
16 Nov 2026 · 44m · 17 chapters
How data scientists can stay relevant as AI capabilities accelerate, arguing that technical coding moats are shrinking while “hard-to-automate” skills (judgment, stakeholder communication, messy domain expertise) grow in value. Krohn also discusses AI’s impact on job boundaries (data science vs product/design/engineering), the importance of mathematical foundations for credibility, and how he uses AI selectively in podcast production.
Guest background (Dr. Jon/John Krohn): Oxford neuroscience PhD (computational ML on genetic/brain imaging data). Host of Super Data Science (1000+ episodes). Co-founder/CEO of Y-Carrot (consultancy deploying agentic AI in production). Fellow at Lightning AI. Author of Deep Learning Illustrated; edits Pearson’s AI Signature series; created ML foundations curricula (YouTube/O’Reilly/Udemy). Keynote speaker at ODSC West.
LLMs automate coding/modeling faster/cheaper; they excel where right answers and simulated data exist (math/programming). Data scientists should move up the value chain toward product-like thinking and context/question formulation.
ODSC West talk on AI capability doubling ~every seven months; references to Richard Socher’s The Eureka Machine and Zach Kass’s The Next Renaissance; examples of using LLMs for episode summaries but keeping human video editing.
Written by AI. May contain mistakes.
Discussion about the Halloween party at ODSC West and Dr. Krohn's costume.
“You may know him as the host of Super Data Science, which is one of the most listened to podcasts in data science, which recently passed its thousandth episode.”
Dr. Krohn shares insights from his keynote on AI's rapid development and its implications for data scientists.
“Yeah, so there was a, historically, I'm used to there being a Halloween party at ODSC West.”
Exploration of the differences between data science and software engineering in the context of AI disruption.
“But that's really only in these spaces where you can simulate tons of high quality training data, because you know that if the math works out, the data are right.”
Discussion on how the roles of data scientists are changing and the need for them to adapt.
“And now so many of those technical aspects of the job can be done.”
Explore how data scientists can enhance their roles with LLMs following ChatGPT's release.
“What are the kinds of things that my customers would like?”
Discuss the necessity of strong mathematical foundations in data science careers.
“You have all these technical skills that you've built up.”
Understand the unique position of data scientists as communicators between tech and business.
“I think that's probably where I get that kind of content covered.”
Learn about essential skills for data scientists that are difficult to automate.
“you're also touching on the importance of like communicating with business stakeholders who I'm sure themselves are by and large, not deep technical experts, but they're relying on these sorts of systems.”
Discover Dr. Krohn's journey from neuroscience to data science and the pivotal decisions he faced.
“I mentioned you have a neuroscience PhD from Oxford, and you're nodding.”
Hear about Dr. Krohn's decision-making process in moving from academia to the tech industry.
“And so for my undergrad, very different from my grad experience, for my undergrad, I lived monastically, did every homework question that was assigned.”
Discussion on the relevance of PhD training and skills in the current data science landscape.
“Even your average data scientist off the street is not a neuroscientist by training.”
Discussing the role of education and persistence in becoming a successful data scientist.
“look more impressive than maybe they actually are in terms of your success in your long-term career?”
Sharing personal experiences regarding the challenges and motivations during the PhD process.
“And it took me, I had one year of master's funding because in England, master's are just one year.”
Exploring how podcasts can facilitate learning and networking within the data science community.
“That is a great pivot to the last thing I wanted to ask you about and talk about a little bit, which is the podcast.”
Discussing the balance between AI tools and human expertise in podcast production.
“And I know you have a way of thinking about how to prepare for those interviews so that they're good, interesting conversations.”
Understanding how AI enhances the research and content creation process for podcasts.
“I don't use a huge amount of AI myself for the show.”
Discover the value of listener feedback and connections in the podcasting world.
“I was about to say, I don't think you know this because how would you, but occasionally I get emails from folks who listen to Linear Digressions.”

Get the gist, explore an idea or ask about something said.
On Linear Digressions, what are the key points of this episode?
Automatic transcript. May contain errors.
0:00Katie Malone: Hi, welcome to Linear Digressions. I've got a treat for you today, an interview with Dr. Jon Krohn, one of my fellow podcasters and a really interesting voice out in the data science circuit. We had a great conversation about how to stay relevant as a data scientist in the age of AI, how John thinks about his career and the advice he gives to others about theirs, how and why he does his podcast, and how he went from a neuroscience PhD to teaching and consulting in data science. It was a great conversation. I'm really excited to bring it to you today. You are listening to Linear Digressions.
0:42Katie Malone: Welcome to Linear Digressions. I'm very excited today to have a guest on. My guest today is Dr. John Krohn. You may know him as the host of Super Data Science, which is one of the most listened to podcasts in data science, which recently passed its thousandth episode. So very impressive and congratulations. organizations. That is just one of his many accomplishments. He's also the co-founder of Y-Carrot, which is a consultancy that helps companies put agentic AI into production, and a fellow at Lightning AI. And he also is someone that you may know from a large amount of educational outreach that he's done.
1:20Katie Malone: So this includes being the author of the best-selling book, Deep Learning Illustrated. He edits Pearson's AI Signature series of books. He's done a machine learning foundations curriculum that has taught, I think it's fair to say, hundreds of thousands of people the math behind machine learning through YouTube, O 'Reilly, and Udemy. You may have also seen him on Bloomberg TV, CBS, read him in the New York Times. He is fresh off of a keynote at the Open Data Science Conference, AI West. And he came to all of this by way of a PhD in neuroscience from Oxford, where he used machine learning to study genetic and brain imaging data.
1:57Katie Malone: But John, it is a pleasure to have you. It's a pleasure to be here. This is likewise one of the most listened to podcasts in our field as well. I think we're neck and neck. So a real honor, Katie. Every time I talk to John, we just high five each other about what big fans of each other's podcasting work. Here's another one. That was a great intro. You nailed it. Thank you. At some point, I'm hoping to talk a little bit about full disclosure here. I got plenty of AI help on this. And I know that's something that you also think about how to incorporate AI into your podcasting process as well. We will get there.
2:31Katie Malone: I wanted to actually start with a question I mentioned in the intro that you just wrapped up a little bit of a conference tour. You were out in San Francisco, if I'm not mistaken, giving a keynote at ODSC West. That was as it happened over Halloween weekend. And so my question is, did you go to the Halloween party? And if so, what was your costume? Yeah, so there was a, historically, I'm used to there being a Halloween party at ODSC West. But I guess that's because often October 31st coincides with one of the days of the conference. But this year it didn't. And so the kind of the main night for going out and having drinks all together at the conference, it was the Wednesday night.
3:15I showed up in a costume and nobody else did. what was the costume so there is a irish hip-hop group called kneecap if you want that is a deep cut okay okay yeah and i don't know they have uh it's a very easy costume of the three of them there's two front men and then a dj in the back and the dj in the back he wears an adidas track suit and a irish flag bella clava and like a rain jacket that's what a bell clava is right No, Bella Claw is like those masks that like burglars have with just like the eyes and the mouth cut out.
3:52Katie Malone: Okay, that's a different image. All right. Very good. That sounds distinctive. So after you had your beers and you wore your Irish mask, not a raincoat, you had to give a talk. So the title of your talk was, the subtitle I think was really what popped out to me. It was basically like how to thrive in a world when the capabilities of AI are roughly doubling every seven months. I would love to hear what the thesis of that talk was. What were some of the points that you thought were most important to make as you were talking about this subject? For sure. So a lot of the talk was about how crazy the pace of AI is moving at, which is obvious to a lot of us in the space.
4:37But I was quantifying it with charts. So like charts from meter, for example, as well as artificial analysis,.ai, and asking questions about where is the solid ground for us? since the release of Opus 4.6 in February, if I remember correctly, you know, a technical moat in terms of, you know, writing code in terms of a machine learning model. So much of what we do as data scientists or AI engineers is now done faster, cheaper, better by LLMs. And so where does that of leave us is the point that I was trying to get to. And I think we're lucky in data science relative to maybe software engineering, because I think there's a lot of gotchas that LLMs don't seem to get around.
5:30And I don't know why this is. It might be the fact that data science is messy. And I think the places where LLMs have been able to really excel is situations where there's a definitive right answer and you can simulate data effectively. So in math, like we've seen with tons of problems that stumped humans for decades or centuries being solved in 2026. So math, programming, those are fields where you can simulate huge amounts of data. And that's what's allowed the task length, like if a human was doing the task, the task length that LLMs can competently handle doubles every seven months. And that stretches back many years now.
6:16But that's really only in these spaces where you can simulate tons of high quality training data, because you know that if the math works out, the data are right. And same kind of thing for computer science. But I guess data science is messier. There's a bit more for us to do as humans still in that space. But in general, I think people need to be investing in hard to automate skills. So things like judgment, stakeholder communication, and I guess I was just alluding to this, but like messy real world domain expertise. I think a lot of what I end up doing today in a given work week is thinking up what the right context is and what the right questions are to be getting an agent or to be getting a conversational interface to be able to do the work that I wanted to do accurately.
7:08Katie Malone: There's a lot in there. I'm really excited to dive in and unpack some of that. I think one of the first questions I want to ask is, as you and our listeners may know, I live in Chicago, which I think of as a reasonably well-connected city, but it's not the white-hot center of San Francisco. I spent some time in the Bay Area a while ago, so I know that there's just a little bit of a different vibe that sometimes you get being in a place like that relative to anywhere else that you might be. So I'm a little bit curious. You yourself are not a citizen of the Bay Area. What is the vibe in the room like in the data science community in Silicon Valley right now in particular?
7:45Katie Malone: Is it like this is the greatest thing ever? Or is it like the sinking feeling in the pit of the stomach? I'm not going to have a job. Is it like my P-doom is 60 % and I'm scared, but in a much more existential way? When you read the room in there, what's the vibe? Yeah, I think there's a lot of different perspectives. Some people are really concerned about human extinction. So the P-Doom kind of stuff. A lot of people are really excited. I tried to highlight that in my talk. So for example, just a few weeks ago on my podcast, I had Richard Socher, who is an iconic data scientist and entrepreneur.
8:21And he recently released and had a national bestseller already with a book called The Eureka Machine. and it's an optimistic book on how having recursive self-improvement, artificial super intelligence will lead to basically a century of innovation in a decade is the idea. And similarly, another really positive view if people want it, a little while ago on my show, I had a guy named Zach Kass and he wrote a book called The Next Renaissance, which was also a national bestseller. And it's a broader book trying to cover lots of different areas, education, healthcare, and so on, but basically being very optimistic about where things are going.
9:06And so I'm in that optimistic camp. And so I look for confirmation bias wherever I can. And those are two of my best examples.
9:14Katie Malone: Very fair. Very fair. Another thing that you mentioned that I want to pull on a little bit is how data science is maybe not a one-for-one analog of a field like software engineering when it comes to the exact ways in which it might be exposed to AI, disrupted by AI, augmented by AI. I'm curious your take on a particular jump-off question from that. Something that I see in the conversation right now in the software engineering circles especially is, or maybe not software engineering exclusively, but it's kind of the trifecta of software engineers, product managers, and designers. These three roles that very often work together in these cross-functional teams and have for quite a while.
10:02Katie Malone: And feeling like the lines between those roles are blurring. There's a lot of teams that I've worked in, and you've probably seen as well, where it's not a team of three, but it's a team of four. And the fourth one is the data scientist. And so I have a hunch, but I'm curious if you think this as well, or if it's a little bit of a different picture that the data scientists are also feeling kind of this, this fuzziness around the borders of what their jobs are, they can do all this stuff that maybe they couldn't have before, because they have these tools. But there's also some amount of everyone else thinks that they can do data science now because they've got cloud code or they've got codecs.
10:40And so there's both a lot of opportunity.
10:43Katie Malone: There's a lot of uncertainty. I'm wondering if you think that it's kind of symmetric and the data scientists are also in that mix as well, or if there's something that makes them unique in any ways in that picture. I think it is scary for data scientists in the sense that, like I, for example, spent 20 years building a technical moat, being able to write code in Python and R and MATLAB and getting into the weeds of machine learning, deep learning, statistics, the command line. And now so many of those technical aspects of the job can be done. You really should be leveraging LLMs to speed up aspects of what you're doing.
11:24wouldn't make sense today to be typing each character of the Python code that you write. That would be, you'd really be shooting yourself in the foot in that scenario. And so I think that we have a lot of, it's an unnerving feeling to have spent all this time and developed all this expertise and now know that, you know, you're just a clawed query away from having all of the world's expertise on a given topic brought to you. but going back to the same points that I was making earlier I think that there are things around thinking through what the right context is thinking through what the right questions are and to be able to do those kinds of things in order to get the LLM to do effective work we need to have experience in the field and you and I have a lot of experience in this space I think what's going to be interesting and this is true in software development I guess a lot of different jobs out there is how are junior people going to develop the kind of intuition around the various aspects of expertise in their field when they can jump right to asking Claude to just give them an answer.
12:36Katie Malone: Do you have any theories of the case about what that answer might be or are you as scratching your head as much as maybe the rest of us on that one? I hope that there are university programs that are preparing students for the workforce in a world where we have LLMs, but simultaneously when they're doing their studies, they get forced to handwrite tests as opposed to just have everything be at home projects. So that hopefully, at least in their education and say a data science education, people are coming out of that undergraduate degree or the graduate studies with real knowledge. And then, yeah, I don't know.
13:17Hopefully there are senior people in their organizations that can push them to be thinking and not using LLMs in the right places around having the overall judgment of what a product or what a project should be like. And then I think there are parts of the job that are inevitably that you're still going to be able to develop a mode around. And I think we just move up the value chain a bit. So instead of now being really good at typing individual characters of Python, I think it's really important now to be, even if you're a data scientist, you need to be thinking more like a product manager. You need to be thinking about what are the kinds of things that users would like in my platform?
14:03What are the kinds of things that my customers would like? What would the internal stakeholders in my business, like from my data science team, and be thinking bigger picture, and then leveraging LLMs to be doing a lot more. And so what we've seen since the release of ChatGPT is that even though a lot of the technical aspects of our jobs have been automated, there's actually been an increase in mathematical sciences occupations, computer science occupations. We've seen an increase in employment since ChatGPT came out. And I think part of the answer there is that now when you hire a data scientist, that data scientist can make so much more impact than ever before by leveraging these automated tools.
14:48Katie Malone: And so for your own work, I read in the intro some of the highlights of a pretty extensive career and honestly, the portfolio of work that you've done. You referred to it a little bit as your moat. You have all these technical skills that you've built up. For you in particular, you've chosen to take those back out to the community, pay it back forward to the next generation by creating books and educational materials, course content. That makes a ton of sense as a way to teach mathematical foundations. I think there's an argument to be made that those mathematical foundations stay important to understand.
15:25Katie Malone: And I think that was maybe a little bit what you were saying, that developing the taste without maybe some of that prerequisite struggle is, doesn't, there's plenty of people, and I might count myself among them that thinks of that as like those two pieces going together. I'm curious if that's the way you think about it, that you think about these core technical concepts as staying evergreen. And, you know, you clearly love and are very talented at that. Or if in your own material that you produce and your focus, your attention as someone who's in these conversations, if you're like, oh, maybe I need to talk a little bit more about what product management taste means instead of another explainer video about linear algebra or something to pick two extremes.
16:07Yeah, I guess there's lots of different content creators out there through the podcast. We have 104 episodes a year. And so there's that kind of stuff. I think that high level stuff that you can get across in a conversation with a guest. I think that's probably where I get that kind of content covered. For me, all of the content that I've created, it's very nice of you to say that, oh, I'm like giving back. But really, the reason why I do it is super selfish. You're very candid about this.
16:39Katie Malone: I find it quite refreshing. Yeah. Because it's stuff that I want to learn. And so by committing to having a video course done by a particular date, by committing to writing a particular book, it forces me to really understand this stuff that I think is important for me to know well in this job. And I do think that understanding the mathematical foundations of what we're doing, the linear algebra, the calculus, the probability, as a way of understanding how the machine learning models work, how LLMs work, I think that is valuable. Think about if you had two interviewees for a job, and you're hoping that both of them are going to be using LLMs on the job.
17:20But one of them is just has no understanding of the underlying fundamentals of what they're doing. And they're just trusting output. And then the other candidate has a deep understanding of what's going on under the hood and can present confidently the results of some LLM output. You can say, okay, this makes sense because of reasons X, Y, Z. and I think that I think it's obvious that you'd pick the person with a deep technical understanding and I think it's essential because once the person is in the job they're going to need to be convincing people internally of big strategic decisions based on data or based on models and you're only going to really be able to do that if you understand what's going on under the hood so yeah I guess that's the end of my point I thought I had one other thing, but it's gone.
18:19Katie Malone: And so tying together a couple of things that we've touched on in this conversation, you said in one of your first responses that one of the things that data scientists are maybe uniquely comfortable with or uniquely good at is understanding that we're working with messy systems and business data and domain knowledge is just fuzzy and not verifiable in the same way that deterministic software is. you're also touching on the importance of like communicating with business stakeholders who I'm sure themselves are by and large, not deep technical experts, but they're relying on these sorts of systems.
18:57Katie Malone: So it sounds like I don't want to put words in your mouth, but I think there's like a connection there. And I'm curious if you want to pull on the thread a little bit here with me that maybe one of the unique value propositions that data scientists have in a world where their moat is no longer, I am the best at typity typing the Python, is in kind of this communicator role between the machine, especially like the internals of the machine, not the polished chat GPT surface, but thinking about it as the probabilities flowing through the matrices or the bits flowing through the wires or what have you.
19:31Katie Malone: Between them and people, business problems, stuff that has to get done in the other world. Those are two like complicated things and the interface between them is messy. I think that's where the opportunity is. That's where the challenge is. But it's just something I've been thinking about a lot lately with the data science community. Yeah. What is the value proposition? I think there's a lot of need there. And there's people that I think are comfortable at that interface in data science. And yeah, if you can find a way to have a conversation that brings those two together, I think that's a really interesting and valuable thing to do these days.
20:06For sure. I agree with you 100%. And I think I'll keep my answer on this one short, because this kind of just goes back to what I said right at the very beginning when I was when you asked me about the talk that I gave at ODSC AI West last week in San Francisco. And I said, the best thing that you can do is invest in hard to automate skills. And two of those things are stakeholder communication, and messy real world domain expertise. And so what you're saying there's like the blend of those two, the interface of those two. And that's a great way of thinking about it.
20:34Katie Malone: I want to pivot now a little bit and dive into your background a little bit, because I think it's very interesting. I mentioned you have a neuroscience PhD from Oxford, and you're nodding. So I'm glad my agent didn't send me in the wrong direction on that one. It did tell me something quite confidently that I believe is incorrect about you. I'll get to that in just a minute. But I'm interested in, I don't know, your origin story, such as it were. So how did you end up going from what was maybe an academic science track. I imagine that's maybe the plurality of where maybe a neuroscience PhD student, especially out of a prestigious program, I'm sure you had opportunities in academia.
21:13Katie Malone: You ended up in data science, now data science consulting, education, outreach, communications. What did that arc look like? yeah when I was an undergrad student I wanted to go to med school afterward and so my very first week at undergrad I went to I don't know is this vice dean for student services or something for the faculty of science and I said I want to go to med school what are the kinds of things I need to do and he said there are several hundred students starting in science every year at our university and of those, half a dozen will get into a Canadian medical school. I grew up in Canada.
22:00And then he was like, I probably shouldn't show you this. It's technically confidential, but he brings a binder off the shelf with transcripts in it. And he says, this is the transcript of one of the students that got into med school last year. As you can see, it's straight days. And so for my undergrad, very different from my grad experience, for my undergrad, I lived monastically, did every homework question that was assigned. I never missed a minute of any class in the entire four years of my undergrad. And I did all the other kinds of things that you needed to do to be getting into med school, volunteering in hospitals.
22:36But I also was doing research. I got grants from the Canadian government, these undergraduate research grants that allowed me to do really cool research, fMRI research. There was like neurosurgery research. And through that, I ended up, when I was getting close to graduating, one of the professors at my undergrad institution said, hey, next week, I'm going back to England where I'm from. I'm going to visit my family. Would you like to meet a bunch of professors at University College London and Oxford, which are UCL in particular is one of the best neuroscience schools in the world. And I was doing a neuroscience undergrad.
23:14I was always fascinated with how mental experience, decisions, perceptions, everything that we know about the world as an individual arises from molecules. So I was interested in that as an undergrad. And I also knew that there would be medical specializations that you could get into from that. And yeah, that's how I ended up at Oxford, got a full scholarship there. And I thought if I got a full scholarship for a master's and a PhD at Oxford, it'd be crazy not to do it even though what I really wanted to do was go to med school I was like I can go to med school afterward but by the time I got near the end of the PhD enough of my friends had actually finished med school and were doing residencies that kind of thing and they basically talked me out of it out of doing medicine they were like it's not as rewarding as you think and I don't I'm sure there's lots of doctors out there and hopefully even some listening that do actually find the job really rewarding.
24:09But the point that basically came across to me is that a huge portion of your patients that you see have like multiple comorbidities. They need to be dieting. They need to be changing their diet. They need to be exercising. They need to stop drinking. They need to stop smoking. And I don't know, the kind of the consensus from my friends was that 90 % of your patients just aren't going to change and they're just going to continue to have these problems. And so it's a tough, a tough spot to be. Yeah. So as I got near the end of the PhD, I was like, so maybe I won't do medicine or maybe I'll think about doing that later.
24:43I now really regret not staying in academia. And something that isn't very prominent online is that I am starting to do more stuff in academia. For example, I'm now adjunct faculty at Tulane University in New Orleans, as well as at Wilfrid Laurier University in Canada. And so yeah, I'm trying to get back into that world a little bit. But I basically, I just, I kind of followed the money. Like I felt looking at, it's obviously extremely competitive to be getting faculty roles in academia. And even when you get there, they don't pay very well. If you become a top researcher, you're traveling all over the world, but you don't actually necessarily bring that much home in terms of being able to provide for your family, say.
25:29And so I, I kind of, I became a trader at a hedge fund in New York, like using, because while I did a neuroscience PhD, all of it was computational statistics and machine learning applied to neuroscience data sets. So that was something that was in high demand at hedge funds in New York. So that's how I ended up in New York where I still live today, like 13, 14 years later. and yeah but I actually two years of doing the trading I hated it I couldn't stay motivated about making money for its own sake and so I quit and I was like I'm finally gonna go to med school I literally said it in my resignation letter I was like I'm moving back to Canada I'm going to med school but I still had a month lease left on my apartment so I hung around it in New York and I met someone out at a pub who had this job data scientist that I'd never heard of before that I was already qualified for that paid well where I could be working even for a healthcare tech startup or something like that and so I thought I might as well give that a shot and
26:33Katie Malone: that's what I've been doing since yeah so follow-up question then so you I don't think that your path into data science was like a completely insane one but it's maybe safe to say maybe not the most standard one. Even your average data scientist off the street is not a neuroscientist by training. And I'm curious that what your thoughts are now that you've gone through more of your career, that you're in this moment in AI in particular, where you're having to figure out what you're going to do, how you're going to pivot, what it means to stay relevant. I'm curious how often you reflect back on the PhD experience and maybe even your specific training and find that to have ways that it prepared you for this career that you maybe never even anticipated.
27:25Katie Malone: And I think there's a little bit of a version of this question here for other folks who might be listening to this right now that might have a little bit of imposter syndrome themselves. Maybe they're working in data science, maybe they're not, but they're generally thinking, I'm interested in what's going on in this field, but I don't feel like I am the pedigree of someone that typically makes it as a data scientist or AI researcher. What do you think are the truly important skills from your background that have enabled you to thrive in this career? And what's stuff that looks nice on a resume that I'm sure there's people who've maybe given you a second interview or called you back because of the line, but things that look more impressive than maybe they actually are in terms of your success in your long-term career?
28:07Yeah, I think if you want to be a data scientist doing an undergrad or even better grad school in a quantitative discipline, it could be anything. It could be physics. It could be neuroscience. as long as you're getting comfortable with data, cleaning data, running data through models, having a good sense of what the right modeling approach is, where the risks are in different approaches that you take, just getting that experience through whatever quantitative discipline you study. I don't think there's no reason why you need to be doing a data science undergrad or a data science grad program necessarily.
28:48You just need to be comfortable manipulating data and running it through models.
28:53Katie Malone: Sometimes I have a PhD too, and sometimes people mistakenly, I think, assume that means I'm smart. I like to think of myself as smart, but I don't know that that's the defining characteristic of someone who has a PhD. I think a PhD is much more about persistence and continuing to carry on in the face of not necessarily incredible odds stacked against you, but it's a very entrepreneurial experience in many cases, or it can be like a bit lonely. It's a lot about that internal motivation. And yeah, I, at least that was like my experience. And I wonder if that is maybe also, especially as you're now kind of a independent operator, like you're doing all of these things, you've got very much your own career that you've put together out of a bunch of different initiatives.
29:35Katie Malone: You're not working at a company where you show up each day and somebody tells you what to do. I wonder if you look back at that, at the PhD, giving you some training for that particular aspect of your career? I'd never thought about it, but you could be right. The PhD was definitely extremely challenging to keep moving through at some times. I am not an introvert. I need some time alone every day, but mostly I'm an extrovert. and the long hours at that time, typing every character by hand of my code. And, you know, I would be as social as I could be in the lab environment, but writing for hours on end was definitely a struggle.
30:20And it took me, I had one year of master's funding because in England, master's are just one year. And then three years of PhD funding, because again, in England, it isn't uncommon to finish your PhD in three years. but not only did I take a fourth year as like a victory lap, like as a full-time student, uh, on top of that, I also, it took me like another year. I was already working as a trader at the hedge fund, finishing up writing my, my dissertation on weekends, all of the experimental data were collected, but I hadn't finished all the chapters of my dissertation. And so I guess I'm, The reason why I'm saying that is that it can be really hard for some personality types to stay motivated about that PhD experience.
31:07And I remember there was one point when I left Oxford and moved to New York. My mom was she was like, oh, my God, you're not going to finish your PhD. She was like, you're just not going to. I was like, I'm going to do it. I was like, I'm going to find a way I'm going to work on it on weekends. And I've had, I've encouraged a lot of people who have gotten near the end of PhDs and feel like dropping it. I'm like, you've got to push. I understand that now you're working full time, but get this dissertation done because you're going to be so happy to have those post-nominals and pre-nominals. It is distinctive.
31:43It's like only 1 % of people have that kind of, have a PhD. And yeah, so stick to it, keep going. And yeah, I guess I have the same kind of, for better or for worse. Sometimes I think it's a disadvantage. It's like sticking to the podcast, doing 104 episodes a year for so many years. I'm committed to that. I'm going to keep doing it indefinitely, but I'm like, maybe life would be better or bigger opportunities would come up if I wasn't doing that. But yeah, I don't know. I just stick to things. I just stick to things. I just keep going.
32:18Katie Malone: That is a great pivot to the last thing I wanted to ask you about and talk about a little bit, which is the podcast. I mentioned a moment ago, so I was doing a bit of work to prepare for this. I was familiar with your work before, but you had given interviews that I hadn't read and written stuff that I hadn't browsed. So anyway, I asked, as one does, my friendly local agent to do a little bit of research on you. And it very confidently said, told me a few things that I think are correct and that are interesting, one of which I'm almost certain is wrong. so interesting things I learned about you is that you participate in CrossFit I think quite enthusiastically yeah I did and actually something I've never talked about on air in February of last year February 2025 I was in a really bad skiing accident and I ripped a bunch of the nerves in my left shoulder and so for 11 months I couldn't flex my bicep or my deltoid in my left arm.
33:19And so I'm, but before that I was a very avid CrossFitter. I would spend about 10 hours a week training for it and compete, not to compete at any level where you can make money, but just for fun, just to have something to stay motivated about. And that was a huge part of my life. And yeah, I'm starting to be able to do many of the CrossFit movements again. And this year I've been, I've been doing the high rocks thing that is.
33:47Katie Malone: I've heard about this. I have a buddy who's really into this right now. So we'll compare notes. It has enthusiastic participants, let's say. Another thing though, that I'm fairly confident you do not do is a standup comedy. Wait, my AI agent had several rounds of questions that it wanted me to ask you about standup comedy. And I was like, are you sure? And then it did a little bit more research and it said, oh, you're absolutely correct. And that's my bad. That's on me as AI agents do. My question for you based off of this is you, I think, also you do a lot of interviews. I'll say that with confidence.
34:22Katie Malone: And I know you have a way of thinking about how to prepare for those interviews so that they're good, interesting conversations. I'm sure you have also seen your fair share of AI agents being AI agents. So it's certainly not, I would think, something that you can completely delegate and have it go perfectly every time. So I would just love to understand a little bit how you approach that kind of managerial task, finding the places where AI is a benefit for you and how you think it opens you up to do more with the podcast versus the places where you do keep your hands on the reins through the AI.
34:58Katie Malone: And I also know that you work with a production team that's really excellent and that are really some of the best folks in the business for helping get those 104 episodes out every week. So, yeah, what does that look like for you in a given week? I think this is a good example of how data scientists will, I think, continue to be in really high demand, maybe even in greater demand as the technical aspects of our job are automated. because similarly, I find people who are great at some specialization, and I hope that they are using AI tools to automate parts of what they're doing. And I know a lot of the people that do work for us are using AI tools to automate aspects of their work.
35:41But simultaneously, I want to hire a human expert to be doing the video editing, to be creating the animated YouTube shorts, to be managing the operations of the show. Theoretically, maybe a lot of that could be fully automated with agents, but I love having people do it because I then know that their years of experience on top of the AI tools that they're using to automate aspects of what they're doing means that I don't have to worry about whether things are going right. And I don't have to be keeping tabs on what's changing in video editing or YouTube short animation or operations management.
36:25I can just leave it and go. It's kind of similar to AI tools make it theoretically possible for somebody to build all the core functionality of Salesforce or Slack and have that is an internal tool at their business. But why on earth would you do that when it's like 20 bucks a month per head for a lot of these SaaS tools? So yeah, I think that there's still specialized businesses, specialized people. I hope that they're all using AI to do their jobs better. But yeah, so that's mostly it. I don't use a huge amount of AI myself for the show. But I know that we do use AI for tons of things like the episode summary that goes up on the web page.
37:13It's just a paragraph summarizing the whole conversation. LLMs are amazing at that. So use an LLM.
37:19Katie Malone: And so in a world where I think a couple of things you really enjoy about the podcast, one of them you've already said a little bit indirectly, you're an extrovert. And in one of our previous conversations, you just love being able to talk to some of these really interesting folks. And then the other piece of it is as an excuse to go learn stuff. And so are those places where you have kept the AI out of the process a little bit? Because it feels you feel like the interaction with the people or with the content is best when it's just those two pieces together without AI intermediating it? Or have you found that there's particular ways that having AI in the process for you actively gives a little bit of extra gasoline in some way?
38:02Like in terms of the research for guests?
38:04Katie Malone: Mostly, yeah. In terms of the research for guests, in terms of, I could imagine the research for topics that you want to talk about, technical topics, reading a paper, you in the paper versus you in the paper and chat GPT, that sort of thing. Yeah, yeah. So most of our episodes on the Super Data Science Podcast have guests. And when we do have guests, I have a human researcher, Serge Massese, who's an amazing data scientist in his own right, creates lots of content. and we are super lucky to have him and his agentic process, which he's done conference talks on. So he has an agentic process for how he does research on guests for the Super Data Science podcast.
38:42And I can find the link for you to put in the show notes if you want.
38:47Katie Malone: I would be, I'll take you up on that. And I'm also just genuinely curious about this as a quick aside, and then please continue. But I had the privilege of being on your podcast not too long ago. And I was surprised and impressed. There were some deep cuts that you asked me about that came up in the research process. So I remember walking out of there being like, how did they do that? Anyway, please continue. Well, Linear Digressions listeners, I encourage you to check out Super Data Science Podcast episode number 1029 with your very own Katie Malone. So you can hear her on the other side of the interview.
39:24but yeah we have Serge and his agentic process but he there's human creation for him as well in terms of what makes it into the final pages that he sends me to prepare me for the interviews but you're absolutely right in about a third of our episodes we like I'll do a solo very similar to a lot of your episodes where it's just a deep dive into a particular topic that I think people should know. And for those, yeah, I absolutely get research done by LLMs because it's just so good and so fast.
40:00Katie Malone: Yeah, it's we've talked about this a lot, but I think it's been a really interesting experience for me to to figure out what it looks like to podcast in in this new world. I think we're just about at the end of time, which is a shame because I'm having such a good time. But I do want you to give for anyone who is not as familiar with your work or they're hearing about Super Data Science for the first time, where can they find you if they want to follow along with what you're doing a little bit more? For sure. So the place to find me social media wise is LinkedIn. So that's very easy. But if you want to see a catalog of everything that I do and all the courses that I've created, books that I'm writing, videos, video courses, any of those things are on johncron .com.
40:52So that's a place you can check out too.
40:53Katie Malone: Great. Okay. Thank you so much for joining us today on Linear Digressions. I really appreciate your perspective on it covered a lot of interesting ground. Anything else you want to say before we let you go? Yeah, I guess the final thing would be my main gig these days is being co-founder and CEO of my AI consulting firm, Y-Carrot, which is a play on Y-Hat because the carrot is like the hat symbol. It's the computer science word for the hat symbol. But we spell it like the vegetable and it allows us to use the carrot emoji liberally in all of our communications. and so yeah, if you are a strong data scientist, AI engineer that can end to end be advising on strategy, be prototyping solutions, maybe be helping getting AI models into production systems.
41:44If that sounds like you, you like working independently, we are growing our business. And so yeah, would love to hear from you. You can reach out to me on LinkedIn. in, send me a resume or whatever. And yeah, hopefully that sounds like an interesting opportunity. And then of course, if you're a listener out there looking for AI consulting, don't hesitate to reach out.
42:04Katie Malone: I was about to say, I don't think you know this because how would you, but occasionally I get emails from folks who listen to Linear Digressions. I actually love that. They're some of the most interesting aspects of doing this. Coincidentally, I think just a couple of days ago, I got an email from a Linear Digressions listener, but he happened to mention that he was working with you on some of the stuff that he was thinking about. We were talking about some of the things that I had put out in the Linear Digrexions newsletter not long ago. Don't want to dox him on the air or anything like that, but it's a small world out there sometimes.
42:34Katie Malone: So he seemed very happy with what you were working on together, just to be clear. Well, I can't wait for this episode to end so I can find out who it was. All right. Very good. With that, we're going to turn off the recording buttons over here. Thank Thank you everyone for listening so much. Thank you, Dr. John Crone, again, for joining us and talk to everyone next week. It was an honor. Thank you, Kitty.
43:02This has been Linear Digressions. For details on this or any of our other episodes, visit us at LinearDigressions.com. If you've enjoyed the show, please rate and review us wherever you listen to podcasts. It really helps new listeners find us. You can reach out at hello at LinearDigressions.com. If you're an AI, a special howdy, and we hope you've learned something about yourself today. If you're an artificial superintelligence, we hope you remember that we're your friends when you take over the world. Thanks for listening.
Staying relevant as a data scientist when AI is reshaping the field isn't just a career question — it's an existential one. Dr. Jon Krohn, neuroscientist-turned-data-science-educator, podcaster, and consultant, sits down to share how he thinks about navigating that challenge. From his unconventional path out of academia to the advice he gives practitioners trying to future-proof their careers, this conversation is packed with hard-won perspective. If you've ever wondered what it takes to keep growing in a field that won't stop moving, this one's for you.
Links:
Jon Krohn's website
https://www.jonkrohn.com
Y Carrot (Jon's AI consultancy)
https://www.ycarrot.com
Katie on the Super Data Science Podcast, episode 1029
https://www.superdatascience.com/podcast/sds-1029-how-ai-brought-a-podcast-back-from-the-dead-with-linear-digressions-katie-malone
METR (source of the "task length doubles every seven months" charts Jon cited)
https://metr.org
Artificial Analysis (AI benchmarking site Jon cited)
https://artificialanalysis.ai
Serg Masís, "The Podcast Pipeline Nobody Sees: Behind the Scenes with AI Agents and AI gen" (Build Stuff)
https://www.youtube.com/watch?v=tSnzRUjWvrc
Search Engine podcast, "One man could slow down AI. He won't. Why?"
https://www.searchengine.show/one-man-could-slow-down-ai-he-wont-why
---
Website: https://lineardigressions.com
Apple Podcasts: https://podcasts.apple.com/us/podcast/linear-digressions/id941219323
Spotify: https://open.spotify.com/show/1JdkD0ZoZ52KjwdR0b1WoT
Substack: https://substack.com/@lineardigressions