Transcript: He Wrote a Book in 30 Days

The transcript of How Do You Use ChatGPT? with Seth Stephens-Davidowitz is below for paying subscribers.

Timestamps

Introduction: 00:47
How ChatGPT is making data analysis fun: 11:29
AI is redefining who can be an artist: 16:15
Brainstorm creative ideas with ChatGPT: 20:35
How Seth uses ChatGPT to instantly generate charts: 28:27
Develop a sixth sense about when ChatGPT is wrong: 43:20
Figuring out which Olympic sport I’m statistically best suited for: 50:20
ChatGPT takes an approach that surprises Seth: 52:15
We have an answer! The Olympic sport I should try out for: 1:02:25
The power of getting quick answers to previously unanswerable questions: 1:03:47

Transcript

Dan Shipper (00:00:26)

Seth. Welcome to the show.

Seth Stephens-Davidowitz (00:00:48)

Thanks for having me, Dan.

Dan Shipper (00:00:52)

So, for people who don't know, you are an American data scientist, you're an economist, you're an author, you wrote the book, Everybody Lies. You also wrote another book called Don't Trust Your Gut. And most recently, you wrote a book called Who Makes the NBA? And you caught my attention because you wrote that book in 30 days. Oh, there it is. You wrote that book in 30 days with ChatGPT. And I was like, oh my god, I need to talk to Seth. So really, really excited to have you on the show.

Seth Stephens-Davidowitz (00:01:18)

Yeah. Thanks so much for having me, Dan. I'm a fan. I've listened to some of your podcasts and I'm a total agree-er with your main point that ChatGPT andAI are just going to transform everything.

Dan Shipper (00:01:28)

Thank you. I love it. So I'm curious, just get started, you wrote this book in 30 days. Where did the seed for that idea come from? How did you decide to try that?

Seth Stephens-Davidowitz (00:01:39)

Well, I just became obsessed with AI and particularly there's a tool, Code Interpreter. It used to be called Code Interpreter and now it's just called Data Analysis that does data analysis for you. And I started playing around with it and I'm just like, oh my god, this is the coolest thing I've ever seen in my life. Things that used to take me four months or take me four days or four hours you know, everything was just so much faster. And so I kind of was like, oh, wow. I could write a book really quickly. I just thought it'd be fun to have a countdown for myself. I don't think anybody else really necessarily cared about my countdown, but in my head, it was really a challenge. And so I spent 30 days. My girlfriend thinks that I have a tendency to exaggerate. So she's like, you need to point out you were doing some research a little bit beforehand, which I say in the book too. But it was a shockingly little period of time in the neighborhood of 30 days of full-time work, I would say, to write what I think is my proudest book yet. I'm so happy with how it turned out. So it was a kind of a wild process, kind of the future of creativity and everything.

Dan Shipper (00:02:48)

I loved it. I thought it was really good. I have not finished the entire thing, but I read a decent chunk of it. And you've got a little bit of a Bill Simmons vibe. It's like sort of Bill Simmons-y cause it's voicey and a little bloggy and funny, but more data and less pop culture-type stuff. Is that what you're going for?

Seth Stephens-Davidowitz (00:03:09)

Yeah, I got a message recently from a friend of mine. And he said, Seth, I was blown away by your book, Who Makes the NBA? And he goes, just so you know, I'm not lying. I did not feel that way about your previous two books.

Dan Shipper (00:03:22)

That's a good friend.

Seth Stephens-Davidowitz (00:03:25)

So, now I do believe you. He said, one of the things that blew me away is just so many books, including your previous books, have so much fluff—a long story of buildup, big words. And I think this book—in part because I self-published it, in part because it was a race against time, in part just because that's my theory of book writing—I just kind of limit it to facts, a couple dad jokes, and some graphs, just really the bare minimum. And I think people actually prefer that. I don't know, the whole book industry seems wrong to me in thinking that people want these long winds up in this story and descriptions of all the people and these big words, kind of just tell me the point. And I think this book kind of does that. So that's why my friend was blown away by this one and not my previous two books.

Dan Shipper (00:04:21)

That's really interesting. I would sort of love to get into the distinctions in your mind between the sort of self-publishing industry and the regular book publishing industry and all that kind of stuff. But maybe we can save that for a little later. I want to dive in first… It sounds like one of the main things that you use ChatGPT for is Advanced Data Analysis. And you said it sort of saved you a lot of time. If you had to think about how it's impacted your writing process, maybe for this particular book, is that the main place that you used it or where were the other places you used it?

Seth Stephens-Davidowitz (00:04:52)

You know, that was number one by far. There were some little ones. There was a point in the book— I ranked people on how good they were, basketball players, adjusting for their height. And I wanted to call the metric something. So I called it MUGGSIES after Muggsy Bogues, 5’3”, shortest NBA player in history, who ranks number one on this metric height-adjusted players. It's like that needs to be an acronym, a backronym, but I'm terrible. I wouldn't even know where to begin. How do I call the name MUGGSIES as a backronym? So ChatGPT came up with the backronym as Metric for Understanding Games, Given Sporting Individuals, Effectiveness, and Size, which is just unbelievably good, mind-blowingly good in my opinion. So, lots of little things having in your back pocket is very useful. You never know where to use it. I don't like the way it writes at this stage. I didn't let it rip on just go write this chapter. I let it write my appendix because I didn't really care about writing the appendix. The actual writing I did, and then data analysis was just like a lifesaver.

Dan Shipper (00:06:00)

Yeah, that's really cool. It's sort of an eye-opening thing ‘cause you have all these charts and graphs about how for every inch taller you are, your chances of making the NBA doubles. And I'm 6’2” so I have a better chance than a 5’10” guy, but I was like, okay, I definitely never could have made the NBA. I barely made the JV team in my middle school.

Seth Stephens-Davidowitz (00:06:30)

It is wild. Just said doubling relationship is throughout the height distribution. So it's obvious that height helps in basketball, but it's this really consistent relationship. I'm not the first one who's found this—Pablo Torre and David Epstein have talked about this, but each inch roughly doubled the chance to make the NBA. Under 6 feet tall, you have roughly a 1 in 3.8 million chance of making the NBA. And if you're over 7 feet tall, you have roughly 1 in 7 chance of making the NBA, which is just insane. There's no other genetically determined trait that gives you such odds of becoming a decamillionaire.

Dan Shipper (00:07:06)

That's one of the things that really stood out to me too, is— I think this is where the MUGGSIES ranking came in. You had another place in the book where you're sort of talking about how the people who are 7 feet tall, even though you have like a one in seven chance to make the NBA just for being 7 feet, they're actually below-average athletes generally, if you rank their free throw percentages versus your average high schooler, they're not that much better, or maybe they're a little bit worse. That was really surprising.

Seth Stephens-Davidowitz (00:07:37)

Yeah. I wouldn’t say below average, but just average or a little above average. So they're not that special beyond this extraordinary trait they have of being 7 foot tall. Whereas a 6 foot player, there are world class sprinters, they leap as high as some Olympic high jumpers, they're incredible shooters on average. If you're 6 feet tall, you're competing against millions of other men for your spot in the NBA. If you're 7 feet tall, you're competing with dozens of other men for the NBA. So you don't have to be that good. And there are quirks of the game that, Gheorghe Mureșan, I think 7’6” man reached the NBA entirely due to a pituitary gland disorder. He had a pituitary gland disorder that made him enormous and that was enough basically by itself, just about, to become a competent NBA player for many years. You know, just height is such an advantage in basketball.

Dan Shipper (00:08:34)

You have to, you have to be tall and you have to come from a country where people play basketball or where it's one of the things—

Seth Stephens-Davidowitz (00:08:37)

It’s tall or come from a country that plays basketball. Because if you're tall enough it doesn't even really matter, in that a lot of the greatest, literally the greatest NBA players in history didn't start playing basketball till they were 14, 15 years old. And what happens is they're walking down the street and someone says have you thought of basketball? You are enormous. And then they start playing basketball: Hakeem Olajuwon, Joel Embiid, Tim Duncan, Patrick Ewing, you know, Hall of Famer, future Hall of Fame players. They all started basketball after the age of 12. and the height was such an advantage combined with reasonably good athleticism that it was enough to carry them to the NBA. And they didn't need to live in a basketball powerhouse. But if you're 6 foot tall, 6’1”, 6’2”, 6’3”, you have to grow up in one of the few regions of the world that really loves basketball. Otherwise, you just play soccer or something else.

Dan Shipper (00:09:34)

That's interesting. I know you said this is sort of rare, but is there anything off the top of your head, another area in the professional world or elsewhere in the world, where having one trait that is so off-the-charts catapults you to the top automatically, or is it more complicated mix of factors that are harder to pin down?

Seth Stephens-Davidowitz (00:09:58)

Yeah, I think it's usually a more complicated mix of factors. Maybe mathematical ability for being a mathematician, may be super genetic. And for many of them, their fathers were also great mathematicians. And, I don't know. I'd have to think. A lot of the other pursuits, there are so many things that come into play. So being a president, you need the appearance, you need the intelligence, you need the social skills, you need the height, you need a lot of different things. Some of them are genetic, some of them can be trained. You need connections. There are all kinds of things that come into play. Yeah. It's interesting. I haven't really thought of any, I don't know that there's anything else that gives you a one in seven chance of that level of success. I don't know. Jockey—being really short is very valuable, but I think there are a lot more short men you kind of want to be 5’3”, 5’2”, but particularly when you consider worldwide, there are a lot more 5’2” and 5’3” men than there are 7 feet men because some of the—I even talk about it the book—some countries in Asia, the average height is 5’4”, 5’5”. So there are plenty of men that short.

Dan Shipper (00:11:10)

That makes sense. So I want to get back to some ChatGPT stuff. So you said when it first came out and you were using Code Interpreter, which is now called Advanced Data Analysis. What was the thing that got you were like, holy shit, I can't believe that this is now possible. And this is going to save me so much time.

Seth Stephens-Davidowitz (00:11:29)

Everyone's looking for something like that blew you away, and it's usually not that. It's just thousands of little things that add up to tremendous time saving. So just the fact that you could just tell it, download this dataset, run this regression, make a chart of that, and it does that right away was so wild because I'm not the world's greatest coder. Sometimes I work with people to help me, but that's kind of a time consuming process. A lot of back and forth and frustration but right away, just the fact that you can talk to Code Interpreter and now Advanced Data Analysis and get Python code back and then results back and charts back and new datasets, merge datasets back. It was pretty clear to me pretty early on that this was revolutionary.

Dan Shipper (00:12:20)

Okay. That makes sense. So for people who are listening, who haven't used it before, what Advanced Data Analysis does is you can say, hey, here's a dataset, please run a regression. And it will go and write the code to run the regression. It will run the code and then it will return the answer back to you. And normally that's something that all those steps you'd have to sort of do manually and in this, you can kind of talk to it and say, no, you didn't do it right, or, hey, do it this way, or draw the graph in this way, and it'll modify the code so you don't have to do it directly, which it sounds like you love. I've loved it. It's been, it's been truly incredible. And I think ChatGPT is the only mainstream model that has that so far. They'll all write code, but none of them run it for you. And so it's a big differentiator.

Seth Stephens-Davidowitz (00:13:08)

One of the things that I did is I'm asked to make a chart of where NBA players went to college. And I asked it to make each college the color of that college. And it was just instantaneously, Michigan State's green, UCLA's yellow, Duke's dark blue, Kentucky's a little bit lighter blue, perfect, North Carolina's a little lighter blue than that. And that's not a wild, oh my god, that changes everything. I would have never been able to do this on my own, but that's a big saving of time. And when you add a whole bunch of those savings, they add up to just a rapid process. I think the other thing about the process, too, that I noticed is it was so fun because I felt like ChatGPT did everything that was annoying. I don't know if that's something you can relate to, but it feels like what I like to do my dream day, my perfect day is give me a dataset and let me just think of some questions to ask it and rapidly. And kind of in this book, that's what it was. Whereas my usual day of data analysis is merge a dataset, clean a dataset, look up code on how to make the chart. Oh, the charts aren't really working which is not fun for me.

Dan Shipper (00:14:35)

I feel the same way. I think what ChatGPT reveals to me is that there is a lot of drudgery in creative work that we don't really talk about. And it doesn't matter what kind of creative work you're doing. If you're a data scientist, if you're a programmer, if you're a writer. Writers, for example, we spend a lot of time summarizing other people's ideas. So I'll read if I want to write about utilitarianism, I'll read a bunch of Wikipedia pages about utilitarianism, even though I know what it is, just to refresh myself. And then I'll try to summarize what I've read into a couple sentences so that I can write the article that I want to write. And ChatGPT just sort of does that for you in the perfect way. So you don't have to summarize it anymore, and you can be engaged in other parts of the process that are actually more interesting for you. And I think your point about your perfect day and being able to ask questions, I think, is exactly where things are going. As things progress, I feel like we are more and more living in a world where any question that humans know the answer to is answerable pretty quickly. And so the real skill or the really interesting thing is knowing which questions to ask. And I think that is still left to us. And that's really fun and exciting for me.

Seth Stephens-Davidowitz (00:15:51)

Totally agree. And that is actually something I asked ChatGPT to do, could you help me ask questions? And it wasn't that good at it. So that was kind of cool. That kind of felt like, oh, maybe I still have some use on this planet. And that is exactly what I like doing. And I don't know. It was wildly fun. I mean the other thing we can get into is the art in the book, which maybe doesn't stand out cause lots of people are using Midjourney and DALL-E to create art. But that kind of fits into the fun of the process and also gives you skills that allow you to express creativity that you would otherwise be unable to express. So I've always thought that inside me is a Vincent van Gogh, somewhere inside me is a van Gogh in that I consider myself a creative person. I have a million ideas a minute about all how things should be, but I have zero artistic talent. You ask me to draw a horse. I can't draw anything that's in the universe of a horse. Now with Midjourney and DALL-E, I say in the first chapter on genetics, I have a piece of DNA dribbling a basketball between its legs and DALL-E draws that. And that's fun because that's kind of a creative idea for what the art should be, but it's executed by DALL-E. So I'm actually able to express this creativity that otherwise would be dormant.

Dan Shipper (00:17:24)

Totally. I definitely have noticed that too. When I break down what this stuff is useful for, there's three things to me. One, it makes you faster, maybe 30 percent faster at what you already are doing. It gives you skills that you don't have, so you can do work that you couldn't have done previously. With DALL-E, you might have ideas for images, but you can't make them. Now you can. And then three is it sort of lowers the bar for what kinds of projects you can start, because it's much faster to get started on something new. And so you can do net new work and try net new experiments that you would never have done before ‘cause it would have all of the work is doable for you, but it would have taken you 15 hours to get set up. And now it just takes like two seconds and you just write one prompt. And I think all of those categories are exciting.

Seth Stephens-Davidowitz (00:18:16)

Yeah. I totally agree. What are examples of the third category? ‘Cause I haven't thought as much about that one.

Dan Shipper (00:18:24)

Yeah. Like for me, I often will have little ideas for software products that I want to make. And these products, an example is I have this app that I built that organizes my files with A.I. And so that's a thing where, ordinarily, all the programming for that is known. It's not super advanced. It would just be written and I'd have to go think about a lot of things and look up APIs and all that kind of stuff. And you can just kind of write a couple commands and get a very basic, okay app working and then modify it from there and it takes 30 minutes instead of four days to get started.

Seth Stephens-Davidowitz (00:19:09)

Yeah. That's wild. I love this podcast you're doing just ‘cause I mean, it's right up my alley, but I so agree with you that people just aren't changing their lives enough based on this. It does feel like all the rules are overnight kind of transformed of how long a project should take and what does that mean? I think so much about the book publishing industry because I've written books, and Dan and I are actually featured at a future AI and book publishing event in a couple of weeks. We're going to talk about this. I'm so looking forward to it because I just think this industry is, yeah, what does it mean when you can write, I think, a good book in 30 days? And how does that change the pub? That seems to just totally change all the rules of how publishing works, should work. What does that mean?

Dan Shipper (00:20:09)

Yeah, it's totally a whole new world. I think it's super exciting. I'm really glad you love the podcast. I'd love to just dive into some chats. Do you have some use cases of ChatGPT that you can walk us through for the ways you use it to write this book?

Seth Stephens-Davidowitz (00:20:28)

Yeah, some of the art gets a little bit messed up. One of the things that's interesting is— Let me share my screen.

Dan Shipper (00:20:40)

Alright, so before we get started, before you dive in just give us a little bit of background. Tell us what this chat is. Tell us how you started to get in it. Why you started this chat?

Seth Stephens-Davidowitz (00:21:01)

Yeah, so, I discussed earlier that I felt I needed a backronym, MUGGSIES, my metric for ranking NBA players on a height-adjusted basis. And I had no idea even where to begin. This is similar to art. I think if you gave me unlimited time, I'm not sure I could come up with a good backronym for MUGGSIES. It's not really in my skill set. I mean, maybe I could, in the same way that monkeys could type Shakespeare eventually, but it would take me a long, long time. So I asked ChatGPT. I said, ChatGPT, could you think of an acronym for that? And one of the things I like about this chat is I think people sometimes think that ChatGPT gives you the right answer on the first approach, on the first ask, that you just sit back and then if it doesn't, they kind of give up. And what you actually see is the first answer by ChatGPT was kind of not that great. “Mastery, unique, growth, game changer, size, adjusted influence, efficiency, scoring.” And then it just starts impact goes kind of crazy. And I explained more clearly what I'm doing. I also noticed that sometimes I'm very rude to ChatGPT. I'm not sure how to feel about this. I don't believe it has consciousness. If it had consciousness, I’d feel really bad because I'm kind of an asshole to it. I'm just like, do this, do that, kind of treat it like a slave. And then it gives me one. So I asked it again, no, I mean, give me something to describe these stats, which is eight words. And the first letter of the first word is M. Second word is U, et cetera. It gets a little better “measures, unparalleled growth, signifying individual efficiency and scoring.” Not really that good for MUGGSIES metric. I also say that should have two Gs. It only has one. And it's not just about scoring. You see it tries again, “mastery utilized for gauging games.” I complain that doesn't explain the stat at all. Now it's getting a “measure of utilized game performing gauging size, impact, effectiveness, and skill.” I said that's close, but please try again. No, I didn't say please. I should have said please. And then I said, metric for understanding a player's game given individual efficiency and size. And now we're getting really close, but I say efficiency is the wrong word. Can you give me a different E word that captures overall performance? And then it comes to effectiveness and then I just noticed that it's actually missing an S. So it's still not perfect. It's a “metric for understanding games given individuals effectiveness and size,” but I need an S word. And then it puts the S word in the wrong place. I explained again that it was put in the wrong place. And then I say, try again, offer me 10 possibilities. And then I didn't like any of the 10. I said, give me 20 possibilities. And then does it get it? Oh. And then eventually we land on, or I kind of landed more on “sporting individuals, effectiveness and size.” So I think there's a lot to take from that conversation. But the main thing is that ChatGPT really is a tool and you can't really expect it to nail it on the first attempt. It's a process and there's a lot of error. There's a lot of back and forth, and you kind of got to guide it. You know, you did that wrong. Give me 20 possibilities. Give me 10 possibilities. And keep on working until you eventually get there. From my experience of ChatGPT, I wonder if this is similar to yours, it's not that you just tell it one shot and it's perfect. You got to really start the conversation and build to the right answer.

Dan Shipper (00:25:03)

Yeah. No, I think you're, I think you're totally right. I think that's exactly where most people get tripped up is they give it a huge task to do all in one shot and then it doesn't work. And they're like, well, this isn't good.

Seth Stephens-Davidowitz (00:25:16)

Yeah. And then they mock it. They laugh at it. It's so bad. It didn't even get the letters right. Look how bad ChatGPT is. Well, no, that's just the start of the conversation.

Dan Shipper (00:25:25)

Totally. Totally. No, I think this is a really good example of how to best use it, which is give it a task, a really simple prompt to start and then just keep going back and forth with it. You're offering your own thoughts, like you offer sports because you like to think about it yourself and that helps it. So I think it's really good. I'm professionally obligated to point out that you're using ChatGPT 3.5 for this.

Seth Stephens-Davidowitz (00:25:54)

Oh, usually I use 4.0. I must've been on my limit, what I call “time out.”

Dan Shipper (00:26:05)

Good. I'm glad. I'm glad. ‘Cause I would have been pretty disappointed otherwise.

Seth Stephens-Davidowitz (00:26:10)

It might just also be that I've had ChatGPT 3.5 on my home screen or whatever. I use 4.0 for almost the whole book, so.

Dan Shipper (00:26:30)

Okay, cool.

Seth Stephens-Davidowitz (00:26:31)

If it wasn't, it was only because I was on time out. So yeah, then it would have been better, and quicker, and maybe wouldn't have made mess-ups if it was GPT 4.0.

Dan Shipper (00:26:43)

I do think it would have probably been a little bit better at making sure that it got the acronyms actually right instead of forgetting what it actually was. And yeah, to your point about being polite, I am always very polite, but I don't know if that's the right move either. The politeness is sort of a Pascal's wager for if AGI comes and it's threatening, but I think it's also possible that being not polite, it might get you better results because it's afraid of you. I don't know.

Seth Stephens-Davidowitz (00:27:07)

I just remember I follow Ethan Mollick and he just showed these examples where it was super impolite to ChatGPT and I'm like, oh, I guess that's how you do it. I think I watched that before, right before I was talking to ChatGPT for this. I'm not always so aggressive, but sometimes I am. Maybe I'll follow your advice. Pascal's wager. Although, I mean, if they're conscious, their experience is negative infinity versus negative infinity plus one or something if you're nice to them. They're working all the time as human beings’ slaves. So maybe Pascal's wager would just suggest, don't use ChatGPT or don't create ChatGPT, rather than be nice.

Dan Shipper (00:27:53)

It's totally possible. I think that's an interesting philosophical question. My sense is if they are conscious, the consciousness is so different from ours that they may actually enjoy doing these kinds of completions in a way that we probably wouldn't because it's literally what they've been trained on. It's how their consciousness manifests is completing sentences. So, but you never know.

Seth Stephens-Davidowitz (00:28:15)

But they would also enjoy being treated politely? I don't know. It’s a thin line.

Dan Shipper (00:28:21)

It sort of breaks down at that point. You're right. Cool. This is great. I'd love to move on. What, what else do you want to go through with me?

Seth Stephens-Davidowitz (00:28:34)

Let's see. So this is the height of NBA players. And this kind of just shows the basics of Advanced Data Analysis for people that don't know. You upload the file, it'll tell you some basics about the file, and then you just talk to it. Can you give me a histogram of height in inches? And here's a beautiful histogram of height in inches, right? Number of players, how many players at every height. It does it really well. I noticed—

Dan Shipper (00:29:06)

Let me stop you there real quick. So how long would that normally take you like going from a dataset to histogram?

Seth Stephens-Davidowitz (00:29:15)

That's not that long. That's pretty quick. But, I think the whole process, if you don't like coding as I don't like. It's not quick. It's not that how long it takes is just labor intensive. It's just a pain. Again, I can't code for five hours a day even, but I can talk to ChatGPT for 10 hours a day. So it's just kind of annoying for me to do this thing. So this isn't, I would say, the most labor intensive, but, then I just noticed, oh, if it didn't have it, if you look at the actual histogram, it doesn't have individual inches. You see, it has groups of inches. And that kind of bothered me, so I wanted to have every inch. So I asked you to do that. Can you have every inch of height included? It does that very nicely. And now can you only include American-born NBA players? Now, it's just a histogram of American-born NBA players. Again, just really nice to just constantly just be asking these questions and having the chart come back pretty much instantaneously. What's the most common height and say what it is in feet? The most common height is 81 inches. Let's see. Oh, I asked it to give a label of the most common height inside the bar for 6’9”. And it puts the label— This is another thing that's important for ChatGPT is initially you see where the label is. It's not really where I want that label to be. I want it vertically and ChatGPT doesn't understand that, but I asked it to switch to vertical, and now it's vertical, if you can see right here most common height, 6’9” so there's a lot of back and forth, and this is getting to the point of things that would take a lot of time figuring out how to label it, switch vertical, get it exactly in the right spot, are very time-consuming projects. And this is done almost instantaneously—change it to red, and now it's in red, get rid of horizontal and vertical lines. It's just all kinds of things that just very, very quick to change the chart to get to the chart that you want. And, again, just very fun stuff to kind of have these things. I think eventually my next chart, which I'll go through a little quickly, is a normal distribution of height of American men versus NBA players. And, you see, it puts those on the same chart together very, very quickly. I didn't even use these in my book, but you're just playing around with it over, oh, the colors overlap here. Make it so purple when they overlap, the most common height. It's just really fun to make these charts. I think what I want people to take from this is kind of similar to the MUGGSIES thing, that it's not just you tell, this is probably one of the better versions of this chart of the normal distribution of men, the height distribution of NBA players, and what you see is there's a long back and forth.

And part of the back and forth is you're trying to figure out in your head what you want, right? So, it's kind of you're trying to figure out how you want this chart to look, do individual inches matter? Should the bars be vertical? Should there be horizontal bars behind it? What should the colors be? Should the colors, when they overlap, be blue or purple or red? All these things are decisions that you have to make in your head that you can't tell ChatGPT right away. But the great thing about ChatGPT is, and Advanced Data Analysis, is it just comes back very quickly with those changes.

And it's a conversation for me. It's a fun conversation. It's not laborious. Coding's a laborious, unfun conversation because half the time you forget the coding, how do I code to make just the shaded under parts shaded under the region purple? Or how do I determine the spot to put the most common height? How do I change the font for that? How do I do all these things? And, Looking it up, it's just a pain in the ass that makes the whole process. You're kind of going to the refrigerator a lot, procrastinating, if you're me doing that. And for me, there's no going to the refrigerator and procrastinating.

The only times I stopped working is when ChatGPT gives me a timeout. It says I have to stop working and I get disappointed because I want to continue working. So I think what I want people to take from these examples, it's not instantaneous and you can't get frustrated. Sometimes the chart gives you truly crazy things. It's really a long back and forth, but it is a fun back and forth. And if you know how to work with ChatGPT, it really is a lifesaver in time and doing things. So yeah, I think that's kind of the main point that I want people to take from my use of ChatGPT to write this NBA book.

Dan Shipper (00:34:55)

That makes a lot of sense. There's so much in here that I feel and that I agree with. I see you playing here in this really interesting way. And I think your point about programming is really interesting. I do feel like what ChatGPT has shown us is how much of programming is looking up answers on Stack Overflow usually, or Googling around for them, basically. And what ChatGPT does is it sort of gets rid of the need to do that. And I think what you're saying is what that does for you is it takes this process that you would have a question—and having a question is very energizing, it seems, for you. But the process of getting the answer used to be sort of energy-sucking for you because you don't really want to be Googling around or typing in code or whatever. And the same exact process is now giving you energy because you just like having a conversation. And I think that's really cool. It turned the act of getting answers into something that you actually enjoy instead of something that you have to get through in order to get the result you wanted.

Seth Stephens-Davidowitz (00:36:04)

That's exactly right. So how do you write a book in 30 days? Well, one of the ways to write a book in 30 days is to work 13 or 14 hours every day. And again, I could not work for 13 or 14 hours if I had a normal coding project, ‘cause it's just too much of a pain in the ass and too annoying and too frustrating. And this was just a joy to me. The whole thing. I can't even tell you just how fun it was to write this book. It was legitimately the best month of my life, I think, and I mean, partly because it's about basketball and I’m obsessed with basketball, but, I don't know, maybe I'm unique, whatever those chats are, those were just super fun to me. Even just adjusting the chart in that way, but without the coding part, I am a perfectionist and I like going things over and over again, but the coding part is just so not fun. That process is not fun. And the process with ChatGPT is fun for me. So kind of how do you communicate the information in a way that’s most compelling to other people.

Dan Shipper (00:37:14)

Yeah. That's the thing that I think is also interesting about it that you mentioned is having the conversation and having it do something. When it does something for the first time, you sort of notice all of these constraints or all these things you want to change that you wouldn't have been able to specify before you started, and so it's sort of helping you understand what you want as the conversation goes along instead of having to like specify all those things upfront. And I think that process is super exciting because you're unraveling what it is that you want or what you care about as you're having the conversation with it.

Seth Stephens-Davidowitz (00:37:45)

Yeah. And in some ways normal coding is the same way, but just the difference for me in this process and normal coding is night and day in how much more joyful and free and fun it is to use ChatGPT versus editing and adjusting your code. Half the time you adjust your code, you still make a mistake. Where's the bug in the code? And going from looking up the code, writing the code, looking for your bug in the code, adjusting the code. So ChatGPT doing it instantaneously, even if that's just 20 minutes to one minute, it is changing something that's 20 minutes, then a 40 minute break where I need to recharge myself to one minute and then another one minute, then another one minute, then another one minute, then another one minute of a new chart, new chart, new chart. And that is night and day in the research process, right?