Deep Dive with Daniel Bourke | Will AI Take Over, Worst Part of ML, AI Anxiety & 20+ More Topics

So yeah, the blog post is called AI Anxiety, and I feel like it was just a, it's a term I was experiencing it, uh, from, to define it would be AI is coming out. Is AI going to take over humanity at a high level? Is AI going to, at a lower level, like just replace everything that we do? How do people derive meaning from life? XYZ. So I was seeing all of these and what I kind of, I'm kind of past that feeling now, but there was, because I, the reason why I wrote it was to sort of work through my own thoughts. And that's what a lot of my writing is basically is just to, hey, I've got this thought. Is this true or not? Can I just write 1,000 words on it and argue with myself to see if it's, if I should be thinking this way. So that's the main reason I wrote it, but I also wanted to publish it. It was originally like just a journal entry, but I'm like, oh, I'll publish it because there was a vibe on the internet, which was, and still, and probably will be for the foreseeable future, was these new AI announcements are coming out. And because there was so, I mean, I think February and March was just, there was something every week. It was kind of crazy.

Yeah.

And it was just a lot coming out at the time. And of course, with things like ChatGPT, it seems like magic. And it's like, oh my gosh, if this is good at doing that, is it going to take my job? Oh, if it's so good at just taking my job, isn't it going to just get super smart, learn on its own, take over humanity?

Big concern for a lot of people. Yeah.

Right?

What you just heard right there is a concern that I've personally felt, and many of you have probably as well. And who can blame us? AI is really all the buzz these days with a lot of doomsayers trying to freak us out. But that's why I'm pumped that I got the opportunity to sit down with ZTM instructor and machine learning engineer Daniel Bourke. Not only did he make me feel better about this whole AI anxiety thing, but I was surprised to hear that he's even felt it himself as well. Daniel is such an interesting and thoughtful guy. This conversation was truly a blast. I know you're gonna love it. We covered a wide variety of topics, so you're definitely gonna wanna stick around until the end. But like always, if you're looking for a particular portion in the interview, please feel free to use the timestamps below. One last thing, if you'd like to see more insider secrets from professionals in tech, or just show us some love, it would mean a lot if you'd hit that like button. Okay, I lied. Last thing for real this time. If you had a question for Dan that we didn't cover, drop it in the comments below, and I'll try to get Dan to answer as many as possible. All right, enjoy the conversation. How's it going, man? I really do appreciate you coming out here and doing this sit-down interview with us. Before anything else, can you introduce yourself to our audience and let us know what you do as well?

Sure thing. It's good to meet you, Aldo. Talking before on Discord and whatnot, but now in the flesh, well, kind of digital flesh. But hello everyone. My name is Daniel Bourke. I'm a machine learning engineer slash writer slash depends what day of the week it is. I might be a farmer, but I live in Brisbane, Australia and a machine learning engineer, but I also teach machine learning online. I create YouTube videos and yeah, I think that's me in a nutshell.

And you live in Australia, man. Many would say you have the perfect life. Look, you know, we work online, you live in a great area. I mean, Australia, I've heard great things, so it's definitely—

I'm obviously pretty biased. I do like— I really do like, uh, Australia. I live in Brisbane, and then in Brisbane, I live in North Brisbane in a suburb called Brighton. And Brighton is— it's like a 5-minute 10-minute walk to the waterfront. It's not quite a beach where I live, but it's definitely a large body of water. And a lot of my coding problems get solved down there. I go for a walk and then bugs seem to fix themselves in my brain and then I come back and fix them on the actual computer. So yeah, it's a good place.

I'm glad you said that because I feel like, again, I've only coded for like a year, but I feel like those walks are super meditative for me. Almost all, just outside of the technical stuff, outside of coding and stuff, just problems in life. I feel like going on those walks just helps me out.

So I agree. Well, dude, I, I, uh, I, I totally agree. I love walks. I mean, like, I wrote a book about walks. It's called Ballad of the Walk.

Are you serious?

Yeah.

Yeah.

Well, that's, that's my, uh, that's like my, um, that's my novel. So, um. It's, yeah, it's called Charlie Walks and it's basically a guy who likes going for walks. Where can we find you? There's a few more details than that.

Uh, charliewalks.com.

Or, uh, um, on my website, of course.

So beautiful. And what was your website again, dude? Just, uh, put that out there.

Mine is mrdbourke.com, M-R-D-B-O-U-R-K-E dot com. But if you search Daniel Bourke, I should be Somewhere there, it should pop up.

There shouldn't be many Daniel Bourkes, I hope.

Well, there is. There was another guy I actually went to school with, um, not, not at my school, but the same grade as me, but the school on the other side of town was another Daniel Bourke. And so we'd play cricket against each other, and I, I see his name on like all this stuff. And, uh, so we'd never met. We still haven't met, but it would be like That's, there's another Daniel Bourke out there somewhere.

You didn't mistakenly grab a stud?

No, I just saw the cricket bat like with, with Daniel Bourke on the back and I'm like, that's not mine.

That's funny. That's funny. Um, back to the, uh, so you said you work, um, as a machine learning engineer, correct? Or did I miss you?

No, that's correct. Yeah.

Beautiful. What made you get into the field? Like what sparked that curiosity for you in the first place? And around what age did, I guess, did you start diving into that kind of stuff?

Um, so 2017, 20— or probably more so 2015, I guess. Uh, like, I've always been a tech nerd, um, computer nerd, like, growing up and whatnot. Into Xbox, into hacking my dad's school laptop and just, yeah, doing a whole bunch of— because he was a teacher at my school. So, uh, yeah, it used to just— he used to— he never learned to use a computer. Like, uh, he just got help from, um, my mom and like we would help him here and there. Uh, so it's kind of a different lifestyle, but I like that sort of juxtaposition. However, machine learning, I started listening to some podcasts and a lot of them were sort of around tech. And then of course, 2015 till now. Now, basically machine learning was just really coming into the mainstream. I mean, in 2012 was the big moment for deep learning, which is a subset of machine learning. It can get quite confusing in terms of nomenclature and whatnot because it's such a broad field. However, I was working at Apple at the time, not in building the products in the Apple stores. So I was a genius is what they called it, the computer technician. So I would, I would repair the computers and the software on the machines, but I wasn't actually building the machines or, or the software. And then I got to a point where I'm like, I kind of want to, uh, learn how to build things for these. I want to build my own apps. And at the time, of course, machine learning was all the rage, but I started with web development. Quit Apple, started like, wanted to work on a startup with, uh, my a guy I used to work with at Apple, and my brother and stuff like that. And we were building a website called Any Gym, and we wanted the idea to be— so we were all working out, but then you go to someone else's gym, you want to work out with your friends, but you've got this kind of just massive friction. You just imagine Airbnb for gyms. We were like, yes, this will work great. Like, we can just— if I want to go for a session with you, Aldo, and you, you go down to Gym X and I go to Gym Y, we can just, I'll just book a pass and we'll just walk in like it's any other gym. However, we kind of quickly realized like that to do that, we needed gyms on board. And so we were walking to different gyms, talking to them. And then we also realized that the whole, a lot of gyms' business models rely on people not showing up.

Yeah, they want that.

They want people up. Yeah, they want people not showing up because then it's like you've got like 2,000 members, but only maybe 100 show up a week. So if you had 2,000 people in there at one time, it just wouldn't fit. So you just have to have this long amount of people. Anyway, that's— was web development. We were building a website similar to Airbnb, but for gyms. And then I kind of got sick of web development because I was lazy. I was like, I'm building all these web pages and like, I don't like the fact that you have to design every little piece and every little workflow. And again, this was just me being lazy. So I looked into machine learning and I'm like, hmm, I started realizing, oh, okay. So you just, it's kind of reversed normal programming. It's you with normal programming, you have the steps and the design and you get the output. Whereas machine learning, you start with the input and the output and then you, it learns the steps. And so I'm like, I want, I want that. I want to, I wanted the machine to just learn the steps in between what I want. And so I started researching that, watched a few YouTube videos, found a fun character online called Saurabh Raval. He was making probably all of the AI videos on YouTube at the time, or most of the ones that I saw. And I just got I just got the excitement bug. And then everything just started to be machine learning back then. It was just, people were talking about it on tech podcasts. Every company seemed to, every big company that is, started to publish information about how they're using machine learning. I went to Japan. And while I was in Japan, Google announced that their translation website translate.google.com or whatnot, and I was using that at the time, was now powered by neural networks and deep learning. And so I was like, huh, that's cool. Like it's starting to actually, like this technology I keep hearing about is, is I'm using it already. I'm using it right now. So how do I get involved in that? And then it just kept going from there. 2000, I think it was 2000, late 2017, I started my own AI master's degree by just, I wasn't working at the time. So I quit Apple with, and was like, had savings and I sort of, I was like, I want to quit Apple. I want to get into software development of some sort. And I had savings to live on. And so I created my own AI master's degree and I just studied full-time for 9 months or so and then got offered an internship at a local company. And That was 2018 and now we're here 6 years later or 5 years.

Wow. Something like that. You could definitely say though that you got in or your process from like, I guess, learning it was definitely before the whole craze started because recently in these last couple of years, obviously now more than ever, but now people that never knew anything about how models are made or how they're trained and all this. Know about it, but you definitely got in early. I was going to ask you, so would you describe yourself as like a self-learner? Did you teach yourself? Was it through the YouTube videos, that guy that you mentioned, or that you just—

Yeah, so I wouldn't describe myself like I think it's, there's a balance, right? Like with self-learning at the end of the day, you came alone by yourself, but all of the, I was lucky to have such great materials. I was lucky to have the internet. I mean, so although I was for the majority of the time spending time alone in my bedroom studying, the majority of the resources were made by other people. So I can't really, like, it's like, yes, I was doing it by myself. However, it was great resources. Yeah. Thank you for the teachers like Andrew Ong and like some of the greats of, you people. Like, I was doing courses on Udacity, uh, YouTube videos. I mean, YouTube courses weren't as prevalent as they are now back like 5 or 6 years ago, but that just shows how amazing YouTube's getting because you can find 15+ hour-long videos on freeCodeCamp. So yeah, I would say I did self-learn. However, a lot of it was I put it out there And it kind of, it kind of snowballed into people all around the world were finding that I was studying this and they're like, hey, I'm studying this too. And so I got to meet people from all around the world who were interested in the stuff that I was interested in.

Yeah, no, I'm glad you bring that up too because I feel like people forget that the, I mean, there is obviously you can learn by yourself, but there are, you can't just, there has to be something for you to learn, you know what I mean? There has to be some, exactly, some kind of So it is, I do get what you're saying. Now I have a fun question for you just because I'm not the biggest machine learning guru. Like I don't know much about what's going on in machine learning. I keep up, but like definitely not too deep. So how would you explain, if you were to explain what you do for a living to a young kid or a super old person, I want to hear how you would explain it.

Oh, that's a fun one.

It's a bit of an odd one, but it's an effective one.

I was just having this conversation with my mum the other day. She was kind of— I was just talking about what we're doing. My brother and I are working on an app together. And of course, I teach on ZTM and make YouTube videos. She's like, it's just amazing, isn't it? What you can do these days. Half of— I don't understand what you just said, but A lot of that didn't even exist when I was your age. So it kind of, and I have this problem even not even just explaining it to people. I would say the majority of the time I just, I say I'm a teacher. So that's sort of universally understood. So I say I'm a teacher and then they say, oh, what do you teach? And I'm like, well, I'm part of an online school called Zero to Mastery. And they're like, oh, so what you teach in the classroom? I'm like, Well, not really. It's a digital classroom. I make videos and post them and then people around the, all around the world can, can watch them. And people know, a lot of people, younger people sort of know that this exists, but I, I'm yet to meet anyone in real life who does something similar to what I do. And so, yeah, from that point, the teacher, and then I also say that I'm a writer. A video maker. I used to have a small tagline on website was something like, I'm a writer, a fighter, a video maker, and something else.

I'm glad you bring that up. I'm glad— sorry to interrupt— I'm glad you bring it up because I know that you, uh, we were talking before this interview. I was like, uh, Dan mentioned earlier, this is the first time we sat down, and, uh, you're a big jiu-jitsu practitioner, correct? How long have you been doing that?

October 18th, 2018 was the first session I did.

That's so cool, man.

Yeah. So that's nearly 5 years at the end of this year. October 18th will be 5 years. Yeah. It was perfect. Yeah. Perfect timing. I remember the date off by heart because I was traveling in the US because that was my initial plan actually when I started learning technology and machine learning and whatnot was that I was going to move from Australia to the United States. And so I went over there to, uh, my uncle was living there at the time and I went there to see him and sort of get a lay of the land. And because that's like the US is the holy grail kind of thing. Everyone looks at there for, for technology. It's the home of many of the different pieces of technology that we use today. Of course, I'm talking to you on a Apple Mac. And I mean, that's just the home is over there. So I'm like, if I go there, I'll find a job in technology somewhere, somehow. That was my mentality. However, um, I finished that trip and came back to Australia. And while I was over there, I was like, in my head, I was doing a lot of gym workouts and stuff like that at the time. But I was like, I want something. The UFC was just starting to get really big. And like, I want to, I want to start learning a martial art. And then I get back to Australia and literally the day I got back, I got back at say 8 AM in the morning and the gym that I started jiu-jitsu at opened that day at lunchtime. The first session was at like 4 hours after I got back. And so I'm like, well, this is perfect timing. And it's, it was at the time, like a kilometer from my house. So I'm like, well, this is a silver lining.

And you went that same day? Yeah, the same day.

Yeah, got back that morning. My brother and I, Josh, and my friend Dave, who's my training partner, we, we wanted to do a martial art, and it was just— it just literally was perfect timing. Got back from the US, I decided, I'm like, when I get back, I want to do martial art. Got back that morning, October 18th, 8 AM, and the gym opened at 12, and there was a lunchtime session for the first ever session.

And that's great. I think, I think many people would agree you probably picked— I mean, this can start a whole controversy in the comments, so I'll take this with a grain of salt for all the viewers, but many people would probably agree that jiu-jitsu is one of the top martial arts to, to get started with. Um, and I also, I, I also don't, uh, I think that especially for people, especially for people that work in a computer or work kind of sitting down mostly. There's so many benefits to just getting out there and, and getting that type of— because it's a rigorous exercise, it's, it's definitely a tough one. And, uh, I don't know, I just, I'm glad I took it up too, and I haven't looked back, man.

It's addictive, it's just fun. Yeah, I always say like to people who don't, who've never done it and can seem it can obviously seem quite intimidating, but I feel if you just imagine if you put 10 kids in the same room, or 5 kids, whatever, and they, they get bored, what do they start doing? They start wrestling. They start just play fighting and rolling around on the ground and tackling each other, and that's just what kids do. And so I'm like, it's kind of just the adult version of that, but you just put fancy names on different positions and whatnot, but at the end of the day, it's just wrestling. It's play fighting. Of course there's competitive aspects to it. Yeah, for the majority of the time, like with your training partners, it's, it's— once you get into it, it's a dance. You're not— you're never really trying to hurt them. You're just trying to go through the motions and, and, uh, um, have fun with it. So it's like, that's what I say, it's like adult, adult, uh, play wrestling.

I like that. I like that. I know, uh, DDH, um, Daniel, the one that does our, um our blog post, he was telling me, um, that jiu-jitsu is kind of like adults wrestling in pajamas. So shout out Daniel. It does kind of make sense in a way. I mean, we're just— we're people that I guess never really grew up still enjoying that part of our lives, you know what I mean, in some aspect. I love it. Um, all right, so the next thing I wanted to ask you is, so you brought up, um, that you've made a couple courses. How many courses have you made here at CTM?

3.

3 courses. And what are the— can you tell me like the, I guess, some of the key differences between the 3 courses? What makes them different?

So the first one was the— or is the— it's still live— the Data Science and Machine Learning Bootcamp. So it's basically, it's 0 to 1 of— or 0 to mastery, however you want to put it— of, you machine learning and data science. Of course, it's, it does not cover everything because the field, as anyone will find out, technology is very broad, especially machine learning these days. You can apply it to almost anything. That's what excited me about it.

Yeah.

But so that was built out of when I was learning myself online, I found a lot of the materials and resources were quite I don't know, let's just say they weren't entertaining or not that they have to be, but they, they were kind of a slog to get through a fair bit of the time. And then I went and I went and worked as a machine learning engineer in a company in my hometown. And we talked, so it was a machine learning consultancy business, but the, the beautiful thing about that was that we had one team that was building machine learning based product, but I was on the team that, because I could talk to people, I was on the team that would go to other businesses and they're like, hey, we want to implement machine learning. We'll be like, okay, well, let us know what data you have and how we can work with it, X, Y, Z. So I got the benefit of seeing several different companies instead of just sort of working on one project. Yeah. And I kind of worked out that there's a, there's a fair few things, obviously every problem's different, but there's a fair few things that are quite similar across multiple different projects. And then I also realized that the majority of the time we were writing code to process data, to modeling was actually probably the easiest part. However, a lot of the time was writing code instead of and working with the business rather than researching like machine learning papers and new algorithms, that sort of stuff. So I'm like, hmm, a lot of the stuff I've learned is very academic-based, very research-heavy, a lot of math involved. But when I got to the job, and of course every job will be different, if you're going for a job in a tech company versus an actual research job, you'll need different skills. I realized I'm like, oh, okay, so I'm actually spending 99% of my time coding. And so that's what I started to make materials and write about online, was that a lot of the machine learning stuff I was doing— because again, this is like 2018, 2019, machine learning is still kind of new, definitely. And I say that in inverted commas. A lot of companies had never really even heard of it. And I guess now it's like AI is the the keyword. And I kind of made that course for people who were similar to me in the beginning, as in starting to learn by themselves online, because that's just the way the world's going. Online education is so prevalent now. But I wanted to incorporate things that I actually did as a machine learning engineer into the course and teach machine learning via that way. Whereas I'd kind of learned more of a bottom-up approach to machine learning. I was teaching top-down, as in, here's the problem, here's how we're going to work towards that. Whereas instead of learning all the building blocks, which again, there are different resources to learn that if you really want to learn that, but I was kind of like, okay, if someone was to do this course and then go into our machine learning business, they're going to at least know or have heard of many of the different concepts that people talk about in the workplace. And it was very code hands-on. So that course finished, it's about 45 hours or so. It's a big one. Yeah, it's a big one. But it's all code like videos. So every single video, like we're coding side by side. And I think that's probably the ZTM way is kind of it's, it's hands-on from the beginning. Yeah. And I really, really like that. And then that course ended with, uh, a little bit on TensorFlow, which is, um, so a deep learning framework open source by Google. So the fundamentals of the first course actually was if you have zero coding experience, like zero, you can start with that. And by the end of it, you should have some familiarity, or you will have some familiarity with many of the baseline things in machine learning.

So there was no prerequisite?

No, it's for, yeah, it was an absolute beginner's. Got it. Whereas the TensorFlow course, which was the second course that I put out, was a follow-on from the first one. And I kind of tried to make it as beginner-friendly as possible, but you probably would like at least 6 months of coding experience before that, like Python coding experience in particular. But that being said, I always encourage people to just try it out and just, because all the code, all the videos are hands-on. So it's, you can learn as you go along and I try to explain things as I write them in code. And then the third course, the most recent one is PyTorch. And so PyTorch is another deep learning framework, which is open source by Meta, but it's actually now part of the Linux Foundation. So PyTorch is almost completely decoupled from just Meta, which is beautiful. And so that's actually the machine learning way kind of thing is a lot of this is, I sort of stumbled upon this as well. A lot of the research and best materials are just open source so everyone can access it. That's the beautiful thing about it. And the PyTorch course is very similar to the TensorFlow course. It's because they, they, they both offer similar things, PyTorch and TensorFlow. And there's an age-old question of like, which should I use, PyTorch or TensorFlow? And I mean, there's new frameworks these days called JAX, there's another one called Hugging Face Transformers. And it's, at the end of the day, they all offer similar things. So it's, it depends on what you want to do. And so they're more targeted to deep learning workflows rather than just standard machine learning. Well, not standard, but machine learning workflows. So the first course. Complete beginner. The second two courses, you could choose either one because they offer similar materials, uh, for like level 1 to 2.

Okay, I understood, understood. Um, now actually, now, so you have three courses out and, um, you have a full-on beginner's kind of dive fully into this world of machine learning and data science, and then you have two that would you say are kind of like, um, they're definitely like a little bit more advanced, but Um, they're, they're more, I guess, niche. They're not as like the, they're not the complete basics like your first course, correct? Did I understand that right?

Exactly. Yeah. So the, the number of people who will, this is sort of after teaching online for a few years now, that of course beginner courses are gonna, you're gonna end up with more students because a lot of people are, I mean, even myself, like I've signed up to probably a dozen beginner courses over the years, but I've probably completed maybe 3 of them. So because you just get excited, you sign up to something and you're like, oh, I want to, yes, I'll learn that. But again, at the end of the day, you only have a certain number of hours in the day and you only have certain motivations. And so you sort of work it out. Yeah. So 0 to 1 is the Machine Learning Data Science Bootcamp. And then level 1 to 2, you level up from there is the PyTorch and TensorFlow course, by the time you finish those, you'll have replicated machine learning research papers. So you'll see what it's like to like translate the latest state of the art from a research paper that a company like Google has just published into actual code that you can run on your machine.

That's crazy. Actually, and that, that brings me to my follow-up question. Um, with 3 courses out, what was the inspiration behind making courses? Like before you came on at ZTM and decided to make these 3 courses, What made you get into it? Is it because, I mean, from, from what I've heard from you so far, what I think is like you learned kind of in a similar way through people that had great resources online. You were willing to put in the time and all. Was it almost like you wanted to kind of do the same thing for the people, the new viewers? That's, that's the feeling I'm getting, but you definitely let me know kind of what inspired you to get into it.

Yeah, that's, that's definitely one reason. The other reason is of course financially, like money. Like, of course, I'm not gonna lie. It's, it's, I think it is a thing in society these days. Like, people have like the glorious reason, which sounds really altruistic. And I just would, of course, I want to help people with making courses. But at the end of the day, I keep saying that. I want to stop saying that. Anyway, at the end of it, um, I needed a way to make money too, because I wanted to leave my— I, I had an amazing job, machine learning job, but frankly, I wanted to work for myself or work anywhere I wanted, wanted to create on my own terms, create what I wanted to create. And I left my job, machine learning engineer. I posted an article of like 10 things I learned as my first year of a machine learning engineer. And that's how Andre, who's the creator of ZTM, uh, found me and said, hey, we're starting machine learning courses. Do you want to teach them? And I was like, sure. Because the, the transition period of when I left my job— so this is kind of— I've had in the last few years, I've left Apple and then lived off my savings, and then I left the Brisbane company I was working for, Max Kelson, and I was living off my savings. But the second time I left, I was making things on YouTube, and I'm like, I want to work for myself, I want to start a business. And it happened— it so happened that teaching I stumbled upon this. It just happened because I left, I didn't have it lined up when I left my job. I had, I'm like, okay, I'm going to leave my job. I've got some savings here. I'm going to figure it out. I'm going to start my own business and we'll see how it happens. And then literally within 2 weeks, I got an email from Andrej. So, and then I started working on the CTM bootcamp. And so I'm like, I want to make this really good. So that people will buy it and people will learn from it. And if it sells well, then I can use that money to work on the projects I want to work on as well.

Of course. Yeah.

It kind of feeds into, so yes, of course, I want to teach people around the world the magic of machine learning. It's beautiful. But I think people will learn that without me regardless. So even if I didn't teach it, people are still going to learn it because it's exciting. Yeah. So I'm just riding that, the trend at the moment, which is amazing. Of course that, yes, to teach people. But yeah, then I had to have a way to earn money. And so I'm like, well, I've paid for courses online, so maybe I can be the person to make them now and sell them.

Of course.

And that, it worked out really well. No, it's grown incredibly over the last few years. The courses, from what I'm told in terms of ratings and I guess comments on my YouTube videos, people like my teaching style. So at the end of the day, I keep saying that, I got lucky. I got one, I got lucky. And two, I kind of stumbled upon by just putting stuff out there, I stumbled upon something amazing, the opportunity. To teach online. And now it kind of gives me the freedom to learn what I want to learn, which is machine learning, then also create amazing resources through the form of courses, through the form of videos, and also fund my own business on the side, you know, separate to ZTM, which is called Nutrify, which is what my brother and I are working on. So That's where I get to really study the latest and greatest of machine learning. And then the courses are where I put what I've learned into a way that others can understand, hopefully.

Beautiful, beautiful. I definitely want to touch on that Nutrify a little bit later on in this, uh, in this interview. Um, but one question I do have— well, first, before anything else, actually, I could speak for most people. We're glad you're definitely making courses. The feedback I've seen on those courses have definitely been Extremely good. So I'm, I'm sure I'm speaking for most people when I say thank you on that. And one question I did want to ask you is, so we spoke about the course you have. We spoke about the reasonings on why you started, which are fair, definitely. Do you have anything in the pipeline coming up? Is there new courses that you're thinking about doing for ZTM? New— anything that you see in the future?

Yes. So ZTM-wise, there is always something that I'm thinking of creating. However, this is like my, one of my, not, I wouldn't say problems, but chief problems in life, for lack of a better word. I get too excited about too many different things. However, of course, you can only create a certain amount. So I've, within the last couple of months, I've, without revealing too much about what's in the pipeline, I've kind of jumped back and forth between creating about 3 different types of courses. And of course there's the latest and greatest hype that's going along at the moment. I wouldn't say, well, hype's probably not the greatest word, but let's call it hype of large language models and ChatGPT and all that sort of stuff.

Of course.

You could probably guess. And then the other hard part is around that though, is I also like to teach things that I feel are going to stick around for quite a long time. And I'm not saying that like the new stuff is going to go around, but it's because it's so new. It's probably going to, like if I made a course today, who's to say it's not going to be out of date in a month?

Definitely.

So because it's, it's, it's changing so much, there might be a new model. Like OpenAI releases something new and the whole system changes and people are like, oh, okay. Yeah. Whereas I taught TensorFlow and the machine learning bootcamp are things that haven't really changed dramatically in the last 5 years. The way you use them has changed, but the fundamentals of it is they're pretty set in stone.

Similar.

Yeah.

No, that's fair. Um, There's a few things. That's exciting stuff. I'm glad to hear that. I bet there's going to be some rampage going on in the, in the comments section below. I, I wanted to ask you, so I heard that you said you're starting it. So right now, just so I get it straight fully, you worked for Apple and then you worked for two other companies, correct? Before you went and worked for yourself.

Just one.

Just one. Yeah. Okay. And right now you're fully like self-employed, correct?

Yeah. I work with, yeah, I'm part of ZTM. That's the majority source of my income now.

Yeah. Yeah. Of course. No, I get that. And then you also, you also do make YouTube videos on machine learning, right? Really popular YouTube channel from what I've seen.

Yes. So YouTube is, is part of it. I haven't, I've been lazy on the YouTube side of things lately because I've been making other things. So yeah, YouTube has been a big part of it the whole way through, just because I, I like doing what I— I'm monkey see, monkey do. I see someone doing something and having fun, I'm like, I want to try that out, right? So YouTube is— is— I like watching YouTube, of course. And so I want to get— I wanted to get involved, started making my own videos, and They're on pause at the moment because I like to make good stuff and it takes me a little while to make good stuff. So I can only focus my energy in certain places. But yeah, uh, there'll definitely be more YouTube videos in the future.

Cool. And that's, uh, Daniel Bourke, correct? Just so I have it. Yeah. Cool.

Yeah, just about.

People check it out. I'm assuming most people who are probably subscribed to ZTM are probably subscribed to Daniel Bourke. They're probably subscribed to Daniel Bourke first. But, uh, if they're not, go definitely check out his YouTube. Um, now I want to touch on what you brought up. So, um, you worked for one other company besides Apple, and then you started this business with your brother. Um, what was— what, what's the business or this company with your brother? Correct? Is that—

yeah, well, it's, it's still really new.

It's—

we started properly, properly working on it 3 days a week at the start of this year, so it's barely 6 months old, but we're making some great progress. So I've kind of had this dream. I wrote in this, I wrote an article. I write an article every year. Uh, I'll get it up. It doesn't need to come up later. I write an article every year on my birthday, which is basically the last year worth of learnings and things I'd like to do for the next year. And I wrote one on my 25th birthday. And so I turned 30 this year. So 5 years ago, nearly, that I'd like to build a food education app with my brother. And we're finally doing that properly. We've kind of built prototypes over the last few years because I studied, uh, the background story to all of this is how I got started was I studied food science and nutrition in university. So nothing to do with coding and all that sort of jazz. So my heart, my first love is food and then computers and stuff come after that. Naturally, started learning machine learning. I want to merge them in some way, shape, or form. And I figured, well, computer vision is really good now. Back 5 years ago, it was getting really good, but now it's really good. And so I was like, I'm looking at food in a different lens after I've learned about nutrition. And so I'm like, well, can I just take the phone and take a photo of food and inject what I've learned about nutrition into the phone or into the app so that if someone else wanted to start learning more about food, they could take a photo of their food and The app would tell them different tidbits and information. What they do with that, well, it's completely up to them. And so that's what we're working on. So the whole concept of Nutrify is to take a photo of food and learn about it. And in the future, it will be video-based. Right now it's one photo, one, one image at a time, but it's a, it's a merger of everything I love about machine learning. Of building models that people can actually use and a merger of everything I love about food, cooking, nutrition information, uh, what the hell is on the back of a food label, that sort of thing.

Yeah.

All in, uh, all, all like congregated into one. And so that's what we hope to launch. Well, we will launch. I don't know when, but within the next couple of months, likely. So, oh, that's great.

In the next couple of months.

Yeah. So it's going really well, but it's a Pokédex for food. That's what we're calling it. So if anyone's played Pokémon before or Shazam, if you use Shazam on the, on your phone to recognize a song, it's the same thing. And Apple just had a big event the other day. They released the new goggles and that sort of thing. And so of course we're nerding out. Yeah, you could just walk into a kitchen and have everything displayed in AR and that sort of stuff. But that's probably, uh, we'll get it working on the phone first.

That's great, man. Honestly, I, I'm, I'm excited for that because I feel like I've traveled a lot and I've been more passionate about the food and not that passionate about the nutrition. But it'd be nice to know. Sometimes I get too caught up in, uh, in the actual just how good the food tastes, but it'd be nice to know what's going on about it.

We want it, we want it to keep, we don't want to take away from that. We don't want to Um, it's, yeah, we don't want to take away from the food is magic. Food is life, right? So we don't want to, we don't want it to just reduce it to like a bunch of numbers on a page. So I want to try and make it fun if you're gonna, if you're gonna be using it. So it won't just be like a MyFitnessPal type setup where it's just turns it into numbers on a page. You want it to, yeah, yeah, be, be fun. That's exciting. Food's already fun, so you can't take away from that.

Yeah, you can't make food less fun just to make the app more fun.

Exactly right.

I like it though.

Keep the fun.

No, that's a great idea, man. And I look forward to keeping tabs on that, see how you're doing on that, and eventually, hopefully getting to use it myself. Back onto some of the nitty-gritty topics of this conversation is, we spoke about how you would explain the role to someone if they knew nothing about it, but What does a day-to-day look like for you? Like, as a machine learning engineer, I don't know if maybe you can think about when you were working at that company full-time, or what— and what's the difference between that versus now, maybe kind of what you do, especially working on your own application? What are the differences from a day-to-day perspective?

Yeah, this is a great question because it is— and I say this to a lot of people. Anyone who asks, my generic answer is it depends because the roles, the amount of different things I've done as a machine learning engineer is completely different depending on what project, depending on who we're working with. Everything could be different. But for me, most of the time, I give another generic answer is I'm writing code and I'm trying to format data. In a way that I can build a model for. So whatever that looks like really depends. So for Nutrify, it's, it's startup mode. So it's where we're collecting data. I'm doing the machine learning part, my brother's doing the iPhone app. We're collecting data, we're organizing it, we're going, how do we build a computer vision model around this? So I've got a I have a computer in my closet which trains all of the NutriBio models. So it's, that looks really different to what I used to do as a, well, actually not dramatically different as a machine learning engineer at the, at Max Kelsen, which is a company in Brisbane. Day to day there, we would do a lot of talk to a business. I guess it's different if you're working on an actual product versus if you're sort of like doing freelancing or consulting or something. Because then as a consultant, we would talk to the business, work out what their business wants, and then, uh, start to build the product in that way, or the proof of concept in that, in that sort of way. Whereas with Nujify, the kind of— the goal is very clear. It's we want the computer vision model to be better. That's whatever we can do to do that is just make the model better. And I guess it's similar to like Tesla's Autopilot and that sort of stuff with their computer vision models in the car, the goal is very clear, but the steps to get there is maybe a little bit more, of course, much more complex. But the goal is, of course, have a car that can see the world and can drive. That's it. You know, that's it.

That's the high-level goal.

Yeah. Correct. So whereas as a machine learning engineer working with different companies in consulting and proof of concept, the whole thing was Can this even be done? Do they even know what their goal is? Because a lot of the time is many of these people had never heard of machine learning, or if they did, they thought it was like magical and it could solve all of their bits and pieces when a lot of the time it generally, it was sort of, it could help, but it's viewed in a different lens. And so I guess that might be the same with AI at the moment is yes, a lot of businesses are going to be having some sort of AI. Department or whatnot. However, if it, it may not be, I mean, the number one rule of machine learning, right, is don't use it if you can create a rule-based system that's going to work, or a simple rule-based system. So day to day, I could just tell you what I did yesterday. Like, Apple have had a big event and I'm, I'm watching WWDC, so it's Worldwide Developer Conference, and I'm trying to work out do any of the new technologies that Apple are offering integrate with our app? And of course, yes, there are. This is tech nerd disease, is that seeing every new technology can be used in some way. However, a lot of the time when you're building a product, it's— and this is what I'm finding out— it's, it's a very, I guess, intricate skill as to how much time you should dedicate to each thing. Because at the end of the day, of course you want the product to be better. Of course you, you would have I would spend all my time making the computer vision model. But if I spend all my time making the computer vision model, what happens on the other end? Because if there's no information display, the computer vision model is really good. Well, then it doesn't sort of, it doesn't fit. So right now it's a bit of a balancing act of, do I work on the computer vision model? Do I upgrade the information that the app displays? Do I collect more data for this specific food? Do I? Contacted the designer to get something made for this. Um, so yeah, it's quite fun.

It sounds like you're juggling a couple different things. It's definitely a busy day today. Um, and out of that, out of the different things that you're doing, what are some of like the, uh, um, like your favorite tools or maybe softwares that you're— that you would say you use on like a daily or a weekly basis? Do you have any that you can like name off the top of your head?

Yeah, well, generally now I'm either use mostly Google Colab or VS Code, one of the two. And if I'm in VS Code, I have GitHub Copilot. So I find that really helpful.

There we go.

Yeah.

I've played around with that myself. I like it a lot. Yeah.

So I think it's a beautiful tool. Of course, ChatGPT here and there.

Google.

I'm trying to think, I'm trying to look at my computer, like what I use every day. I'm trying to keep it quite simple. Yeah. VS Code, Google Colab, GitHub Copilot, Jira itself. And then of course Discord for communication. They're probably like the main tools. And then, oh, Notion. I keep, I keep notes in Notion.

Notion's like the, like the least technical, but like the biggest one I see with everyone I talk to. Everyone uses Notion. Um, yeah, whether it be the most simple things or they're doing something really advanced in the, in, in the application.

It's a, it's a beautiful product. I, I love it. I, I remember seeing it years ago and I was sort of like, oh, I don't need another note-taking app. But then I tried it and I'm like, I don't know what it is, but I like it. Of course, it's gotten way better over time. So it's, uh, I'm a big fan.

Perfect. And then the next thing I wanted to get into is I do have a couple technical questions. They're not too technical, nothing out of this world. It's just for people that don't know much and hear these, these people that hear these like common phrases, like a model, for example. For most people, at least my friend group, a model would not be what you were thinking of and more like the ones you see on the TV or that kind of thing. You know what I mean?

That's probably the better model.

That's probably the better one. But, um, actually, that's actually the first one for a complete beginner like me. Can you explain to me what a model is when, when you guys, when the machine learning engineers mention a model, what's a model exactly?

Well, I could describe it with another M word, which is map. So a model in a machine learning context is you have some sort of input and the model is going to be your map to the output. And the map can be incredibly detailed or it could be a straight line.

That's a great analogy though. I like that one. The map.

Well, well, your inputs could could be, and this is the magic of machine learning, your inputs, if you frame it in a certain way, your inputs could be almost anything. It could be structured data like you find in an Excel spreadsheet. It could be unstructured data like you find in video, images, sound, anything. But if you structure it in the right way, you can create a model or a map to your idealized output. In the case of Tesla, their Their input will be vision from a car. What does the car see? And then they have the model, which is going to be some sort of map, quite a complicated map. No human, individual human could understand this map, but that's the reason we have computers. Yeah. Their output, their ideal output will be actions in the car. So turn left, turn right, more gas, brake. Et cetera.

And that map leads to that output.

Yes, ideally.

Understood. Yeah, yeah. In the perfect world.

Yeah, exactly. Yeah, exactly.

Now this—

oh, sorry, you go.

No, no, no. Finish off, finish off.

Well, it's similar to what, like, the analogy is always— people will argue to the end of the earth for this, but it's like the analogy of neural network is the dominant model people will hear about these days. It's analogous to like how our brain works. And however you want to take that is completely up to you, but it's, it's if you see something with your eyes, if you ask me a question, the model in my brain is, uh, I've taken in your audio, your sound waves, and I've processed it in some way. And then hopefully if my brain's working, my model's working, I'm going to reply to that question in some way, shape, or form. So the model is the the, the map between input and output.

And the output. No, I love that explanation. You actually, uh, pretty much answered my next question as well because I've heard the term neural network and we kind of have, like you said, we have like, uh, this neural network going on in our brain that's very similar to what a model is trying to mimic, I guess. I don't know, correct me if I'm wrong, but that's how I see it. Like we have this.

Yeah, well, I think it's— I, I'm not a neuroscientist, uh, I, I know hardly anything about the brain, but if you look at the diagrams, they look similar. But then again, it's, it's quite, we know a lot about the brain, right? But there's still, I guess, hmm, the mystical part of me is like, maybe there's something that we're missing. I don't know. Of course, right? That's the fun part. But the neural network, it's, yeah, It's got, you have the way it's constructed is a lot of smaller functions, smaller models contributing to one bigger model. Got it. And so the analogy to the brain is there's a lot of individual brain cells that combine to output the entirety of our existence. So if you look at GPT-3 or GPT-4, we don't know how many parameters that has, but GPT- GPT-3 has 175 billion. I'm not sure how many neurons the brain has. I think it might be trillions. So we're not quite there yet with the number of neurons and whatnot. But if you draw the very, I guess, rough line between, well, the model has 175 billion smaller models trying to predict one output. And our brain has a trillion brain cells all combining in some way, shape, or form to predict something, then there's— that's the similarity there.

No, and I definitely understand that. I think that's— for me, I'm looking at it from not like a less technical way, and it seems quite similar. Um, so I definitely— I could see that. Um, another thing I was going to ask you was, so if models— models nowadays are getting more capable, more capable, is there ever going to reach a point where these models can train other models? And if so, like, um, Does that affect machine learning engineers, like in a way? Like if, if, if like a model gets so advanced and this is coming from a complete novice perspective. So apologies if it sounds like really, but you know what I mean? If the, if the models get so advanced, is there going to reach the point where they can train other models?

Uh, yes, I would say that's, that's already happening. So there's a technique called model distillation. Which is basically if you take a big, big model, say 100 billion parameters, just for an easy number, and you, but that's, that model's too big to fit on a phone. It's really good, but it, you can't fit it on the phone. It would take an hour to process one thing if it can even fit in the memory. So too long to people. No, no one's going to wait an hour. But you, so that big model is really good, but you might want to go, okay, so it's 100 billion parameters, but we can only fit one with 10 million. So what's that, 1,000x less? Yeah, that might be wrong, but a lot less on the phone. And so what you can do is, and that model will run in under a second, much better. So what you can do is you can have the big model, the outputs from that train the smaller model. So you can basically compress them. And so there's the setup for this is a little bit more intricate than that, but the smaller model won't be as good as the big model, but it will at least take a fair bit of its outputs and compress it into its potential. So that's what we're doing with, with Nutrify actually is we have we have really big models that are really good at predicting what's in food, but then we, we can't deploy them to the phone. It needs to be a faster compressed model because again, no one's going to wait for a few seconds, even a few seconds these days to— there's a lot of weight. It needs to be, needs to be instant. So yeah, we take the big model, train a smaller model, deploy it to the phone. But probably what's happening nowadays is there's a concept of agents where you have a large language model and you sort of just give it a task and it's going to try and complete that task no matter what. And so whether that task might be training a smaller model to help it complete that task, it will try to do that. I think we're quite a fair bit, fair way of of that actually happening. But who knows? Again, could be completely wrong. The field just seems to completely change every 6 months. So, but yeah, that technique of larger model training a smaller model is definitely happening.

Wow. And actually, now that we're talking about, you mentioned developing, deploying models. What is like the, if you can give me like a high level, doesn't have to be too detailed, but a high-level overview of how you develop and deploy a model, a machine learning model. How does that, how does that work?

So let's start. Let's say you have a trained model, as in you've, you've created a dataset of some sort and you've trained a model on that dataset and it's, it's performing well. And again, let's, let's take Nutrify as an example. You train it a computer vision model to understand what's in food. So if it takes a photo of a red apple, hopefully it says red apple. We want to take that. There are a couple of different ways to deploy models these days, and there's— I've got quite an in-depth section on this in the PyTorch course at the back. I can give you the, the link to the tutorial for that. However, I'll just describe it. So say you— there's online and there's offline. And then there's, I think that's enough to actually say it. Online is you will have someone is taking a photo on their phone and that photo, if our model was online, it would be on the cloud. So that photo would have to go from the phone to the cloud. The model would make a prediction there on say a GPU hosted in a computer somewhere in the US, anywhere. And then the prediction would come back to the phone. So that's online. The, the benefit of online is that you can use a really big computer. That's what ChatGPT is, is you chat with it and it goes to Microsoft servers or OpenAI servers and they have like warehouses full of computers. And so it can crunch the numbers and then it shows you back to your screen. So that's online. However, you can imagine the downsides of online. Imagine if you were trying to use an online model for Tesla self-driving car. What happens if you have no internet connection? Or what happens? Yeah. What happens if your predictions need to be really quick? So ChatGPT is quite quick, but of course, sometimes it's slow because it's bottlenecked and that sort of stuff. That's not going to fly in a car. You want the car to make decisions almost instantaneously, faster than you, in fact. And so that's where offline deployment comes, is that the model is completely on the device. Say it's completely on the phone, which is how Nutrify works. It's the predictions never— the data and the predictions never leave the device. So you get a couple of benefits from that. So the benefit of online is you can use a big computer. The downside of online is that there may be a delay. And then you have offline, so on-device models. The benefit of on-device is that it's generally really quick. Privacy, if you need to be worried about privacy, sometimes you don't want to send data to OpenAI. And, but the cons of that is that because you're, uh, an iPhone doesn't have as much computing power as a full-blown GPU in a server. So you don't have as much compute. So they're the two like main ways of deployment, online and offline.

Wow. No, that's, that's in depth, man. I got to watch this course. I got to fully. No, but it all makes sense, especially when you use the, uh, or when you talked about the Tesla, um, or the car situation where in, uh, in a situation where you need live and quick, you know, data, it's, it's, it's—

yeah, you wouldn't want to be waiting for your car to upload video of a red light and then a prediction to come back and be stopped. So all of Tesla cars have these huge compute engines on there, um, that are really, really fast optimized. And this is Apple as well. The chips are optimized, the Neural Engine and whatnot. Uh, and Google are doing it too, and their new Pixel phones is optimized to perform a lot of machine learning on the device.

Nice, nice. Um, well, that— I, I appreciate that explanation. The next, uh, set of questions that we're going to get into, I would say, are some of them are more so like struggles and successes, um, kind of throughout your your career. So hopefully we can highlight more goods than bads, but you never know. Um, this first one's an easy one. Um, what are some of the biggest changes? And I don't even know if I'd label it really successes or, um, struggles for this first one, but I'll ask you, um, what are some of the biggest changes that you've seen in the industry, I guess, for the amount of years? It's 2015, you said, right? 2015 is when you got started.

Yeah, I started hearing about it, yeah, in 2015, and probably, yes, really started studying it at the end of 2017.

Got it.

But so biggest changes. Yeah. I would say it's not called machine learning as often anymore. It's called AI.

You work in AI.

Yeah, that's the, that's the, um, like all the, all the news, news, news stories and stuff. Now it's, it's, it's AI. It's officially AI instead of, I think a lot more like it was called machine learning for quite a few years, but now, uh, Definitely running with AI.

Dan, you know what? I'll take that over. I've heard some of my friends, unfortunately, some of my friends have said that, like, do you have anyone when they're talking about AI? They're like, do you have anyone that works in ChatGPT? Yeah, that's like the— no, he's the—

he's the AI.

Yeah, it's the poster boy for the posting. Yeah, it's, it's, it's funny, man. That's a big one, actually. It gets kind of gets annoying and I'm not even inside the space, so I can imagine.

Hey, ChatGPT is a magical product. I'm a big fan of it. That's great. The only thing I would say is that it's like, it shouldn't, I don't know, these chatbots shouldn't really, I guess it's a private company. I don't know. It's, you kind of, you kind of delve into the details here. It's like fudging the outputs for a particular to suit a particular narrative. Like, it won't say certain things. Of course, you don't want it to just be swearing and like being racist and all that sort of judge, but I don't think— I don't know, I don't think it can— it needs to take as much of a political standpoint as it does. But I guess that's, that's me having more faith in humanity, is like, I think my idea is like people I guess, and I'm wrong probably a lot of the time, is that people are fundamentally good. So, and if they're smart enough, they can sort of decipher answers for themselves. So yeah, but no, I think it's a very useful tool.

Now we will get, we will get into ChatGPT in a bit. Um, but that, that's, I mean, yeah, I, I tend to agree as well with that because I feel like, I mean, just in general, even outside of these chatbots, just you speaking to someone else. A lot of people are trying to push these, like, whatever narrative it might be. So it— I wish that it was kind of left out on some of these things. But you can tell when you, uh, when you talk to these bots, um, sometimes the responses I get are a little bit more—

yeah, it's like, um, I don't know, I don't know. I guess that it'd be cool if there was a mode like it was like The baseline one is like safe mode. It's like you can use it at work, all that sort of jazz. But then it'd be cool. It's like, if you go, like, you go to the Wild Wild West mode where it's, it's, it's just going to output. So like, beware, like, this is, this is the Wild Wild West.

This is free speech mode.

Wild Wild West. Yeah, this is, this is Wild Wild West mode. So, and you've, you've actively clicked that button. So you can't hold us accountable for anything that this says. So I think that'd be cool.

Oh no, I agree. I agree. We spoke earlier about the different contents that you're making. Like we spoke about the YouTube videos. We spoke about the 3 courses that you've done here. One thing I didn't mention is that on one of your recent machine learning or ZTM's recent machine learning newsletters, there was this AI anxiety post you did, if I'm not mistaken. I read a little bit on it and I was quite curious with it because I feel like with all this new like AI and especially all this buzz recently. There's a lot of people that, that are going through that anxiety. Can you kind of touch upon that a little bit more? Um, explain kind of what you were talking about. Don't, don't give us all the way— don't give away all the goodies because I want to link that in the corner, um, so these people can go— so that people can go check it out. But definitely give me some high level of that article.

So yeah, the blog post is called AI Anxiety, and I feel like it was just a it's a term I was experiencing it from, to define it would be AI is coming out. Is AI going to take over humanity at a high level? Is AI going to, at a lower level, like just replace everything that we do? How do people derive meaning from life? X, Y, Z. So I was seeing all of these and what I kind of, I'm kind of past that feeling now, but there was because I— the reason why I wrote it was to sort of work through my own thoughts. And that's what a lot of my writing is, basically, is just to— hey, I've got this thought, is this true or not? Can I just write 1,000 words on it and argue with myself to see if it's— if I should be thinking this way? So that's the main reason I wrote it. But I also wanted to publish it. It was originally like just a journal entry, but I'm like, oh, I'll publish it because there was a vibe on the internet which was, and still, and probably will be for the foreseeable future, was these new AI announcements are coming out. And because there was so, I mean, I think February and March was just, there was something every week. It was kind of crazy. Yeah. And it was just a lot coming out at the time. And of course, with things like ChatGPT, it seems like magic. And it's like, oh my gosh, If this is good at doing that, is it going to take my job? Oh, if it's so good at just taking my job, isn't it going to just get super smart, learn on its own, take over humanity?

Big concern for a lot of people. Yeah.

Right? However, if you look back at history, basically every 10 years there's been the same fear ever since like technology like became a thing of, uh, the, the the car being invented, all that sort of stuff 100 years ago. Every decade there's been some sort of robot that was going to take over humanity.

They probably thought the same thing about a fridge.

Yeah, right? Like, it's like, yeah, what are butchers going to do? Like cut up meat and whatnot? Like when we can just store it for months, we don't need it. And so, yeah, it was just a feeling that I had for like 2 weeks and like just looking at everything. Basically, I'd just go for a walk and like, look at that. Oh, could AI do that? Oh, could a, what if we, oh, what if we don't need a coffee shop person to run this coffee shop? Like that, that you could just get a robot. Oh, what are people going to do? What am I going to do? Look, if I can just, what, what are my courses like? Do they, my course is just null and void now. People are going to just chat to ChatGPT to learn to code. And I was sort of, yeah, I felt that feeling for like 2 weeks and then I sort of like wrote through it, researched more, learned about how ChatGPT works. And because I knew from a high level, but I hadn't read that like real technical implementations. And I've got to update that blog post actually. There's a few resources I'll probably link at the bottom and started to learn more about it. It sort of changed my viewpoints from the initial anxiety feeling. But I'm actually glad that I had that feeling. Because now I'm more optimistic than ever in terms of the future. And I think that's just the natural ebb and flow of life is of course, to have the feelings of excitement, you do need to have the feelings of anxiety. Like if you can experience joy and happiness, you need to be able to experience sad and anger and that sort of stuff. And I think it's just a natural way. And of course, anyone who's new to the field, new to AI technology in general, I think that's going to be a lot of people in the next few years, especially with the last year or so, is that they'll probably feel at some point some form of AI anxiety. And so that's the reason I wrote that to go, hey, I feel this too. Here's different thinking patterns that help me navigate that feeling.

Yeah, I'm glad you say that, like, and what you said earlier as well, because I feel like that's the way of the mind almost, you know what I mean? With, um, in just in anything, even outside of tech, the mind naturally worries first and thinks about everything that could possibly be bad, and then it starts finding solutions and finding ways to, to, to come out of that. So that's really cool, man. And that's on your, um, Machine Learning Monthly newsletter, correct? Is that—

that's— yeah, it was on the I linked it on the newsletter. So the newsletter is Machine Learning Monthly. That gets published once a month to ZTM.

And that's brand new topics or what's the buzz around machine learning, kind of new things that are going on. Is that just like a high-level overview of everything around?

I like to call it the latest and greatest in machine learning, but not always the latest. Because there's sometimes, it's basically what I've found in the last month of things that I found interesting in machine learning. So it's about 10 or so things. Some of them are brand new. Some of them are a couple of years old from blog posts that I found. I've gone, wow, this is, someone wrote this 5 years ago, but it's still valid for people to know in terms of machine learning today. So I collect, yeah, about 10 resources or so from the internet. And then if I've done something in terms of machine learning, I'll put that at the top as my work. So I figured I've linked— so the AI anxiety blog post itself was on my personal blog, and I just linked it in the, the Graves letter.

No, I'll link, uh, the newsletter and your blog in the comments below. So if anybody wants to check that out, make sure to check it out. Um, one thing that I was going to ask you related to that is, um, what do you like— what, what— well, a little bit off topic, but what do you dislike most about kind of what's going on now? Now we talked about the anxiety, and I guess that's normal with just anything in life. But if there was anything in particular that you'd say you dislike, um, from the field— I know that's a hard question, but I feel like a lot of people also, they are— a lot of people like to glorify things.

We—

this is something we spoke about, but there's always, there's always negatives and bad things about anything in life. So, um, that is something that interests me as well.

So The main one would be, I love, I love a lot of things, but the main one would be the, uh, the closed nature of it. The, or sorry, because machine learning, well, just so from my perspective, ever since starting in the field, it's been very open and very I didn't experience this when I was studying at university. A lot of the papers were sort of behind paywalls and that sort of stuff from different journals, but machine learning, a lot of them are just available for free and the code, like these days you can get on GitHub and even now the models on Hugging Face. However, there's been a trend sort of of late and I guess I can kind of also understand the closed nature of it because it's a business and there's competitive like you got to compete. If you, if you publish everything, how, and everyone else knows how to do it, you know, just copy it. It's kind of, it's a fine line. Um, so I don't know the full solution to it, but like OpenAI, as much as like they create amazing, amazing things, they really have almost become closed AI because it's like they've gone from just being completely open to, uh, and publishing things to like Now we don't tell you how we made it, but we just charge you.

We just made it.

Yeah. So like, that's kind of like the real dig, but I can also understand at the end of the day, like, there's that saying, businesses have to make money if you want to survive.

Yeah.

You, that's just the world we live in is yes, of course, it would be amazing for things to be offered for free to everyone. And I believe Hugging Face is doing some incredible work there. And what they're doing, that's probably my favorite machine learning company at the moment. But to train giant models, you do need millions of dollars and that sort of stuff. So the question is, where is that going to come from?

Yeah.

So that's my main gripe is if I had my way, it would be all different models, like large language models would be open. I can understand that And I think they will in time. I think they will, like, because the open source community is just amazing. And that's kind of the rule of machine learning is that if there's anything, and like Open Assistant, they're doing amazing jobs of basically replicating ChatGPT. And how can I phrase it? Ah, so my gripe of it is the closed nature and also potentially peddling narratives. Such as AI safety, after learning more, of course, I'm not ignorant. I know there are chances of it being a problem in terms of people using it in mis— like, not ideal ways. And then there's an even smaller, smaller, smaller chance, very, very small of it, sort of just waking up one day and deciding, well, it's time to go Terminator and destroy the human race. I think that's very, very unlikely because, yeah, we can't even get a car to drive itself yet.

So at least not in any of these, you could say probably in these decades and even beyond that, probably next couple thousand years, maybe we, like you said, I really don't know.

That's my answer on that. Of like long-term AI sort of safety. I really don't know. And so that's why I think it's not dangerous, but kind of maybe pushing the wrong thing to kind of say that we have to close these models down out of a fear of safety. I think it's a tightrope, right? It's like one part of me, one part of me, the creator goes, no, just open source everything, make it available to us. We want to build things. Most people are good. So if, if, if the, it's, it's kind of, I don't know, it's a very, it's a nuanced problem, right? It's fun to talk about, but then the other side of me is going, hmm, you know what? Yeah, it's probably smart to sort of not publish everything straight away. You can test it, make sure it's safe, and then release it. But on the other hand of me, it's kind of like, well, I'm a big, I like having the freedom to do whatever I want. So it's like, give access to this. I really don't know. No, that's, that's, yeah, if I had, if I had one thing that I don't like, I don't like the pushing the narrative of AI safety as a way to go, we're going to be the sole arbiters of controlling everything that goes on.

No, totally. It's funny that you mentioned the the AI safety and kind of touch in general, just touching that AI in the future, because actually up next, what I'm going to ask you is impact of AI questions. So we're going to get into the questions that you see just about everyone asking the people that are inside the field or out, outside the field. And the first one is actually, how do you think the most typical classic question— I bet you've heard this a ton of times, Dan, but how do you think that AI is going to impact jobs around tech in particular? And how do you see that? I want to scale that down a little bit because that's, to me, that's kind of hard to answer at this moment. What I will say is, how do you see the impact in the next 10 to 15 years? I think that's more so—

Wow.

That's something that like, 'cause I feel, you know, down the line, like you said, there could be, there's going to be many more advancements, but right now, at least in the next couple decades, what do you think is going to be a big, factor? Is anything going to change really? Do you think we're still going to be kind of playing the same rope where it's like, you know, we're still in control? And I'm not saying in like an AI taking over kind of way. I'm saying like, are there going to be jobs that are going to just get automated like many jobs in tech, would you say, or no?

I think it will almost— things will look Hmm. I think back 10 years ago and how much has changed. And I was 20 then. I could never have predicted what I would be doing now when I was 20. So I think the same thing is just going to happen again. And of course you can have, if you made enough predictions, some of them are going to be right. Some, most of them will probably be wrong.

It's the better chance though.

Yeah, right. In a, okay, I like to bet on things. So if I was to bet, I think the rate of automation will be far, far lower than what we think it will be. That would be my bet. I think some things will be automated, of course, but the sort of the outlandish thoughts of that thousands of engineers, tech engineers will be automated. I think that's very slim. I feel like if I just use Copilot today, it makes me faster. It does not make me a perfect engineer. It makes mistakes like I make mistakes. So Copilot, Google Copilot to help me code, it does make me faster. So I would say I can write 30% more code, whether that code is better or worse, it doesn't, like at the end of the day, I think a good engineer writes negative code, as in the number of lines you write trends to the negative because what you're doing is you write 1,000 lines, but then you edit it and make it better and reduce it to like 300. So Google Copilot helps me write that 1,000 lines. but then I have to make it go back. And who knows, AI systems in the future may be able to help it go back. However, that will also be helpful, still won't automate a job because you still need someone to write the code. Uh, Colab maybe, or to put it into place. Google Copilot by that stage may be able to just write a whole script verbatim. However, I also believe that the world, we do have a limit, a shortage of software. So think about how many people have ideas of like, if only I could create this app or create XYZ, I think that would be really helpful. However, the barrier to entry to code and machine learning and all that sort of jazz is quite high. You still, it takes a while to learn. Like I've been doing it for 6 years. I'm still learning something new every day. And I'm by far not the best engineer. So I think if more people could program, could develop their own apps and whatnot, create whatever they need to create, well, that's a beautiful thing. I think there was an article recently, I haven't read it, but I did see the headline. It was back in 2015 or somewhere thereabouts when computer vision was getting really good, is Geoffrey Hinton, who's one of the godfathers of of deep learning. If you look up his name, he'd be on every sort of founders of deep learning post ever and has done some amazing work. However, he made a prediction in around 2015. Again, I might be butchering the dates, but basically said because computer vision is getting so good, it can understand X-rays and then the job of a radiologist will be automated within the next 10 years. So radiologists, someone who looks looks at X-rays and tells you about it. So if you went to an X-ray, here's the, the technologist viewpoint is you go to the X-ray, you get the scan and the AI will just tell you whether you're healthy or not. You don't even need a radiologist. However, that prediction has turned out to be not only wrong, but catastrophically, not catastrophically wrong, but the rate of radiologists, the number of radiologists is being hired is going up. Instead of down. So computer vision has gotten much better than it was in 2015, but more radiologists have been hired now as a percentage than when the prediction was made. And so here's where I think it's a very, it's a, this is why as a bet, I think again, automation will probably be far less than what we think, but as a technology nerd, I would say it would be higher. So this is the problem though. Is because a lot of people who are sort of too embedded in technology and research often predict that the output of the technology is going to be much larger than what it actually turns out to be. And don't get me wrong, of course AI is going to be everywhere. It already is. Like, we've probably used 10 AI services this, this morning just waking up, uh, whether that's machine learning on the phone, whether that's sound compression, XYZ. So from a purely economic standpoint, I think automation will be far less than what we think it is. It will probably make us more productive in many different aspects. But as for in 10 years' time, there'd been like, yeah, 30, 30% less software engineers than there is now. I think that's completely wrong. I think it'll probably be more closer to 300% of the software engineers. So 10x instead of 10 less. But from a technology nerd standpoint, I would say that a lot of things would be automated. But again, viewing things through the eyes of technology all the time is far more narrow. I mean, this is my AI, that was what I was feeling AI anxiety for was because I was like, wow, I'm viewing this from just one point of view. It's very narrow. It was like technology is going to be everywhere. And of course it will be. And it is, we're talking on different screens. However, you just got to go for a walk or go for a drive in a neighborhood of like manufacturing and different things and just see how much is just actually going on. There is so much going on that you don't even know. And so the point of view, I think it's very sort of self-centered, the point of view to go, AI is just going to take over all of that. When, how long have self-driving cars been on the horizon? And I mean, Tesla and other companies are doing amazing work there, but what we're starting to really find is you've got thousands of engineers working on this one problem, just get a car to drive it by itself. Yeah, it's, it's, it's really good. Tesla's Autopilot is really good. Full Self-Driving is really good now if you watch the YouTube videos. However, there are still so many little edge cases that it just gets tripped up by. Whereas a human who learns to drive for 20 hours can probably do a lot of those things.

Of course.

So there is like something missing that we may figure it out eventually, we may not. The, the religious person in me goes, well, we're never really going to figure it out because it's— there's just a spiritual component that can't be reduced to, uh, numbers and code and whatnot. Uh, the engineer in me thinks Yeah, that's just a, just give it enough time, technology will get at it. So it's, I think it's just, it's just a matter of from all different points of view, I come to different conclusions for what's going to happen in the next 10 years.

How does someone get started in this field? Um, for someone that's interested in machine learning, maybe hasn't really learned much about it, doesn't really know much about it, um, what are the steps you would tell them to like to, to learn, uh, How to get into this and actually look for a job.

So the real answer to this is that there's probably a thousand different ways to approach it. And I'm going to answer it if you're a complete beginner, because the steps are kind of relative to any position. You're a complete beginner. You're seeing all the, the, the new awesome things being done with AI. Machine learning is everywhere and you want to learn it. Of course, there are abundance of resources online. However, as I said before, you're going to come across thousands of different things. However, that's normal. I do that daily. You're going to get overwhelmed, but you have to just know that you're going to, you can learn it. If you have that fundamental belief, you can start to narrow things down. Pick the things that interest you the most. Of course, there's going to be the fundamentals that relate to anything. So narrow 1,000 down to 3. It's a very hard thing to do. That's why a lot of people can't actually do it. So the art of learning anything new is to reduce the number of things you're trying to learn because we're humans after all. We can't learn everything. So reduce it down to a— that's kind of why I created my courses. They could be 1,000 hours long, but the 42 hours is kind of reduced down to what I thought my point of view was the best things to get started and learn. Now, again, you don't even have to— you could just ignore that, ignore my courses. That's because there's so many beautiful resources out there. So you have to narrow it down to about 3 things. But then you have to be diligent enough, have discipline. If you want the freedom to work in a field that you'd like to work in, you need to have that discipline to learn the skill. And that's with anything, any field, machine learning, web development, anything. You have to dedicate the discipline, the time to learn that. And so I find discipline is a lot easier to have when you enjoy the thing that you're learning. This is why I always advise for follow your interests. And the fundamental thing that businesses and people are looking for when they're hiring is skill. However, another sort of caveat of that is that it happens with a lot of tech people, engineers, is you have to market your skill to get a job these days. It's kind of, it's kind of strange in the tech field. In a lot of fields is you have to almost— the skills for getting a job and the skills for doing a job are, are different. Yeah. As in, there's a lot of resources out there of just how to prepare for an interview. Now, I've never really gone through that path, so I can't talk from experience, but the fundamental thing, the no questions asked, is the skill. You need to have skill.

Yeah.

So you need to have the discipline to build skill. So let's say you've gone your big funnel of 1,000 different things you found online. You've narrowed it down to 3 to 5 and you've really gone through them. You've applied some effort. You've got some skill. How do you go from having just skill in studying in your bedroom like I was doing to getting a job? That's where marketing your skill comes from. And my favorite thing to market your skill is to start the job before you have it. I have a blog post on this on my website, which talks about this more, but it's very simple, is that if you want a job at said tech company, or you want to work with someone, or you want to, there's an awesome repository on GitHub that you found, and you want to do things that they're doing. My advice for that is to start doing it. And you might think, what do you mean? How do I start doing it? If I don't have, like, if I'm not working for their rate. That's the thing with a lot of, this is part of, so courses teach about the foundational knowledge, which is important. You need the foundational knowledge, but then working on your own projects teaches specific knowledge, knowledge that can't be taught, knowledge that you can only gain by actually doing a certain thing. This is like with martial arts or jiu-jitsu. How good could you get at jiu-jitsu if you just read a textbook on jiu-jitsu?

Not very good.

Yeah, exactly right. So you need to actually practice the jiu-jitsu in the field. So the same thing with, with, with university. You do a lot of textbooks reading, you do a lot of, sometimes you do a lot of practical, but my, in my experience, online courses, a lot of them are similar. They're getting better these days, especially ZDM stuff, but they will only teach you foundational knowledge. It's then up to you to take that knowledge and practice it. So like reading the textbook on jiu-jitsu, you'll never actually be good at jiu-jitsu unless you start to roll with it.

Yeah, I agree.

The easiest part of— oh, sorry, it's not going to be easy. Start the job before you have it because you, you're not going to know what to do. But that's the point of figuring it out because any job that you get you actually get, you will spend the majority of the first few weeks, months upskilling yourself to, to learn their processes because it's going to be different. Every company is going to have different processes, is going to use different styles of this, different styles of that. The skill you really want to develop, the meta skill, is the ability to adapt, to learn things on the fly. So if you have an GitHub repository of like an open source company, say you wanted to work for Hugging Face, amazing machine learning company right now. Start going into their GitHub issues and figuring out how to fix what's wrong with their codebase. It's going to suck because you won't know a lot of it. You'll spend days reading things that you don't really understand, but you're going to build skills doing that that you can't learn through a university course or you can't learn through an online course. You need to kind of have that practice of just going through not knowing how to do something, but then working out how to do it step by step. And that's starting the job before you have it.

Yeah, no, I agree. I agree. I think sometimes you just have to dive into it, right? You have to just jump in the water. I like the way you said that. Another thing that I've gotten asked actually a fair bit of times from some viewers, and I don't know myself, but there's this big thing about to become a machine learning engineer, data science, machine learning and data science in general, you need to have this high-level math, mathematics skills, mathematician skills, or these high statistics skills. Is any of that like, what's the truth to that? Like, do you need to be a math wizard to be in this field? What's your take on it?

So I'm going to say no, but I don't want to take away from the fact that machine learning is math. It's math at scale. So although you don't need it, you shouldn't shy away from it. And I think this is just a fundamental zeitgeist or culture view of how math and whatnot is perceived in school. And statistics is viewed and whatnot, just through the eyes of the culture. However, to me, it is, is just another skill set that you can learn. So for me, the majority, the majority of my day is spent coding. 99% coding. I don't read mathematical equations. I don't solve mathematical equations. I don't If I read a machine learning paper, there are mathematical equations there, but often I'm still interfacing with the paper by code. If you can learn to code, you can learn mathematics. So that's my sort of take on it. Now, to deploy a machine learning model, you can do so with zero math. To build a machine learning model, you can do these days with zero math. To apply a machine learning model, you can do with zero math. You can literally write 5 or 6 lines of Python code and have some of the best machine learning models in the world working on a sample of data. Will that be a product? Probably not. Do you want to build a product? You should be quite good at code. You should be quite good at business acumen. But again, these things are skills. You can learn them on the fly. I'm learning a lot on the fly. I'm building Nutrify. I know how to code. I kind of know a thing or two about business. I'm trying to build a product. We're not using any math for Nutrify. We're working with data. We're working with machine learning models. All the math that we're using is quite basic. It's like, is the model— we measure with metrics. Is the model better than last time? Yes or no? Very simple. However, if you want to be a researcher and go to a university or be hired as a research machine learning scientist at one of the big tech companies, you'll need a lot of math. So it really depends. Do you want to build products? Math, I would say, 30% important. Do you want to do research? Math, 100% important. So that's the caveat, but don't be scared away from it. It's just, it's math is another language. It's beautiful. And what enables This is why I don't want to seem like I'm sort of knocking it, because machine learning isn't. If you're running a short code, it works because it's running just functions on a GPU thousands of times per second.

Yeah, no, that makes a lot of sense. But it is nice to hear from an outside perspective that it's not like you need to have this graduate's degree in math. To just start trying to dive into this?

Uh, no, it's a good start. No, not at all. I'm a big fan of learning what you need on the fly.

That's great. Yeah, I love that as well. I love that as well, especially with jiu-jitsu. Uh, but sometimes it takes, you know, you know how it is, you, you start rolling and you pick something up that first day, then the next day you come back, you just said, I don't, I don't remember what I was doing.

Yeah, exactly. That's the same thing with learning to code. Anything is, is just You'll have an idea, whatnot. You won't know how to do it, but then you just practice, practice, practice. And then eventually you get better or hopefully you get better. And then you can start to do things that like, I forget if I'm learning something new, I struggle with it. I got to do it 100 times before I can do it like verbatim by memory. But now it's like, reflect back on 5, 6 years ago, I could not write a single line of code, but now I'm building full-blown machine learning applications.

Yeah, no, that's great to hear. This next segment is actually something that people are also very interested. I think it has this like great cloud around it. I want to say every time, regardless if it's machine learning, if it's whatever topic when it comes to, and that topic is compensation. People always sometimes get scared of talking about money and all, but it's a big thing, especially because like you stated at the beginning of this interview, part of the reason you got involved with making courses is because you were looking to somehow make money for yourself so you can finance your life pretty much, you know what I mean?

Yeah.

So I think it's an important thing for people that are thinking about getting into machine learning to, and have those questions, maybe you can give us some insight on it. And the first question of that is, from your perspective, when you worked at that company back in the day, what is typical for an entry-level machine learning engineer to make? If that, I know that there's so many ifs, ands, buts, but entry-level, I would say like junior.

Yeah, junior in Australia these days is probably 90 to 120,000 Australian dollars. So that's USD, oh, sorry, um, 90, so about 65,000 USD to 100,000 USD. So that's for a junior. And then of course, like the salaries like go up from there. Um, I personally don't know like what, what, what it would be these days. Um, that was when I was, and I'd say it might be a little bit higher than that. I would say this is now, this is at a smaller company. This was not a big tech company. So, and I think it's a very valid, okay thing these days. I think it's talked about or not talked about. Oh, sorry. It's kind of looked down upon if you just say, oh, I did it for the money, XYZ. I think that's just a valid reason as any. And I don't know why, like I've had job interviews before where people are like, why do you want to work here? Especially some of my first jobs. And I think one was for Subway, like a sandwich shop. And it was like, I think the guy kind of expected me to say, it's my dream to make Subway sandwiches. It's like, no, I'm 19. I just need a job for some money. Like that's, that is, that was it. Right. Um, so no, I don't think it's, it's, it's fun. That's why I say like I'm doing, I teach one of the reasons I teach courses because it's good money. I think it's a very good thing to, uh, or a very smart thing in, in, in your life to, to earn money doing something you enjoy doing. And so I agree. Um, now rates-wise, if you have skill, you're going to get paid more. So if you're a beginner, don't expect to just jump into massive amounts of money. You need to build skill. Then as you get better over time, you get paid more relative to your skill. I'm a big fan of that. I think Netflix did this with like, instead of having 10,000 engineers, they just have 1,000 or so. And instead of paying 10,000 engineers $150,000 each, they pay the 1,000 engineers like $1 million a year. But you have to be really damn good because they keep the team small. Instead of having so many people to manage, but they pay the people more because they're good at what they do. So I like that model. It depends where you want to work. Do you want to work at a big company? A big company will probably be able to pay you more, be able to offer you more benefits. But then again, I'm sort of anti-big company, partly because I generally wouldn't have the skill to get a job there. Partly because my skill set, I prefer doing different things and partly because, and I'm not the greatest engineer, but also because I just like the smaller team. Like, of course, FAANG is amazing. It seems like this holy grail, like everyone in tech wants to get a job there. But I think I just have it for some reason. I just like the underdog. I like the people working for smaller companies, you know everyone you work with, you, you're working on something that you all like love and enjoy. Um, so I don't know, I'm biased towards that, but you might not get paid as much.

No, I, I think that it's, it's for some people and it's, it's not for others, the whole FAANG thing. But I know around the web development space, that's also, it's also very big, as you know. So yeah, I'm, I'm more of a person, I love the situation that I fell into, and I love a smaller team. I always have. But You know, if the big thing, if, you know, if some people want to go for a FAANG company, you know, by all means go for it because nothing's stopping you there, right? You put enough time in it, you'll get there. One thing I did want to ask you is what's the way or how would you respond to someone that simply told you, hey, Dan, like I want to make as much money possible. That's kind of like I like this machine learning thing, but I want to make as much money possible. That's what I'm going for. Is there anything that you would particularly tell them to niche down in? Any particular topic in deep learning, maybe deep learning or in any of this kind of machine learning umbrella, data science umbrella? Is there anything that can potentially lead you to get paid more if someone had that intention?

I don't know what specific thing I would say to niche down in because that specific thing would probably change because that's the nature of technology. Yeah. Like, uh, maybe 5 years ago, if you were, or maybe longer now, if you started a self-driving car company, you could probably get a lot of money through funding. But whether your company would still exist now, that's another question. So I feel like the principle is if you want to get really rich, you need to start a business of some sort. You can work your way up at a company and, or be part of a business, part of a small business. That's how you, Naval on Twitter has this great thread of how to get rich without getting lucky. And it's basically you need to own a part of a business or you need to own the whole business, something around business, because then you're going to leverage economies of scale. And so that's what I'm trying to do with like Nootify, with CTM. It's part of a business, right? It's instead of just rocking up to a place and maybe getting a raise once a year, all that sort of stuff. Yes, you can get rich doing that. There are plenty of people who are, probably worked at Facebook for 10 years and have a lot of money. If you want to really, really get rich, you need some sort of business equity. And the other thing is just time because like at the start, your business might not make that much money, but then as it grows, it will make more. And then if you are lucky, do things right, it's going to go up and up and up. And the other side of that is if you just want to work a job, well, you have to save and then you have to invest in stocks and whatnot. These are tried and true things that have been done. Nothing new under the sun, right? Tried and true for hundreds of years. But what you're doing with investments is you're investing in a business. So you, as that business grows, your stock portfolio grows. So either way, if you want to make lots of money, one, develop skill. You need skill of some sort. Whether that skill is, it doesn't have to be technical skill because there's a lot of, that's another thing about AI taking over is that there's a lot of, um, super smart people out there that don't take over, that don't take over things. Like house cats. House cats basically dominate humans, uh, for many people, right? So there's a lot of people out there who like literally pander to anything that the house cat does. The house cat has— runs the house basically. Now that's an extreme example, but going back to it, there's a lot of companies where I know that if I was to hire engineers to work with me on Nudgefy, that there's be a lot of engineers out there that are smarter than me, but there's a difference between like technical smart and product smart, business smart. So Mark Zuckerberg is a smart guy, but there are probably engineers at Facebook that, or Meta, that are smarter than him on a purely intellectual basis. Basis. However, he's the richer one because he's running business. Yeah. Does that make sense? So it's like—

no, 100%. Yeah, you need skill to get rich, and one of those skills can be running a business, and sometimes that tends to be the most lucrative one, um, for the people that are looking to make the most money. Uh, another thing I was going to ask you outside— so we, we've spoken about, um, kind of like the number side of things, but Are there, 'cause I know for, you know, for these companies, I'm assuming it works similar for a machine learning engineer, but there are like these benefits, right? So for example, at a FAANG company, if you were to go in as a level 1 web developer or a software engineer level 1, you get a certain amount as a base salary and then you'll get certain like stock in the company. It's not stock in the company, but these different bonuses that you get. Is that relatable in machine learning as well? Is that safe to say?

Yeah, well, I feel like, yeah, with any large tech company, their benefits are generally quite good. I can't speak from experience because I've never worked there, but I've been to, I went to the LinkedIn head office in New York. I've got a video on my YouTube channel about that. And there was just like multiple floors of different food and it was free. So they give you lunch and that sort of stuff. Same, I went to the Instagram office in New York. Um, there was a restaurant there, basically it was free food for all the employees. Uh, but that's kind of a side effect of— it's like a gotcha in terms of if they give you free food, you spend more time there.

Yeah. And you stay— they have a gym there. Now a lot of these, I, um, I forget which one, but, um, I was doing— I was watching a video on, on, I think it was Google's, uh, head office in New York City. And, uh, they have a gym there. They have all these different amenities there. I think the purpose is to try to get you to work all day.

Hey, well, it's, it seems, yeah, that's a great experience for a lot of people and many people would, would love that. I do. And that's, it's just not, I would love that too, but I think it's just not like, it's like levels of what you would love. It's like, I love what I do right now a lot. Would I enjoy going to a facility like that, uh, and having free food? Yes, I would enjoy that. But that might be an 80 out of 100, whereas I'm like at a 95 out of 100 doing what I'm doing now.

Yeah, I'd say your, uh, your situation is pretty sweet. Uh, one, one other question in this, um, compensation thing is for the freelancers out there. What happens if there's this person that just is not doesn't want to work for anyone. Can you make money as a freelance, uh, knowing the skills that you know, machine learning, data science? Um, is this something that is profitable as a freelancer?

Yes. So this is the same principle would apply if you're being a freelancer or if you're, um, going for a job at any company, is start the job before you have it. So a lot of my opportunities have come to me instead of me applying to them. The reason being is because I've put a lot of information online. I've created like this portfolio, but say, oh, if you research my work, you go, oh, this guy must be pretty good at machine learning. I'm going to ask to work with him. And yeah, I am pretty good at machine learning. There are definitely people that are better than me out there, but I've kind of got the marketing. There's this like building skill and marketing skill. Similar, but they're like two different things almost. So you can build skill in private, but then if no one knows that you have that ability to do that, like, are they going to come to you and hire you?

Or like, it's, it's almost like a self-branding, right? Like you have to brand yourself at a certain point. Um, yeah.

So whether that be if you're looking for freelance gigs, you start writing blog posts about what you know, you start adding code to GitHub, you start doing pull requests, you start applying to jobs, you start looking at job posts and going, before I apply, I'm just going to spend 2 to 3 days doing this, and go, hey, I've already built you a proof of concept. And then they're like, oh, actually we've decided to go to someone else. Is that wasted? No, it's not. You just write about it and put it online and go, hey, here's what I did for this company. I didn't get the job, but this is what I would do. Then all of a sudden you've got a project that you can refer to. You go to the next job, what have you done in the past? Oh, here's this thing I, I built for company Fern. I've got a Fern here, so I'm just trying to think of it for company Fern. It's like, oh wow, that looks great. How can you do this? Oh yeah, I've actually done that as well for, uh, company Headphones. So that's the, the starting the job before you have it. You get the system going. In the beginning, it's probably like you just, I'm doing all of this for nothing, but over time is where the It compounds.

Yeah, 100%.

To build that skill of, of, of seeing something you don't necessarily know how to do to begin with, but you figure it out. Because that's what a lot of companies are trying to hire people for, is like, can this person figure it out? We don't have the answers. That's why they want to hire you.

Yeah, 100%.

Proof that you figured something out.

I like that. Um, now I know that this interview has been a long one already, or this chat, this conversation we've had has been a long one, but I have Before we close it out, I do have this last section that I wanted to try out. We didn't do it in the last episode. If you didn't see that, that was with Daniel Scifano. You can link to watch that over there as well. But this is a true and false section, Dan. So this for here, what we're doing is we're trying to go as quickly as possible. I'm going to say some things and you're going to tell me if, if it's true or if it's false. And I looked up different things. I even talked to some chatbots for this. I was in ChatGPT Plus. I was, I was seeing, uh, Ipiq give me some. So I'm, I'm going to test you here. I got 5 of them. You ready for them?

Mhm, I'm ready.

Let's go. The more complex an ML model is, the better it will perform.

True or false?

Heavy hitter right off the bat. Was that honestly true? All right, all right, I'll take that. Machine learning models are completely unbiased in their decision-making.

False.

All machine learning algorithms are made equal and can be used interchangeably for any type of problem.

False.

Adding more data to an ML model will always improve its accuracy and performance.

Mostly true. Okay.

I thought you were going to give me like a 50-50 on that one.

Yeah, well, yeah, it's mostly true.

Mostly true.

Yeah, mostly true.

I like it. And then the last one, the goal of machine learning is to create models that are 100% accurate and can predict everything perfectly.

True, but you probably won't ever do that.

True with an asterisk. Probably will never get there.

True depending on what your desired metric is or desired output. Oh man, but it won't ever happen.

Out of those 5, which one do you think was the hardest one to answer? Just out of my personal curiosity.

Is it the, uh, uh, the first one, the more complex and machine learning model depends. Of course it depends on your definition of complexity, but the, the latest models, uh, yes, they're complex, but when you really break it down, there's, there's only a fair, like a handful of moving parts. So, but then the patterns they learn are complex. So it's, it's kind of, so it's like, and then I see the, the unbiased is easiest was the easiest one. All machine learning models are made equal. That was easy one. And then, uh, adding more data to ML model will always improve its accuracy performance. It, of course, it depends on what data. So it is a pretty broad term. Like you can add, maybe you're trying to take photos of food, but you add photos of cars. It probably won't help that much.

It won't do much for you. No, I like that. Um, yeah, I like that. I like that little segment. I think I have to keep doing that. Those little true and false.

That's a good one.

Um, all right, man. So listen. Yeah, but I, that's why I had to hype it up a little bit. I had to be like, Daniel, you got to be ready for this, man. I'm going to hit you at 5. Um, man, it's honestly been a great conversation. I've thoroughly really— I just really enjoyed it. Um, but one thing I wanted to ask you, a couple more things, but, um, one of the most important ones is if you were to give your, uh, yourself, your younger self advice, um, with kind of everything that— even outside of mostly about, you know, your professional career, but just what you're doing in life now, what's one piece that you would tell yourself, just looking back at everything you've kind of done?

Have fun.

Have fun. Australia is a good place to have fun. I love that one, man. And any last piece of advice that you would give for our viewers out there, the people that are interested in diving into the machine learning world, kind of like a final capstone of one thing that you would tell them?

Same again, have fun, dive in. There's a lot out there. You'll feel overwhelmed, but the overwhelm is important to eventually feel free to learn, free to explore the depths of machine learning, the magic of AI. And yes, You can learn it.

Yeah, I love it. I love it. Um, well, you're always constantly pumping out different content, man, and, uh, on different, on different platforms. Can you give us like a, of the, uh, social medias that you're willing to put out here? Can you just, uh, to say it one more time for our viewers, the different, um, social medias that they can find you at? And we'll make sure to link them down below as well.

So the best place is my website, mrdeeburke.com. Mrdbourke.com. Daniel Bourke should be one of the first if you Google that. YouTube, Daniel Bourke. Twitter, Mr. D Bourke. My handle everywhere is Mr. D Bourke. And then there'll be a bunch of links on my website to other places and what, but they're probably the big three. LinkedIn is another one. Daniel Bourke on LinkedIn. So yes, my website's probably the best.

Perfect. I'm glad that you said that because I'm a big fan of just keeping everything the same. All of my social medias are the same handle. I wish I did that. To our viewers out there, start doing that. It just makes it so much harder to find someone when I have to check different names. But in all seriousness, Dan, it's been an absolute great conversation, man. I appreciate you spending the time. I know it's been a long interview or a long conversation, and I appreciate you taking the time to do this. And I look forward to more chats in the future, man.

Beautiful. Thank you, Aldo.

Thank you, bud.