In this episode of Data + Curiosity, I had the opportunity to chat with Randy Au, about how he came to count things for a career, why he stresses out about data quality, and how he got started in his latest hobbies.
I learned so much from Randy, and am excited to share the conversation with you:
You can also read a lightly edited transcript of our conversation below:
JESSE MOSTIPAK: I always associate you with the term counting things is hard.
RANDY AU: Yes. I don't know why I say that a fair amount.
JESSE: But I think it's true. I think about this a lot lately in animation because I've been working in Maya. And I find myself counting keyframes all the time. I'll be like, OK, I'm on frame 48. And I know that I want to add six frames. And then my brain will be like, what's six plus 48? And I'm like, oh my. I get stuck.
But I'm constantly counting frames. So I'm constantly reminding myself that counting is hard.
What are some things that you've discovered that make counting hard?
RANDY: Oh, god, probably the thing is that I always say that because I always wind up saying-- people ask me what I do. And I always joke that I just count things for a living. That has been the one constant in my career for 15 plus years now is just counting, finding the right thing to count, right?
If you can figure out the right thing to count, you don't need the fancy stats. You don't need a lot of fancy anything if you can magically just reach into people's heads and count the right thing, right? Like for example, how many people happen to love this obscure sandwich? I don't know. I'm just making something up here.
If you can figure out the right thing to count, you don't need the fancy stats.
Well, like no one's-- think about how you would do that in the real world, like surveys and interviews and sampling and all this complex stuff. But if you had a magic wand, you could just reach into every person's brain and just ask them, do you like this sandwich? And you just got the true answer out of them, right?
You wouldn't need any of that fancy stuff. You literally just count everyone, get the percent, and be done, right? And so obviously, you can't do that. No one has that power. So that's why. But like if you could, right?
So that's why counting is hard. Most of my job is just figuring out how to slice things enough, so that I can use my limited ability to use statistics and data science and whatever to get at the answer. I just spend my life cheating doing that.
JESSE: I don't think that's cheating. I feel like this is coming up. And now I'm going to-- I am not even going to try to attribute this to someone because maybe it's someone else who said the core of data science is counting things.
So what kind of things do you count in your work or have you counted in previous jobs?
RANDY: A lot of times, it's just people with certain attributes, like a very common theme in data science, right? Like people who want to give you money, every business person wants that number. Obviously very complicated in how you define that.
Or people happy with a certain feature, people-- back when my first job was at an interior design company. And they hired me as a consultant. Very strange. Like a small consultancy in New York City. They did space planning for banks.
And they did the-- they literally wrote the Google master plan. As if you build a Google building, here are all the rules for where the maximum number of feet you could be from a micro kitchen. They wrote that book.
And I just joined randomly as a consultant helping them. They were like, we need a data person. And somehow I convinced them that they should hire me. I don't know how. And I was just doing survey work, analyzing data, automating things in Python for them.
And it was just figuring out how, looking at surveys, rows and rows of students at a university. And you start seeing interesting patterns about how students were piling books up next to them in the stacks to get privacy, to get at the table. They were studying. This was like some large Ivy League college. And they just were stacking books.
And you can just like eventually just tease out the fact that a lot of these students just needed quiet places to study. And they weren't getting it. And so they were just figuring it out.
And so I was just helping them to quantify that. It's like, well, you see a lot of these, a surprising number for the amount of responses that we're getting. And this has to lead to something. And of course, that goes into how they design the library. Or should they put more desks or more silent space working or more meeting rooms?
And so that's where I got my start. Then I got thrown into ads, which is all about counting interesting, sketchy things, like fraud. Because I was in a third tier ad place I got to see all sorts of fraud get sent around, saw all sorts of shenanigans going on in that industry when you did scratch the surface of it.
And then went to MeetUp. And we were counting actual users not fraud, which is a change, right? And just like, did they like doing this, do they like going to events? That kind of interesting stuff. And helping product work. So that's just what I've been doing, just counting very interesting things.
JESSE: But it is interesting.
What prepared you for this career in counting things? How did you become a career counter?
RANDY: I don't know. As in, look, my background, my undergrad was I did a couple of degrees. I had philosophy, so continental philosophy, not like the American. If you go to-- there are branches of philosophy in the grand scheme of things, right? In America, you usually do a philosophy degree.
It's like the majority of it-- I don't know the percent any more-- is what they call– it's essentially logic, right? It's like logic and the logicians of like Bertrand Russell and all that stuff about the counting things. Whereas I did continental, which is the European tradition. You read the great philosophers and then critique them and understand them. So it was like Kant. And Aristotle was in there, Hume, all sorts of stuff.
And so somehow I went there. But then on top of that, I had a business undergrad. Don't even know why. But I wound up doing operations research there. So that was like networks and decision support systems and just some basic accounting.
And all of this stuff is totally unrelated. But it becomes useful later on when I'm working with everyone in a company. And suddenly, hey, I know accounting. I can talk to the finance guy. Hey, I know marketing. I can talk to those guys and understand what they're up to.
And all of this stuff is totally unrelated. But it becomes useful later on when I'm working with everyone in a company.
And then the operations guys were useful when I was in ecommerce. And they were making kids clothes. And we actually had shipments going around the globe for cloth and dyes and factories. And suddenly that became useful.
So I just stumbled upon things. Then I go get a master's in communications, human communications, social science. Totally unrelated. I wanted to go into information science.
The chair was a chair of both departments. I couldn't get into the info side because it was a small program. But they're like, hey, if you want, you can go over here and do comm. And I'm like, all right, fine. I'll do comm. So it makes no sense.
JESSE: But it kind of does because it does help you, right? Like this has got-- these are an intersection of really interesting things.
RANDY: I mean, it helped. I totally-- like social science was a-- or I should say a lot of really good data scientists come out of social science programs. Political science, neuroscience, those people-- the thing about social science that I kind of joke about is that they've got a complex literally about being a science, right?
They're like, if we're measuring stuff in people's heads, to what extent is that a real thing or some fake thing that we made up? That is a thing that they are very concerned about and think a lot about. And so they have a lot of rigor and philosophy of science classes as a first-year graduate student to just kind of drill into your head that, yes, we are an actual freaking science. We measure stuff, scientific method, you know?
And we're not just making theories up like the philosophers of old. And that kind of self-doubt, that whole awareness that you are measuring something that may or may not even be real and here's ways that we're dealing with that is really useful in data science.
JESSE: That's such an interesting point. So I don't ever talk about this, but I didn't finish my PhD. But my PhD was in immunology and infectious diseases.
But I was particularly focused on prion diseases of the nervous system, so I did a bunch of neuroscience. And you're reminding me I took a philosophy of science class, right?
We read Objectivity. And I remember reading this book. And it was like, what is the more accurate picture of a flower? Is it an exact drawing of the flower you see? Or is it a representation of all flowers of the same species? And I was like, well, now I don't know what to believe.
RANDY: Welcome to epistemology. Yay. Yep.
JESSE: Even in my dissertation project, I was picking up a small grant that someone in the lab before me had done. And they had gotten positive results. And I did this experiment for 18 months over and over and over again. It took three to four weeks to run. And I kept getting negative results.
And my PI was like, your funding depends on positive results. And I was like, but I'm not-- they're negative. They're negative. I can't. I don't know what to tell you.
And he was like, well, maybe you don't know how to do a Western blot. And he would make people sit with me to do a Western blot. And they would be like, she's doing it right.
And it was kind of that idea of we got positive results, but then I couldn't replicate. And so then you get into reproducibility, right? So if you measure something and then you can't repeat that measurement, like where is the validity?
RANDY: Yep. So all of that. In data science, we don't even do the validity studies.
JESSE: I think some data scientists do.
RANDY: Yeah. Some have the time to do it, right? It's mostly that.
JESSE: That's true. So OK, so you've got these two undergrad degrees. You've got a master's degree. Are there more degrees in there?
RANDY: Luckily, no. After two years, I was like, you know what? Research fun. Publishing not fun. And I just bailed out of there. And I just went straight into industry, which worked out.
JESSE: So where did you-- so you worked. I think your title now is quantitative UX researcher. But you are in the data science world.
Where did you pick up programming?
RANDY: So I started my first programming-- well, high school, we had a joke programming class where we learned C or C++. But back then, and this was like the very late '90s, C++ was weird. Back then it was essentially C with classes. And that way-- they taught it as C with classes.
C++ didn't become the standard library and all that fancy jazz until a little bit later. So I wasn't on that. So it was just C. But you could do templates and classes. It was very weird. So I learned that. I didn't know what to do with it. But it's like, all right. Cool.
And then I like learned, in my decision support class in college, I learned VBA and Excel and did a lot of horrible things with it, which is really funny. I also learned Python at some point. I don't even know where.
So I was a bit of a nerd back then already, like playing with Linux and stuff. But anyways, so I learned VBA. And then for a research project in undergrad my professor was like, hey, I need you to do this network routing thing and just implement it.
And I was like, well, I know VBA. I don't have time to do this. I'm going to build a freaking linked list in VBA with literally just having linked lists and columns in Excel and then the cells just moving things around because I didn't have time to do it the hard way. I knew the algorithm. I just didn't have time for the language for it. So I did it that way. It worked. It was terrifying.
JESSE: You can do, I mean, unholy things with Excel. It is horrifying and incredible and beautiful all in one.
RANDY: Yeah. And then in my first job, I took that even further. And I used Python to automate Excel to make thousands of slides in PowerPoint, save thousands of dollars in consultant time. Because I was just mass generating these templated presentations out. It was terrifying and amazing all at once.
I used Python to automate Excel to make thousands of slides in PowerPoint, save thousands of dollars in consultant time.
JESSE: I think the true power of Python is really kind of incredible, the things that you can-- I mean, many programming languages have that. But with Python, you can do a lot, like a lot with Python.
RANDY: Enough to be horribly dangerous. And it was just like, yeah, this is throwaway code anyway. So it's OK. But now I have fun stories to tell.
JESSE: I mean, that's like half of it, right? I think that was something when I first started out in data science, I got really hung up on doing things the right way instead of just getting them done. And so just wanting like, oh, if everybody's using this database and this programming language, like how do I do it like the pros instead of just delivering value. So I would have had more fun stories, I think, if I just focused on delivering value.
JESSE: So you talk a lot-- well, you mention in your bio, I should say, that you stress about data quality.
Why is data quality important? Why should anyone care about data quality?
RANDY: Oh. It goes back to the counting is hard thing. Because if you don't know what has been counted or, specifically, if you don't know what has not been counted, you can do-- it gets extremely dangerous, right, about what you're actually able to say about this data. Imagine you were doing neuroscience things. And you're like, Yes, I've counted every instance of someone having this disease. And you just did not count the rest of the population that did not have a disease, right?
You get-- the stats get all wonky when you do it that way. And so knowing that-- and then this is true for any data set. Like a lot of data science, the boot camp and all that, right there, like here is dataset. Just go with it.
And I'm like, what? But where did this data come from? Who made this data?
But where did this data come from? Who made this data?
There's all these horrible questions that when you start scratching the surface of, there's no end to it. And you have to-- so you have to balance. See, well, am I willing to engage with this data set and the quality of it. I need to really understand what's going on. And that's going to take me weeks. For some big data sets, like the census, it'll take you and your entire career to get it. Right? And I have to deliver something. So I worry about that a lot.
JESSE: That makes sense. So earlier, I talked with Alison Horst and Alison Hill about the Palmer penguins data set and what they liked about it. And part of it, so it was someone's dissertation research. It is a complete data set.
And you know how it was collected. They talked to the researcher. And there were some missing values. But they knew. Like you could get the ins and outs of this entire data set, which I think does make it really nice.
But even now that you mentioned the census, you're right. That's an entire career. And sometimes we hand intro to data science students census data, and say have fun, without any kind of guidelines on what they should or shouldn't be doing.
RANDY: Oh, yeah. The census one is especially dangerous. Because if you ever listen-- like, because I don't know the census data very well. But I pay attention when people start talking about it, like the researchers you talk about. And they're like citing like, yeah, the ACS is done every X, Y, Z years. And this is how it's done.
And then the sampling change on this year, like they know their stuff. And I'm like, I cannot. I'm not even going to touch this because I can't even come close to whatever it is that they're doing, right?
JESSE: When we're thinking about data quality and census data or whatever data, how-- I guess it would help if we think about this in terms of audience. Let's just say like your average practicing data scientist, right? So they've gone through whatever bootcamp or course or self-taught. They are in the workplace doing their data science.
What level of granularity should one be thinking about data quality? Is it down to every single individual value?
RANDY: Well, it always depends. But I would start with every individual data point, right, every row. It's just there enough. You can have all sorts of questions. Like why is there a row here? And why isn't there a row here?
Just those two questions alone is like a good week of work, just understanding that because you're like you start going into the tech stack, right? Just, why is this row here? What triggered this row? Is it a frontend operation? Did the browser put the row there? Or is it the backend operation?
Because there's a loss either way. Because if it's the browser, 50 million things can go wrong in the browser. And therefore, no row ever shows up. You have to mentally account for that.
If it's the backend, different reasons can happen, where a row will or will not show up. Is there double counting? Why is there double counting? All sorts of things.
I had a fun-- one query I love running every so often is: I go to a database that's supposed to have unique IDs. And I just check if that's true. I've had a couple of CTOs swear at me when I show them that that is not true. Really fun, especially one that was a revenue database.
One query I love running every so often is: I go to a database that's supposed to have unique IDs. And I just check if that's true.
That was like it was the ad click database. And there were just a few duplicates in there. And so he started cursing and ran back to this station to go talk to the programming people. So that was fun.
Those kinds of assumptions, when they're not hard enforced, they're just kind of like supposed to be there, can be really fun and to check on. So that's a basic integrity thing just with data. And then you can always go up a level or down level to individual fields or to why does this table exist in the way that it does, right? But those are questions you can wait on because they're implementation details at that point.
JESSE: I love that idea of starting with a row. But I also really appreciate that you said that understanding one row of data is about a week's worth of work.
RANDY: If you're lucky.
JESSE: I mean, if you're lucky, right? It could be a really, really long row. You could have hundreds of thousands of fields. But I think what that really gets to is this idea that data quality isn't like, OK, I've spent 20 minutes. I've checked all the boxes.
RANDY: Oh, no.
JESSE: This is done. It's great. Because that could be weeks to months. I mean, depending on your data, you might need it. You might need to take the months to make sure it's in the right format and ready to use.
RANDY: I mean, imagine just one row of medical study data. That one row is months of work for multiple people in the medical field. Like a doctor or something, a nurse, had to check on that row who knows the number of times. And there's procedures. The procedures are terrifyingly long. And just understanding that, right? And you have 1,000 rows of these things, which is a very expensive study already. Just that alone can be half your PhD, right?
JESSE: Yeah. So those PhD projects, they're out there. They're just waiting for people, all of this data.
RANDY: There are just questions everywhere, right?
JESSE: I mean, I think that's part of what makes data science such an appealing field is if you are-- even you don't have to be curious about everything all the time. But if you are even just a little bit curious about stuff and good at-- not even good, but like asking questions, there is an endless source of material available.
RANDY: Yeah. It's really good for people like me who love playing with domain knowledge. I love domains, right? I love it if someone's got a weird, obscure domain that they're expert on and they just know stuff about and they're making data about it, I listen to them for hours about just how they're doing whatever it is that they're doing, right?
And so each row of data is the reflection. It's the shadow of a domain. And the more you understand it, the more you understand what's going on.
And so each row of data is the reflection. It's the shadow of a domain.
JESSE: Yeah. So this is the most brilliant segue that you have set up here with all of these obscure bits of domain knowledge. So you have a ton of hobbies, right? And you have, I think, they've come and gone over the years. But I think one thing that you and I have in common is that you don't just have a hobby. You go all in on your hobbies.
JESSE: So you really develop a sense of mastery on a hobby, right?
And so tell me about, what are some of your current domain explorations in terms of hobbies?
RANDY: Probably the most fun is like the gem cutting thing.
JESSE: Look at that. Yeah.
RANDY: I am holding this is a piece of synthetic ruby that you can buy. This is like maybe 50 bucks of ruby. It is like 1,000. I forget how heavy this is, like maybe a couple of grams or something like, 1,000 carats or something like that.
You can cut this into-- so synthetic ruby is really cheap. It's the cheapest synthetic material you can buy. But it's really shiny. It's fun. And it's just like you buy a bag of these for whatever 100 bucks or something. And then you can turn them into like interesting gems that are-- here's one that I can just grab off the shelf, even though this was one of my first ones I made. So it's not that great.
And it's just like this shiny, sparkly thing. And it's not focusing very well, right? And so you can cut this into a gem. And then if I-- and then it's like, well, now what do I do now that I have a bunch of these shiny objects on my desk? My daughter picks them up. She's going to get spoiled because she's going to be surrounded by just gemstones.
And then I was like, well, now I have these. So now I need to do something with them. Therefore, I'm like on the side of my I need to go learn how to work with silver, so that I can make mountings, so I can make something useful with these things. Otherwise, I'm just– I have paperweights.
So these things snowball. I always find excuses to find a second hobby off of a better hobby and justify it somehow. And it just keeps going.
So what got you interested in gem cutting?
JESSE: I am genuinely curious. Were you just one day like, I should cut gems? Or was there a process to it?
RANDY: Well, what hap-- I don't know how. But somehow, I fell into a rabbit hole on YouTube where I found a gem cutter thing. Maybe I was curious about it or something at some point.
And it's just like maybe I woke up one day. I'm like, how is this done? Just how is it done? I was just curious. And I'm like, oh, well, there's a device that you glue. This is like you glue a stone onto a brass stop, right?
And then you just hold. You have a device that holds it against a diamond plate that spins. And you just kind of-- it holds it at a fixed angle. And you can rotate it at fixed angles. And that's how you cut facets.
And it was like I'm staring at this watching it for a couple months or something, you know how YouTube works. And then eventually, I'm like, this doesn't look very hard. What-- can I do this myself? And I look into that. I'm like, well, a machine's not that expensive.
Like a good machine's only-- well, I shouldn't say only. It is a few thousand dollars. But I'm like, well, it's a pandemic. And I have a little bit of disposable income. My wife will not be too angry at me if I do this because I'm only going to do this once. All right.
And so it happened. And then I bought a machine, ordered it from Sri Lanka, took a while to get here. months. Buy some other stuff, blah. Blah, blah. And then suddenly, here I am with gem cutting.
JESSE: I mean, that's incredible. It's kind of like living out an MMO career, kind of, right? Like you're jewel crafting. So is silver-- what is it? Silversmithing, what is that called? Like is that up next?
RANDY: Yep. It's silversmithing. So silversmithing, you can go online and look up. And it was like, well, there are jewelry classes all over the place, right? I'm like, OK.
Then you look at the video. And like, that does not seem extremely hard. Now, it's not easy. But it's like, well, metal, bits, fire. I'm like, I can solder electronics. Soldering silver can't be all that much harder, right?
Spoiler. It's slightly harder because there's like actual torches involved and things. But it's still not that much harder. And so all right, I'll give this a shot. I'm very bad at it. I haven't had time to practice. But it's like, all right, I'm going to keep going.
I also bought a 3D printer that I haven't figured out how to-- I haven't set up yet. So I can get castings done. So I can do CAD much better. I can just CAD up the complicated things that hold these gems and then get a cast, which is probably a safer bet. So yes.
JESSE: I mean, that's amazing. And so how long did it take you to go? I feel like I followed a lot of this on Twitter. It feels like you went from, I will learn how to cut gems. This may have some challenges to it that I wasn't anticipating. To, I mean, you're-- I don't know what the levels are. But like I would-- at least proficient, if not like well on your way to mastery.
RANDY: It's like, one, gem cutting's surprisingly not that difficult. Because all the stuff is holding it for you. It's not like you were-- there are people who are at where professional gem cutters are done like the factory type jobs, and they're usually in fairly poor countries. And they've got very low tech stuff. And it is just all muscle skill.
They're just holding gems on these very basic things that don't have gears. They don't– the angles are set purely by muscle memory, essentially. And they just bang them out, right?
So it's like with all the mechanical aids, it's not that difficult to do it, right? It's just mostly patience and practice and a couple hundred hours or so. You can probably get a decent looking gem. Like your first try, you can probably do it with someone helping you, right?
So it's like, OK, well, now that I've done, become competent at this. And now I'm like, I want to do something else to complement it and just make it more complex. Because I'm apparently just insane.
JESSE: Well, it's just like leveling up. You've got a skill. And you're leveling it up. Do you-- so one, how do you find the time? Just asking for me.
RANDY: I trade off time from other things. So I have to rotate my hobbies, for example. Like woodworking is totally out of the question now because wood is just way too expensive. I don't have room for the thing.
If I build a table, I don't have room to put another table in my house, right? So forget it. So that's been like shoved down the priority scale. But I have all the sharpening tools and all the whatever things. And I have that.
And then I just try to, well, I have this stuff already. What can I do with the next-- how can I reuse it to do the next thing, right? And then I just keep doing that. So there's that.
I also probably don't sleep nearly as much as I should be doing. So there's also that. So I don't know. Somehow I can-- I seem to be able to manage. I don't know how.
JESSE: It's not like you're spending eight hours a day every day working on gem cutting type stuff.
RANDY: It's like a couple hours here or there. A single gem will take me weeks because it's an hour here or two hours there. If you add it all up, it's maybe a day of work. And that's because I'm not good at it. A professional who does this for a living would be able to bang one out in like a couple of hours, right?
So you know, so I'm making up for it with my beginner skills with just time. But I can afford that because it's a hobby. I'm not-- I don't need to be fast. I just have to be good enough to satisfy my own whatever sense of goodness is.
I don't need to be fast. I just have to be good enough to satisfy my own whatever sense of goodness is.
JESSE: Yeah. I love that. I love this idea. You've talked about practice and patience, which I think is key. But also not-- your hobby doesn't have to become your career, right?
You don't need to go forth. I mean, you could eventually become a jewel maker. I'm not sure what the correct term is. But you could. But that would take more time. But there is like a level that you have where you're looking for mastery.
So do you ever have any crossover between something in your hobbies that you've learned or thought about or experienced that crosses over with your work or vice versa?
RANDY: Sometimes. For example, other universe of hobbies, I used to translate essentially for money professionally, right? I was a Japanese to English translator for games. And so I worked with a game dev company. So I know that space a fair amount.
And so it's like, well, so translation, it's like working with other languages, especially working with Japanese because they don't speak English very well, I use that just sometimes professionally. Just like, hey, I can read this. Hey, I can talk to them.
One time, as a UX researcher, I flew to Japan for the Google Next conference because they wanted to do research. And I'm like, no one there will speak English. You have to speak Japanese to have any chance at communicating with these people. And so I just went over there and got permission and just started interviewing people at the conference just what their experiences were.
So that was really, really fun. I also got a trip to Japan out of it. So that was great. So that was an example of just me bringing stuff over. But it's also like, hey, working in game development, I also like project management becomes part of it.
Shipping, I am like I've probably written a couple of blog posts about shipping and logistics because I watched people do that, right? To make goods, make boxed games is this terrifying process with artists, vendors, people assembling the boxes, then shipping it out, and all the horrors involved in shipping, right? And so all that like I just find use for it. Because that knowledge will be useful for some other place.
Randy’s blog posts on shipping
- Logistics: a story of a stack of stickers
- The Complexity of Makin’ Goods
- Reflecting on how system complexity grows
JESSE: Yeah. Absolutely. Well, you have a shipping conundrum with your 100 data counting is hard stickers.
RANDY: Oh, yeah, those counting is hard data stickers that I have. I have these. It's like they're sitting in a bag on my desk because I'm like I want to ship them out. And it costs too much money to ship them out for profit. So I'm just going to need an excuse to just give them away.
JESSE: Just giveaway stickers for everyone. Yeah, shipping is hard. It is surprising. So in my adventures into animation, did you know that it takes like almost a year to produce a 22-minute episode of a cartoon like Bob's Burgers?
RANDY: Oh, wow, a year.
JESSE: A year. I was like, I don't know. Eight weeks tops. Like that's got to be so easy. It is close to a year with all of the hands that go into that.
RANDY: That's kind of crazy. There's this amazing anime, Shirobako. Have you heard of that one, right?
JESSE: I haven't. No.
RANDY: It's a couple years old now. But I have friends in the anime industry because of the game industry. And all of us, we watch it. And we remark on just how realistic it is. I'll link you to it later.
But it is about an animation studio in Japan doing anime, right? And it's about– it follows the story of a project manager there. And she's got spreadsheets with all the cuts, all the scene cuts, right? And she's like, well, this scene cut is from that animator. That one's from this animator. And she's driving around getting the frames because it's on paper and getting them and then bringing them here. And then the passes like bring it to the CG department, editing department, all that jazz.
And we're all just watching it. I'm like, this, it hurts to watch because we know exactly that this is how it is and how this is just made. And it's just terrifying. But it's also very fun.
JESSE: I'm super stoked for that link. Yeah. I mean, there's so much. We call them X sheets or dope sheets with the timing. I was listening to a podcast where this guy was like, my job is basically to be like you have this arm swing at 17 frames. But it's funnier at 12 frames.
And like just thinking about the kind of knowledge you have to internalize. Then I'll put one on the screen of what it looks like. But it is kind of almost like a weird data visualization. And to be able to look at that and be like, oh, yeah, it's not funny at 17. But if you take out five frames at 24 frames a second, that's going to make it funny. And I just, OK.
RANDY: It's like, OK. I don't know. I'm just amazed that someone knows that. I love that someone knows that.
JESSE: It's just like all these obscure details and things that go into it. I have a mentor. And she works at Pixar. And so she sometimes gives us the inside scoop on weird animation things she's had to do or how she had to work on something and how hard it was and the tricks she had to do to get it to work. And I was like, oh, so you mean it's not like 360 degrees beautiful in animation. And she was like, no. No.
RANDY: Oh, yeah. I've seen those where they have to ruin the proportions to get the shot right because it doesn't look realistic otherwise.
JESSE: Yeah. So she was on the Trolls movie. And one of them is the two trolls have to give each other a hug. But their arms are actually like this big.
RANDY: Oh, right. So they have to stretch.
JESSE: Yeah. So they had to get the right angle. And like the arms were stretched all the way out. But it also has to look realistic. And I was like, that's really hard. Like that's really hard. So it's amazing what goes into creating things and not just creating things but then getting them to people, right? Like there's the ideation, the creation, but then distribution. And distribution is whew.
RANDY: Oh, yeah.
JESSE: That's for someone else.
RANDY: Oh, yeah.
JESSE: That's not for me. But you do distribute something. You have a newsletter that you not-- you are better than all of us. In that, you write regularly. What motivates you to keep writing?
RANDY: Almost 95% of it is because I know that if I drop one, it's going to-- the drop is going to continue. And it's been one of those like don't drop it, don't drop it, don't drop it kind of things. And to my surprise and to my wife's surprise, I've gone over two years without dropping a single one.
JESSE: I can't. I don't-- I'm sure other people keep to writing schedules as well. It is so-- this is me projecting my own biases. It is so hard for me to tweet every day. Like I can't even imagine writing. You write weekly, right?
JESSE: Yours is weekly, or is it twice a week?
RANDY: It's every Tuesday, I get a post out. There's the newer paid ones that are shorter. And that's every two weeks. But if I-- though, I don't hold myself to that one. If it slips, that one's OK to slip.
But the every Tuesday one has been going for over two years. And my wife, every time, is like, how do you have so many things to say? And I'm like, I don't know. I'm desperately thinking of an idea. Like every Friday night, I'm like, I need a topic. And I'm scanning Twitter or just looking for inspiration either at work or whatever.
And I can somehow manage to find something. I mean, some weeks are always a little weaker than other weeks. But I've managed to do it. And I don't know how. I have no idea how.
JESSE: That's incredible. Although, these days, scanning Twitter, I feel, gives you lots of material.
RANDY: It's lots of it. I'm like, do I really want to talk about the dumpster fire again like everyone else, right? So there's been a very deliberate I'm not going to talk about the thing everyone's already talking about. And so I'm looking the other way. But that's a very deliberate choice.
JESSE: Well, you did a great post on data friend connections and Twitter. And you had some great advice in there. So we know each other through Twitter. We are both part of data Twitter. And where do we go now? Where are we going to hang out and talk about data?
RANDY: Yeah. That's the thing, right? Like data Twitter has been this constant in many of our lives. I literally owe one of my jobs to data Twitter. I got laid off somewhere for whatever reason.
And then just someone from data Twitter is like, hey, you should interview at this place. They're hiring. And I got a job there. I just– that was magic, right? And that has been true for other places.
I mean, I've given all sorts of resume advice and those sorts of things, and job postings to fly around that place. And so I don't want to see it go. And so I've been kind of banging on the pots and pans saying, hey, guys, if this thing goes on fire, which it seems to be doing right now. Here's some backups just in case, right?
Like you can call me crazy if it doesn't burn down and we're all fine next year. But just in case, here's some backups that I've been just sharing Google Sheets because we're data people, obviously, it's going to be in a spreadsheet right?
And luckily, like we've been moved to Mastodon. Or a lot of us have as a backup. And there's Discord and Slack. So we will probably be fine in the near term. But I don't know about the medium term.
JESSE: I think that's great advice. You have put out great resources about just staying connected. And I think you made this great point about how we may not be able to connect to everybody that you're connected to now. But if you connect to a critical number, you will find everybody and find new people.
RANDY: Yep. It's basic network theory, right? You just need to get to one or two of the hubs. Yeah. I'm dragging up like weird things in my past like studies again. But it's like--
JESSE: This is perfect. I love it.
RANDY: There's some hubs, right? Like if you connect to like, for example, Vicki Boykis or Chris Albon or something, right? They, through their shit posting, will bring you to the rest of the community, right?
So you don't have to grab everyone. But you do need to connect to some of them. Or else, it is extremely hard if it totally all disappears on you and you're just lost, right? So having just a little bit is a good backup.
JESSE: Yeah. I love that. So where on the web is a good place to find you? You can list as many as you would like or just one.
RANDY: My newsletter is probably the easiest one, counting.subtack.com.So obviously, the whole counting theme, I just like went with it at some point. So that one.
Otherwise, RandyAu.com is where I post my links to myself. And that's really-- I mean my mastodon thing, it's so awkward to say, right? All the Mastodon domains are just awkward to say, right?
JESSE: Mm-hmm. Mastodon is a weird word, right? It's MAS-- like it's a cool animal. It's a very cool animal. All in on the mastodon. It is weird to say.
RANDY: Yes. Yeah. It's not– it doesn't roll off the tongue, right?
RANDY: And also like just the domain names and all that, it's like there's no way I can pronounce this thing in a way that someone will type it into a keyboard and have it go to the right place. So it's like, I'll link you to it. And then we'll deal with it later.
Thanks for tuning in to our latest episode of Data + Curiosity! I’d love to hear your thoughts on this interview, how you’re enjoying Randy’s newsletter, Counting Stuff, and what you’re curious about! Let me know in the video comments – I can’t wait to hear from you!
Follow Randy online
Choosing the right horizontal scaling setup for high-traffic models
Horizontal scaling via replicas with load balancing is an important technique for handling high traffic to an ML model. Let’s examine three tips for understanding how to properly replicate your instances to save users time without wasting your money.