Jon Krohn: 00:00:00
This is episode number 899 with Kirill Eremenko, founder and CEO of Super Data Science. Today’s episode is brought to you by Adverity, the conversational analytics platform and by the Dell AI Factory with NVIDIA.
00:00:21
Welcome to the SuperDataScience podcast, the most listened to podcast in the data science industry. Each week, we bring you fun and inspiring people and ideas, exploring the cutting edge of machine learning, AI, and related technologies that are transforming our world for the better. I’m your host, Jon Krohn. Thanks for joining me today. And now, let’s make the complex simple.
00:00:55
Welcome back to the SuperDataScience podcast. We’ve got not so much an interview today as a fun and engaging and hopefully interesting conversation between Kirill Eremenko and myself. We actually intended this to be a short Friday episode, but then ended up having a ton to talk about on air, and so it became a full length Tuesday episode.
00:01:14
Many of you will already know Kirill. He’s been on this podcast many times in recent years, and he was in hundreds of episodes in a row at the start because he founded the SuperDataScience podcast nine years ago, and he hosted the show until he passed me the reins five years ago.
00:01:32
He’s founder and CEO of www.superdatascience.com, the e-learning platform that is the namesake of this podcast with over three million students. That’s crazy. He’s the most popular data science and AI instructor on Udemy. He holds a master’s from the University of Queensland in Australia, and a bachelor’s in applied physics and mathematics from the Moscow Institute of Physics and Technology.
00:01:56
Today’s episode is ideal for anyone looking to advance their data science or AI career, or if you’re looking to break into a career in this field for the first time. In today’s episode, Kirill details why employers are still testing AI engineers on basic machine learning fundamentals, even for LLM-focused roles. The surprising reason why staying in data science as opposed to developing an AI specialization could actually be the right career move for you, how one developer discovered the hidden age bias in tech recruiting and the simple hack to beat it, the two critical skill areas that separate amateur AI engineers from the pros commanding huge salaries, and why the back-to-office movement could give you a competitive advantage in landing a top AI role.
00:02:41
Are you ready for this magnificent episode? Let’s go.
00:02:52 Wow. Special guest today on the podcast, Kirill Eremenko. Welcome back to your podcast. How’s it going, mate?
Kirill Eremenko: 00:02:59
Thanks, Jon. Super excited to be back here. Everything’s going well, and here morning in Australia. How about you? How are you doing?
Jon Krohn: 00:03:07
Yeah. It’s a afternoon here in New York. The sun is at a perfect spot to give me really shadowy lighting in this recording for people watching the YouTube version. But, yeah, great light in the apartment. And for most of our listeners who are the audio listeners, it doesn’t impact them. We’ve got great audio, great mics, both of us, which is nice.
Kirill Eremenko: 00:03:32
Yeah.
Jon Krohn: 00:03:33
So Kirill, I understand you have something special for the audience today.
Kirill Eremenko: 00:03:37
Well as always, always have something special for our audience. Yeah. Today, we’ve got five stories of five people from the Super Data Science platform, from the membership that whom I interviewed recently in the past couple of weeks.
00:03:56
And, yeah, we’ll just go through what they’re going through because learning, machine learning, AI, data science can be a daunting task, can be an exciting task, but also can be a lonely task. And so for those of you out there that feel that you’re going through certain challenges or having certain wins, there’ll be, hopefully, something that you can relate to.
Jon Krohn: 00:04:21
That loneliness is great for us at the podcast here, because people feel so lonely. They have no choice but to find a data science podcast to listen to and feel a little bit less lonely.
Kirill Eremenko: 00:04:31
It’s funny. But it’s interesting. It can be bizarre. But indeed, one of the things that we hear on and on again in the community because we have this interest page. Once you become a member, you type in your intro and you say like, “Where are you from? Why you joined? What your hobbies are,” and so on.
00:04:51
And one of the most common things that why people join, I would say maybe 40% of the time I hear or I read this comment that they want to feel part of a community of learners. They’re learning these courses and tools and doing these projects. But around them, there’s nobody like that. At their work, there’s nobody else who’s learning. You can’t feel they can connect with people, and it’s actually a real challenge for people to find connection in this space.
Jon Krohn: 00:05:21
It’s great. I make the joke, but the platform is great. I’m sure the podcast and the platform are great for people feeling like they’re in a community. And by the way, when you say the Super Data Science platform, yet, we mean www.superdatascience.com-
Kirill Eremenko: 00:05:36
Yeah.
Jon Krohn: 00:05:37
… the website, yeah, which might be obvious, but just to absolutely make that crystal clear.
Kirill Eremenko: 00:05:44
That’s right.
Jon Krohn: 00:05:44
Yeah. And so Kirill, yeah, tell us about these five. I guess you can start with the first one.
Kirill Eremenko: 00:05:50
Okay. First one, so all of these names are changed for privacy purposes, and our first person is Alex. He’s early in his career, and he’s in his late 20s and just recently had a huge win, which is very exciting. He landed a job as an AI engineer. And that’s what our catch-up was about. We spoke about it.
00:06:17
Basically, this is how his interview went. He applied to many jobs. Like a lot of them he got rejected or didn’t hear anything back. One of them, he heard back. They said they were interested to talk to him more, invited him for an interview.
00:06:36
As far as I remember, it was originally like a screening interview. They said, “We’ll be back in touch.” He didn’t have high hopes for that because that’s what usually happens, but they invited him for a technical interview. And basically what he did to prepare for the technical interview, he said… So in the membership platform, for those who don’t know, we have lots of courses.
00:06:56
And he went through our LLM, large language models, A to Z course, again, just in case to refresh some concepts. And on the interview, they asked him specific questions about their data in the company and how he would create an LLM that… Let me find the quote here. How would you create an LLM that can access our data and just answer questions based on that data?
00:07:23
And from there, he basically just went into brainstorming mode and gave them ideas of how he would fine-tune existing models, apply RAG and other feature engineering and other things that would be relevant in that case.
00:07:40
And another thing that really helped him to stand out was that he did a few… No. I think he did one collaborative project in the platform. That’s where we get members to work together on projects. And that experience of going through a real world project helped him be able to brainstorm on the interview, and they were really impressed.
Jon Krohn: 00:08:00
Cool. That’s great. Congrats to quote, unquote, “Alex.”
Kirill Eremenko: 00:08:04
Yeah. Yeah.
Jon Krohn: 00:08:07
Did you anonymize any other demographic data? Have you switched genders on people or-
Kirill Eremenko: 00:08:11
Yeah. In some cases, yeah, that’s right.
Jon Krohn: 00:08:13
Oh, really? So you really anonymized?
Kirill Eremenko: 00:08:17
Yeah. As much as I could, to keep people’s privacy. Even though Alex in this case actually gave us full permission to disclose his story. You can actually read about it on the Super Data Science website. Nonetheless, the other thing I wanted to point out was that, interestingly, he added that after asking about large angle models, guess what they asked him? They asked him about fundamentals of machine learning. How are you going to build a regression? How are you going to for predicting price?
00:08:47
How are you going to build a classification for predicting problems relating to client data? They didn’t spend too much time on it. But apparently, they wanted to know that he knows the fundamentals. What do you think about that?
Jon Krohn: 00:09:04
You can’t be senior in the job. I’m sure there’s an entry-level AI engineer role where you really are just using APIs calling LLMs. But I think in a lot of senior roles, if you’re going to be making a big impact in your organization, you need to understand simpler machine learning models.
00:09:26
Ideally, not even just the fundamentals of machine learning, but also the foundational principles that underlie like linear algebra and partial derivative calculus because you can do… There’s all kinds of magical things you can do. If you understand that stuff, you can cut through the abstractions and come up with really performant solutions that, otherwise, wouldn’t be possible.
00:09:51
A career in data science is an endless opportunity to be learning more and more and more. And so you also should never let all of the possibilities of the things you could be learning prevent you from get and going some job applications off the ground. You should go for it.
Kirill Eremenko: 00:10:08
Yeah. For sure. For sure. But in this current, where we are in terms of AI, it feels to me like AI engineering, LLM engineering, et cetera, all those roles, they’re going through that same arch that data science was going through 10 years ago. So it’s a very blurry at the moment. What does AI engineer mean to you?
Jon Krohn: 00:10:35
Yeah. It’s an interesting thing because, theoretically, up until two years ago, if you said you were an AI engineer, that probably meant that you were like an AI researcher. So maybe at a frontier lab like Meta, Google, OpenAI, Anthropic and your engineering, you’re figuring out how to get transformers to work together more quickly across a bunch of GPUs, how to automatically clean up data in some way that it improves the outputs of a trained large language model.
00:11:12
That’s what I would’ve up until recently, up until a couple of years ago, thought an AI engineer was. But now, it seems to mostly mean somebody who is using existing LLMs and calling APIs. And so there’s still thoughtfulness that needs to go into making sure that data are clean and consistent. You have guardrails up. But you’re working at a more abstract level, at least in the simplest way of thinking about the job.
Kirill Eremenko: 00:11:44
So do you reckon you could get that job done without any knowledge whatsoever or fundamentals of machine learning or even the underlying deep learning tensor… What’s it called?
Jon Krohn: 00:11:58
Yeah. TensorFlow, PyTorch.
Kirill Eremenko: 00:12:00
Yeah, TensorFlow, PyTorch, even transformer architectures.
Jon Krohn: 00:12:04
Oh, yeah. Yeah. Like I’m saying, you could get… Yeah, there’s lots of LLM jobs out there where you basically need to understand how to be evaluating data that are going in as inputs and outputs. You need to be able to do some exploratory data analysis in Python, those kinds of data science skills. But you wouldn’t necessarily need to understand how stochastic gradient descent works or reinforcement learning works just because those kinds of approaches were used to train the LLM that you’re using.
00:12:39
It’s probably more important to just have lots of experience with prompting LLMs and seeing what they can do, understanding, experimenting with, okay, if I use a three billion parameter LLM, how does that perform relative to using the latest and greatest Claude 4 from Anthropic?
00:12:58
There might be fine-tuning involved. So understanding the approaches that exist out there for fine-tuning. So things like LoRa, Low-Rank Adaptation, being aware of those kinds of things to be able to take a three billion parameter open source Llama momdel from Meta, and then be able to fine tune it to some specific task. You might actually be able to, with a three billion parameter model, running on your own infrastructure or running through some cloud provider like Hugging face or PyTorch Lightning.
00:13:29
You can have this very small LLM running on some very specific task or some small number of very specific tasks. And because you fine tune it to those tasks, it can outperform the latest and greatest like Claude 4. And so those are the kinds of things, that kind of empirical experience with playing around with LLMs is probably more important.
00:13:54
You don’t really need to understand how LoRa works to use LoRa. And by the way, that LoRa, it’s L-O-R-A–
Kirill Eremenko: 00:14:02
That’s right.
Jon Krohn: 00:14:04
… if you’re Googling that. I’ll try to remember to put-
Kirill Eremenko: 00:14:06
Low-Rank Adaptation.
Jon Krohn: 00:14:08
Low-Rank Adaptation. I’ll try to remember to put a link in the show notes to our episode on LoRa, which is back in episode 674. I give an introduction to fine-tuning.
Kirill Eremenko: 00:14:23
Yeah. I remember that one. That was a great introduction. Yeah. Interesting. The analogy that comes to my mind is I love cooking. And for me, it’s like, let’s say, for cooking, you’re using a blender or other tools like an oven and things like that. You don’t need to know how an oven works in the backend or a microwave or… That’s lazy cooking. Or a blender. You don’t need to be able to pull apart a blender and put it back together, but you can use it.
00:14:51
So same thing here. AI is going towards that direction where these tools are actually just tools you can get away. And in fact, as you said, you don’t have to go into the depths of understanding these tools to be able to use them. And I’m just wondering what percentage of jobs are going to be for AI engineers who know how to use these tools versus the percentage of jobs where you actually need to understand the underlying technology and be able to tinker with it.
00:15:27
What should people be focused on learning? I know it depends on their interest, but in terms of supply or demand for jobs, I wonder how it’s going to play out in the coming months and years.
Jon Krohn: 00:15:38
This episode is sponsored by Adverity, an integrated data platform for connecting, managing, and using your data at scale. Imagine being able to ask your data a question, just like you would a colleague, and getting an answer instantly. No more digging through dashboards, waiting on reports, or dealing with complex BI tools. Just the insights you need – right when you need them. With Adverity’s AI-powered Data Conversations, marketers will finally talk to their data in plain English. Get instant answers, make smarter decisions, collaborate more easily—and cut reporting time in half. What questions will you ask? To learn more, check out the show notes or visit www.adverity.com.
00:16:22
Yeah. I think we move more and more towards the abstract. I love that analogy that you just gave with the blenders and microwaves and stuff. You can more and more rely on the abstractions, but I still think you can probably get… Ultimately, as you progress further in your career, I think you can get higher paying roles. It depends on exactly which way you go because you could kind of say, “Okay. You know what I’m going to do?” I’m riffing here, by the way, Kirill. This is just my thoughts.
00:17:00
But I’m thinking if you want to focus on commercial impact, you could actually say, “You know what? I’m going to use training there is available in and use that to become proficient at building LLMs, and I’m going to figure out how to make those look nice in a Gradio App or something,” some kind of user interface you can quickly put together so that you can have a click and point interface for people to be using your AI models in the background or some kind of AI solution that you come up with through using LLMs.
00:17:41
And you might not understand how gradient descent works or fundamentals of machine learning, but you’re able to put together powerful commercial applications. You could do that on a small team or on a big team or just on your own, and you could potentially be enormously successful. In today as well as more and more, the further we go into the future, the more you will be able to have agents, teams of agents working on different tasks for you. And you could build a big business. You could be a solo entrepreneur and have hundreds of agents working on different tasks with different clients, and that’ll get better and better.
00:18:22
So you could potentially have a lot of success that way using just the abstractions. But where I was originally going before I thought of that second idea is that, similarly, as you advance in your career, you could say, “Okay. I’m going to peel back layers of the onion more and more. I’m going to understand what’s going on under these abstractions more and more and kind of just chip away.” Over years, over decades, you become more and more expert at understanding machine learning fundamentals and mathematics and physics and engineering,, maybe electronics. There’s all kinds of related fields that you could dig more and more into.
00:19:04
And as you dig more and more, your value to your clients, your users, your employer, I think, does increase. And I think that that will continue to be the case in the future, even though those kinds of things like doing mathematics, being able to solve data science problems, computer science problems.
00:19:27
Even though that’s something that LLMs will be able to do more and more and more, I think there will still be, and maybe I’m just a dinosaur with outdated ideas, but I think if you understand that stuff, one, it’s interesting. Yeah. It’s an interesting thing here. If you’re a chef who’s able to understand how a microwave works and make a better microwave, that somehow you can put a raw pizza ingredients in there and it turns into this, it seems like it’s like this great pizza oven cooked pizza in just like 10 seconds. That kind of magic would only be possible if you learned the nuclear physics of how the microwave works.
00:20:20
There’s magic and possibility if you do dig deep. Yeah. So I think either way. You can follow your passions. If your passions are deep in the nitty-gritty of what’s underlying machine learning models, I think you can have a huge amount of success there the more and more you learn. But equally, if you’re more interested in applications and just making a big impact and you want to stay with abstractions, you can also have a lot of success that way too.
Kirill Eremenko: 00:20:44
Yeah. It feels like that transitionary period or that intermediate period. I love the example about the chef who knows how to make a new microwave, but I don’t think there’s a single chef like that on the planet. But for AI, it’s the case.
Jon Krohn: 00:21:05
Who knows? Yeah. There probably isn’t. To give an example, at the time of you and me recording this episode, I’m not sure the episode will be out or not yet, but I recently recorded a long Tuesday episode with Shaun Johnson, who is a renowned AI investor in San Francisco, and we were talking in that episode about how there’s only a few thousand people in the world that are at the cutting edge of AI.
00:21:39
And those people are typically more interested. I’m talking about these are the AI researchers that up until a couple of years ago, you might’ve called AI engineer. They’re like the chefs who know how to take apart a microwave. They can be in a big organization and they don’t necessarily need to worry about the downstream commercial impact. They just need to worry about making a better microwave.
Kirill Eremenko: 00:22:08
Yeah.
Jon Krohn: 00:22:09
And by focusing on their piece, they can be making a seven-figure base income. So that is making an impact. Yeah. It’s interesting, I guess, like a chef, maybe the incentives aren’t aligned as much for them to be learning nuclear physics.
Kirill Eremenko: 00:22:33
Yeah. I think we got to also think about the quantity of jobs. There’s definitely space for people making better microwaves in AI. But what is currently and what is going to be the majority of demand from employers around the kind of people employ.. And in my view, it’s really hard to tell at this stage, but from anecdotal evidence like Alex’s story, I can see that even though this abstraction layer is gaining popularity, employers still want to hedge their bets and want to vet their candidates by requiring that you know the fundamentals of machine learning. They’re not ready yet fully for candidates that are just operating on the abstraction layer without an understanding the underlying fundamentals.
Jon Krohn: 00:23:28
And I think part of that, I’m thinking about this from the perspective of a hiring manager for a role like that. If you were to hire somebody who had spent just a couple months learning about LLMs, prompts, inputs, outputs and didn’t have an understanding of anything beneath it, the barriers to entry there are relatively low. You’re able to quickly get up to that point.
00:23:55
And so yeah, I think it’s like hedging their bets, like you said, and also just trying to make sure that they’re hiring somebody that’s more well-rounded has been invested in this space for a while that are really committed to a career in this area.
Kirill Eremenko: 00:24:11
Yeah. I think that’s a good summary. Should we move on to number two story?
Jon Krohn: 00:24:18
Yeah. I thought this might be a Five Minute Friday episode.
Kirill Eremenko: 00:24:22
But this conversation is just too interesting. I’ve enjoyed this. All right. Story number two is Ben, their mid-career switching from a career in process engineer to data science learning. That’s their goal. They’re in their early 30s. And the interesting thing about Ben is I had a conversation with Ben in August, 2024, which makes it, what is that, five months plus another five months… 10 months ago.
00:24:53
And at the time, it was interesting, Ben told me that they’re learning a lot about AI, machine learning, data science, specifically data science and machine learning. That’s their goal of their career. And that in five months from August 2024, basically, they were going to be job ready. They were aiming to be job ready to apply for jobs in that space.
00:25:27
Five months is a decent amount of time, especially for somebody in process engineering who’s also been studying machine learning data science for past couple of maybe a year or so.
00:25:36
Interestingly, so when we caught up just recently, Ben’s comment was back then in August, I thought I’d be job ready in five months, but the field evolved faster than I could keep up. And at the moment, Ben feels scattered as job requirements keep shifting. He’s looking through these different jobs all the time, and they’re different to what they were five months ago. He’s learning one thing, but then by the time he’s finished with that course or that series of courses, job requirements have now changed again, and he feels like he’s always playing catch-up.
00:26:13
Plus, on top of that, he’s got a full-time job, he’s got personal family commitments. So he can’t just focus eight hours a day on preparing for these things. And also, he’s got the comfort of the income coming from that job. He just wants to change because that’s no longer his interest.
00:26:33
He’s interested in other things. He wants to be following his passion. But, yeah, that’s the kind of fearful state we find a lot of people are in these days with AI evolving and machine learning, this whole field evolving so fast that they can’t keep up and just can’t get a toehold on this whole job application process because things are changing so quickly.
Jon Krohn: 00:26:56
Nice. Sorry. I’m piecing this together just like our listeners as we go. So each of these five stories just gives us a different glimpse of different kinds of situations that people can be experiencing. Nice.
Kirill Eremenko: 00:27:18
Exactly. And I think a lot of people would be experiencing a similar kind of feeling. I’ve heard this from several people. This is an actual story, not just a aggregated story. It’s an actual story of an actual person, but I’ve heard the same story from other people as well where it’s evolving so rapidly.
00:27:38
Think about even LLMs like LangChain, LangGraph where all everybody was talking about a year ago like, “Where are they now?” They’re no longer as popular as the hottest thing right now. Or what about, what are they called, prompt engineering?
00:27:57
Everybody was talking about prompt engineering one and a half years ago. That was the hottest new thing. Now, people are talking about MCP, Agentic AI and things like that. What’s going to be happening a year from now, we can’t predict with certainty at all.
Jon Krohn: 00:28:11
I would say you can predict some things. Some things are fundamental. Yes. There are exciting new trends, absolutely. MCP right now, Crew AI, those are exciting trends. But simultaneously, there are undercurrents that you can see long term and be like, “Okay, that is a safe thing to be learning. This is going to be valuable to me for a long time.”
00:28:36
For example, all of those things that we were just talking about happen in Python. And so learning Python is a great skill. And then that also means if you want to go even deeper, you can say, “Okay. Well, learning data structures and algorithms is going to be useful because understanding the computer science that this Python code is written to be able to do gives you lots of options.”
00:29:04
So even if we somehow move on from Python, we’re not going to move on from Python in the next couple years.
Kirill Eremenko: 00:29:11
No. Probably not. No.
Jon Krohn: 00:29:11
But maybe five years, 10 years, everyone’s using Rust or something. And in that scenario, it’ll still serve you well to understand data structures and algorithms. There’s these trends, these mega trends in the same way that with prompt engineering, to me, that always seemed like an obvious thing that was going to go away quickly because all of the frontier labs developing the cutting edge LLMs. They are creating huge data sets and fine-tuning LLMs to be better and better at taking whatever prompt goes in and predicting what output someone was looking for.
00:29:50
So it becomes less about being a specialist and, “Oh, how do I hack this LLM to do what I want?” And every six months that goes by, the leading edge LLMs are going to just be way better at just anticipating your needs.
00:30:04
So yeah, so these kinds of long-term trends, you can bet that microchips are going to be cheaper and cheaper per unit of compute. That’s the ultimate mega-trend that is making all of this magic happen.
00:30:23
And so those kinds of things, you can look for those kinds of big long-term trends and feel confident about some things. So yeah, I agree with your point that there’s definitely always hot new things, and that creates anxiety for sure. But simultaneously, you can find some peace. You can find some stillness in these long-term things like SQL. That’s been around for decades, and it’s not going away.
Kirill Eremenko: 00:30:50
That’s a good point. Yeah. I like that. It probably adds calmness if you separate your learning into… I guess it’s like exploitation and exploration. You exploit the existing, as you said, mega trends, like 60% of your learning. And then 40% of the learning, you focus on new hot things. At least, you’ll have that 60% of ground, that calmness, and slow steady progress.
Jon Krohn: 00:31:14
Yeah. I like that idea.
Kirill Eremenko: 00:31:16
Pretty cool. And what you said about the cost of chips going down reminded me of a comment by Sam Altman recently. He said that cost of intelligence is going to converge to the cost of energy over time. And, yeah, I don’t know why I thought of that, but relates to that.
Jon Krohn: 00:31:38
Yeah. And related to that, so the Anthropic CEO, Dario Amodei, a few months ago wrote a blog post which I-
Kirill Eremenko: 00:31:45
You got to have him on the podcast, man.
Jon Krohn: 00:31:48
Yeah, I’d love to.
Kirill Eremenko: 00:31:48
That’ll be perfect guest.
Jon Krohn: 00:31:50
Yeah.Sam Altman, Dario Amodei-
Kirill Eremenko: 00:31:52
Sam, Dario, if you’re listening.
Jon Krohn: 00:31:55
… Bill Gates, the Kardashians. Let’s get them all on.
Kirill Eremenko: 00:32:02 Got it.
Jon Krohn: 00:32:05 Joe Biden’s retired now. Should get that guy on.
Kirill Eremenko: 00:32:08
Yeah. So you’re saying CEO of Anthropic?
Jon Krohn: 00:32:11
Yeah. So Dario Amodei, a few months ago, had a really popular blog post, and one of the things that sat… It’s related exactly to this cost of intelligence going down so much that he was describing a situation where in the not-too-distant future, you have a data center with a million agents like the equivalent of a million human brains. And those human brains are Nobel Prize winners.
00:32:38
And you have these million Nobel Prize winning intelligence brains just in one data center working away. They don’t need to sleep. They don’t need to take care of their kids. And that’s coming. That’s going to change the world. And related to the energy comment that Sam Altman made there, what I think is really interesting is AI, this abundant intelligence is playing a role in helping us get more energy, clean energy, including things like helping us contain the plasma in a nuclear fusion reactor, which if we can crack that, then all of a sudden you have basically unbounded energy and unbounded intelligence because… So that’s a pretty wild world that-
Kirill Eremenko: 00:33:30
Incredible.
Jon Krohn: 00:33:31
… we could be going into. Yeah.
Kirill Eremenko: 00:33:32
Incredible. You said Dario is the CEO of Anthropic.
Jon Krohn: 00:33:36
Indeed.
Kirill Eremenko: 00:33:37
He strikes me as a bit of a futurist, like his comment that by the end of 2025, all code, all code or 90% of the code will be written by LLMs. To me, it feels a bit far-fetched. Kind of reminds me of Ray Kurzweil in his predictions, but Ray Kurzweil’s predictions mostly come true. So we yet to see.
Jon Krohn: 00:33:57
This episode of SuperDataScience is brought to you by the Dell AI Factory with NVIDIA, delivering a comprehensive portfolio of AI technologies, validated and turnkey solutions, with expert services to help you achieve AI outcomes faster. Extend your enterprise with AI and GenAI at scale, powered by the broad Dell portfolio of AI infrastructure and services with NVIDIA industry-leading accelerated computing, it’s a full stack that includes GPUs and networking; as well as NVIDIA AI Enterprise software, NVIDIA Inference Microservices, models and agent blueprints. Visit www.Dell.com/superdatascience to learn more. That’s Dell.com/superdatascience.
00:34:43
And a big part of the way that Ray does that is from that big mega trend that I was talking about of compute cost, basically, you can model… It’s been consistent over decades, what compute costs over time, and you can extrapolate that forward to get the big even five-year, 10-year, kind of 15, 20-year predictions that Kurzweil’s been making have been based on that hardware basis to say that if that’s true… I don’t know that quote from Dario Amodei about us having all code be generated by the end of 2025. That’s not true. I guess a key thing to remember is that when Dario or Sam Altman are talking, they’re also pitching.
Kirill Eremenko: 00:35:29
Yes. Yes. I heard that podcast with you with the guy from LinkedIn, I forgot his name. He mentioned-
Jon Krohn: 00:35:34
John Rose. Yeah.
Kirill Eremenko: 00:35:36
No. Not John Rose. The influencer-
Jon Krohn: 00:35:40
Shirish Gupta?
Kirill Eremenko: 00:35:41
Nope. Nope. I forgot. Very big following. Here, you guys were talking about the CEO of NVIDIA, talking about the importance of microchips, and he mentioned that you got to remember that he’s pitching his company whenever he makes one of these kind of claims.
Jon Krohn: 00:35:59
Oh, yeah. Jensen Huang.
Kirill Eremenko: 00:36:01
Yeah.
Jon Krohn: 00:36:01
There’s another guest that we should just… Let’s phone him.
Kirill Eremenko: 00:36:04
Oh, yeah, why not? Why not? Yeah. Anyway, what do you think of this advice that I gave to Ben in his situation? Well, he was debating that he wanted to get into data science, but he also sees AI as the future, and he was thinking, “I’ll go into data science from process engineering,” which he’s currently in. I will learn everything about data science, then I’ll move to AI.
00:36:29
And I told him that he should aim straight for AI. If AI is his end goal, there’s no need to go through the path of becoming a data scientist first and then going into AI. Maybe, that was the case two years ago when AI was very deep learning everything, you have to know those things.
00:36:49
But now, there’s a straight path to AI. What do you think? Would you recommend that for people as well?
Jon Krohn: 00:36:54
100%. Something that I’ve talked about on the show, maybe even with you, I can’t remember, but this feels like a conversation we’ve had on air before. Let’s see, but I think you can think of an AI engineer, an LLM engineer as being a specialized kind of data scientist where 15 years ago when data science was a brand new term and people were starting to do it, there was this… It hadn’t evolved. We didn’t have all the kinds of different tools, all the different kinds of specializations.
00:37:26
And so as a data scientist, there’s this joke that what’s a data… A data scientist is someone who’s not good at statistics or programming. It’s like you’re able to do a little bit of a bunch of different things. You were able to understand enough about statistics at that time, maybe machine learning, data and analytics, a bit of SQL, maybe R at that time, some Python and-
Kirill Eremenko: 00:37:59
Presentation skills.
Jon Krohn: 00:37:59
… you have to do some visualizations, presentation skills, getting buy-in from management. Originally, it was kind of this idea that to be a data scientist, you might need a PhD 15 years ago. But now, it’s evolved.
00:38:21
Actually, it’s related to the same idea. People talk about how AI could take everyone’s job. And maybe, there is some timeline where that kind of happens. But the thing that has happened historically with all other automations is that more roles are created. LLM engineer could not be more of an embodiment of that truth which is that because AI is so capable, now all of a sudden, you need all these humans to be able to glue together all of those intelligent machines in order to do something that’s useful in order to create a product that provides a solution that’s commercially valuable.
00:38:55 And so, yeah, AI has created all of these additional kinds of data science specializations. So now, you have your data engineer, your ML engineer, your LLM engineer, your data analyst, your database specialist.
Kirill Eremenko: 00:39:11
Prompt engineer.
Jon Krohn: 00:39:12
Yeah. Maybe, last year. Exactly.
Kirill Eremenko: 00:39:21
Yeah. Interesting. Yeah. On that topic of creating new jobs, I love what the CEO of Dell said on one of your previous podcasts that I never thought of this, that a lot of new jobs will be created in construction, building those data centers and infrastructure for data centers. And that’s going to last decade if not decades.
00:39:42
That’s huge. That’s technology impacting the non-tech sector in terms of number of jobs. That’s incredible. And I love that that. That creates opportunities for people who are not even in the tech space.
Jon Krohn: 00:39:56
Yeah. Exactly. That’s a huge guess that we had recently the CTO and Chief AI officer of Dell, just hundreds of thousands of employees, John Rose. And I think the other episode we were talking about, I think that might’ve been Greg Michaelson from Zerve that was talking about. Yeah.
Kirill Eremenko: 00:40:14
I can find the person, but what I was going to say is such a good episode with the CTO of Dell, incredible episode. Loved it. So well spoken and also so many great ideas, especially anybody looking for commercial applications of AI and how it applies to business and industry.
00:40:38
I’m recommending that episode to people around.
Jon Krohn: 00:40:41
Nice. Yeah, super popular. All right. Are we on to number four now or what?
Kirill Eremenko: 00:40:45
Yes, yes. Almost. I’m just finding this guest’s name for, yeah, there we go. Andriy Burkov is
[inaudible 00:40:54]
Jon Krohn: 00:40:54
Andriy Burkov, yeah, yeah, yeah. So that’s going back a little bit. Yeah. Andriy Burkov, it’s completely insane. He has his LinkedIn and newsletter. At the time of recording, it was like 980,000 subscribers or something. He’s super close to having a million subscribers to his AI newsletter on LinkedIn.
Kirill Eremenko: 00:41:15
Episode 897, I believe.
Jon Krohn: 00:41:17
Yeah. 867.
Kirill Eremenko: 00:41:19
I think people like him because he’s just raw, no filters, just says his opinion, doesn’t matter if it’s going to offend people, not offend people. I’ve seen some of his comments on, “They removed my post from this post because they didn’t like what I said, ha ha, ha. Here’s a screenshot of how it looked.”
00:41:39
I know in this day and age when there’s so much fakeness and in non-genuine people or non-genuine presentation of themselves, I think people value that the role. And whether you like it or not, that’s a separate question. But having access to somebody’s raw personalities, character, I think it’s nice.
Jon Krohn: 00:42:01
Super, super lucky to have him on the show. He never does videos, podcasts. He writes books which are extremely popular.
Kirill Eremenko: 00:42:08
And free.
Jon Krohn: 00:42:13
Are they free?
Kirill Eremenko: 00:42:13
I think. You paid if you like [inaudible].
Jon Krohn: 00:42:15
[inaudible] you have to pay to get the physical version, obviously.
Kirill Eremenko: 00:42:16
Yeah. But then you get the one online for free and you pay however much you valued it for or something. I was very impressed by that.
Jon Krohn: 00:42:23
So he’s famous for 100-page machine learning book, but then he was on the show talking about his 100-page LLM book.
Kirill Eremenko: 00:42:29
Yeah.
Jon Krohn: 00:42:30
So we’d reached out to him. I’d been reaching out to him through various means over several years, and he never responded. And then when he finished the 100-page LLM book, he reached out back out to us, and he was like, “I’m still going to do this.”
00:42:41
And so it’s really cool to have somebody like Andriy on the show who so rarely does those kinds of appearances, and he does them so rarely that here’s something hilarious.
Kirill Eremenko: 00:42:50
Totally.
Jon Krohn: 00:42:51
So he never hears recordings of his voice. And so he’s from Ukraine, and so he has this strong Slavic accent. It’s not that strong, but he thinks that he sounds like… Because he’s been living in Montreal for decades.
Kirill Eremenko: 00:43:08
Yeah.
Jon Krohn: 00:43:09
And so he thinks that has the kind of North American kind of sounding English accent. So he was like, “What the hell?” When he listened to the recording, he’s like, “Who is this guy?”
Kirill Eremenko: 00:43:21
That’s funny. Oh, yes, yes. Good times. I think it’s important to hear yourself radio on air sometimes to understand yourself.
Jon Krohn: 00:43:35
Are you doing a bit or something? You sound the same.
Kirill Eremenko: 00:43:40
Nice. Nice. All right. Let’s move on to number three.
Jon Krohn: 00:43:44
Okay. Wait four. Surely four.
Kirill Eremenko: 00:43:48
I wish.
Jon Krohn: 00:43:50
Are you kidding me? We’re on number three?
Kirill Eremenko: 00:43:52
Number three. That’s right.
Jon Krohn: 00:43:53
Wow..
Kirill Eremenko: 00:43:54
This is Five Minute Friday screen going for 37 minutes.
Jon Krohn: 00:43:58
Maybe, we’ll have to convert this to a Tuesday episode.
Kirill Eremenko: 00:44:00
Maybe. You’re the host. It’s your call. I’m just having a fun time. That’s the benefit of retiring from the podcast. Don’t have to make these decisions. All right.
00:44:11
Number three, Clara. She’s a senior developer living in LA and in her mid-40s aiming for roles in the 200,000-plus salary range. So this person, she has been working already for two decades or more in this space and has tons of experience, in fact, has done all sorts of roles in software engineering, developing apps, developing programs, developing different things for different companies.
00:44:54
Most recently for the past, I think it was five… Sorry, three or five years, I don’t remember, let’s say, three years. She has been creating software using Python, interestingly, software that processes data, lots of Excel, lots of CSV files using Python in the medical space.
00:45:11
In fact, a lot of our members, I don’t know the exact percentage, but a lot of our members I speak with work in the medical space supporting companies, whether it’s hospitals or pharmaceuticals or other medical space related, like medical equipment companies, procurement companies or supply chain companies and so on.
00:45:32
Anyway, so she’s been creating all this software using Python, specifically Pandas and other tools to process lots of data. So lots of Python experience and recently has done four of our machine learning courses, machine learning A to Z, machine learning level one, machine learning level two, machine learning level three. And she wants to get into the space of machine learning and AI. Why?
00:45:56
The reason is because Clara is in her mid-40s. She predicts that she’ll be in the workforce for at least another 15 years, and she can see that the current role that she’s doing is while pays well, and she’s very good at it, it might not be as relevant in the future.
00:46:18
As we discussed with Clara, it’s not a role that’s a self-fulfilling prophecy. She’s not learning new skills in the role that will open up more doors for her in the future that will keep her growing with the growing trends in technology. She’s very selective about applications.
00:46:35
In fact, she left her job a few months ago to focus specifically on studying and preparing for the new role. She’s not in a rush. She wants to take things slowly and basically goes mostly through her network, not applying to thousands of jobs through LinkedIn and so on. Mostly goes through her networks, very selective.
00:46:59
So that’s her goal to get into this space. And the interesting thing, the pain point that Clara has is she’s finding there are thousands, literally thousands of job applicants per job. And even at her level of experience, expertise, and background and all these projects that she’s done, she’s finding it difficult to break in and to land the job that she’s looking for.
Jon Krohn: 00:47:26
Oh, yeah, interesting story. Why do you think that is? Why do you think she’s having trouble?
Kirill Eremenko: 00:47:32
That’s a good question. I think it’s probably related to this phenomena where there’s lots of jobs, but there’s also lots of applicants, and it’s really hard to stand out. I think it’s been the same for the past 10 years when there’s lots of people applying through the direct means of just submitting a resume and all of them get pre-screened with AI tools.
00:48:03
And if the hiring manager had a conversation with Clara directly, magically, then they would realize she’s amazing, and they would hire her in a heartbeat. But because it’s really hard to get in front of people this direct way, I think that’s the problem. And I think Clara has got the right idea of going through connections and going through networking to get in front of the people quicker. What do you think?
Jon Krohn: 00:48:35
Networking ideally in-person is, I think, easily the best way to get your professional opportunities. Not everyone can do that. You might have a family situation or just where you are geographically. If you want to get a job in data science or AI, maybe there aren’t in-person things you can be doing. Remote’s really the only option. And there are probably then in that kind of scenario still things like www.superdatascience.com, these kinds of platforms where you can get involved. You can do collaborative projects together, get to know people.
00:49:09
That gives you that kind of collegial feeling. You’ll remember the projects that you’ve worked on, the people you’ve been with, their expertises. And that’s something like working with someone in an office and understanding what they can do and maybe they’ll open a door for you some years from now.
00:49:27
So the more you do that, the more that you’re working with people online if you have to. But ideally, you are meeting people in person. In the US ,there’s something called meetup.com
00:49:48
With meetup.com, in any major city in the US or Canada, you can find meetups for whatever you’re interested in. It’s not specific to tech. I’m sure there’s like microwave, reprogramming, chefs meetup. There’s all kinds of specific things out there.
00:50:10
But in data science in particular, there’s lots of different of these kinds of meetups. And you go and you could be at any stage. You could be just getting started. You could be thinking about it. Maybe you’re a medical doctor and you’re tired of just dealing with one human at a time. And you have a vision for some kind of medical AI system that you want to build to scale up your impact.
00:50:36
So you can start going these meetups and meeting people and decide, “Okay. How can I take further steps into this? Should I be joining a platform like www.superdatascience.com or do a master’s in person at a local university?”
00:50:52
So you could be at that very early stage where you’re just exploring if a career in data science or AI is something you’re interested in all the way through to being a big expert. You might participate in giving the talks if you’re an expert. Often, these meetups have that. It could be based around one or two speakers talking about real world projects or some open source library they’re developing.
00:51:15
But around these, you learn stuff from the speakers. But around these, you also have lots of social interaction. There’s drinks at a lot of these. Pizza is often the food that they order. And sometimes, it’s sponsored by some local data science or AI company, or maybe there’s some small fee, like five bucks or 10 bucks that you pitch in to be able to buy the pizza and the beer or whatever.
00:51:42
And yeah, it’s in those social interactions that you meet people and some people just you clique with them, and you chat with them more. You see them there a few times, and you might find your next job. You might find your romantic partner. You might find your best friend. You never know in a way that I think those kinds of things, those kinds of connections. It’s a little bit harder to make them online, but it can happen.
00:52:07
You and me, we’re going to meet soon in person for the first time, but we’ve known each other for over five years. I feel like I know you. You’re one of my closest friends, but-
Kirill Eremenko: 00:52:16
I know.
Jon Krohn: 00:52:16
… I’ve actually only talked to you through pixels.
Kirill Eremenko: 00:52:21
Yeah. Well, thank God it happened before all the deep fakes, so we know we are real. Yeah. Interesting. You mentioned the medical field, like doctors looking to apply LLM’s AI to scale their impact. I could probably name right now off the top of my head, I could… Well, I had to look it up, but at least five people in the SuperDataScience platform, we’ve had five people who in their interest have posted, “Oh, I’m a doctor or I’m a nurse.”
00:52:49
And there’s this point in the hospital databases, there’s this problem. We’re constantly facing it. I’m eager to solve it. I’m on a mission. We’re going to use AI to solve it. That’s why I’m learning AI. There’s so many opportunities in the medical space to apply AI for whether it’s scaling impact to patients, improving existing systems processes, improving administration of hospitals and things like that.
00:53:15
It’s incredible. I’m always surprised at how many people are finding active pain points all the time in the medical field. It’s by far the highest, the most popular almost commented on to the point that we even had a quote, unquote, “meetup,” a virtual meetup inside SuperDataScience. It was actually more of a mentorship session where we hire a expert in the medical field in, I think, it was bioinformatics. And she ran a workshop for people. And then they could ask questions and all discuss these kind of things because that’s how much demand we had for that space.
Jon Krohn: 00:53:58
Yeah. I think people in the medical field, often, they want to be making a big positive impact in the world. They’re typically very intelligent people, hardworking people. And so there’s a lot of overlap in that Venn diagram with the kind of people who want to be interested in data science and AI.
00:54:13
But you can come from any field. I don’t know if you’ve come across this woman… I hope I’m pronouncing her name correctly, Adriana Salcedo. She is in Bavaria in Germany. And she’s been working as a flight attendant for seven years, and she’s actively… Her full-time job is flight attendant. And she has been studying… She’s listening to the podcast for a couple years to this podcast, the Super Data Science podcast.
00:54:38
And she is learning data science skills. And recently, she started posting little projects she’s been doing things. She actually recently completed Ed Donner’s LLM engineering course, which the Super Data Science team was involved in creating. Yeah. So it’s really interesting. There’s so many possible topics for podcast episodes, but she’s someone that I’m like, it’d be interesting to get her on the show.
Kirill Eremenko: 00:55:10
Yeah. She’s been a member of SuperDataScience for over a year now. Yeah.
Jon Krohn: 00:55:16
Oh, she’s a member of the platform too?
Kirill Eremenko: 00:55:18
Yeah. Yeah.
Jon Krohn: 00:55:18
I didn’t even know that.
Kirill Eremenko: 00:55:20
Yeah, passionately, enthusiast.
Jon Krohn: 00:55:22
There you go.
Kirill Eremenko: 00:55:23
Yeah.
Jon Krohn: 00:55:23
Cool.
Kirill Eremenko: 00:55:24
Okay. Awesome. Okay. Oh, one piece of advice I gave Clara, let me know what you think about this is Clara’s like, “All right. I’m a software engineer senior developer. I’m going to learn all these skills about AI machine learning to then apply for these jobs.”
00:55:45
And the advice I gave was work backwards. Go and talk to some recruiters in LA and ask them, “What are employers looking for in an AI engineering role? What are their main requirements? What boxes do I need to check for you to be able to get me a job?” And then learn those things. And you hit two birds with one stone there.
00:56:09
You narrow down your learning scope, and you can focus more deeply on those things. And also, speaking of in-person networking, you get in front of these recruiters who then now know that you already have a great background. You just need to touch upon a few things. They’ll take you a couple of months, and then you can reach back out to them and say, “I’m ready.”
00:56:33
And it’s in the recruiter’s best interest to place the best candidates because they get paid a commission for that. So it’s a win-win. You don’t have to go looking for jobs. And then these recruiters, as long as they have trust in you that you have checked those boxes, they can go and help you get those jobs.
00:56:53
So it seems obvious, but sometimes, we don’t stop to think. We just learn everything under the sun or whatever the hardest thing is. We think that is going to help us. But really working backwards, especially in your area, what companies might be looking for in LA could be different what they’re looking for in Montreal, or could be different to what they’re looking for in Berlin. Maybe because of the industry that’s dominating that space or maybe because of the cultural aspects or where technology is heading or trends, local trends. I think there is a lot of benefit to working backwards that way.
Jon Krohn: 00:57:33
100%. There’s really interesting. There’s lots of different websites that give breakdowns of what skills in an area. I know LinkedIn has done surveys, and even I’ve talked about those on air recently. I don’t know how quickly I’d be able to find. Oh yeah, episode 856.
00:57:52
I talked about how AI engineer is the fastest growing job of 2024. And everything in that episode is from LinkedIn surveys done all across the world. And there’s interesting trends like AI researcher is a particularly popular role in San Francisco where there’s lots of frontier labs, whereas something like AI consultant is very popular in New York where there’s fewer places that are working at the cutting edge of developing LLMs and more places that are working with clients to make a big impact with those models.
00:58:30
And so you’re exactly right. There’s definitely regional variation. And then also, of course, there’s variation by industry. If you want to be working in the healthcare sector or the finance sector, there’s different tools. Maybe, if you’re working in finance, you need to figure out some way of doing all this stuff on a Windows computer.
Kirill Eremenko: 00:58:49
And Australia is probably more heavy-industry focused, like mining and things like that. And agriculture really depends. A cool website that one of our members recommended, thank you, Ricky Singh, for recommending this to me recently is called hiring.cafe.
00:59:10
Recommend checking it out. I was surprised at… It’s like an aggregator of jobs. You can filter by location, job titles and things like that. And, yeah, it’s very raw type of layout, user interface. It’s not one of those fancy websites like, I don’t know, Indeed or SEEK. It’s very accessible, and they do a fantastic job at aggregating all the jobs. So anybody looking for roles or even just to get information on roles in your local area or wherever the skills are required, I recommend checking out hiring.cafe.
Jon Krohn: 00:59:49
Cool. Thanks for that tip.
Kirill Eremenko: 00:59:51
Okay. Moving on, number four. Story number four is David, an experienced professional who’s been in the data science space, and they’re planning on staying in the data science space. They’re in their mid-40s. They have a background in gaming and analytics and consulting, and they’re now staying sharp in data science and not planning to transition to AI.
01:00:15
And I specifically like this story because not everybody is going to become an AI engineer or it just shows that you don’t have to become an AI engineer. And the reason I asked David like, “Why don’t you want to become an AI engineer? It’s the hot thing right now. Everybody seems to want to get into AI.”
01:00:31
And David explained that, first of all, he’s not that technical with or not just not interesting getting that technical. That’s not his passion. At the same time, he doesn’t see the role of a data scientist getting replaced by AI because he sees huge value in being the customer-facing data science person, basically helping translate insights into business outcomes. He’s interested in using AI, but he is not interested in building AI and fine-tuning and Agentic, and LLMs ,and things like that.
01:01:08
Yeah. So what do you think of that? Is there value in people staying focused on data science and using AI to their advantage, but not really diving into AI engineering and those other roles that we talked about earlier?
Jon Krohn: 01:01:25
For sure. I mean that think that ties into this idea that I was talking about earlier where there’s these long-term mega trends that you can find relatively solid ground on. And so one of those big mega trends is that AI models are going to become better and better and better at being able to help us out on data analytics, data visualization.
01:01:48
There’s very little reason today why you should be typing out every character in the code that you write. And so you should definitely be leveraging these tools wherever you can. And then, yeah, the other thing to think about with respect to mega trends is if there are certain spaces where you can find solid long-term ground that don’t involve staying up to date on the latest things with LLMs. You can become expert at other aspects of data science.
01:02:24
You don’t need to be like, “Wow. Everyone now is learning how to call LLMs and stitch them together.” There’s a lot of demand for that, but there will continue to be a lot of demand for data visualization and being able to write performance SQL queries, being able to tell a compelling data story from the results that you have.
01:02:45
And so you can focus on those kinds of approaches. You could become expert in Bayesian statistics, which has a lot of applications, and that isn’t anything to do really with AI or LLMs. And you could be making huge impacts. You could become a huge expert in Bayesian statistics and use LLMs to help you learn that stuff and make it easy to do the work. But yeah, does that answer your question?
Kirill Eremenko: 01:03:14
Yeah. Yeah. Yeah. For sure. For sure. And there’s always going to be room for that bridge people connecting the technical insights and takeaways to the non-technical audience because at the end of the day, you got to drive business outcomes. And there’s going to be a lot of people.
01:03:36
Sometimes, I get caught up because I’m listening to this podcast. I’m speaking with our members. I’m learning things in AI. I’m teaching things, and I get caught up. And I think that everybody around me knows AI. I feel like, “Okay. Everybody’s on.”
01:03:51
But realistically, it’s probably a small percentage of people in the whole world may be like… It might be an overestimate to say 3% of people on the whole planet understand LLMs and understand even what a regression is and how classification works and things like that.
Jon Krohn: 01:04:13
That’s way less than 3%.
Kirill Eremenko: 01:04:15
Yeah, I know, but it feels like that. It feels to me like probably 30% of people. But I have to consciously tell myself it’s probably less than three. So I personally get carried away. But then I get woken up from this dream state when I speak to someone just of my friends at a cocktail party or something, and we start talking and they’re like, “Oh, what do you mean?”
01:04:40
And then I have to bring myself back to, “Oh, actually, I’m talking to a lawyer who at this stage is not yet using Agentic AI and things like that.” So there’s always going to be room for people who translate from this small percentage of experts and from this world of tech to the non-tech people, insights how things work, how things should work, explore their pain points, problems, and things like that.
01:05:05
So that’s definitely an area of data science that if it interests you, if you feel excited about talking to people and helping people, then that’s a great area to follow. And you definitely don’t need to go super technical if you’re in that space.
01:05:27
A couple of other things that David also said was that interesting tip. He’s in mid-40s, and he has experienced age bias in recruiting. And what he does is he actively limits what’s visible on his LinkedIn to avoid age bias. For example, he’s removed his dates of graduation. He’s removed, I don’t remember what else, his birthday and things like that. So people and algorithms cannot bias against him in terms of age.
Jon Krohn:
01:05:56
You said he’s 40?
Kirill Eremenko: 01:05:57
Mid-40s..
Jon Krohn: 01:05:58
Mid-40s.
Kirill Eremenko: 01:05:59
Yeah. Interesting, huh?
Jon Krohn: 01:06:00
Yeah.
Kirill Eremenko: 01:06:01
So that’s a tip. Unfortunately, it’s sad, but it probably does happen. So if you want to protect yourself, especially against age bias, whether you’re young or old, whatever you feel is maybe affecting you, that’s one thing to go about it. And then once you get into the interview, it doesn’t really matter what age you are, it’s all about the skills. It’s about the value you can bring to the business.
Jon Krohn: 01:06:25
Yeah. I think in data science, there really are people who, if they’re making a transition from some other career, they really are willing to take some big pay cut to get some experience. But it is interesting how I’ve encountered several times in my career maybe like someone in finance, for example, is coming to mind for me right now, someone that I worked with in finance.
01:06:57
She was in the finance department, and I was talking about negotiating offers. And I said, “This person would be coming over from this different career where they’re being paid a lot more,” but they’re willing to take a pay cut to work with us because love the work that we’re doing. They want to have these skills. And they were skeptical that that’s even a thing.
01:07:19
They were like, “No. It won’t work. They won’t stay.” Everybody wants more money. This is a finance person. So, yeah, it’s just interesting. You definitely can because I think that age bias, I think, fundamentally, it isn’t like somebody in their mid-40s, you’re not like, “Oh, they’re too old to learn.”
Kirill Eremenko: 01:07:40
You’re never too old to learn. I think we have somebody in their 80s or something in the 70s learning in super data science. No problem at all.
Jon Krohn: 01:07:50
That is cool. No, for sure. But I think you get this expectation. Some people in an organization, not me, and probably not you, but some people get this thing in their head that people are always looking to be making more, to be more senior. And so I think that’s the thing is that the thing about age is that it becomes associated with some expectation of some level of compensation, which for some roles that this person might be applying for that you’re describing, they might be applying for roles that are like, yeah.
01:08:25
So the hiring manager sees it and is like, “This person, they’ve been working way too long. Their salary expectations are going to be too high. I’m not even going to talk to them.”
Kirill Eremenko: 01:08:32
Yeah. Yeah. That’s unfortunate. It happens. And there’s a tip how you can protect yourself. So finishing up about David, I forgot to mention that he’s looking at a salary between 200 and $250,000 roles. They’re also based in, I think, mid-US somewhere in the Midwest, it’s called. An interesting trend that he’s observed is this back-to-office trend.
01:09:04
It’s getting more and more traction, and he sees it to his advantage because he is willing to go back to the office. And that to him means that there’s less going to be less competition for roles that he’s in his area from people from all over the world.
01:09:22
And he also recommends to look at places where companies are because of this back-to-office trend where places where companies are opening up office. For example. he mentioned IBM struck a big deal, a multi-billion dollar deal with a hospital somewhere in Ohio. And he predicts that there will be growth in terms of AI jobs in that space, and there isn’t that much. There is talent, but there is going to be more opportunities there than there is a supply of talent.
Jon Krohn: 01:09:55
Yeah. It makes a huge amount of sense. Great strategy. Yeah. If you want a fully remote job today, it is increasingly competitive.
Kirill Eremenko: 01:10:06
Why is that happening? Why is this back-to-office trend happening?
Jon Krohn: 01:10:09
So I have this hypothesis that when you are an executive or a manager, you feel powerful when you can come in. Imagine you’re in New York, you’re an executive at Goldman Sachs. You want to come in at 8:30 and all of your hundreds of underlings have already been slaving away since 7:30. When you have guests from your friend from the sovereign wealth fund is visiting New York. You want to be able to take that friend to office and be able to show them all of your minions.
01:10:53
Whereas when you’re like, “Oh, we have a large workforce. Everyone’s online.” It’s so much harder to feel the power of that you have. Because it’s the more senior you are… I read the stat years ago, I’m sure it’s still the same, I read the stat like 2022 that the more senior you are, the more likely you are to think that people should be back in the office.
01:11:21
And that’s when I originally hatched this hypothesis. But I also recently read that younger people, like recent grads, they’re also increasingly happy to return to office because you learn so much more. I think those are the people that miss out the most, I think, are if you’re starting off new in a career and you’re stuck on Zoom meetings, you don’t develop the same kind of rapport as you do in hallways around the coffee machine, going out for drinks after work. If that kind of stuff is happening all the time, I don’t know, you learn a lot more about your industry, I think.
Kirill Eremenko: 01:12:00
For sure. Speaking of junior roles, I think they’re becoming more and more at risk with AI, like Agentic AI, automating, let’s say, junior lawyer tasks, the whole research of case law and stuff like that, or accountant tasks. I heard of this theory 10 years ago. But now with this kind of AI, it’s becoming even more prominent that AI’s going to automate the junior tasks first and what’s going to happen next?
01:12:32
Junior people are not going to have an opportunity to train and grow into senior people. And so we’re going to have this whole layer or a slice of the workforce cut out in certain roles that are easily automated with Agentic AI. And then we will face the consequences of that 10 years down the line where we will have no mid-level people or senior people that would’ve come from those junior people.
Jon Krohn: 01:12:56
All kinds of things changing, changing quickly,
Kirill Eremenko: 01:12:58
Yeah. Changing quickly. Got to keep up but also focus on those fundamental, what are they called, mega trends as you mentioned to keep your sanity.
Jon Krohn: 01:13:08
You must keep listening to the podcast still.
Kirill Eremenko: 01:13:11
That’s a mega trend. The podcast is a mega trend. All right. Number five, Evan, he’s an experienced engineer upskilling in ML deployment. So for full disclosure, Evan has decided to move on from Super Data Science from the membership, and this was part of his exit interview.
Jon Krohn: 01:13:31
Oh really?
Kirill Eremenko: 01:13:32
Yeah.
Jon Krohn: 01:13:33
Thank you for the transparency.
Kirill Eremenko: 01:13:36
I want to be honest. It’s a learning platform. It works for some people. It doesn’t work for some people. Evan was here for a year. He got everything he wanted to get out of it and moved on.
Jon Krohn: 01:13:47
It would be kind of wild to expect that people are members for life in something like that. I understand that you want different perspectives. Yeah. I totally understand that.
Kirill Eremenko: 01:14:00
But anyway, so we had a great catch up. I messaged Evan, I said, “Hey, you’ve been one of our most vocal and members.” I’d love to talk to you and understand what your goals are and things like that. And basically, Evan is very experienced. He’s based on an island in Europe. I think that’s ambiguous enough not to give away the exact location of the person based on an island in Europe, but he does have a company, a consulting company. And he works with US clients on machine learning, AI, software development. So he comes from a software development background, but he has upskilled a lot.
01:14:46
So in the past year, he’s attended several of our collaborative workshops. I think he attended two collaborative workshops where he worked in teams of people to build machine learning and AI and deploy them like models and projects and so on, deploy them.
01:15:03
He’s also attended a lot of our labs, a lot of our courses, a lot of our mentorship sessions. He’s one of those people that we have mentorship sessions where you select a career path and you then get assigned to a specific mentorship group based on your career path. The beginner level advanced or expert level mentorship group. He actually asked me to put him into all three mentorship groups just so he could interact with all three different of our mentors.
01:15:31
Anyway, that was a funny thing about him. So basically, the pain point that he’s facing is that increasingly, he is finding that companies where he’s working on LLMs or that he’s applying jobs to work on LLMs and Agentic AI, they are looking for specifically cloud skills, production-ready skills and deployment skills.
01:15:59
So he sees that an AI engineer needs to increasingly know how to deploy production-ready systems into cloud-based environments. What are your thoughts on that? Do you think that’s a compulsory skill for an AI engineer these days?
Jon Krohn: 01:16:16
I don’t know about compulsory, but one thing I will say for sure is that there’s huge demand. I’ve said this on the show many times before. For any kind of software engineering skills, if you learn those alongside data science, AI skills, you broaden by so much the scope of possible jobs that you could get. If you could be straddling the engineering team and the data science team and figuring out how to make the AI models work for the particular circumstances or some production use case and some cost-effective performing way for the users, that is valuable.
Kirill Eremenko: 01:17:02
Yeah. For sure. But interestingly, I am hearing more and more. It’s been super surprising to me. I’m hearing more and more from our members. We are learning production-ready skills. We’re learning deployments because that’s why, for example, our collaborative projects are so popular because the final step in the collaborative project, the final phase, which takes a whole week, is the deployment part.
01:17:29
And more and more members. And I was speaking with funny enough, your friend Ed Donner about this yesterday, we were exchanging emails. And he said as well, he thinks, and I agree with him that in my view, and he says that the science of AI is supposed to be the important thing, like understanding your data, experimenting with your embedding model, validating the relevance of your context and things like that.
01:17:56
But alas, people studying and learning this space more and more, especially at the more advanced level, are looking to deployment skills.
01:18:05
In fact, Ed told me he ran a survey literally last morning to his current students, and what’s the topic they would like to hear more on? And sure enough, production deployment is in the top spot. So it’s kind of like this in my mind, because AI engineer is such a vague term, vague role at the moment, still shaping up, in my mind from these conversations, shaping up in a way that an AI engineer, yes, there’s that abstract level of AI engineer we talked about earlier where you can just use the tools and not really get deeper.
01:18:40
But as you get deeper, there are two main areas that you should understand well. The first one is the science of AI, and that’s what we talked about like your embedding model, your experimenting with your data, understanding what model to use, and even to the point of LoRa and things like that.
01:19:00
You can go very deep into that AI part. But also, there’s the second part, which is the production ready system deployment and cloud. And that’s understanding CI/CD continues improvement and use deployment pipelines. That’s understanding how cloud works, what kind of tools in the cloud you’re going to use, understanding whether it’s right to use model context protocol like MCP or not, understanding how to build Lambda functions on AWS, to put your Python code into them, how to use step functions, how to make all of that work together and tie into a system that is working, that can be updated, that can be used by companies and so on.
01:19:42
That’s a whole separate area of an AI engineer. Like we were talking about, an ideal AI engineer would know both of those things really well and be able to build a great model so that it’s efficient and it delivers the business outcome that needs to deliver and be able to deploy it, monitor, set it up so that it’s running in a cost-effective way. There’s security properly set up that it’s secure and things like that.
01:20:13
So I think those two areas, if anybody’s looking to build a long-term career in AI engineering really go deep, I would really focus on developing skills in both areas.
Jon Krohn: 01:20:24
And am I correct in understanding that you actually have in www.superdatascience.com, you have a bootcamp for these kinds of skills coming up?
Kirill Eremenko: 01:20:31
Yes. Yes. Now’s, the time to plug our bootcamp. So basically because of these things that we’re observing and all this demand, we’ve just literally launched a bootcamp. So applications are open. You can go to www.superdatascience.com/bootcamp and apply there. So it’s an eight-week intensive program where you get to work with experts in the field that are actually doing these things.
01:21:04
So people who are an AI scientist and is good at that first area, that’s the first three or four weeks of the bootcamp, understanding that and going deep into that. And then there’s also that the last four weeks is the deployment part. We work with AWS, understanding all those things like Lambda functions and how to structure your deployment, how to do the CI/CD and everything else like security and things like that. So you work with an expert in that space.
01:21:32
You’re in a cohort. They’re very small cohorts. It’s going to be between five and 10, maybe maximum 12 people per cohort. There’s going to be a capstone project at the end. There’s projects along way. They are requirements though. In order to join, you have to have certain minimal level of skills. So I just want to be upfront with everybody. You have to have Python foundational skills. We want to make sure that we can move fast in this bootcamp.
01:21:58
So you have to know how to work with classes and methods, how to manipulate data in Python, things like Pandas and NumPy. You do need to have basic cloud skills. So unfortunately we won’t be teaching basics of AWS. You need to know foundational understanding of how cloud works and introductory experience with AWS.
Jon Krohn: 01:22:16
But if people need those skills, they can go to cloudwolf.com. Isn’t that right to you?
Kirill Eremenko: 01:22:21
Let’s not overwhelm people. Yes, you can learn AWS at cloudwolf.com. And also in www.superdatascience.com, we teach some introductory level of AWS. So CloudWolf is more focused on certifications for AWS. Super Data Science, you can get basic machine learning in the cloud, like things like SageMaker and Bedrock.
01:22:42
Also, another requirement is you need to be able to commit eight to 12 hours per week. That’s the bare minimum. So there will be three or four hours in-person sessions and then you have to still do homework assignments. And, yeah, that’s pretty much it. It’s obviously advantageous if you’ve played around with LLM APIs and some Agentic workflows, RAG, and stuff like that, but that’s not really required.
01:23:06
So those initial things I mentioned are absolute requirements. We’re very excited about it. Our first cohort is launching soon. We’re aiming to launch it in June. And then from there, we’ll be launching. The next one’s probably going to be September or October.
01:23:20
So if you’re interested, apply. There’s a waiting list. You need to put down $100 deposit. And we interview every single person who applies for the bootcamp to make sure it’s the right fit because if it’s not the right fit, then… We help people, and we deliver value. We don’t want to waste your time or money or our time for that matter or space in the bootcamp.
01:23:43
As I mentioned, they’re limited. So if it’s not the right fit, whoever’s interviewing you, me or Adlan or maybe someone else, we’ll let you know right away that’s not right fit, and this is the things that you can brush up on in order for it to be a right fit. Anyway, would love to see people in there if you think this is the right thing for your career.
Jon Krohn: 01:24:02
Very cool, Kirill. I would love to take a course like that myself.
Kirill Eremenko: 01:24:07
Me too.
Jon Krohn: 01:24:09
Yeah. I’d never get past the interview though. You guys would be like, “This guy is a liability.”
Kirill Eremenko: 01:24:16
Overqualified.
Jon Krohn: 01:24:17
Well, this has been awesome, Kirill. So for full transparency to you listeners who are still listening to this episode all this time later, we set out… Were in the middle of Kirill and I had a business meeting just to talk about the podcast business, and he was like, “Can we record a quick Five Minute Friday episode?” I’ve got these five people I want to talk about.
01:24:44
And now, we’ve been recording for almost an hour and a half.
Kirill Eremenko: 01:24:46
[inaudible].
Jon Krohn: 01:24:48
Yeah. It’s going to be a Tuesday episode, and we’re probably going to have to have our business meeting next week.
Kirill Eremenko: 01:24:57
But it was fun. it was good fun. I enjoyed-
Jon Krohn: 01:25:00
I really enjoyed this. I hope listeners enjoyed it too. I felt very relaxed with you. I know you so well. I really enjoyed getting questions from you because I feel like this was a next level kind of back and forth conversation.
Kirill Eremenko: 01:25:15
Yeah. Yeah.
Jon Krohn: 01:25:16
It’s the kind of thing, I’ve never really listened to the all in podcast very much, but I imagine it’s something like that where there’s four hosts on that show, and it’s always the same people. And so you get all that rapport, which I think you and I have.
01:25:28
So hopefully, people enjoyed this episode.
Kirill Eremenko: 01:25:30
Hopefully.
Jon Krohn: 01:25:31
Kirill, and that means now this is a Tuesday episode, I have to ask you the usual ending questions. It’s been a while since you’ve been on the show.
Kirill Eremenko: 01:25:44
Yeah.
Jon Krohn: 01:25:45
It’s been, well, since January. Your last episode was 853. You and Adlan were on the show. Do you have another book recommendation since then, by chance?
Kirill Eremenko: 01:25:53
Yeah. Can I do two books I’m listening to?
Jon Krohn: 01:25:58
Well, I guess you are one of the owners of the show, so I got it.
Kirill Eremenko: 01:26:05
Okay. All right. Thanks. So in terms of fantasy, I’m loving Joe Abercrombie right now, the trilogy called The Blade Itself, The First Law, and the first book is called The Blade Itself. I’m about to finish the third book, the reading, you got to listen to it on Audible though, because it’s read by, I forgot, Steven Pacey, I think. And it’s incredible. It’s kind of like Lord of the Rings, but much funnier and it’s kind of like Game of Thrones. They’re like that much funnier and bloodier than Game of Thrones, if you can get more bloodier than Game of Thrones.
01:26:39
Really funny book. Really fun book. So that’s fantasy. And in terms of a self-improvement, listeners who’ve heard me on the podcast before will know that I’m quite excited about growing myself as my character, my personality, and discovering my psychology and things like that.
01:26:58
The most recent thing that I’ve listened to in that space was an amazing podcast on the Tim Ferriss show, episode 798 with Terrence Real, who’s one of the best relationship psychiatrists or psychologists in the world. And I’ve listened to that episode twice now. Why I mention it as a book recommendation because it shares five chapters of his book called Fierce Intimacy. I haven’t read the book yet. I have just purchased it two days ago on Audible, and I’m going to listen to it.
01:27:29
But listening to that episode 798 on the Tim Ferriss show really has helped me uncover certain things. He talks about that in relationships you’re always in one of three states, which is harmony, disharmony, and repair. And it’s all about how you manage repair, and what are those childhood or inner child reactions that we have. There’s five of them, and he covers all five, and how we go through them and how to notice that and not let yourself fall into those patterns to not ruin your relationship because in that repair time, that’s when you can actually grow or you can destroy a relationship.
01:28:08
So again, the book’s called Fierce Intimacy, but Terrence Real, I’ve only heard five chapters out of it. But so far, it’s been really transformational for me. I’ve enjoyed it a lot. I’m really looking forward to reading the full book.
Jon Krohn: 01:28:19
Nice, Kirill. Thank you for those. And then for following you, obviously, people can go to www.superdatascience.com, join the platform. I know that that’s the number one place to reach out to you.
Kirill Eremenko: 01:28:29
Yep. We have a free trial. You can check it out and reach out to me while you’re in the free trial. No problem.
Jon Krohn: 01:28:36
Nice. I love that. They probably get extra attention because you’re trying to convert them.
Kirill Eremenko: 01:28:42
I wish. Yeah. I wish. We have quite a lot of people joining. I think, yeah, thousands of people joining every day. But of course, I reply as much as I can.
Jon Krohn: 01:28:51
Amazing. All right. Fantastic, Kirill. Thank you so much for being on this Five Minute Friday episode.
Kirill Eremenko: 01:28:56
Thanks a lot, Jon.
Jon Krohn: 01:28:59
We’ll catch up with you again soon.
Kirill Eremenko: 01:29:00
See you in a bit.
Jon Krohn: 01:29:07
Such a fun episode. And Kirill covered how combining LLM knowledge with machine learning fundamentals can be key to landing an AI engineer role, how our field is evolving faster than individual learning pace, but you can find peace by focusing on long-term mega trends like Python and SQL rather than chasing every new framework. How staying in data science is a business-facing translator of insights can remain invaluable. Not everyone needs to become an AI engineer.
01:29:34
And we talked about how age bias is real, but you can remove your graduation dates from LinkedIn to obfuscate your age a bit, how back-to-office trends create regional opportunities and how in-person networking remains the most effective jobs search strategy.
01:29:51
All right. As always, you can get all the show notes including the transcript for this episode, the video recording, any materials mentioned on the show, the URLs for Kirill’s social media profiles, as well as my own social media profiles at www.superdatascience.com/899.
01:30:08
Thanks to everyone on the SuperDataScience podcast team, Kirill himself who’s the founder of the show, our podcast manager, Sonja Brajovic, our media editor, Mario Pombo, Nathan Daly, and Natalie Ziajski on partnerships, our researcher, Serg Masis, and our writer, Dr. Zara Karschay. Thanks to all of them for producing another magnificent episode for us today. For enabling that super team to create this free podcast for you, we’re deeply grateful to our sponsors.
01:30:34
You can support the show by checking out our sponsor’s links, which are in the show notes. And if you’re interested yourself in sponsoring an episode, you can find out how at jonkrohn.com/podcast. Otherwise, share, review, subscribe, but most importantly, just keep on tuning in. I’m so grateful to have you listening and hope I can continue to make episodes you love for years and years to come. Till next time, keep on rocking it out there, and I’m looking forward to enjoying another round of the SuperDataScience podcast with you very soon.
