drive. It's okay if it's in your RAM, basically. Ha.

HERCULES: There were times when people looked at me funny because I would forget basic things. Even when I was, I wrote a paper in astrophysics, and a colleague or somebody who's reviewing the paper before I submit would review a thing, would just catch a thing and say, Hey, how about this thing? How would you get around that? and I would say, I don't remember, but it's going to be in my notes. I'll come back to you later. and they thought I was the stupidest person alive, because how can you not remember a thing you put in your paper? I got the paper written. That was my objective. I had to get over that thing. I looked it up. It probably took me a while to look it up; I ticked that box. I said, This assertion is correct. I wrote down some reasons why I think this assertion is correct. So I can refer back to it, but I didn't have it off the top of my head. and that that was fine. The paper was fine. Yeah.

TIM: Copy, I guess it's good. I just think of myself as also used to writing a lot of Jira tickets for our product. And if I went back to it a month after I'd written it, I wouldn't know my logic of why it needed to be exactly the way I'd written it. But I know at the time. I'd solve the problem. There was a very good reason why it had to be XYZ. I can't necessarily remember it now, but as long as you get it down, that is all good.

HERCULES: Absolutely. Transcript.

TIM: One theme of our discussion so far has been

HERCULES: transcript.

TIM: You mentioned yourself as a scientist; you just have to give it, and you're going to have, you could be able to acquire all these new skills and knowledge. You'd have to go and do it, which is a great mindset. You've also mentioned. in this process, like

HERCULES: I'm

TIM: That skill, that mindset, is going to be even more crucial now. video as opposed to maybe some other bit of the world where you might've lived your 50 years in the dark ages and nothing really changed. Now we're at the opposite end of the spectrum. Things are changing all the time. Do we all have to learn and adapt even more quickly?

HERCULES: I think the thing we, especially people who learned in the previous version of this world, the pre-transformer world, think that we need to be really good at, get good at, is to not be precious about solutions and skills that we've acquired that become obsolete. That I noticed about 10 years ago when I was working at Atlassian, we were running a few things, and it was the emergence of certain auto ML things. It was a profit by then for Facebook. That was a wonderful AutoML suite. And of course, all of us people trained in proper stats and whatever, most people were not keen on profit taking away their enjoyment of writing a beautiful model. It's real I mean you hone these skills for so long then suddenly they just automate and that's that takes that can take away a lot of the Soon takes away the time for craft that we make we have to make for job satisfaction That's a big adjustment that I've had to make, and it takes away the satisfaction of being a person who knows how to do the thing. So I think, I would imagine, for people training now, junior people, I don't think that will be as overdeveloped as it has been. With my age, and I hope not because it's hard to shed, I'll tell you that. That was the test that convinced me and changed me. We ran, so there was this kind of auto ML; there was this task we needed doing. and it was something we could automate, and we, as a test for profit, this is probably 2016 now that I'm talking about, so quite a while ago. As a test, we're thinking, okay, maybe we can automate this when we can trust an algorithm, a gray box to do this, but maybe they, maybe the technology is not quite there yet. Let's see. But we went into it with open eyes and open minds. So I took on the task of making the old-fashioned model, the proper one, actually tune everything, actually roll it out, write all the code that automates the actual underlying machine learning process, testing, and do all kinds of stuff. Then the other data scientists on the team just did the profit implementation. Of course, he was done much faster. And, okay, there were edge cases where my bespoke algorithm was much better. But it wasn't really worth the time I spent on it, and it certainly wasn't going to be worth the maintenance of it, because I built the thing; I have to maintain the thing. You're outsourcing the maintenance of the thing, just deploying, and it's great. So that was the test that kind of sold me, and it well sold me. It convinced me that this is the world we're in now, and these things are going to get so much better. That's it. Now. I'm on the implementation side, and the important skill is figuring out how to put all these things together and not necessarily You know, it had already been some time since I'd written linear algebra. Maybe not. That's not true. I had, I remember coding regression models in SQL for some reason in a

TIM: Wow, that's tough.

HERCULES: We just had to. this really niche application. A proper developer and I locked ourselves in a room for three days and came up with horrendous stuff in linear algebra and SQL. We did it, so I was that close. That was maybe a couple of years detached from going down to a pretty low level to build these things. And then suddenly for that, an equivalent task of that, and a little bit superior in terms of complexity. Yeah. It's a one-liner now; just do it.

TIM: As someone who shuddered in their econometrics class back in undergrad, I, if anything, am happy with this development and how much easier it's become. I wonder if part of the challenge we're going to have to overcome is that people's identities are caught up in their jobs. And so people would say, What are you? I am a coder. If that's what you've been saying for 20 years, and now that bit of your job, which is according to the way you describe yourself, all of it, if that is surely on the precipice of being eliminated, then people are going to have to really just rethink mindsets and rethink who we are and what we do because if not, you're going to be pretty miserable, I would have thought.

HERCULES: Absolutely. I've definitely had to go through that adjustment, and I've done it a couple of times also because, you know, I moved from. research science to data science, which was already a big step. And now recently, the past few years, I am a data scientist. And I actually had a penny drop in my head just now, as you said that, because I've been thinking about this framing, which is a bit trite, but it's real that startup people talk about all the time, which is that you're not building a product; you're solving a problem. And I, that we're having this discussion, I know why that's on my mind as much because I've had to take other people on that journey as well to tell them that, hey, it doesn't matter for us to craft this beautiful code. We can take the shortcut because we're supporting me. We are actually here to solve this problem. I guess that's a really good framing for that now that I think of it, because I've had, again, I really had to let go of this craft person thing. And at some level, I make myself, my identity as a professional, as a problem solver; that's been the shift that works for me. Again, it's a, there's, we associate prestige with things and being a craftsperson, being a coder, being a data scientist, being somebody who does the, I don't know, even down to the linear algebra, that gives you prestige, that gives you a sort of thing, oh, this is a very smart person, this is because we operate in a, in tech, we operate in a domain where intelligence is currency, where really we make our living by solving problems, not by virtue of being smart, but not by token, if

TIM: Yeah, a bit of a dose of humility and a bit of a slap in the face maybe is helpful every now and then to bring us back down to earth.

HERCULES: Yeah. Yeah. No, I mean that, again, it's, for me, it's been easy because it's not been easy, but it's been, I've had the path because I did get those slaps in the face. I remember the first project I started running after I left academia, and I went so deep. Somebody asked me, Hey, can you make a little widget that predicts how much electricity somebody might generate with solar panels on their roof? This was quite a while ago. So now there are a hundred tools. I'll do that online for you because everybody's got survey data. Everybody knows what pitch your roof has and whatever. In 20, I don't know, 14, whatever, 15, there was no such tool. So I'm like, okay, great. How do I go about building this? And a week later, the product manager comes to check in with me, and I'm like knee-deep in building some GIS system for even fetching the data to begin with. It's whoa, man. No, like, just let's talk next week and just tell me if we can make it. Don't make it. Just tell me if we can make it. So that shift, that was week one in my first kind of real job, so to speak, outside academia. And immediately I'm like, oh, this is a different context. So that really took me and put me in a different mode. Then the technology is changing again; give me the next, yeah, the next little bit of a wake-up call. So now I'm pretty, let's solve this problem in the best way possible. And if that's paying a couple of bucks for a thing that already exists, obviously, and it's cheaper than building it ourselves, that's a no-brainer.

TIM: Yeah. And you mentioned startups, and that's the mindset you always have to have, is it? What? Oh my God. Why build it ourselves? If there's just a package we can install, a company I used to work at, there was a phrase that was used slightly sarcastically there, which is. N B H is not built here because it's like, Just please, we have to make it ourselves. No matter what it is, we're going to build our own email service, and we have to. I'm sure that mentality has changed these days, but it's so easy to get stuck in that mindset. And if you just think about you're solving the end problem for your end user and customer, that gives you a sense of clarity and makes everything easier.

HERCULES: It's easy to do that for me now. It's easy because we're a service team. We're a science support team. Satisfaction also is, it's an alignment. It's a classic alignment thing. Incentives are aligned. The customer wants a solution to their problem. If we solve more problems, we get more job satisfaction. It's a pretty nice alignment. But it always creeps in for me as well. Yeah, I can import this package up. Wouldn't it be nice to write our own thing because we can do this niche little thing? And you do consider it. You always consider making your own stuff. We do make a lot of our own stuff because sometimes even the packaged things are too fine for a niche. Like that also happens, though. The package that you get off the shelf is way too much in a specific paradigm that we don't think has legs. For more than a year or two. So yeah, okay. We'll build our own, but it doesn't, hasn't gone away, but you have to shift your criteria a little bit.

TIM: Hercules is a final question today. If you could ask our next guest one question. What question would that be?

HERCULES: I'm being a little scientific, and I have to say, who's the next guest, and what domain are they in, and all that? No, I'm kidding. I would ask, Oh, I am such a novice thinker. Tim, would you give me the hardest assignment now? Because I've already gone into what kind of people will be on this podcast and what kind of things will they know? What would I ask the next guest? How are you doing? How's the world for you? Has everything turned upside down? I think I would ask that first.

TIM: Yeah.

HERCULES: See where I go. I just talk to people all the time who are like, I don't know how this is going to turn out. I don't know how this is going to turn out. I talk to people who are anxiously anticipating AGI so they can slot their startup into it. So I'd ask, has everything changed for you? Has your entire world changed in the past week or month?

TIM: Wow. That's a great one. Okay, I will level that at our next guest, whoever that may be, and see what their reaction is. I guess they'll give me a profound answer. Maybe it's my feeling. Anyway, . Hercules, it's been a great chat today. We've covered a lot of ground. We've taken a trip around the world, learning about you and your background and all your different experiences, and you've given us a lot of food for thought and a lot of insights. So thank you so much again for joining us. for joining us.

HERCULES: Thank you, Tim, it's been a pleasure chatting with you.