image: thumbnail.jpg
Teaching out loud and becoming StatQuest | Josh Starmer | Data Science Hangout
We were recently joined by Josh Starmer, Founder and Creator of the massively popular YouTube channel StatQuest with Josh Starmer, to chat about his process for simplifying complex technical topics, overcoming imposter syndrome and feeling cringe, the enduring value of foundational statistics over fleeting AI trends, and how to use AI tools responsibly without sacrificing true understanding.
In this Hangout, we explore Josh's unique approach to teaching and learning difficult concepts. He reveals that he specifically avoids watching other videos on a subject to ensure his explanations remain original. Instead, he focuses on finding the "simplest possible example" that still highlights the method's utilityβsuch as creating a neural network with only three data points to visualize the math. Josh emphasizes that the hardest part of his work is constantly asking, "Can I make it simpler?" and extracting the essence of an equation so that learners can "feel" the statistics rather than just memorizing them.
Resources mentioned in the video and zoom chat:
π StatQuest with Josh Starmer (YouTube) β https://www.youtube.com/c/joshstarmer
π The StatQuest Illustrated Guide to Neural Networks and AI β https://www.amazon.com/dp/B0DRS71QVQ
π The StatQuest Illustrated Guide to Machine Learning β https://www.amazon.com/dp/B0BLM4TLPY
π The StatQuest Store β https://statquest.org/statquest-store/
π Introduction to Statistical Learning (ISLR) β https://www.statlearning.com/
π Grokking Machine Learning by Luis Serrano β https://www.manning.com/books/grokking-machine-learning
π Hands-On Large Language Models by Jay Alammar & Maarten Grootendorst β https://www.maartengrootendorst.com/book/
π Statistical Methods for Research Workers by R. A. Fisher β https://link.springer.com/chapter/10.1007/978-1-4612-4380-9_6
If you didnβt join live, one great discussion you missed from the zoom chat was the "Smug Bayesian" vs. Frequentist debate, which sparked jokes about p-hacking and a community demand for "Smug Bayesian" t-shirts. Will Mike K. Smith answer this call?? Would you wear a Smug Bayesian shirt? π
βΊ Subscribe to Our Channel Here: https://bit.ly/2TzgcOu
Follow Us Here:
Website: https://www.posit.co
Hangout: https://pos.it/dsh
The Lab: https://pos.it/dslab
LinkedIn: https://www.linkedin.com/company/posit-software
Bluesky: https://bsky.app/profile/posit.co
Thanks for hanging out with us! π
Timestamps:
00:00 Introduction
01:47 "Who are you and how did you get to the YouTube space?"
05:38 "What was it that you really wanted to do as a career?"
07:52 "Have you got any top tips on how to make video editing easier?"
10:30 "What is your process for breaking down a difficult concept?"
15:56 "What's the most common statistical or machine learning mistake you see in real projects?"
19:40 "How do you keep up with the hands-on stuff?"
21:43 "What is the one book that every data science professional should read?"
26:13 "What statistical concept do people think they understand but almost always misuse?"
30:42 "What would be your advice for a new grad to break out as a data science professional?"
35:55 "What are your tips for generating confidence to teach out loud?"
40:18 "Do you ever feel cringe? How do you deal with that?"
42:25 "Do you have any career advice?"
45:28 "How have your musical talents helped with your work in data?"
47:25 "How would you integrate AI as a responsible assistant and where would you forbid it?"
47:43 "What data tools and IDEs do you use?"
52:25 "What is a memorable concept that you remember having a personal moment through teaching?"
53:10 "What are you most excited for in 2026?"
Transcript#
This transcript was generated automatically and may contain errors.
Hey there, welcome to the Paws at Data Science Hangout. I'm Libby Herron, and this is a recording of our weekly community call that happens every Thursday at 12 p.m. U.S. Eastern Time. If you are not joining us live, you miss out on the amazing chat that's going on. So find the link in the description where you can add our call to your calendar and come hang out with the most supportive, friendly, and funny data community you'll ever experience.
Can't wait to see you there. I am so excited to welcome our guest today. It's Josh Starmer of StatQuest. Hi, Josh. I would love it if you could introduce yourself to everybody if they don't already know you. I'm so happy to be right here today talking with the data science meetup with my favorite company, People at Paws. It makes my software that I use every day. Hooray. Hello. Hello. Oh, look at all the hearts coming through. I see so many smiling and laughing faces, too, in my camera on crew.
Thank you so much, Josh. Everybody, if you don't know Josh, Josh has millions of followers on YouTube, and he runs a YouTube channel called StatQuest that if you are a statistics person like me who was ever in school and kind of terrified of some of your statistics topics, you probably found Josh's channel to get friendly introductions to complicated topics. Josh, I would love it if you could give us a short rundown on who you are and how you got to the YouTube space and kind of what the ethos of your channel is.
Yeah. So, hello. I'm Josh Starmer. My background is in music and computing. I kind of always have loved them both. The computing has often paid better and had better hours. So, I've focused a lot of my professional life on that side, and I have a strong musical hobby. I, golly, I took a statistics class in graduate school, and I fell in love with statistics, even though the class was absolutely horrible. It's like one of the worst classes I've ever taken. But it opened my eyes to things, to think about things in ways I'd never thought about them before, even though these are obvious things.
Like, it's kind of like, I almost feel embarrassed to say that before I took statistics, I'd never really thought about variation. I never really thought about like, oh, we can quantify. I mean, I knew that like, snowflakes were all different, but I didn't know we could quantify variation, and I didn't know we had tools out there that could help us make sense of that variation and help us make good decisions with it. And that's what I fell in love with, was this realization that like, yes, everything's different all the time, and I can use that information to my advantage. And I was like, wow, that's really cool. Because up until then, whenever I'd done math, math had always resulted in a very, a point estimate was what we call in statistics land, like a single number would be the result. And I was always like, well, that's the way math is. And then statistics was like, no, it's not. This is reality. This is the world. This is not heaven. And I was like, okay. And I just love that.
But I'll be honest with you. So statistics, even though it was a terrible class, I've always struggled with it as a subject. It's even struggling with that very first concept of like, oh, there's this thing called variation that we can quantify. That was hard for me. But everything's been hard for me. And so I kind of started creating, for many different reasons, I started creating a YouTube channel as a way to teach myself statistics in a way that made sense to me. I'm very visual. And I'm not someone who can just look at an equation and go, oh, that makes sense. I have to break it down into sort of a visual interpretation. And then I can go back to the equation and go, oh, this part of the equation corresponds to this part of this illustration or something like that. And I can understand it that way. And my video is my way of sort of teaching myself, but also not to skip too much of the background in this short introduction. I worked in a genetics lab doing data analysis for a long time, for 13 years. And I created the channel also to teach them and my coworkers sort of that what I was doing wasn't magic.
So statistics, even though it was a terrible class, I've always struggled with it as a subject. It's even struggling with that very first concept of like, oh, there's this thing called variation that we can quantify. That was hard for me. But everything's been hard for me.
And I can talk for hours and hours and hours about this. So I'll stop you. I'll save you. I absolutely like the things that I want to call out from that to highlight are that you struggled with anything at all, right? Because I think that people look at your YouTube channel, they look at people who they think just know a lot of stuff and they think this person must just get it easily. It's so wonderful to hear you say, I am good at explaining this because it's hard, because I had to work through understanding it myself. And now I'm just sending that forward, right?
The only other thing I wanted to ask to make sure, maybe you said this and I missed it, was this was not your dream job, right? What was it that you really wanted to do? Where were you before this? Well, I either wanted to be a rock star or you are now achieved. And I actually, I succeeded at that and realized it was not the job I wanted. But the other job I wanted was, I wanted to be a professor at a university. And I specifically wanted to have a laboratory that focused on sort of mathematical biology, did computer simulations of biology and organisms and organs as a way to prevent having to use animals as test subjects. I was like, what if we can have computer programs as test subjects? Can we model the biology accurately enough that we can use that instead of animals for understanding toxicity or whatever? And I had that dream. It's like a weird dream. But I think part of the dream came from me when I was like six or seven years old, misinterpreting what my dad did for a living, because I wanted to do what my dad did. And he kind of told me once and I didn't really understand it. So I came up with this own vision and that really appealed to me. Like I thought it was his vision as well, but it turned out not to be. But I was like, oh, that is what I want to do. So ever since I've, as long as I can remember, that's something I wanted to do. But eventually, for better and for worse, there's something I had to say goodbye to because I'm a terrible grant writer. Much better at making YouTube videos, it turns out.
The horrible reality of academia is that you're actually sometimes just a grant writer. I think a lot of people here could probably identify with that.
Making video content
Well, I wanted to hop right into the Slido and start with a question from Mike K. Smith. Mike had a fantastic question about videos. Would you like to ask that live, Mike? Yeah, I'm sorry. This isn't a stats question. It's much more of a YouTube-y, video-y question. But I'd love to do more in terms of video training and stuff like that on various topics. But the editing and the kind of post-production side of things takes so much time. So have you got any top tips on how to kind of make it so that it's a little easier or shorten that time from idea to final video? I really wish I could.
That would make my life so much better, too. There's no shortcut. I'm just terrible. The way I make my videos is so primitive. I just make a PowerPoint presentation. I use a program called Keynote, but it's essentially a free knockoff of PowerPoint. And I just export the slides, and I import them into iMovie, and I just record my overdubs. It's a long process, and it takes forever. But I've been doing it for so long, so many years, that when I sit down, it's what I expect to do. So it never dawned on me that I could do something else or make it faster. So yeah, unfortunately, I can't give you a good answer for that.
I think that that's a fantastic answer because it's, oh, there's no shortcut, actually. Editing is just really hard. And if you get fancier editing tools, they crash all the time, and they won't just crash themselves. They'll crash your entire computer. My husband is a video producer, and he's the most patient person on the planet. I don't know how he does it. I will throw a shout out out there for Snagit, which is what I use to record code reviews and stuff and tutorials. I really love it. It's just enough tools to do the stuff that I want to do and annotate my screen and have me in the bottom in a cute little shape without it being overwhelming with so many options. So if you happen to like Snagit, I see some plus ones for Snagit in the chat. Go for it. It's worth it. It's by TechSmith, the same company that makes Camtasia. If you really want more fully fleshed out editing, then you could do Camtasia. It crashes on me all the time, too, though.
Simplifying complex concepts
All right. Thank you so much for the question, Mike. Let's head back over to Slido for a question from Dylan, Dylan Polson. You had asked a question in Slido. Would you like to ask that live? And everybody, don't forget to put your name on your question so I can ask you to ask it live. And if you can't ask it, just put a little asterisk or a note in there. I'll ask it for you. Yeah. Big fan, Josh. I am just so impressed with your ability to explain difficult concepts. I've taught a machine learning course and had students watch some of your videos. And then I think to myself, this could be explained to a high schooler. So I'm wondering if you could go through your process of how you take these really high difficulty concepts and turn them into something that's accessible for so many people.
Yeah, sure. I'd love to. So like I said, I'm a pretty slow learner. And so the first thing I have to do is sort of read everything I can about the subject. And I'll be honest, I specifically avoid trying to watch videos about subjects. I'm afraid if I watch another person's video, I'll copy them. And that's the last thing I want to do. Because if somebody else is already doing a great job the way they're teaching, then that's great. I want everyone to go to that video. I don't want them to come to mine, which is just an echo of that one. And so I try to avoid personally, and it's just for personal reasons, because people always ask me like, what's your favorite YouTube channel for learning new concepts? And I'm like, I don't use any of them.
So I read everything I can. And then I try to, whatever it is, I try to come up with the simplest possible example that still highlights one key reason that whatever the method is, is useful. So for example, for when I made a video on neural networks, neural networks can fit straight lines to things. But that's not very interesting, because we've got linear regression, we got a million other tools that can fit straight lines to things. So what I wanted was to come up with the simplest example that could illustrate something that was a little more exciting. And so I came up with a data set that had three points. So one, a low point, a high point, and then going back down to a low point. So we had a nonlinear sort of effect. And it just has three data points. And then I was like, what's the simplest neural network that I could come up with, that will then fit a nonlinear shape to that, and accurately make predictions based on it.
And, and this was actually very challenging, because training a ridiculously simple neural network was non trivial, because I kept getting stuck in, when I was trying to find the optimal parameters for it, I kept getting stuck in really bad local minimum. And I just had to write a program that would just change the seed, the rent for random numbers, it's because neural networks start off with random numbers associated for the parameters, and then it optimizes from those random numbers. And so I had to, I read a program that just iterated through different seed numbers, like for 1000s, until I finally found a combination of parameters that would work. And, and what was cool about this, what the goal was to come up with something where everything could be visualized, and I could do all the math in front of people. And there was no sort of mystery. And there was no just sort of like this matrix times this matrix equals this matrix, and just sort of, which is cool, and an awesome and compact notation, but it's harder to glean sort of exactly what's going on, if you can't follow the math one step at a time.
And so I succeeded. And what was nice is this, this neural network had one input, which was corresponded to the x axis on a graph. And it had one output, which corresponded to a y axis on a graph. And so for every x axis value, I could look at the corresponding y axis, and we could actually see exactly what was happening, we could see how two different shapes, two different curves were getting added together to end up with this nice bell shaped curve that fit the training data. And it took a long, long, long time to come up with that. And I'm probably giving you infinitely more details than you actually need right now. But the idea in general is that I do struggle to come up with as simple an example as I possibly can think of. And that's actually probably the hardest part of what I do is coming up with an examples like, and I'm always challenging myself, can I make it simpler? Is there something I can remove from this? Can I is there is there more of an essence that I can extract from that equation? That really just gets at just the important things. And I'm always asking myself those questions. So yeah, that like, how can I make this simpler? Yeah, use that for your repractice. How can I make this repractice simpler? And then by doing that, you'll probably answer your own question, won't you? Yeah, that was a fantastic
question. Thank you so, so much. And I love the, the breakdown of like, this took a really long time. That's such an empathetic thing to say, wow, this took me a really long time to understand. These are the things that really helped me figure it out. And I'm going to take the time, you know, in the beginning, like for free out of the goodness of my heart to like make this whole video explaining this people understand. That just makes my heart glow. I love it so much.
All right. Thank you so much for the question. We're going to head over to a question from Edward. Edward Clayton Leland. Would you like to ask your question live?
Hello. So it's cool. You did your own like optimizer thing because I'm always like, well, we have Adam that's been made, we can optimize. But my question was, what's, what's the most common statistical or machine learning mistake you see in real projects that doesn't show up in textbooks? Maybe it's people like me just being like, oh, Adam is invented. I don't need to reinvent the wheel. Yeah, I use Adam all the time, by the way.
What's the most common mistake? Oh, golly. That's a, that's a hard question for me to answer. And I'm a little embarrassed as to why, but I will tell you why it's hard for me. And the, the, the issue is, is, is although I do think about these things all the time, I don't, I, I have, it's been a long time since I've actually done it in practice or been around other people doing it in practice. And so I see, you know, when I go to, when I go places and I talk to people, I, I, I see what they're working on, but they don't give me all the details of, of, of, of things.
Although, I don't know if this is a good answer to your question or not. It's not a good answer. I can, I'm just going to change the subject.
Assumptions of linear regression
I think. And the subject, the subject I'm going to change it to is actually something I'm really proud of. So I'm writing a book on statistics and I'll be honest. I've been haunted by the assumptions of linear regression for probably the last 20 years. And the reason why I've been haunted by them so much is that I memorized them, but I never really understood them. I never, I, I, I could, I could repeat what they are and I can repeat basically what they mean, but it's not something that's internal to my being. And, and when I wrote my book, I actually left them out. And the reason why I left them out is back in the, don't tell anyone this, but back in my 13 years as a professional bio statistician, I often just ignored them because we just didn't have enough data to even care, you know? So in practical settings, I know a lot of people freak out about all those, those assumptions. And for the most part, nine times out of 10, your data is going to be fine. Um, and don't worry about it. And, and so I, for, for a while I was like, I didn't even leave it. I did not include it in my book because I didn't want people to worry about something that in practical settings isn't actually worth worrying about that much in rare settings. Yes, sure. It's a big deal. But oftentimes for just really simple data sets, which is 99% of what, at least the stuff I was doing was like, it just, it was, it was above and beyond like even possible. You just didn't have enough data to test all those things. So the, um, but in my book, I'm so pleased. I'm so, so, so pleased. I came up with a way to at least for myself that I can now look at data and just go, Oh, I know that this is not a good, a good data set for regression. And I can tell you exactly why. And I can point it out. And it just means these predictions in this region are going to be terrible. And I came up with a way to like, feel it. And I no longer have to memorize these things. And instead I can just look at the data and go, yeah, this is why don't worry about it. You know, or, you know, this is, this is terrible. I don't know. I'm just super pleased with this sort of revelation. I like, I overcame this thing that had been sort of like an itch I couldn't scratch for years and years and years. And I'm like, I finally, finally got it. So that's, that's my pivot on your question of what people mess up. That is okay. It is totally okay to pivot.
back in my 13 years as a professional bio statistician, I often just ignored them because we just didn't have enough data to even care, you know?
Keeping up with hands-on work
And that was one of my questions that I sort of asked Josh when I first met him was like, well, how do you keep up? Because I stopped doing data science and stopped, you know, doing all of the stuff that was like hands-on practitioner work every single day in order to do this, right. To be here for the community. And that's something that Josh is doing too. He's like, I'm doing all of this for all of these people. So how do you keep up with the hands-on stuff? Is it a worry that you will like slip behind, you do personal projects?
Yeah, always. I'm always worried. And I've almost, so I used to work in a lab and I still go back to the lab on a fairly regular basis to present rough drafts of videos that I'm working on because they were my original audience back in the day. And there's still, I think they're the hardest people to explain because they basically hate the subject. And so they're just tough in so many ways. So if I can get something that they like, it's great. But what do I need to do? Yeah. So that's, I'm actually, so my old boss, the boss that let me go because I was terrible at writing grants, we still actually get together. He's one of the best influences I've ever had in my life. He's the one that said, look, you are bad at writing grants, but you're good at this YouTube thing. You need to give it a try. And so he pushed me out of the lab and I've, instead of sinking, I was able to swim. And I owe a lot of that to him. Anyways, me and him are going to start doing some longitudinal data analysis in the spring. And I'm so excited about it because I've also wanted to do a series on time series, longitudinal analysis techniques for forever. And so I'm just going to dive in it with him. He's got some data that he's been working on. So I'm going to get back into the game a little bit and we'll see how it goes.
All right. Well, I think everybody should get themselves a boss that will kick them out for their own good. That's a rare person.
Book recommendations
All right. I have so many questions in Slido about books, book recommendations. I don't know if you have any, but let me sort of bring some of these together because there are so many. Satkar asked, what is the one book that you think every data science professional should read? Which is a great question. And let's definitely go with that one. What's the one book that every data science professional should read? But there were so many others that were like, what's the best book for somebody who's doing specific things like ML, specific things like AI development. What would you recommend as a book for data engineers or AI engineers? So, so many that are along the same lines. So let's go with the one book if you had to choose one.
Yeah. So a shameless self-promotion. I've written some books and I've actually, I've written two and I've got a book on statistics coming out and I absolutely adore my statistics book. I love it. But I also, I've got a book on neural networks and AI and I have a book on machine learning and all of my books are just, they're almost like, like, like graphic novels. It's all pictures. Cause that's, I think in pictures. So I just draw pictures and then I annotate the pictures. They're going to show you what's going on. So I do love my own books. I do have some good friends though, that have written some great books that sort of depend on the, on the field. I've been for the past four years, I've been in the AI ML world sphere and I'm trying everything I can to get out of that sphere and get back into like hardcore statistics. Cause I, to be honest, I find that actually more interesting because the statistical methodologies are, I just find them more interesting at a fundamental level. Whereas AI, I feel like a lot of AI is just optimization. Like, can we do this with less RAM? Can we do this faster? And those tricks I think are cool, but they're very ephemeral, right? Cause five years from now, someone is going to even have a faster method and a better way of using RAM. But statistics has a way of being useful for a very long time. Like, like the, the, the discoveries that we make in statistics and the methodologies that we have, have staying power. And, and for that, I have a ton of respect for it. And that's why I'm trying to get back to it. So it's not just like the AI du jour type situation that I had been in, but I've got a lot of great recommendations of good AI books that I'm a big fan of. I've got my friends, Jay Alomar and Martin, I can't remember his last name. They've written a fantastic book on how to use agents and how to use LLMs that I really like. And I've got a, a good friend who, Louis Serrano, that's written a nice book on machine learning called rocking machine learning or something. He always uses really great examples.
But statistics has a way of being useful for a very long time. Like, like the, the, the discoveries that we make in statistics and the methodologies that we have, have staying power.
Him and me, I like to actually like to do a lot of presentations with him because we're almost like complete opposites. But, but I love that because I, I, because I learn a lot from him and he learns a lot from me. And, and, and I think between the two of us, pretty much anyone can learn anything if, if, if we're presenting it. So I don't actually have that too many book recommendations other than a shameless self-promotion and some AI books. That's all right.
Well, I put the longest link ever in the chat to one of them. I found both of them. The rocking machine learning, and then also the other one. So I put those in the chat if anybody would like, Oh, it looks like Rachel's got a link in there too. I wasn't sure if I had the right one, but I found an illustrated guide to AI agents, but it looks like it was the, both the authors you mentioned. Yeah. Yeah. It's Jay Alamo. Yeah. Yeah. Awesome. And I think that one's coming out pretty, or maybe it's already out. It's, it's brand new. Yeah. I think it's brand new. Cause I found their LinkedIn post talking about it coming out. So cool. Not sure if it's available yet. The very long one that I linked to is hands-on large language models, language, understanding, and generation. Same, same authors. Yeah. The kangaroo on the front.
Okay. If you are watching this on YouTube later, these are going to be linked below. Do not panic. I'll put them in the, in the resources for you. I'm Sarkar. Thank you so much for asking that question. Thank you to everybody who asked a book question. Put your recommendations in the chat so everybody else can see them too.
Misunderstood statistical concepts
Yeah. Okay. We had an anonymous question that got the most uploads of any question on Slido ever. And it was what statistical concept do people think they understand, but almost always misuse in practice. This is another practice question. I'm not sure you feel free to pivot. Well, there's, there's always the P value, right? I mean, that's notoriously bad. And actually, interestingly enough, like confidence intervals are so bad that I just left them out of my book. I don't want them to be part of the conversation because even if you understand them, nobody else will. So just do everybody a favor and don't ever publish a confidence interval. That may be a controversial opinion, but I just think it's for the best. I just know that people will misinterpret that thing. And, and just don't do it. Just let's just like start a new campaign. Just like no more confidence intervals. Maybe I'll put a, like a, I get a shirt that has a CI with a, like a line through it saying don't do it. Those are just notoriously awkward. And, and so P values along those lines, people, a lot of people, even if people understand the P value and they know exactly what it means, they still treat it as if it's all you need to know. And you don't also need to know the effect size, right? Because if the effect size is like minuscule, then it's who cares what the P value is, right? The P value is, is, is only useful in the context of knowing like what the R squared is or what the difference in means are. You have to give it context for it to have any kind of reasonable interpretation. And a lot of people just present it on its own. And, and you're like, no, don't do it.
Yes. It's more nuanced than a threshold. And I shout out to Nick Drew who like put those exact two things as his reply in Slido. He said, plus one to this topic, especially P values and or confidence intervals. That's great. So hard. I think that they're, they're hard because if you don't know anything about statistics, you can look at them and just so easily infer something else because those are English words that mean something to everybody. And the things that you think they would mean is not the thing that they mean, right? So it's always tough in case anybody has a non-statistics person here and hears us talking about this stuff. It can be tricky. I saw lots of shout outs in the chat for it. This is why I'm Bayesian. So I figured I would ask you that question. Yeah, of course. All Bayesian people are like, oh yeah, oh yeah. We recognized that years ago. We don't do that anymore. And I think, and that's legitimate. I am, I think that's fine to, to be a smug Bayesian, but the, but let's be honest in the real world. And let's be, let's talk about the real world, not this little tiny Bayesian bubble we live in that where everything is obvious and easy to interpret in the real world. People are just doing those P values left and right. And that is the world we actually have to deal with in a practical way. And it's one of the reasons why I've, I'm planning on doing a series on Bayesian statistics because it is awesome and I'm a big fan of it. But it's one of the reasons why I've, I've, I've waited so long to do it is can you do Bayesian statistics in Excel easily? No. Okay. End of discussion. Accessibility, I guess, is the point there.
I have rarely felt like I needed to mute myself because I was in such danger of laughing as when you said smug Bayesians. I'm losing it. Y'all, I'm losing it. And we have a consensus in the chat that we would all buy a shirt that says smug Bayesian if it existed. I feel like my finger pointing to myself. Mike Smith, please get on this. You need a smug Bayesian shirt.
Okay. This is another great question from Satkar. So Satkar, if you are available to ask this live, I would love it if you did. I'm going to highlight it in Slido so it goes to the top so you can see it and everybody else can see it. But I'm also going to copy this into the chat right now.
Yeah, for sure. Thank you so much, Libby. Bam. And I appreciate Joss very much because I think your content actually got me through grad school and I just graduated yesterday.
Triple bam. Amazing. Satkar got a triple bam, y'all.
So my question is like, what would be your advice for a new grad, especially someone without industry experience, to break out as a data science professional, especially since the job market is brutal and competitive right now?
Yeah. You know, again, I wish I had lots of personal experience that I could reflect and go, oh, I know this works or this doesn't work. I can just tell you what I've heard from other people, which is practice your skills out loud on GitHub. You know, build a portfolio of data analysis projects. You can get data sets from wherever and just show that you know what you're talking about. I think that's a, I mean, I've heard a lot of people say that that's a good thing. And I've talked to some hiring managers that say, yeah, that kind of stuff is impressive.
I've also talked to some hiring managers that have said, don't worry. They said, the one thing they said was everyone thinks they have to know the latest AI thing these days. And they say that the reality is that it's not like that. You know, the reality is like, it's just like we were talking about with Bayesian versus the rural world. Most organizations are doing their statistics and their predictions in Excel, right? They've got a spreadsheet with two columns of data, you know, and so it's, you know, they were actually emphasizing that a lot of companies are actually looking for people that are just really solid on more conventional data analysis techniques. And maybe they've, you know, they can use AI tools to work faster, but they're not like, you know, putting prompt engineer on their resume or something like that.
So I don't know if that's helpful or not. I think it's very helpful. And a lot of people would agree that portfolios are super helpful, but also like hard. If you are struggling with creating a portfolio, I feel you, I do not have one. I'm going to put the link in the chat right now. Everybody, everybody chill out in the chat. I'm going to put something in there. Go look at it. This is our survey for the end of the year for the data science community. And one of the questions you wanted to know, which is how I built this survey is, do you have a portfolio? I put a little place in there that's like, yes, I have a portfolio and here's the link. So I can put a portfolio gap, you know, gallery together. So you can see other people's portfolios.
Scott Carr is sharing his portfolio there. Thank you so, so much. Zach says, sorry, we're talking of today. It's, it's the Josh effects you guys, but let's be real. We're always wild.
So please go take that survey. It's going to take you like 10 to 12 minutes. It's like about 40 questions depending on how you navigate it. And I wanted to shout out Nathan, Nathan Jeffrey. He is here. He's a big part of the data science hangout crew, and he helped me make this survey better. If you don't like it, it's my fault, but it's way better than it was when it first started. And that's thanks to Nathan. So thank you so much, Nathan, for your reviews and your help. I am not a survey writer, y'all. I do not like analyzing survey data, but I'm going to do it for you.
So we will put out the results to that and have it in the new year. Please stick around and join the discord server so that you can see the results when they, when they come out. Okay.
Thank you so much, Scott Carr. And let's pivot to Noor because I saw Noor had a question and I would love her to ask it live. Hello. There she is. Hi. Long time no see. Good to see you. Holy smokes.
Wow. Yeah. For a very, very, very short context, I worked in, when I was in grad school at UNC, I was in like the lab next over from Josh's. So when he was like being a rock star and starting his life, I was being a very, very depressed little grad student, but. You made it. You made it. That's the important thing. I did. I did. Still on the market, but like, you know, life works. Life moves on.
Teaching out loud and imposter syndrome
But yes, this is more of a more generic-ish question. Let me try and dig up said questions. Oh, I highlighted it for you, Noor, but I'm also going to stick it in the chat right now. Because I have like 10 windows open. Yes. Okay. We talk a lot about learning out loud and I'm working on that whole process, but arguably you took it further with the whole act of teaching out loud. And especially because at least when I knew you, I was like, I didn't know a biology postdoc was going to be teaching about stats because I, when I was at UNC, was one of the people who was often, my fellow bio grad students came to me because I was one of the few people who knew what stats was because I came from a math and stats background. I was like, we don't do p-values that way. No, that's not how you do it. But how did you, you took it further with teaching out loud. What are your tips for like generating confidence with that act? But it's like, I do know these things, even though technically formally my background is X per se. Yeah. How do you avoid imposter syndrome is basically. That's a lot more elegant than what I was saying, actually probably more accurate.
Yeah. So one thing that a lot of people assume about me that is not correct is I've actually never taught a course. So I've actually, the only teaching I've ever done has been a sort of informal, you know, you know, it during a lab meeting when I was at UNC or, and now making YouTube videos. And occasionally I'll give like a seminar and, but I've never actually taught a class. And so I've, I've had a lot of imposter syndrome myself. And, and how did I get it over, over it very slowly. And I still have it to a certain degree. I actually just had to present basic statistics to the statistics national honor society. And that was one of the most stressful things I've ever done in my whole life, because I was not a great statistician. And, and this is when I, and again, when I was a student doing statistics, it was, I would have, it would have, I was like, if you told me then like one day you will be inducted into the national statistics honor society, I would have said, no, that's, that's impossible.
But you don't know you're, you missed me in another Josh Dahmer. So, and so that was kind of like stressful for me. Cause I actually do, I think of myself as a statistician, but fundamentally I think of myself as more of a, just a computer person. Like I think in terms of programming and iterating and for loops and while loops. And when, when I think about statistics, I don't think about the theory so much as I think of experiments that I can do to sort of confirm my hunch that this P value is significant. And so, so one thing I do, you know, with, you know, like, instead of just like, cause a lot of statistics is based on like theory that was developed or at least originally published a hundred, a hundred years ago.
And so in 1925 Fisher published this book called like, and the analytical techniques to use with data or something like that. It's got this really generic title, but it introduced basically all of linear models in one book. Although it wasn't sort of a, sort of a sort of an overarching, like this is linear models. It's a similar thing. He, he, he did a terrible job teaching it because he treated each thing as if it was like its own sort of like set of rules. So it's kind of confusing and hard to read for that matter. And actually that set off the tone for how statistics would then teach that subject matter for the next hundred years, which is a horrible thing to have done. But so when I think about statistics, you know, you guys look at the theory and I can say, well, without a computer, you'd have to do it this way. And, you know, and it's very complicated. You have to do a lot of calculus and it's kind of a nightmare. And, but then I was like, but computers are so fast nowadays. We can do it. We can actually test the null hypothesis and the frequentist assumptions by just doing tons and tons and tons of these tests and just see what happens.
And it's a trick that Bayesian people have been doing for years. And it's like, why aren't frequencies doing the same thing? Like frequencies are kind of smug. They're like, well, we've already done the math 50 years ago. So why would we do it again? But the, the thing is, is, is when you do it, when you do, when you test it out yourself, I can feel it. And when I can feel it, I can be confident in it. I'm like, look, I, uh, not only does the theory work out the way it's supposed to, but when I did that, right, wrote this program to like test the null hypothesis, the way, you know, and do it a billion times and blah, blah, blah. I got the exact same number. And when I do it that way, I can feel it. And when I can feel it, I feel real comfortable talking about it in front of other people, unless it's the national statistics honor society. And even then my knee will shake.
is when you do it, when you do, when you test it out yourself, I can feel it. And when I can feel it, I can be confident in it.
I like, uh, knowing that you get shaky too, because I get shaky every day before the hangout. And there was a question, an anonymous one as well. That was like, uh, do you ever feel cringe? How do you deal with that cringey feeling? It's not exactly imposter syndrome. It's more just like, Oh, recording myself is yeah. Yeah. Have you just completely gotten over it or do you still feel it sometimes?
Well, uh, so, uh, here's a, here's a story kind of that might sound unrelated, but I have, I'm definitely afraid of snakes. Um, I've always been that way. The snakes just, I have a, it's just fun. Whenever I see a snake, I have a heart attack. Whenever I see a stick, I have a heart attack. When I see a rubber on the street, I have a heart attack. I just have heart attacks all the time. Snakes, snakes, snakes. I do not like them, but I once went to a zoo and they said, here's a snake you can hold. It won't bite you. And if it does, don't worry about it. Um, and I was like, okay, I'm going to touch a snake. I'm going to hold it. I'm going to, I'm going to put it in my hands. And I held it with the hope that it would eradicate my fear of snakes. It didn't. Um, I'm still definitely afraid.
And I'll be honest. I do these silly songs at the beginning. And every time I do one, I'm completely embarrassed by myself. Uh, it's, and they never turn out the way I want. I always come up with different words in advance. And then when I'm doing them, everything gets banged, mungled in my head. And I, and I always mess it up and I'm always like, Oh, you know, I'm always regretting that. Uh, but, but I always try to remind myself that, that nobody cares about that stuff. No. And in the chat, everyone's saying the songs aren't silly. We love the songs. Yeah, exactly. And that's the thing is like, I have to remind myself that it's like, the important thing is that we're having a good time and we're learning something new and, and, and that's, what's important. The details of like, Oh, I wanted it to be this way. And it ended up being slightly different. Doesn't matter. What does matter is like I said, yeah, we're having fun. We're learning. And as long as we're doing those things, it's all those little details of like expectations versus reality. None of them matter. And I just have to remind myself that a million times over and over again. And no matter how many times I do it, like with the silly song, I always have to go, Oh, I can't believe I just did that. And I go, look, you know, just, you know, you'll never get over it, but, but just don't worry about it because that's not important.
Career advice: finding what's missing
Do you think that that is like the career advice that you would give somebody at this point? Or do you, do you have anything else in your pocket for career advice that might be good? Yeah. So I do have I do have career advice. It's not for everyone though.
Um, but it's worked for me, which is, um, I always look when, when I, whenever, in every job I've ever had, I've always looked to see what isn't being done that needs to be done. And I've always tried to find those things that nobody's doing that need to be done. And I've done it very successfully in a lot of different, um, avenues in my life. I've done it. Um, I've done it with music. I grew up playing the cello and I loved the cello and I wanted to be a classical musician, but I wasn't quite good enough for that, but I was more than good enough for rock and roll. And it turned out that the instrument nobody was playing was the cello. And so that was my way of like looking around and seeing what people are doing and saying, well, what are they not doing? Nobody's playing the cello. So I'll start going up to rock bands and say, Hey, do you guys need a cello player? And a lot of them said no, but then one of them said yes. And then we were hugely successful and that was awesome.
Uh, likewise when I worked in the lab, um, one of the things that, you know, I could have just done the math and just given people P values, but instead what I did is I looked around and I said, well, show me how your experiment works. I want to, I want to see how it works and I want to see what's missing. I want to see what, what details you're telling me and what details you're not telling me about how this is done so I can get a better sense of what the data are going to be like. And so they would, they would show me stuff. And, and I'm very, now then I'd be like, Ooh, what just happened there? And it's like, they're like, Oh yeah, a lot of stuff happens there. And then they would, you know, I would discover things that, that, um, that they weren't telling me that needed to be corrected for. And so I was always looking for things that were left unspoken that needed to be attended to in some way. And I've found that's been, that's, that's, that's made me sort of critical on a lot of situations where, um, you know, once, you know, sooner or later the organization catches on that to, to the fact that even though they hired me just to do the data, that I'm actually doing all these other things that are critical to just doing it right. And, and, and getting stuff published and stuff like that, uh, that they're like, Oh, you're very useful. Um, and so I always try to look for the things people aren't saying and the things people aren't doing, um, as a, and then trying to focus on those because it's easier to compete with
nobody than it is to compete with everybody. I love it. And I, I love the anonymous question that, um, that came up that was, it's bubbled to the top as well. That was like, you know, what are some transferable skills? Like transferable skills come from unexpected places.
How have your musical talents helped with your work in data? And I know so many musicians in data. Yeah. Yeah. So one surprising thing is, uh, with the musical talents is, um, yeah, is I, I just like being on stage. I'm a performer. And so I take that attitude when I speak in front of people of, of instead of like holding the microphone all the way down by my belly button and kind of mumbling into it, I get up on the microphone and, and speak in a commanding way because I'm comfortable being on stage in front of audiences. And that's came from being a musician and came from playing in front of people and sort of putting my heart on my sleeve and saying, okay, I hope you like it. Um, and, and that's, that's been useful. The other thing that's very useful about music is it's, um, it's also good for clearing my head. Uh, so if I'm stuck on a problem, I can always pick up the ukulele and strum. It's hard to sing and strum and think about math at the exact same time. Um, and so it's a great way of sort of resetting my brain and starting over. Um, and so I use it as a, as a, as a kind of, instead of like doom scrolling, you know, or something like that, I'll, I'll, I'll, I'll do a little silly song about whatever my problem is or, or even just whatever I had for lunch.
Um, what a good piece of advice. I think doing, doing anything that is silly and doesn't have a point is really cathartic for me. And I'm just like, I'm allowing myself to do this. Um, okay. I love that. I had a question in the Slido that, um, I lost. Hold on. Let me find it. There it is. Let's highlight it so that it goes to the top here. Um, and it was an anonymous question. So thank you. Anonymous asker, whoever you are, we have about eight minutes left. Um, and so I'd like to roll this question into one of my own. The question is Josh, if you were building a StatQuest learning system today, how would you integrate AI as a responsible assistant and where would you forbid it to protect true understanding? So the question that I had was about tools. What tools do you use as far as data tools? Do you work in R? Do you work in Python? What IDEs do you like? If you use, um, coding assistants, like this question is asking about what do you use? Where do you integrate them? Um, so I would love it if you could start with tools and then I will remind you of the, where would you put AI as an assistant? Where would you forbid it? Yeah.
Tools and AI in learning
Yeah. So I use, um, I use R all the time, uh, for all of my simulations of P values and things like that to make sure that I understand the math. Um, cause it's just so easy. I, I use Python for doing neural networks. Um, but, uh, but basically I use R for everything, even just like calculating basic things. Um, and I use RStudio and I use Positron almost daily. Both of them. Um, I love notebooks. I, I, in general, I just do everything in notebooks these days, uh, unless it's super basic. What is your notebook of choice?
Cause I like Google Colab a lot. I work a lot in Quarto. It depends on what I'm doing and who I'm showing how to do stuff. Yeah. So I do, I, I, yeah, exactly. So on my own, I just do everything in sort of like Positron or even just Jupyter, you know, just to on, on, on, you know, just to have it on myself. But I love, I love, uh, Google Colab because it can, it actually can do R as well. Like I can do Python, I can do R in, in Google Colab and, and I, and then I don't have to worry, you know, it's everything's installed. It's like, I can share it with everybody. And I, to me, that's one of my favorite teaching tools ever. Um, cause it's, it's instantaneous. It just works. You can play with it, you can break it. Uh, and it never touches the original file unless it's, yeah, you can literally just type in a cell, pip install something and it's going to install it if you need it. It's one of those accessible tools where I can be like, you know, okay, if we're not worrying about installation, I can make this work right now and I can show you how it works. I can share my screen right now. I think that that's fantastic. Um, also you can use Google's TPU to run your deep learning models and they'll run really fast, way faster than they would on your machine.
Um, okay. So let's get to the question that was anonymous, which was where is AI a beneficial, responsible assistant and where would you forbid it to protect true understanding to make sure people don't, uh, you know, circumnavigate the learning part? Yeah. So I'm a big fan for using anything you can to write, write the software. To me, a lot of the understanding doesn't come from writing the software. The understanding comes from interpreting the results, uh, correctly. Um, and so I, you know, if people want to use AI to write their software, great. That just gets it done faster. I mean, make sure you at least can read it and annotate it, like go back and like annotate or at least read the comments that the AI wrote and make sure it works the way you want it to. Um, but in terms of interpreting the results, I think it's really important to do that yourself. Um, because that's, you know, that's because statistics is the world, the mathematics of the world, because there's variation because nothing's definitive because you're looking at probabilities of like, of ranges of things. Interpretation is really critical and nuanced in a way that AI might be able to do. And I might be able to do, and it might be able to do well, but if you can't do it yourself, then you, you don't understand it. And I, and, and, and if you don't understand it, then it's, I don't think you can confidently make recommendations to your boss or to other people. Um, I think it's critical that you want, you know, that you'd not just, you're not just a parrot for the AI because if you are, you're going to get fired, right? Immediately because the AI is doing your job, right? So you, you have to come up with some way to add value to that. And I think the way you do that is understanding all the nuance involved in the decision-making that statistics requires you to do.
Um, I think it's critical that you want, you know, that you'd not just, you're not just a parrot for the AI because if you are, you're going to get fired, right? So you, you have to come up with some way to add value to that. And I think the way you do that is understanding all the nuance involved in the decision-making that statistics requires you to do.
Absolutely. Fantastic answer. Fantastic question to our anonymous asker. Thank you so much for asking. It clearly spoke to me. Um, and I think that we have to wrap up very soon, but we have maybe like one minute left to ask one super quick one. So our sentence had asked, what is a memorable concept that you remember having a personal aha moment through teaching it or creating content for it? Do you have one? The very first one for me was R-squared when I could visualize it and feel it inside. I was like, Oh, I get it. And that was one of my first videos I ever made. We got nine views in the first year, but for me it was transformational. Oh my gosh. You guys, even if you only get nine views, keep, keep going, keep making your stuff. And if anybody wants any encouragement for me on like, you know, being cringe or doing it scared anyway, listen, I do all kinds of weird stuff on the hangout live on YouTube. I touch my hair all the time. I, I make, I think my voice is ridiculous. Go ahead and do it anyway. Um, and maybe one more since that one was pretty quick. Is there something that you're just like the most excited for in 2026 since we're at the end of our year?
Oh yeah. I'm excited about so many things. I got my statistics book coming out in February, which I cannot wait for. I'm so excited. Um, but I'm also excited to kind of like break through free a little bit from doing AI and doing lots of videos on all kinds of like time series, getting diving into actually doing time series, but also making videos about getting into getting my hands dirty, but also learning a lot of new things. Nice. Are you doing the, the full, like super simple rolling averages all the way to Arima Sarima, Sarah max, all of that stuff. What are your favorite time series? Oh, I'm adding questions. Sorry. We have a minute left. I know what a rolling average is. So we'll start there.
Okay. Sorry. We're going to pivot. Okay. I would love everybody in this last second here. I'm putting it in the chat. Go take our survey, please, please, please, please, please. So that I can get the results real quick. Josh, somebody said, is there a pre-order link for your book? Yes, please. But there isn't. So keep, keep tabs on LinkedIn. I'll, I'll publish it as much as I can bear to do. I'm very shy about that, but I'll, I'll try to publicize it. Yeah. I will put it in a data science hangout channel on the discord. As soon as I see it, I do have your book, but it's on my iPad. So I couldn't bring it to show. Thank you so much for coming, Josh. I hope you had a good time. It's wonderful. Thank you very much for inviting me. I had a great time.
Oh, we did too. This was super fun. I know everyone, this was not long enough. We needed like a double time hangout. I hope everybody enjoys their holiday. If they're taking a break, don't forget that we do not have a hangout next week or the week of new years, but we will see you back in January when we are going to be joined by Adrian Perez, who is the head of people analytics at GitLab, which is super exciting. GitLab is just like the most amazing open company that shares so many good things about, um, the way that they work with the world. So I'm really excited to have Adrian on. Thank you everybody. I'm going to give you a second here to save the chat. If you want to top right corner of your chat, there are three dots. You can click those and save the chat so that you can save all of our links that we shared. And you can go find Josh on YouTube to follow him, obviously, and on LinkedIn. All right. Hooray. I love it. Thank you so much, Josh. And thank you Libby for everything you do too. Have a great holiday, everybody, and see you next year. Bye everybody.