Jon Krohn: 00:00:00
This is episode number 467 with Noah Gift, founder of Pragmatic AI Labs.
Jon Krohn: 00:00:12
Welcome to the SuperDataScience podcast. My name is Jon Krohn, chief data scientist and best-selling author on Deep Learning. Each week, we bring you inspiring people and ideas to help you build a successful career in data science. Thanks for being here today, and now let’s make the complex simple.
Jon Krohn: 00:00:42
Welcome back to the SuperDataScience podcast. I’m your host, Jon Krohn and holy guacamole, what a guest we have today, Noah Gift. Noah has worked in countless technical leadership roles in his 25 year career, including as CTO, cloud architect and consulting chief data scientist. He held the roles at companies ranging from tech startups he founded to prominent institutions like ABC, Caltech, and AT&T. Today, Noah’s founder of a consultancy called Pragmatic AI Labs and he devises and teaches data science curricula at several of the most prestigious American universities, including Duke, Northwestern and Berkeley.
Jon Krohn: 00:01:24
He’s written on top of all that eight books. He has eight of them, including Python for DevOps, and the forthcoming Practical MLOps. Our conversation today centers around how data scientists can make a massive impact in their career with relative ease. We discussed the pros and cons of various educational options inside and outside of formal university programs for becoming a data scientists, as well as for upgrading your data science skills while working.
Jon Krohn: 00:01:52
We talk about how massive global problems like cancer, hunger and climate change can be solved by operationalizing data flows and machine learning, and we talk about how you can attain more autonomy and freedom in your career. Despite Noah’s enormous depth of technical knowledge, this episode is relatively light on technical specifics, making it ideal for listeners thinking of getting started in data science, all the way through to senior professionals who are looking to learn how they can make a bigger impact than ever, in their data science career and in the world, with bonus guidance for anyone who’d like to diversify and grow their income sources. All right, so many deep, exciting subjects to cover. Let’s get into it.
Jon Krohn: 00:02:41
Noah, welcome to the show. I am absolutely delighted to have you here. It’s been far too long since you and I have caught up. I can’t wait to hear all of the latest. Where in the world are you today?
Noah Gift: 00:02:55
I’m in the beach area of North Carolina, working near the border of South Carolina and North Carolina and on one of the barrier islands, which is always a fun place to be.
Jon Krohn: 00:03:09
That sounds pretty pleasant. Have you been doing that throughout the COVID pandemic, or is this a new thing?
Noah Gift: 00:03:15
It’s kind of a new thing in that basically, one of my ways to be creative is to spend time at the beach and whether it’s in Hawaii … Actually, when I wrote the book, Pragmatic AI, I actually was in Hawaii for three months and I was just kind of walking on the beach. I was like, hey, I should write another book. So for me, it’s very conducive to independent thought and creativity.
Jon Krohn: 00:03:43
Nice. That sounds great. Nice. Not a whole lot to do in New York here these past few months. We’re filming mid April, and I guess it’s going to start changing. So we’ve had a lot of vaccinations on Manhattan and I just noticed in this kind of app, you can see the charts, locally as to how testing is coming along and for months and months and months and months, we were over a 3% positive rate in Manhattan. Just over the last week, it precipitously diving down below 2%. So vaccinations seem to be working. It still won’t rival being on a North Carolina or Hawaii beach, I’m sure but at least I’ll have something to do.
Noah Gift: 00:04:25
At the beach it’s pretty nice because there’s wind blowing things around and you’re getting vitamin D and helps your immune system and there’s nobody out there. It’s not bad, I will admit. I feel very happy when I walk outside and look at the ocean.
Jon Krohn: 00:04:44
I am not surprised. That sounds idyllic. So we’ve known each other for a while. I don’t know how long now. Maybe three years. So we were introduced to each other because we were both lecturing in the O’Reilly Learning Platform which we both still do today and I actually can’t remember exactly what it was, but [inaudible 00:05:04] at Pearson, who manages the Pearson side of the live trainings in the O’Reilly Learning Platform, there was a specific reason why she was like, “You’ve got to meet Noah Gift. I’m going to introduce it in it,” but I can’t remember why.
Noah Gift: 00:05:18
I think it was something to do with Jupyter Notebooks or something like that.
Jon Krohn: 00:05:21
Oh, yeah. Because you were using Jupyter Notebooks integrated, I think, into the platform at the time. That was a new thing. You were using Jupyter lab, I guess.
Noah Gift: 00:05:28
I think a lot of the material that I train with at the university or in books all use actually the Google variants. So Google Colab.
Jon Krohn: 00:05:43
Colab, me too now. Yeah. But at that time, I don’t think … Or maybe that’s what it was. Maybe Colab was new. Oh, that could be it. You might have introduced me to Colab.
Noah Gift: 00:05:52
I think I did. I think I did introduce you to Colab.
Jon Krohn: 00:05:55
I love Colab. I use it all the time. All of my teaching is done in Colab now. It makes life so easy. The only downside, and I hope that someday, this can somehow be ameliorated. I don’t know how tricky it would be from a DevOps perspective, which we’re going to be talking about a lot in this episode, but if I could specify my library versions, like specify a Docker file somehow, that would really then, Colab would be an absolute slam dunk all across the board, because I worry that I’m going to be teaching online and some line of code isn’t going to work and I’m not going to know how to fix it and I’m going to have to stop the class to figure it out, because a library version has changed.
Noah Gift: 00:06:38
I think the trade off is that it’s a managed platform. It’s kind of like a Mac in a way, where it’s like it’s probably going to work. So they’ll probably have the version you care about but if you need to get tricky, then things could get a little bit dicey, but then one fix is you could always just uninstall, and reinstall using the shebang line, or the exclamation point. You could just say pip, uninstall. Which is not ideal.
Jon Krohn: 00:07:11
Basically, exactly. So first of all, for listeners, in case you aren’t aware … Probably most listeners know what a Jupyter Notebook is. So it’s a way to easily execute, typically Python code, but also the name Jupyter actually comes from supporting Julia, Python and R Jupyter and it’s a very, very popular data science tool that allows you to play around script, print out charts, share notebooks with other people, and I absolutely love them as a teaching tool.
Jon Krohn: 00:07:47
I think they’re the best. It sounds like you agree, Noah and Google Colab allows you to run them in the cloud. All you need is a free Google login, and you get access to a pretty beefy cloud compute server, and the only two downsides are what we just talked about, where you don’t have complete control over library versions, any control of library versions, but typically, everything you need is in there. The other thing is that at some point, without interacting in the session, it will timeout. So after half an hour or an hour, and you actually, you’ve used Colab Pro, I’ve noticed. So that gives you a longer timeout, maybe a few hours.
Noah Gift: 00:08:26
For me, it’s worth the 10 bucks a month to actually get a better GPU, and then also a higher memory and then it doesn’t timeout as often. So I use it so much, that I’m probably getting more than my $10 worth.
Jon Krohn: 00:08:44
I bet you are. You’re a disaster for them. Because you’re like the person who pays for the gym membership and always shows up and is using the equipment.
Noah Gift: 00:08:54
Exactly. One other more of like a trivia type thing about Colab and Jupyter, is that the first book that I wrote for O’Reilly is called Python for Unix and Linux, and it was in 2006. So I’m dating myself a little bit here, and believe it or not, we used ipython for the whole book, and ipython is Jupyter. So that was the one of Jupyter was ipython. So I’ve been using “Jupyter,” since almost 20 years ago.
Noah Gift: 00:09:34
In fact, when we wrote the book, it’s pretty funny that people said … I think there was a negative review that was like, “This ipython things going nowhere. I can’t believe we covered this,” and it’s like, boy, were you wrong. Of course, it took 10 years for them to know that they were wrong but I always think about that when someone says any kind of critique is I just go back, think about the fact of like, well, sometimes people that are critical actually just don’t know what they’re talking about. Sometimes they do, but it was just a funny side story.
Jon Krohn: 00:10:14
So that name there. So it was developed originally just for Python. So they called it ipython, lowercase ipython. Then somebody had the idea to expand it to supporting Julia and R as well. So hence the Jupyter name, but I actually don’t know anyone who uses it for Julia or R. Do you?
Noah Gift: 00:10:33
I don’t. No.
Jon Krohn: 00:10:35
It’s funny. Eliminating unnecessary distractions is one of the central principles of my lifestyle. As such, I only subscribe to a handful of email newsletters, those that provide a massive signal to noise ratio. One of the very few that meet my strict criterion is the Data Science Insider. If you weren’t aware of it already, the Data Science Insider is a 100% free newsletter that the SuperDataScience team creates and sends out every Friday.
Jon Krohn: 00:11:12
We pour over all of the news and identify the most important breakthroughs in the fields of data science, machine learning, and artificial intelligence. The top five, simply five news items, the top five items are handpicked, the items that we’re confident will be most relevant to your personal and professional growth. Each of the five articles is summarized into a standardized, easy-to-read format, and then packed gently into a single email. This means that you don’t have to go and read the whole article, you can read our summary, and be up to speed on the latest and greatest data innovations in no time at all.
Jon Krohn: 00:11:48
That said, if any items do particularly tickle your fancy, then you can click through and read the full article. This is what I do. I skim the Data Science Insider newsletter every week, those items that are relevant to me, I read the summary in full and if that signals to me that I should be digging into the full original piece, for example, to pour over figures, equations, code or experimental methodology, I click through and dig deep. So if you’d like to get the best signal to noise ratio out there, and data science, machine learning and AI news, subscribe to the Data Science Insider which is completely free, no strings attached at www.superdatascience.com/DSI. That’s www.superdatascience.com/DSI. Now let’s return to our amazing episode.
Jon Krohn: 00:12:38
Anyway, great tool for Python, and I guess other popular statistical programming languages as well.
Noah Gift: 00:12:45
I would actually, it’s actually probably, in this point, really the go to tool for MLOps as well, when we get to that.
Jon Krohn: 00:12:55
Nice. We’re going to talk about a ton about MLOps later on in the episode, but first, I want to get to a question that I think will be interesting to tons of our listeners. So you have a lot of people who are relatively early on in their data science careers, they might be switching over from another technical field or maybe getting involved with quantitative things in general for the first time. Excited about what you can do with data science and machine learning in the world and are jumping on the wonderful train. This journey, excited for all of you who are getting started. So I get a lot of questions from audience members, listeners to the SuperDataScience podcast, who say, “What’s the best place for me to get started?” I specifically had a question earlier today on LinkedIn, from someone who was wondering about specific master’s programs for data science.
Jon Krohn: 00:13:54
So I think typically here we’re talking about somebody who has already completed an undergrad in a quantitative discipline, probably and now they’re weighing up the various options. So you have taught at tons of universities in the US. Very well known universities, Duke, Northwestern Berkeley. So do you have recommendations for people as to what programs they should be pursuing, and in particular, what about the cost trade off? So some of these programs are much more expensive than others. I didn’t actually research this, but in the question the person mentioned to me, it’s about $70,000 for the Berkeley data science program, whereas you could spend $10,000 to Georgia Tech. So, what are the advantages of paying more, if any?
Noah Gift: 00:14:43
It’s a very complex question, but I’ll see if I can tackle this. So I think if you look at a car, like if you look at Mercedes, for example, I used to have a Mercedes C34, which was a pretty awesome car. It was zero to 60 in I don’t know, 4.1 seconds and auto driving and all kinds of fancy stuff and it’s very expensive, but it was awesome. I got rid of it, because in COVID-19 I didn’t need to drive anymore. So I was like, I’m just going to get rid of this thing. Likewise, there’s cars that are not $80,000. There’s cars that are $20,000 that are incredible, but if you go to the car dealership, they’re not going to give you necessarily an espresso. The seats aren’t heated or maybe they don’t have the semi autonomous driving mode, which could save your life.
Noah Gift: 00:15:41
So in fact, in the Mercedes in particular, I remember one day I was with my family and I was driving on the Bay Bridge coming from Marin County into San Francisco, and someone with a family was just not paying attention. They must have been tourists and they literally just did one of those things where the wheel goes like this, when they go, oh, and then I saw a 14 year old girl or 12 year old girl look at me. I looked at her and I was like, “Oh, no.” I’m going to basically run into this girl. My car was smarter than I was and it just stopped. I was like, woo, this saved some some kid’s life and obviously, this is not me. This was the person driving the car that did this. So there’s some great uses for technology.
Jon Krohn: 00:16:29
I got it. So you take the Berkeley course, because it will literally save a 14 year old girl’s life?
Noah Gift: 00:16:36
I think it’s that there are there basically, it’s a cost structure that if you can afford a very expensive program for whatever reason, your company is subsidizing it or you’re in the military, for example, at Duke, I think is an exceptionally good data science program, the MIDS program, and there’s a lot of military people in there subsidized or you get financial aid, that you’re going to get a Mercedes level experience. It’s going to be incredible. On the flip side, it’s possible that in the case of my own academic career, I got a master’s degree twice while I worked full time. So more of a work full time type master’s student and I’m very thankful that I was able to work full time because I was able to apply the things that I was learning and also, they’re relatively low cost, because they were all California schools.
Noah Gift: 00:17:32
So UC Davis, the Graduate School of Business, and then also Cal State Los Angeles, I got a master’s in computer information systems while I worked full time. I think it really just depends. I don’t know if there’s necessarily one perfect solution, and I think there’s a lot to like about a low cost one as well. So for example, Georgia Tech, that is $10,000. I think that’s incredible. So I don’t think there’s really a perfect right answer for anyone. I really think that it really depends on the situation you’re in. Whether you’re working full time, whether you’re not, who will subsidize it and I do think there’s value in the premium, just like I think there’s value in a C34 Mercedes, but I also think there’s value in cars, like a Nissan LEAF.
Noah Gift: 00:18:20
It really depends on the situation you’re in. I would lean a little bit towards … Just my own bias would be I like the idea of being able to work while you’re going to school, if you can. I think there’s a lot to be said for that and I think the future of education may be that there could be a change where just like we used to go to the movies two times a year and see Star Wars, now we see Netflix, and it says subscription, I wonder if that might be something that’s kind of cropping up. So in a way, I think that that might be one of the things that’s going to occur is that what if instead of getting just one terminal master’s degree, what if you got a subscription and you got the equivalent of four.
Jon Krohn: 00:19:11
I love it. That sounds so useful. In a career like data science, machine learning, any kind of software related, computer hardware related work, any of those fields are constantly evolving. To pay a subscription fee, and get access to extremely high quality training, part time on your own terms, wow, that sounds like a brilliant idea. It sounds … Hopefully, I’ll have you on the podcast in 10 years, and it’ll be like the ipython thing where-
Noah Gift: 00:19:45
That’s literally what I’m doing. So the stuff that I’m working on with Duke is literally that which is that I’m trying to create this concept of lifelong learning, where if you’re an athlete, for example, you don’t just train for the Olympics and then say, “Okay, I’m done, I’m in shape.” That’s not how it works. You have to keep training and hopefully you do it in a way that’s healthy and you don’t destroy relationships or cause too much stress or have to live in a tent somewhere, because you can’t afford housing, because you don’t have a job, which athletes do, but if you can do it in a maintainable, safe and repeatable way, I think that’s the ideal way for education.
Jon Krohn: 00:20:32
So you wouldn’t suggest getting so deep in the weeds on AI research that you don’t even have a job and you live in a tent, and just you’re watching Udacity lectures on your phone?
Noah Gift: 00:20:44
Yeah, I think that’s just the over optimization part of human nature that it’s easy to potentially get too caught up into, I need to study for four years before I can do anything, versus why don’t you just try something and see what you don’t know, and then study those things.
Jon Krohn: 00:21:08
I think, if you can get a relatively entry level job in the space, a data analyst, a business analyst or something like that, getting that hands on experience during the day, and spending some time on the evenings and weekends towards attaining the master’s or the quadruple Duke subscription masters, any of those, I do think that that’s a really great option. We’ve had extremely positive experiences hiring people in my company, who … So in one case, I hired a data scientist. He already had an engineering background, biomedical engineering. So quantitative smarts, and had been doing Udacity courses in his spare time, and had almost finished a master’s, actually the Georgia Tech master’s with an AI specialization.
Jon Krohn: 00:22:02
He wasn’t done, but we were able to convince him to come join us and work with us and he’s one of the most brilliant data scientists that I’ve ever worked with. In fact, actually, he was on episode 459. He was the guest on episode 459, Vince Petaccio II talking about how you can use machine learning to fight climate change. Then similarly, we have a front end developer who works for us, who obtained her high school diploma, worked for a few years as a personal assistant, and was reading software development books and she was working as a personal assistant, and did something like a 12 week or 16 week, full time software development bootcamp and she’s amazing, she is great. So anyway, if there’s people out there who were wondering whether you have to take a four year degree to become data scientist, you don’t.
Noah Gift: 00:22:56
Again, back to the athlete analogy is, imagine if you’re on your way to training for the Olympics and it turns out that there’s a specific regimen that your coach is telling you, because he’s Olympic trainer, and then he says you need to run, I don’t know, the 300 meter. You need to run these 300 meters every month, at this certain time for the next 12 months. Then once you get to the 12th month, then we’ll submit you for Olympic trials, but at month 11, what if you’re ready? It doesn’t even make sense.
Noah Gift: 00:23:38
If your body is capable of producing whatever, a 400 meter interval, or 400 meter run in, I don’t know, 46 seconds, then who cares? Who cares if you’re supposed to complete more. So there’s some things I think about education and the actual diploma itself, that are a little bit shaky, and I think we’re going to see that change.
Jon Krohn: 00:24:04
I don’t disagree with you. I think it makes a lot of sense. So let’s talk about this a little bit more. Let’s talk about learning outside of the university entirely. So you do have all of these experiences creating data science programs at top universities in the US, but you’ve also created a lot of programs outside of universities. So some of the prominent names that you have created tons of data science, machine learning, software development content for include DataCamp, Udacity, Coursera, and O’Reilly. All the big names. So what do you think about those approaches? Do you think that there’s added value to the structure of a master’s to getting that diploma or getting that bachelor’s degree in data science?
Noah Gift: 00:24:54
Kind of going back to the original thing that we talked about. I think they’re all valid approaches. So, for example, I’ll take something I know something about as a hobby. So I’m really into Brazilian Jiu Jitsu, although I haven’t done it in a year because I couldn’t do it.
Jon Krohn: 00:25:11
BJJ, it’s hard to socially distance.
Noah Gift: 00:25:14
Yeah, that’s pretty much the worst possible [inaudible 00:25:18] and I want to be responsible and not cause a problem. That’s kind of an interesting one, because there’s so many different ways to learn Brazilian Jiu Jitsu. One is to just do it, and then you really, actually physically feel pain. Then you want to quit feeling pain, because people keep basically getting you in bad positions, and it doesn’t feel good. So you just kind of learn, and then you can also go to the training sessions like they do at most gyms, where they go through a lecture and you go … To be totally honest, I found those to be almost useless that when someone lectures you, which again, is often what happens in university.
Noah Gift: 00:26:07
Then the other one that I found the most value is really, in a way, going somewhere to a gym. So going to Hawaii. I’ve trained there. Brazilian Jiu Jitsu with black belts, or going to another city, and then just asking somebody there, “Hey, can you show me some things, and I’ll hire you as a trainer, and just show me some things that you know, and tell me what I’m weak at and what I can get better at.” So I think it’s really, if I had to pick the two approaches that are the best is that practice, which would be those learning platforms, or your own job experience, is there’s no substitute for it. You can’t learn your way out of practice. There’s no substitute for feeling a 240 pound man on top of you who has a black belt, who is crushing your oxygen, and you can’t breathe and you’re panicked.
Noah Gift: 00:27:00
There’s no training for that, you just have to experience it. Likewise, in building software solutions, or building systems, you just need to do it. If it’s experiential, I think a lot of the learning platforms are heading this way. I think that’s absolutely the best way to learn and you just can’t substitute it. Even if you’re getting a master’s degree at Duke or you’re getting a master’s degree at Berkeley, you have to do it. On the flip side, and this is kind of going back to the value of some of these programs. If you do get experience with somebody who really knows what they’re doing and you can have one on one attention, there’s no substitute for that. So I think they’re both very valid approaches.
Jon Krohn: 00:27:40
Great answer. I think you covered a lot of the big pros and cons beautifully and with a wonderful analogy. Love the BJJ analogy. So if listeners aren’t already blown away by the many things you do, and I am summarizing. When I list the universities that Noah has taught at, the online platforms that Noah has created content for, I’m only scratching the surface of all of them. I didn’t want to spend the podcast listing all the various places he’s taught.
Jon Krohn: 00:28:15
Then so next point, is the number of books he’s written. So if you’re not blown away by all of the courses he’s taught, he’s written eight books, four of them self published, four of them with publishers, including the very well known Python for DevOps published by O’Reilly. I think I’d love to focus on talking about your upcoming book. So practical machine learning operations, Practical MLOps, and you’re co writing that with Alfredo Deza, who was an Olympic high jumper. We have lots of sports analogies coming in here, I have a feeling.
Noah Gift: 00:28:53
Really, I met Alfredo in early 2000s in Atlanta, Georgia and at the time, I was bragging about how I was a good athlete in college and, “Oh, look at this.” I could high jump, I don’t know what I said, 6’6 or something like this, which is okay. Then he goes, “Oh, well, I high job 7’3.” I was like, “Come on.” I was like, “Were you at the Olympics?” He goes, “Yeah, I was.” Then I look in Wikipedia. I’m like, whoa, he was in the top 10. In fact, he was the number one high jumper in the world at one point at 18 and I was like, wow, pretty impressive guy. Then even the reason I know Brazilian Jiu Jitsu was I accidentally learned it because I was working at a startup in the Bay Area, and everybody at the startup would go to the gym, and one of the guys there was an Olympic wrestler and pro fighter, and we just kind of became friends with him and then he was like, “Hey, you should learn grappling.”
Noah Gift: 00:29:56
I just did really get along with athletes because they have self discipline and I found that it’s just really … And resilience. Self discipline and resilience is really what I find from athletes, and they’re just so pleasant to be around. So I’ve done a lot of work with Alfredo. He’s writing the book with me and in particular, I think the focus of the book is really, I think, maybe different in a way than what people may think it is, in that I have a challenge, I guess, to humanity in that, I think a lot of times, we are not focused enough on results. I think in particular, this will go into maybe a book recommendation that I’ll mention is the book Code Breaker by Walter Isaacson is just incredible book. He’s obviously a fairly famous author and written other books, biographies, Steve Jobs, Leonardo da Vinci.
Jon Krohn: 00:31:02
Yeah, I know the name and now you mentioned the Steve Jobs. We had a guest on not too long ago. I think it was Michael Segala in Episode 447. I’m stretching a little bit of my memory here, but I think he recommended the Steve Jobs biography by Walter Isaacson.
Noah Gift: 00:31:21
Great author. One of the best authors of the modern era. What I really enjoyed about his book about Jennifer Doudna, I believe is how you pronounce her name. I apologize if I didn’t pronounce it correctly, but she’s a Nobel Prize winning scientist. She won the Nobel Prize in 2020 and she co invented CRISPR and they use that CRISPR to create both vaccines, Moderna and Pfizer. So we’re talking about not only someone who’s got the skills, she won a Nobel Prize. That itself is a big deal.
Noah Gift: 00:32:03
Then there was a real crisis, which is it turns out that she needed to save the world, and then she did it. Then also the person that created the, I forget what his name is, but a scientist at MIT that helped create the Moderna one. Basically they operationalized their vaccine. So just like we’re doing with data science, we’re talking about, hey, all this research is awesome and it is, but then there are times when you really do need to actually save the world, and save hundreds of thousands and millions people’s life and she did it and this other person did it as well. I think that’s the real challenge with MLOps is that, why aren’t we doing that? Why aren’t we actually doing what she did and actually thinking, there’s some problems that are pretty big problems.
Noah Gift: 00:33:00
One of them is cancer. Why don’t we actually really, really focus on using some of this technology, in terms of looking at things like radiology, looking at the chest, or we’re looking at whether someone’s going to have a heart attack. There’s all these things that really could help humanity and I feel like if there was a little more of a sense of urgency and a mind frame, a different way of looking at things that that’s how you get … Here’s where I’m coming with this is that if, for example, COVID-19 didn’t happen, would these new vaccines, how long would they have taken to be developed?
Jon Krohn: 00:33:46
A vaccine for COVID if COVID hadn’t happened?
Noah Gift: 00:33:49
Or using that technology. Because it’s a newer-
Jon Krohn: 00:33:53
A vaccine for anything and using the CRISPR and these RNA techniques. Because that’s something that’s new here about both of those vaccines is that we are injecting RNA into people to create proteins that look like the proteins of the coronavirus and that’s never happened before. So really good question, and I absolutely love where you’re going with this. I can’t wait to talk about this more. This is amazing. This is something I think about all the time. Exactly the line that you’re going down here on why isn’t there a sense of urgency that everyone is sharing around climate change, or rather, I think a lot of people have a sense of urgency.
Jon Krohn: 00:34:35
Not everyone, especially in the US, I know that there’s climate denialism stuff going on, but a lot of people are aware of the issues and they’re like climate change is bad. I would like to contribute in some way to not having an issue, but so few people take that same fear that that gets you clicking on clickbait, about climate change and like, oh, flooding here and Manhattan’s going under and southern Florida is going under and those kinds of things. People read about it, you like reading about it. So news outlets like writing about it.
Jon Krohn: 00:35:15
A shockingly small fraction of people take meaningful action. I think there’s consumer choice. So there’s consumer choice things that people do. I think, if you were presented with two options that were the same price, and one of them was using clean energy, the other one was dirty energy, I think a lot of people would make that decision, but we’re not often presented with that decision. So good intentions aside, very few people are taking action. So anyway, I’m interrupting you, but I love where you’re going with this.
Noah Gift: 00:35:46
I’m a fan of clean energy for multiple reasons. Just from a technical perspective, it’s just a superior technology. Forget, for a second, even climate change. I have Tesla battery backups at my house, I have solar panels, and basically, I don’t pay for energy and also, I cannot ever need power. Why would you not want that? It doesn’t even make sense to not want that and in fact, everything should be like that. It’s just a better technology. I think that’s kind of the sense of urgency even if we shifted to things like clean energy is that you can build optimization solutions that know what the weather is, and then decide how much to preheat your house and all kinds of technology.
Noah Gift: 00:36:35
I do think that it’s the operationalizing of things that really is, in fact, its own skill. I think that it’s easy to get into analysis paralysis with things, because there are arguably some really tough problems with deep learning that really does take deep, deep expertise, but it doesn’t mean that you can’t do the parts that aren’t challenging in that way, and just get something done, like implement a solution.
Jon Krohn: 00:37:05
I absolutely love that. If listeners are interested in … I don’t think I’ve mentioned this yet, since we’ve been recording. I think I mentioned it to you only, to Noah before we started recording is that episode 461 with Sam Hinton is … There’s a number of topics covered in that episode, but roughly the second half is focused on using machine learning operations for renewable energy purposes.
Jon Krohn: 00:37:27
So that’s something that we’ve already talked about a bit. So that’s probably why it was on top of mind for me, and I brought it up just now. However, do you want to talk about more of these applications? So kind of the cancer one, I love that and the Moderna vaccine, the Pfizer vaccine, well, the BioNTech vaccine that Pfizer is mass producing. These RNA-based vaccines, there’s tremendous potential, even in cancer. So maybe that’s why you brought that one up specifically is-
Noah Gift: 00:37:56
Exactly. Yeah, exactly. In fact, I’ll give you a little bit of insight. I know some people that work in biotechnology, former students. I have former colleagues from Caltech when I was working at Caltech that are now in the NIH. In fact, at the NIH, I’ll just speak broadly, because I don’t want to necessarily be specific about a certain person, but broadly, that one of the issues is that they give, I believe, $50 billion per year in funding and that one of the things that would be great is if the data was more easily shareable. So at this exact moment, there is no data lake. There’s no kind of like, “Hey, we’re going to all share all of this research in one spot.”
Noah Gift: 00:38:40
Every single research lab is a spoke solution and I think this is part of this sense of urgency is like, let’s say for a second, you really did have to fix that problem. Could you fix it? Absolutely, you could fix it, pretty quickly, or you could you create a centralized system where people could create research and share it and et cetera, et cetera. The context isn’t correct and I think the easiest context to think about if you ever get a chance is Netflix. There’s World War II in Color, and I watched that during-
Jon Krohn: 00:39:13
I’ve watched that. It’s awesome. I recommend it.
Noah Gift: 00:39:15
So that’s the context which is, I was just putting myself into the mind of somebody in like England or something is like, there are the worst people that humanity has ever created and they’re coming for us and it’s not like squishy. They’re coming. They want to destroy everybody. They’ve already destroyed everybody and now they’re coming here. We need to have a very specific plan of action that has to work, and we can’t take a long time to do it and then they did it.
Noah Gift: 00:39:52
So it’s not like we can’t, if the conditions are correct, as humans build incredible things, but I think part of it is that it’s the context and the sense of urgency, and I think with NIH in particular, and health and cancer and all these things, that it would be great if there was more of an operational mindset, where we really did … Maybe the Biden administration with the infrastructure bills, let’s solve this. Let’s actually get stuff working and there are problems that are really, they’re just execution problems. That’s what I think MLOps really is, is less that doing everything with AutoML is the end all be all. In fact, I would say AutoML is probably 1% of a solutions, but it’s operationalizing everything is really, I think, the point of doing MLOps.
Jon Krohn: 00:40:49
Nice. So there are so many things that I want to talk about. You’re absolutely blowing my mind and it’s crazy how aligned our thinking is on this. I don’t know how much I’ve had these conversations out loud to people. The kinds of things you’re saying are the kinds of things I think about in my head all the time. Exactly that World War II is something I think about all the time. It’s something that like the amount of innovation that came out of World War II, because of urgency, because of absolute need that if you didn’t, freedom could be lost and another example that I think about a lot is, so I can see the new World Trade Center outside my window here in Manhattan.
Jon Krohn: 00:41:28
I don’t remember the exact stats, but I remember reading that 20 years ago when 9/11 happened, infrastructure, telecommunications, the severity of the destruction was enormous, but essentially the telecommunication companies and the key utility companies, water, they were able to get things up and running again in something like 24 hours or 48 hours, despite this incredible amount of destruction. I often think like, what are all the engineers doing the rest of the time? Why can’t we, every once in a while have these transformative bursts of creativity and innovation without the destruction.
Noah Gift: 00:42:19
Funny, I was walking on the beach today and I was thinking about something similar to what you just said, which is, and I would say this can maybe go into a deeper topic, but that for some reason, a lot of times when I’m walking on the beach, I think of this expression, desire leads to suffering and it’s this-
Jon Krohn: 00:42:40
Buddhism.
Noah Gift: 00:42:42
Yeah, Buddhism. If you think about the workplace, it’s full of suffering because of the fact that you want a promotion, you want to a better job, or you want your boss to approve your plan. I would say that’s one of the problems of capitalism and also any hierarchical organization, education institutions, government institutions, is that you have to ask people’s permission to be successful. I think that maybe one of the solutions to their sense of urgency is more solo entrepreneurs, where you’re basically, you don’t ask for people’s permission to be successful. So if you really think of some of the really urgent solutions that were positive solutions, they were done where there was a lack of hierarchy. Like in the case of COVID-19 vaccines, there’s a real lack of hierarchy when it’s like, hey, by the way, everyone’s dying.” There’s no, like, let me ask for permission to solve these problems.
Noah Gift: 00:43:46
Then similarly, even if you look at venture capital companies, my experience is that there’s a lot of talk about, “Hey, we’re saving the world,” but in practice, there’s actually not a lot of saving the world happening. It’s really a lot about making money and there’s nothing wrong with making money, capitalism, but it feels like there could be a different form of capitalism that is more impactful and focused on doing things that while you’re still making money, you’re doing unambiguously good things and that’s my focus of life right now is to do unambiguously good things.
Noah Gift: 00:44:27
Funny, I was even thinking of like, maybe I’d write a fictional book at one point called The Developers Shrugged and basically kind of like a little bit making fun of Ayn Rand, but then saying like what if all the talented people quit working for venture capital companies or quit working for big tech companies and then say, “You know what, we’re not going to do anything for you anymore because we don’t have to and we’re going to work on impactful things that are unambiguously good.” That’s one way you could do this is quit working for people that are not doing things in an ethical way, if possible. That’s a very preachy thing to say, especially if you have incredible talent, because I’ve met tons of people who are incredibly talented. Maybe think about the fact that you owe the world something and maybe avoid a hierarchal organization.
Jon Krohn: 00:45:23
I think also something that you didn’t mention, but maybe is one of the ideas behind your thinking there. So everything you’ve said, I think is brilliant and maybe that sense of urgency. If you go out on your own and start your own business, it forces a sense of urgency upon you to solve the problem that you’ve set up to solve.
Noah Gift: 00:45:43
Exactly. If you can’t create revenue, then you’re dead in the water. I think revenue is really solves a particular business problem, the sole entrepreneur’s problem, which is, can I make a product? That’s question number one and does it work? Then also will someone pay me for the product? Those are kind of some of the fundamental questions of solo entrepreneurship and in fact it’s kind of embarrassing. If you think of other professions like roofers or people that do lawn maintenance, do they have a product? Yes they do. They mow your lawn and does someone want it? Yes they do. A lot of times there’s a lot of focus, I think on the venture capital side is that it’s just this silly game where people are like, “Hey, look, I raised $30 million.”
Noah Gift: 00:46:36
It’s like, why don’t you just brag that you have a trust fund? What does that even mean? This means nothing that you raised $30 million. Instead, I think with, and I’ll even encourage students to this. I said, “You should at least consider being a solo entrepreneur and building something while you’re in school so that you have the autonomy.” Let’s say that you were able to create a small machine learning SaaS company while you’re in college and earn $5,000 a month. You might be done for the rest of your life. You might be done. Then you could actually have the autonomy to really think about what you’re building, the kind of solutions and then work on something that’s really helpful to the world.
Jon Krohn: 00:47:21
Nicely said. That kind of option falling to that possibility, maybe falling more to the people who can afford to go to the Berkeley in the first place, but definitely something worth considering.
Noah Gift: 00:47:36
Well, I would say, in fact, definitely, my background is that I remember when I was growing up. We had so little money that someone gave us a couch when I was like eight. I still remember that. I was like, we didn’t have a couch. So, I think that there is definitely an alternate non independently wealthy way to approach things, which is if you’re able to, from a very young age, start building. Not that you’re trying to get rich, but you’re trying to build autonomy, which is a very different thing because autonomy could mean that your cost of living is incredibly low and you just have some form of small income so that you have the ability to not be forced into bad decisions. I think in particular that’s one of the problems is that the more money you make, then the more likely you are under someone’s control.
Jon Krohn: 00:48:37
All right. So I didn’t intend on going down this route and talking about this, but you just reminded me about something that you’ve definitely written about online and I think you might’ve also incorporated it into a book maybe even into this Practical MLOps book, but I know you talk about, you have three colors of income streams, right?
Noah Gift: 00:48:55
Yeah. So in particular, and I’m still kind of developing this concept. I think about it still a lot, but I call it red money, yellow money, green money and in particular, a salary is red money. The reason why it’s red is that I think many people think it’s safe. “Hey, I got a job at a big tech company and I make $800,000 a year.” That’s true. You, you have a lot of money, but it’s more complicated than that because now, and not that it’s bad to have a job, but you should be aware that it’s actually less safe than you think, because now they’re providing your income stream.
Noah Gift: 00:49:34
Yellow money would be doing consulting work. So I would encourage someone that is only earning a salary to also have some form of consulting work. It could be doing what you and I do. We’re writing a book. That’s consulting. We’re doing some kind of work with other people or having side projects or whatever. What’s nice about that is that you can potentially think of things more as a computer scientist and a computer scientist designs for failure.
Noah Gift: 00:50:02
Think about red money or a salary is a monolithic application on one server. It’s just sitting there. It could be really awesome, but it’s on one server and what happens if it goes down? Well, we know many examples of that. Where consulting, you’re designing for failure and I would say my opinion with consulting is never, even if somebody’s giving you a ton of money and I’ve had these situations where all of a sudden people start throwing a ton of money at you, and I just cap it at 25% of my consulting and say, “Nope, you’re not going to turn into red. I’m going to keep you at the yellow color.” Then I would say the green side is getting into, can you wake up and walk on the beach and then continue to walk on the beach for four months and not do anything? Green has that-
Jon Krohn: 00:50:58
Which is passive income.
Noah Gift: 00:50:59
Passive income. It’s either MRR, monthly recurring revenue from a SaaS product, for example, a rental from a property, index funds, dividends, royalties. There’s an unlimited amount of things, where it doesn’t even have to be technology. It could be anything. No one told me that when I was growing up, but I’ve kind of figured that out along the way and I think that not that I’m like Paul Graham who has written some really controversial essays lately.
Noah Gift: 00:51:30
Someone said, “Get ahold of him and say, ‘Hey, quit writing these essays. I don’t think you’re doing what you think you’re doing.'” But instead of saying everyone should be a billionaire, I think instead everyone should be autonomous, which is if you’re able to have autonomy with … And doesn’t even have to be a lot. Let’s say maybe you have some stocks and they pay you dividends each year and let’s say that it comes out to be $30,000 a year in dividends. Just that might be enough that you can make good decisions for the rest of your life and really focus on thoughtful work. So, my son, for example, I say the exact same thing to him is that you want to have the ability to choose to do things, not have to do things.
Jon Krohn: 00:52:23
I love all of that. I’ve read pieces of it and I love hearing you say it out loud. So another big thing, and I’m actually going to use this as a segue back to the MLOps book and the big impact that we can make. So when I’m at a dinner party, that hasn’t happened in a year, but you’re out for drinks, you meet people, people ask me what I do. I always take a big inhale before I start explaining it, because I haven’t figured out for people who are completely non-technical, if somebody is a lawyer or a friend of my parents, I feel lost. When I do, I often don’t know what to say. It’s so hard to kind of explain what data science is, if you have no awareness of it as a starting point.
Jon Krohn: 00:53:19
So something I say, and then often regret is I say, “I work in artificial intelligence.” Now people jump right away. You get most people, I think it’s fair to say most people who are not technical, which is most people, their impression of what artificial intelligence is, is from movies and TV. So, a lot of people have this sense that right around the corner, there’s an artificial general intelligence that has all of the learning capabilities of a human, which we don’t even have a roadmap today to get there. So it’s interesting, like this huge expectations disconnect between the popular perception of AI and what we have today.
Jon Krohn: 00:54:13
All right. So where I’m going with this is that for some reason, and maybe it’s related to the automation of jobs, that kind of thing, this conversation often then leads to conversations about how unhappy the world is. How unhappy people are, why can’t we make things better for people, everything’s so bad. I have people sometimes who have to hold me back. My girlfriend will have to hold me back on things because I’ll say, “Well, what do you mean? How is the world worse now than ever before?”
Jon Krohn: 00:54:51
If you look at the data, there’s very things that are worse on almost any measurable outcome. Health, longevity, autonomy, ability to put food on the table. Nothing has ever been close to how good we have it today, and that’s actually around the world. Then if you think, okay, if you’re especially lucky to be in a highly developed country, then really, it’s crazy that there is this perception. If we look at politics, there’s a lot of issues in politics in the US today and many of these ideas are around, things are unfair, I’m hard done by. Meanwhile … So anyway, where I’m going with this whole thing and bringing it back to the MLOps thing is that people today, objectively in terms of, measurements of how people spend their week or spend their day, we have way more free time than ever before.
Jon Krohn: 00:55:59
We have way more leisure time. We have way more capital available as a society to invest in projects and ideas. The possibility for people to be making yellow money or green money, in your definitions, it’s never been greater, not even close. As far as I can tell, things are only going to get better and better on those fronts. So it sounds to me like a lot of what you’re getting at in your Practical MLOps book is, look at all of this spare capacity we have as individuals in this society today and how can we be harnessing that spare capacity to be solving the big problems that are facing us as a society? Is that right?
Noah Gift: 00:56:44
Yeah. I think that’s definitely a theme of it. I think that part of the thing that’s holding us back is that it reminds me of a story from when I was working at Disney feature animation in Burbank, which is pretty cool building. I don’t know if you’ve ever seen it before, but there’s like a big hat. It’s like from The Sorcerer’s Apprentice. It’s a really familiar-
Jon Krohn: 00:57:11
It sounds vague and familiar.
Noah Gift: 00:57:14
It’s an iconic building. The dwarfs are holding up the building because Snow White actually made the whole Disney franchise. It’s like the premier spot for feature animation in terms of … Disney is it. So it kind of a cool place and there’s a lot of union people there. I remember at the time that one of my jobs was to go in on a Saturday and put a CD or a DVD into these editing machines that were like $200,000 and push a button and then run some like cleaning tool, like a software, like fix things on the computer tool and I was like, “What?”
Noah Gift: 00:57:58
I was like, “First of all, this is code.” So I reverse engineered these systems that weren’t designed to be multi-user and then I made them multi-user and then I also mounted a network volume on them, and then I created a network drive so that you could basically either remotely or up go up the machine, just hold down the in key and it would turn on network boot and it would automatically give it a script when it booted up and it would reformat the machine.
Noah Gift: 00:58:32
It took 3.5 minutes. So basically, I basically automated our whole department into a three and a half minute automated script and people were outraged. A guy that was in his 60s, he pulled me into a room and I remember my boss was in there with him, a middle manager who used to be in law enforcement, which is kind of interesting, but we’re in there and I remember him saying, “You’re going to script yourself out of a job,” and just shaking and just shaking and shaking and just in rage. I remember the first thing that came to mind is like, “I think maybe I could script you out of a job, but I’m not going to script myself out of a job. I work with you.” How can that possibly believe me that I’m going to script myself out of a job. I just wrote code that automated our whole department. I think part of it is that people are thinking about things …
Noah Gift: 00:59:31
That’s a very silly example, but that people are thinking the same things with machine learning, it doesn’t even make sense. It’s like that AutoML and deep learning are not substitutes for each other, but both sides are not seeing things clearly, which is that, in the case of this person who was actually a very talented person. So I kind of said some negative things [inaudible 00:59:57] but a really talented, actually audio and video engineer and extremely talented. I mean like ridiculously talented person. It’s like, “Hey, now you can focus on building really high end audio solutions all of your time.”
Noah Gift: 01:00:14
The thing that I can’t automate, why don’t you just spend all your time on that? Because there’s a need for it and the same things with this kind of false dilemma that, oh, Google has AutoML vision and you can’t do that. Yes, you can. I’ve taught over 1,000 students to create AutoML with computer vision on Google’s platform and they’ve made iOS apps and Android apps and they worked great. So you’re thinking about the same thing. There isn’t AutoML for self-driving cars. People are kind of talking past each other and I think that’s the real issue, which is that, of course you should automate dumb stuff. Of course you should, because then the real smart people can spend all their time on the stuff you can’t automate.
Jon Krohn: 01:01:04
One thing we should definitely define. Probably people get the picture and it is something I’ve talked about on the show a lot before, but AutoML, automated machine learning is where you take an algorithm that already exists. So it could be a Google cloud API, and you provide it with some training data and it fine tunes. It figures out how to optimize maybe using some existing model weights. So using some transfer learning, but then also experimenting with characteristics of the model, the hyper-parameters and basically AutoML, it’s an algorithm and there’s an infinite number of ways of building an algorithm like that but this algorithm figures out how to optimize for your problem so that you don’t have to.
Noah Gift: 01:01:53
So in the case of very simple single label detection of images, like, is this a flower, is this a tulip or is this a rose, that’s a perfect problem for something like AutoML that really could solve a problem. Let’s say you’re a gardener, for example, and you keep having problems with disease crops. Wouldn’t it be awesome if you, yourself, as a gardener knew nothing about machine learning and you just say, “Hey, I have all these pictures of diseased tomatoes, and I would love to build an application where I can just upload the images to Google cloud, train it. I’ll put my own labels on there, and then now I can deploy it to my phone. I don’t even need to put it in the app store.
Noah Gift: 01:02:43
Nobody even needs to care about it, but now when I’m in the critical time period where a lot of my crops get diseased, I’ll just put this over it and then uh-oh, there we go. The fungus is spreading. I’m going to isolate this tomato.” You know what that’s like? That’s like someone building a shelf in their house, and then to tell someone you’re not allowed to build a shelf. Only master carpenters can build a shelf. It’s like, no, you can build a shelf. It’s just false. No, can you build the house? No, you can’t do that. So I think people are just talking back at each other and they’re lacking a sense of urgency for like, why would you not let people solve simple problems with automation?
Jon Krohn: 01:03:28
So circling back, the idea with the Practical MLOps book is to allow people to take advantage of extremely powerful machine learning models through techniques like AutoML to allow people who may not be able to build the machine learning house, be able to take advantage of the shoulders of giants, that hundreds of thousands of people that have contributed to the machine learning capabilities that we have today. Allow somebody with relatively simple scripts to take advantage of all of that existing knowledge, all of that existing training data and do powerful things with it and change the world.
Noah Gift: 01:04:14
That’s the industrial revolution. Think about, there’s been at least two industrial revolutions. What was it? 1760 to 1850 or something like that. The first one, steam powered, automatic looms and all this stuff and then the same thing happened with automobiles and assembly lines and all this, the second industrial revolution. I think it’s possible we could have a third industrial revolution with all this machine learning and basically it’s, again, it’s like imagine being the person that said, “That automobile, it’ll never replace my horse.” It’s like, come on. Of course, if something’s a better solution, we should use it, but it doesn’t negate these really tough problems. So I would say in the book, there’s actually two core concepts that are slightly different.
Noah Gift: 01:05:04
One is that, yeah, if there’s AutoML solutions, great, use them, but I would say that’s maybe even 10%. The other 90%, which is let’s even take the hard problems. Let’s also operationalize all of the hard parts, all the parts that can be operationalized so that the real experts can get their work and do something with it, like in the case of the Pfizer vaccine. Let’s get into production, let’s save human lives. Because I think that’s a separate different problem, which is that many experts actually are not putting their code into production enough because they’re getting too precious about things and that instead of building one model a month, why are you not building 100 a day?. So it’s not that we’re trying to replace you or saying you’re not valuable. It’s that why don’t you just amplify your own work at 100 or 1,000 times?
Jon Krohn: 01:05:56
Right. That makes a lot of sense. So, as another example that I wanted to circle back to a long time ago, and I think maybe now is the right point. So we were talking about how the National Institutes of Health, the NIH, which is the largest federal funding body in the United States for scientific research, they are funding lots of different labs that set up their own compute clusters, their own data centers, build their own models. So something that has happened a little because of the COVID pandemic and probably could have happened a lot more is sharing of data, say between hospital networks, research institutes.
Jon Krohn: 01:06:38
So for example, Sam Hinton, who was in episode 461, he was living in Australia working on a COVID project where almost all of his data, because Australia didn’t actually have that many COVID cases, almost all of his data came from the US and Europe and that kind of sharing might not have happened if there wasn’t a pandemic. So I guess that’s an example of how we can operationalize aspects that allow … Operationalize data operations, automate data operations. You know where I’m going with this.
Noah Gift: 01:07:14
Yeah. Basically in a nutshell, I would say that it’s, again, almost like what’s happening with politics where you were alluding to this, I’m, I’m reading the tea leaves here is that basically, there are good points that Democrats have. There’s good points Republicans have. There’s good points politically on both sides, and that if you only take the most extreme aspect of what someone’s saying and use that as the argument, you’re not using critical thinking.
Noah Gift: 01:07:43
So the same thing goes with operationalizing machine learning is that, in fact there are both very good approaches. Like of course self-driving cars are incredibly complex problem that actually, I would think will be solved and it’s great that people are working on it, but that doesn’t mean that just because there’s all these other things you can do to automate aspects of machine learning or DevOps, automated testing, automated deployments monitoring, scripting, they’re both equally valid points that should be combined together so that we have more solutions.
Jon Krohn: 01:08:28
Beautiful. So I think that is a great point. So it’s kind of an overarching summary of what we’re discussing today. You can save the world with machine learning and machine learning operations, and you … Going back to our early partner conversation, you don’t need to stop everything you’re doing right now, go back to school and pursue a degree full time. You can do this piecemeal. You can bit by bit, chip away in your evenings and weekends solving interesting problems and it sounds like your forthcoming book, Practical MLOps with Alfredo Deza is a great starting point for people to make a huge impact with the skills that they already have. I’m really looking forward to that book being out. I also wanted to quickly mention you also have something that came out recently, your Coursera course, Building Cloud Computing Solutions at Scale. Do you want to fill us in for a couple minutes about that as well?
Noah Gift: 01:09:23
Yeah. So I spent the last couple years actually working on this course because I’ve taught cloud computing now three times at Duke. These are brilliant students and many of them have gone on to big companies and biotech companies. So I took a lot of the feedback from people asking about this or that and I put it into this course. In fact, I think it’s like what you just mentioned, it’s probably the perfect kind of course I would recommend to lead a foundation for operationalizing things, which if you later then want to get into building stuff that probably you do, Jon, like getting super into the intricacies of deep learning, it’s, you now have a foundation for it.
Noah Gift: 01:10:09
So just like you could make the case for linear algebra and calculus being a foundation for doing machine learning and deep learning, I would also make the point. If you want to get yourself into production, you do need to know cloud computing in particular. It’s really is where else is the data going to go. It’s going to be in the cloud. So that’s really the idea is that I think that I want to help people get the foundation and it’s exact same information really that we’re covering at Duke in their data center.
Jon Krohn: 01:10:43
I love that, and you summarized that perfectly. So I think what you’re alluding to there is that I have something I’ve been working on for the last year and it’s this machine learning foundation series where I focus on linear algebra, calculus, probability, statistics, algorithms, data structures, these kinds of topics that you would study in a lot of these … If you did an undergrad or a master’s in machine learning today and you were really getting into the math, the low level detail with these things, these are the kinds of prerequisite subjects that you would study first and your way of going at it, where you learn some practical things first, I actually think that’s the best way to go.
Jon Krohn: 01:11:21
Where you’re on the job, doing things practically, because then you have a sense of, wow, look at these amazing things I could do. Now, what if I understood it a little better and I could dig in here a bit more. So it’s interesting. I call my curriculum Machine Learning Foundations, but it’s interesting having a foundation in how things work in the cloud and how you can be using MLOps to be doing state-of-the-art things without necessarily knowing the linear algebra underneath. I think that that’s even a better way to approach this.
Noah Gift: 01:11:51
We’ve talked actually about this at Duke with some of the people there that this is my opinion. I don’t want to speak for Duke, but my opinion is what you just said is similar is that I think it’s better to go through the motions and start to feel like what it’s like to build something. Then if you can get the feedback loop and mentality down, which is really what cloud computing does, then you’ll hit some rough spots you’re like, “Hmm, I wish I really understood calculus more or I wish I understood SQL more or I wish I understood algorithms more.”
Noah Gift: 01:12:27
I think that’s actually the right time, versus if the first thing you do is algorithm SQL. It’s like, you don’t even know what you don’t know yet, but it’s good to first understand, do I know anything? Then realize actually I know some things, but I would like to know others more deeply. So if I was designing a master’s degree in data science from scratch, I would … And let’s say it was a three-year master’s degree. The first year would be like pottery class. Like we’re making pots and we’re making stuff. Then what would happen is in the summer, a lot of people would go, “Boy, I really need to know machine learning. I’m pretty bad at it.” Then it’s like, okay, well, let’s go and get deep into the stuff because now you know what you don’t know.
Jon Krohn: 01:13:14
I love it. That makes perfect sense to me. All right. So we’ve come a long way in this podcast episode. I feel like we could go on for hours and hours and hours. We’re going to have to have you on the show again, soon, Noah, but this has been amazing. I love everything we’ve covered from educational options through to MLOps and tying those themes together right at the end here. So thank you so much for being on the show and I’m very much looking forward to the next time.
Noah Gift: 01:13:39
Sounds good. Happy to be here.
Jon Krohn: 01:13:47
I really could have continued that conversation with Noah for hours and hours. We only scratched the surface of his formidable wisdom and technical knowledge. In today’s episode, we covered the pros and cons of various data science education options, including more expensive university programs versus more economical ones and formal degrees versus practical online platforms. The consensus seemed to be that on-the-job experience is the most valuable education of all. Separately, we detailed a number of examples of how high level abstractions of machine learning like AutoML enable data flows to be intelligently operationalized. This allows junior and senior data scientists alike to scale their impact with relative ease, allowing them to tackle the most prominent practical issues facing modern society like cancer, food insecurity, and climate change. Noah outlined his colored money guide to increasing your career autonomy. That is by shifting your income away from red salaried income, toward yellow consulting or green passive income, wherever opportunities permit.
Jon Krohn: 01:14:58
What an awesome set of topics. As always, you can get all the show notes, including the transcript for this episode, the video recording, any materials mentioned on the show and the URLs for Noah’s LinkedIn profile, as well as my own LinkedIn and Twitter details at www.superdatascience.com/467. That’s www.superdatascience.com/467. I’m always happy to connect with listeners. So please do connect and feel free to tag me in posts with your thoughts on the episode. Your feedback is huge for figuring out what topics we cover on the show. Since this podcast is free, if you’d like a hugely helpful way to show your support for my work, then I’d be very grateful, indeed, if you’ve made your way to the Data Community Content Creator Awards nomination form. The link is in the show notes.
Jon Krohn: 01:15:46
Obviously we’d hope you could nominate the SuperDataScience podcast for category seven, the podcast or talk show category. I’d love my name, Jon Krohn nominated for category eight, the textbook category for my book, Deep Learning Illustrated and finally, I’d also love my name again, Jon Krohn nominated for category two, the Machine Learning and AI YouTube category for my YouTube channel, which contains tons of free videos on deep learning, linear algebra applications and machine learning libraries. The Data Community Content Creator Awards themselves are coming up on June 22nd, and I hope to see you there. All right, thanks to Ivana, Jaime, Mario and JP on the SuperDataScience team for managing and producing another amazing episode today. Keep on rocking it out there, folks, and I’m looking forward to enjoying another round of the SuperDataScience podcast with you very soon.
