Exclusive Analysis of ‘How Do You Use ChatGPT?’ With Logan Kilpatrick

My take on the latest episode of How Do You Use ChatGPT? with Logan Kilpatrick and the episode transcript is below for paying subscribers.

My super-secret plan is to eventually get Sam Altman on How Do You Use ChatGPT? I mean, come on—wouldn’t it be awesome to see how he uses it every day?

The first step in the plan is complete: We got our first OpenAI employee, Logan Kilpatrick, on the show. Logan’s great, and I really enjoyed getting to record with him. The main takeaway from this episode is this:

Even someone who is close to the latest and greatest language model technologies believes that they are likely to enable—rather than replace—developers over time. In the episode, Logan says that being a developer is still the most valuable skill you can have, and it’s only going to increase in value over time. He also believes that ChatGPT is going to enable the next billion developers.

You might say: Well, of course he thinks that. He works on behalf of developers for OpenAI—it’s part of his job description.

But I agree with him. I’ve been writing about ChatGPT as a tool to expand who gets to be a developer for a while —and I’ve seen it first hand. In the ChatGPT programming course I run, people who have never programmed before build and launch a chatbot in 30 days. They’re not incredible programmers by the end. But they have joined the ranks of people who are able to build something themselves—and that gives them the motivation to learn the fundamentals of programming that might have seemed out of reach before.

It’s a magical thing to watch, and I think we will see ripples of it throughout the economy. Building is no longer necessarily the province of an elite few: With AI, anyone can be a builder. And that, in my opinion, is a much better world.

Timestamps

Introduction 00:44
Why learning to code is the highest leverage thing you can do 09:18
AI is empowering the next billion coders 13:40
The first prompt in GPT Builder for our video game 35:58
How to increase the chances for your custom GPT to go viral 39:27
Prompt engineering tips while using GPT Builder 43:00
How to use ChatGPT in conjunction with GPT Builder 56:13
Ready to play our text-based strategy game! 1:06:33
How to finetune your custom GPT 1:19:44
Why you should build custom GPTs 1:43:12

Transcript

Dan Shipper (00:00:00)

Let’s brainstorm. What’s the game you want to make?

Logan Kilpatrick (00:00:02)

It would be interesting to see if we can, like, you are a leader of this civilization and then each turn, your people come to you with, here are these choices that you have to take to take the next step in the growth of this civilization.

Dan Shipper (00:00:14)

“As President Kennedy, your challenge is to balance the budget with the pressing needs of a nation on the brink of major social and technological changes. Will you prioritize space exploration, pushing NASA's budget to beat the Soviets to the moon, or focus on domestic issues, ensuring equality and prosperity at home? The choice is yours.”

Logan Kilpatrick (00:00:30)

I like that. I like that.

Dan Shipper (00:00:44)

Logan, welcome to the show.

Logan Kilpatrick (00:00:46)

How's it going, Dan? I'm super excited about this.

Dan Shipper (00:00:48)

I'm excited too. We've been following each other and interacting on Twitter for a long time. So it's super fun to actually get to meet you face-to-face. For people who don't know, you are a member of the developer advocacy staff at OpenAI—the first OpenAI employee we've ever had on this show. Woo! So excited. And, as part of that, you focus on enabling developers to build with the OpenAI API and with ChatGPT. And yeah, I'm psyched to have you.

Logan Kilpatrick (00:01:19)

Yeah, this is gonna be awesome. And I'm super excited that you put this show together to really show—Part of the historical challenge of ChatGPT is showing people the art of the possible. So you doing this is making all of our lives easier. And we're also doing a bunch of stuff from a product perspective to push on this, but I'm happy that you're pushing from this angle as well. It's important work.

Dan Shipper (00:01:42)

Thank you. I really appreciate that. I love the idea of the art of the possible. I feel like that's exactly the thing that ChatGPT opens up for so many people and for me personally, so that's really great. So we were talking a little bit in the pre-production call what we want to talk about, what's most interesting, and I think one of the things that struck me is, your role is to be a developer advocate. But historically, developers have been this sort of small component of the population. And what's really interesting about ChatGPT is it's creating this world where the number of people who can become a developer, who can develop things, is expanding tremendously, both because you can actually code underlying apps with ChatGPT and because of the thing that you just launched custom GPTs, which allow you to create custom versions of ChatGPT for specific use cases. So tell us a little bit about custom GPTs and tell us why you’re excited about them.

Logan Kilpatrick (00:02:40)

Yeah, the being able to build GPTs is probably the biggest unlock since ChatGPT itself came out. We classify everybody who's building GPTs as quote unquote builders to kind of not make people who hear the word developer are scared or say like, oh, that's not me. I don't fall into that persona. And we give this flexibility of using the GPT builder, using the custom configuration when you're creating a GPT, you don't need to know how to code at all. If you do know how to code, there's a ton of stuff that you can do, but you can actually make something that is materially much more useful than base ChatGPT itself without writing a single line of code.

Also, now there's companies like Zapier is a great example of this, where Zapier has this whole guide of how you can integrate Zapier's tools into your custom GPT without having to write any code. I think that's a huge unlock and I do think that's also the direction that we're pointing with GPTs is allowing people to do a lot more without having to write any of the code themselves today.

Dan Shipper (00:03:48)

I think that's really great. I've been using it. I have a GPT specifically for doing dream interpretations. So, when I wake up, I have a little dream, I throw it in there and it just gives me a little bit of analysis. And then it does a picture with DALL-E and it's really great.

And I use it a lot. But I have a bunch of other ones. I've been working this weekend on one, where I want to be able to do a chat with it and then have it save the notes from it from my chat into Notion. And so I'm hooking up to the Notion API. I got a little stuck on it, but one of the things that I think is really cool, you talked about the GPT builder and that is a GPT that lets you build GPTs So it's sort of GPT is all the way down. And I think that's a really important thing for people to notice and realize is that as you start stacking AI on top of each other, you get extra leverage where now you have a GPT builder that lets you build GPTs and that lets you build more things more quickly than you would ordinarily.

Logan Kilpatrick (00:04:56)

Yeah, and it's interesting because when you use the GPT builder, it doesn't feel like you're using a GPT, which is interesting. So it's kind of cool that, as you abstract out of the interface where people are very aware that this is some specific tool, you also kind of lose the fact that it's AI all the way down. It feels like some other system that we built, which I think speaks to the general reasoning engine capabilities that large language models have, which is really interesting.

Dan Shipper (00:05:23)

Yeah, that's really cool. So, we're going to get into, we're going to build something together. We're going to build a custom GPT together, which I think would be super fun. Before we do that, we always like to talk about historical use cases or just ways that you use ChatGPT yourself. And I know there's another custom GPT you might want to talk about, but, generally, tell us how you are using ChatGPT and custom GPTs in your own life and in your own work as a place to get started.

Logan Kilpatrick (00:05:55)

Yeah. I think the biggest, highest-leverage thing for me as somebody who's still doing engineering work on a daily basis is coding stuff. I know there's a ton of really useful GPTs around coding and making stuff. Most of them have been historically focused on low-code type of, give me a single line prompt and I'll go and build you this thing. And that's not the use case that— I'm more around I'm not a Next.js expert and I'm in a Next.js codebase or I'm going to typescript code base and I'm trying to build something or add a new feature and I have a computer science background, but I'm not formally trained as a web developer. I never did that as my job. So I'm always asking ChatGPT the really, really dumb questions. And you'd be surprised the number of features that are exist on our API platform that were built because ChatGPT helped me—the feedback, thumbs up-thumbs down is a great example of that, when there's a outage in the API and there's a little pop-up that shows up dynamically, all of that was built by ChatGPT itself, which is really cool.

Dan Shipper (00:06:57)

Yeah, I think that makes a lot of sense and it sort of dovetails with the thing that I've been writing about or thinking about a lot is: ChatGPT's facility at summarizing, and once you start to see how good it is at summarizing, you see summaries everywhere in your life and in everything that you do. And I think that that is true also of developing where people would be surprised, if you're not a developer, how often being a developer is just summarizing the API docs or summarizing a Stack Overflow answer into your, into your codebase or whatever. And it sounds like that's a lot of what you're doing. There are some parts of programming that really, you're doing fundamentally new stuff that no one's ever really done before. And in that case ChatGPT might be sort of helpful to think it through or whatever, but it's not going to be as helpful. But most of developing is really just stacking bricks that have already been laid in a slightly different configuration than they've been laid before for your specific use case. And that's where I think it shines so much because it's so good at knowing what people have done in the past and just writing a little snippet for your specific scenario.

Logan Kilpatrick (00:08:02)

Yeah, it also has been really helpful to me as I look at 300 lines of code and not— I kind of understand, I read the code that's right around the thing that I'm trying to make an addition to our modification, but I don't in my personal context window, I don't have all 300 lines of code. And it's really helpful when I then go and bring that into ChatGPT and say, here's all the code, ChatGPT now actually has all 300 lines of that code in its context window and it can more effectively solve the problem, honestly, than I can, because it's really difficult to take 300 new lines of code, understand it, and actually conceptualize how you can make some modification to it.

Dan Shipper (00:08:44)

That makes sense. And I guess I have to ask, cause you're in this really interesting position where you're at OpenAI, you're watching all the progress, you have access to internal stuff, you're a developer yourself. How is using ChatGPT, making custom GPTs, seeing the progress internally at OpenAI, how is that changing what you think the future of being a developer is? Is learning to code still a thing? How are you thinking about this over the next year, five years, all that stuff?

Logan Kilpatrick (00:09:18)

Yeah, learning to code is still definitely a thing. I think if you're looking for the highest-leverage thing that you can do in your life, learn to code. I think this will continue to be the case in the next five-to-ten years as well. My expectation is that folks will be able to solve probably 10-to-20x and it's going to continue up this exponential curve of the difficulty of how conceptually hard the products and the services are that you want to build just by yourself. To me today, I feel like if you had taken me back two years ago, I feel like I can solve 10x more ambitious problems. I have some idea in my head. I want to build a restaurant application. I can actually go and build that myself and two years ago, I would have been like, that's probably a little bit too much. Maybe I'll go hire someone to help me do that. And, I think continuing up that exponential curve, soon I'm going to be able to build an entire empire of products and services for my company just by myself. And I don't think three years ago people would not have imagined they can do that.

And that's why I still think developing today is one of the highest-leverage things that you can do, but in two to three years and five years. It is going to be even higher leverage and the gap between people who can who can code and who can't code is going to continue to grow and I think the nice part is AI can help bridge that gap but there will always be limitations on the no-code tools as far as what you can build you need to understand how the systems work.

Dan Shipper (00:10:53)

Yeah, that makes perfect sense. I mean, you're speaking my language here. Cause I'm running a small company where I write a lot, I do courses, and we develop software products and it's a small, profitable lean team. And the thing that I love so much about tools like this is, we can go so far with the small group of smart people that we have, because now we all have ChatGPT. And I think you're totally right. It lowers the bar to getting started on something so much that for me—a project that I never would have attempted before. Cause it would be like, I only have a couple hours here. I can suddenly do it and it's out in the world. And that is the coolest thing ever.

I think that's one thing that people don't quite understand. I actually talked about this a little bit with— I had Tyler Cowen on the podcast last week. And I actually talked about this a little with him where, for really skilled people, it can, it can increase your productivity where you're slightly faster, but it also allows you to do things that you never would have done before because the bar would have been too high to get started. There's an activation energy to making projects. And I think that that's incredibly cool. And I think you're also right. It does something where if you can't code at all, you can suddenly make stuff and it'll be better if you understood the underlying technology and that'll work. But the really cool thing is, I teach this chatbot class where people learn how to build a chatbot. It's not with GPTs. It's the actual underlying, make the whole thing a web app—all that kind of stuff. But you use Chat to help you code and it takes people from, having never developed in their life to, I built and shipped an app in 30 days. And it's so cool to watch them be able to make stuff because they all have ideas, but they've never been able to put them out in the world before. And Chat becomes this thing that enables them to do it. And also it gives them a lot of motivation to learn the underlying fundamentals because previously, if you wanted to learn to code, you spend six months, learning if statements and for loops and all this stuff where you're like, I have no idea how this levels-up into the app I want to build. And what ChatGPT does is day one, you make the thing that you want to make and then day two, you run into some bugs and you're like, well, I guess I have to figure out what an if statement is. But you now have a lot of motivation to do it. And ChatGPT can explain it to you and it’s just really, really cool.

Logan Kilpatrick (00:13:35)

That resonates with me so much. After we're done recording this, I'm going to have to get the clip out and play it back to myself every month or something because this technology is what is going to be the thing, the catalyst to enable the next billion people to learn to code.

There were not enough computer science educators and resources on the planet to teach the next billion people to code. It was not going to happen. And I think it's so crazy to think that because we have LLMs, people are actually going to be empowered to go and do this in a way that's personalized and empowering to them specifically. That might be the real thing that ends up being the most beneficial of this technology, of all the things that it can do.

Dan Shipper (00:14:15)

I think so. Yeah. It expands access to being able to build significantly. And I think that's the coolest thing.

Logan Kilpatrick (00:14:26)

It's beautiful.

Dan Shipper (00:14:28)

Yeah. And with that, maybe we can get into some building. But before we start actually building something together, which we're going to do, I think you wanted to show me a custom GPT just to sort of whet our appetites. Give us a GPT that you built. Tell us about why you built it and what it is and why you think it's interesting to look at.

Logan Kilpatrick (00:14:44)

Yeah, 100 percent. Let me get my screen-share pulled up and then we'll, alright. So we're in the GPT store. So I was lucky enough to be able to build Planty, which is a little play on words. If folks watch the OpenAI DevDay back in November, there was a customer video of people talking about how they were using ChatGPT, how it's impacted their life. And there was this incredibly cute old man who came on, with a really grainy video sitting in his office. And he was talking about how Chatty has sort of unlocked all these use cases. And he was referring to ChatGPT as Chatty, and it was really cool to see him get that value and have that relationship with this tool.

So Planty is a little bit of a play on words with that. And, the whole concept was, literally sitting on the couch with my girlfriend and we were thinking of, what are the cool things that we should build for people? And she's not somebody who's in AI, but she's somebody who loves house plants and gardening and everything like that. The first thing out of her mouth—so all credit goes to her—was we should build Planty. And it's really designed as a houseplant caretaker. And she has a Monstera and she needed to do some stuff with the roots. That was a little bit over my head from a plant technical perspective.

And she went in, as we were building Planty, used it to actually solve that problem. I got my older brother for Christmas, a lemon tree. And as soon as it got into our apartment, all the leaves were gone. It looks very desolate. So, I put out a tweet about how if Planty can save this lemon tree, I'm a deep believer in AI. And, really, you can do things like just ask simple questions, or get really tactical and upload images and be like, here's the color of the leaf of this plant. Why is it like that? And it really unlocks, again, ChatGPT could do some of these things, but providing a bunch of guardrails to allow it to be able to do this. And, giving it this friendly personal feel. I think this actually brings to life what's possible with this GPT.

Dan Shipper (00:17:07)

Tell me about that. How would I know whether or not I should make a GPT versus just vanilla ChatGPT?

Logan Kilpatrick (00:17:16)

That's a great question. I think that this is one of the tricky trade-offs today. I think what will end up over time being the way you answer this question is like social signals. I think people will probably come up with a bunch of—if you have a plant question and there's like 100,000 people who said that Planty was the thing that solved their problem, you'll probably just go there because that's what it seems like you should socially do. I do think the interesting thing for people to keep in mind is, behind all the scenes, behind everything, it is inherently ChatGPT that is still answering these questions. So it's not some net-new capability.

I think the thing that we did with Planty is like grounding it in very specific stuff, like if it needs to go and look something up, it goes to a bunch of specific websites that we pre-vetted as places of authority that you should go to instead of, I don't know, we want to sell you plantstuff.com, which might have a biased perspective. So there are a few nuances to how it's being built, but again, fundamentally, it really is just ChatGPT behind the scenes with like a few layers of abstraction on top of it.

Dan Shipper (00:18:27)

That's really cool. So I actually have a bunch of plants. As you can probably see, I have a fiddle-leaf fig, in my apartment. And I actually do have a question randomly, that might be worth doing. I don't see your screen anymore.

Logan Kilpatrick (00:18:42)

Yeah, I just switched it to Planty.

Logan Kilpatrick (00:18:50)

I’d love to see the question. Let’s do it.I ’m curious what Planty's suggestion is.

Dan Shipper (00:18:52)

Here's the thing. I have this fiddle-leaf fig. Fiddle-leafs are supposedly very finicky and hard to take care of, but I've had it for like four years or five years or six years or something like that. And I just water it every week and a half or something. And it’s doing pretty well. Every once in a while, a leaf will fall off and I freak out, but then it's fine. But one of the things that I don't know is, I think you're supposed to repot these things at some point, but I just don't know when to do that or how to do that. And I'm just wondering if I'm like artificially constraining my plant and I should be repotting it, and how I would know.

Logan Kilpatrick (00:19:39)

You said it’s a fiddle-leaf fig? How and when should I repot my fiddle-leaf fig?

Dan Shipper (00:19:41)

So you correct your typos in chat cause I never correct my typos. I just type like a maniac. Is that a bad thing?

Logan Kilpatrick (00:19:43)

I do type like my maniac, but it's more of a habit thing where like if I stop correcting my typos in this, then I'll stop at everything else. But people will be pissed at me because I'm typing like a maniac on Slack or whatever it is.

Dan Shipper (00:19:58)

That's so funny. It's just funny doing this show. Cause I'm typing stuff a lot in screen-shares that get recorded and broadcast to thousands of people. And I'm just like it looks like I'm in third grade or whatever. And I'm like I swear I can actually type. I just don't do that for ChatGPT.

Logan Kilpatrick (00:20:20)

Yeah. I'm chronically a bad speller, so I need to do it. So I asked the question of how and when should I repot my fiddle-leaf fig. What are the trade-offs from not repotting and, I think, touching on the artificially constraining it. So it's interesting. “When to repot.”

Dan Shipper (00:20:42)

So it said “Repotting your fiddle-leaf fig can be quite the adventure.” Is that part of your personality you gave it?

Logan Kilpatrick (00:20:51)

Yeah, it is. I think it's supposed to be this friendly, exciting thing that's helping you instead of—I think by default, ChatGPT is pretty generic about just trying to give you this information. So I think it brings it to life a little bit.

Dan Shipper (00:20:56)

Okay, cool. And if you scroll up, it says, “Repot your fiddle-leaf, ideally during the growing season, which is spring or early summer. Take a peek at the drainage hole at the bottom of the pot. If you see roots starting to circle around or peek through, it's time for a bigger home.” I mean, I'm going to just check that real quick. Give me one second. We'll see how it goes. I'm super curious. See roots starting to circle around or peek through. Hold on. There are literally roots coming out of the side of the pot. It's not peeking through. It's like they've grown out and they're coming out basically almost onto the floor. So I think I need to repot.

Logan Kilpatrick (00:21:49)

“How to repot it.” This is perfect.

Dan Shipper (00:21:50)

Yeah. “Find a pot that's about two or three inches larger in diameter than the current one. Make sure it has some good drainage holes. Add some fresh potting to the bottom of the new pot. Fiddle-leaf figs like well-draining. Gently remove the plant. Check the roots and place your fiddle-leaf in a new pot and fill around it with more.” What I want to know—So here's my last question: I'm super lazy and I am not going to figure out the diameter of the pot and do all this stuff. Is there some way I can get someone to do this for me?

Logan Kilpatrick (00:22:40)

Where do you live?

Dan Shipper (00:22:42)

I'm in Fort Greene, Brooklyn.

Logan Kilpatrick (00:22:48)

Fort Greene, you said? See, there's my recommendations typo—horrible word to type.

Dan Shipper (00:23:02)

So you said, “Are there services or ways I can get someone to do this for me? Any specific recommendations for someone in Fort Greene?” And Planty says, “Absolutely. Getting some help with repotting your plants is a great idea, especially if you're not up for a DIY project or if your fiddle-leaf is getting a bit too hefty to handle on your own.”

Definitely too hefty to handle for me. “In Fort Greene, you're in luck because it's a vibrant area.” Definitely. Thank you, Planty. “With lots of plant loving folks and services.” And it gave some, I would say generic recommendations, like local nurseries and garden centers, although it did give us a couple of specifics within that saying Natty Garden or Greenery Unlimited.

Logan Kilpatrick (00:23:38)

Are those real places?

Dan Shipper (00:23:40)

I don't know. I'm not up on the local plant caretaking services in Fort Greene, but I would love for you to check it out. It also recommended task websites like Taskrabbit or Thumbtack, or plant care services like The Sill, which I definitely know is real. A friend of mine's wife runs that company. So, definitely got some good directions. Maybe, after this episode is over, I will get it taken care of and we'll do a part two where I update you on the progress of my fiddle-leaf. So it looks like you looked it up and Natty Garden is a real place.

Logan Kilpatrick (00:24:20)

It’s a real place. It looks nice too. I kind of want to go to Natty Garden. This looks like it would be a good time regardless of if you get your plant repotted there or not.

Dan Shipper (00:24:31)

Great. Well, I will be checking out Natty Garden after the episode. And I'll send you pictures.

Logan Kilpatrick (00:24:37)

This episode is not sponsored by Natty Garden, but it looks like an incredible local community store.

Dan Shipper (00:24:42)

Cool. That's great. I love Planty. I think it’s a cool little idea. And, I think my fiddle-leaf is probably pretty happy that you showed that to me as well.

But, now I want to move us on to the main event. So, I told you that one of the really fun things about this show is getting to make something together or do something with ChatGPT where we’re both sort of like, is this going to work? And that's sort of the magic of it. So you proposed making a game in the GPT builder. And I wanted to start with why did you propose that? Why do you think that's exciting?

Logan Kilpatrick (00:25:22)

Yeah, I think if you were to think about how you would have had to do something like this even just pre-GPTs, just with ChatGPT and plugins, I've been brainstorming this idea for a while of, I would love to take people through if you played Civilization or Empire Earth or any of those games that are situation-based where you're building an empire over time.

Those were always fun to me. And, it really brings history to life, especially if you can sprinkle in actual facts about real things that happened. And without GPTs, it was impossible to do this. I, as the developer or user, would have had to pay to generate these images. Each individual image, it would have cost me money. The constraints of building it's actually kind of hard to do. And it's a non-trivial amount of work. I think the cool thing now with GPT is you can use the GPT builder. You can sort of have it take care of a lot of the nuances of game design.

And, we build in DALL-E. We can use built-in browsing to pull in facts in real time, keep track of things for you, potentially use Code Interpreter to do some actual simulations of, I dunno if we can simulate battles or like statistical likelihoods of things, but it's incredible what will be possible with this. So I'm really excited. And also it's one of the areas that I think specifically because there's not a category right now. It's a little bit underexplored. If you look at the trending categories in the GPT store, there's not really a ton of games that are in there, if any games at all.

Dan Shipper (00:27:07)

So I think it's an opportunity for us. That's really interesting. I wonder why, cause I know, in the '80s and '70s, during the first PC revolution, all text-based games were super, super popular. And it feels like an obvious thing to try. I wonder what's holding people back. Have you seen any at all?

Logan Kilpatrick (00:27:24)

I have seen a few. So actually the most interesting one was this game, which was one of the most popular plugins, and it was called Algorithma. And it was this guy and his and his daughter, I'm pretty sure, the daughter learned to code to build Algorithma and ended up scaling it to the point they had like 5 million unique users or something like that, or 5 million like users over the course of a few months who used Algorithma. And it's actually a little kid who built this thing. And, it was actually fun to play. It was crazy. And I talked to the dad and he was saying how she was using AI to help her build it. So it's a great example of the theme of people being empowered to build cool things. Yeah, I think that was probably the biggest one that I haven't seen their GPT, but I'm guessing it's doing a bunch of really cool stuff.

Dan Shipper (00:28:16)

That's awesome. So opportunity, some white space for us. Maybe this is the next big thing. Maybe we'll do it live. I guess to start, let's brainstorm. What's the game you want to make? If you have a couple of ideas, throw it out and I can help refine.

Logan Kilpatrick (00:28:33)

Yeah, I think it would be cool to—and I think this is, maybe the decision for us to decide is do we want it to give it a purely fun and exciting type of vibe or is there some educational component? Would we want people to be able to learn about history as they go through this? I think there's a trade off either way. In some sense, people just want to escape and play some fun game, but there could be some value— And, maybe we can have a toggle where like we let people at the beginning of the game decide are you here for fun? Are you here just to actually learn about stuff as well as you go through?

Dan Shipper (00:29:14)

Here's my feeling. I think the core of a game or really most content has to be about fun. It's about creating an amusing experience. And if you happen to learn something along the way, I think that's amazing. This podcast is sort of similar. Ideally, it’s super fun to listen to. And you happen to learn something along the way, but we don't want to jam in the educational stuff because then that gets boring. So but I think what I hear you saying is history is kind of an interesting thing for you and having a part use history in some way, it would be kind of fun. Do you have any specific ideas for how to incorporate history into a game like this?

Logan Kilpatrick (00:30:03)

Yeah, I'm curious if the idea of—and maybe history is too dark for this and it won't work, but I was playing as different world leaders or something like that. And maybe we need to make up something so that we avoid all the darkness because history is a very dark thing. Maybe that's the route that we want to go so that we avoid, like I was thinking of Julius Caesar or something like that. I'm sure Julius Caesar's done a bunch of horrible things and we might want to try to recreate that in a game. But it would be interesting to see if we can follow, like you are a leader of this civilization and like you get to choose, and then each game turn is like your advisors or like your people come to you as here are these choices that you have to take to take the next step in this, in the growth of this civilization.

Dan Shipper (00:31:01)

And I think that's great. I love that. And it would be fun. It's like, maybe you can pick any, like a couple of big historical turning points or events and see if you can do it better. It's like, you're, you're JFK deciding about the Bay of Pigs, whether to do the invasion and each turn, someone's coming to you and you have to make some decisions and you have to make all the decisions to see if you can make the Bay of Pigs—either just don’t do it or make it work out or something like that. Maybe we could pick a couple historic events where it seems like things went wrong. And it's like, can you do it better? Do you think you could do better than JFK?

Logan Kilpatrick (00:31:46)

I like that. I like that. I think the key for me is finding things that are uncontroversial enough, or ones that we both have enough information where I just have so little context on the Bay of Pigs, other than that it happened where I don’t know if I—

Dan Shipper (00:31:54)

Me neither. I'm sort of making that up, but yeah, let's find something that feels like it would be easy to do or not easy to do, but find something that feels like it would be good. And not get too dark.

Logan Kilpatrick (00:32:11)

Yeah, I'm curious if we can do—

Dan Shipper (00:32:13)

This is a good thing to ask ChatGPT about, by the way.

Logan Kilpatrick (00:32:15)

Yeah it is. I think the two that come to mind that would be really interesting is potentially some of the land purchases that the U.S. made back in our early founding history of play out the scenario of the world if we don't make certain land purchases. And, how does that structure, if we're sharing North America with other people or potentially the—I don't know if you've seen the show on Apple TV about what happens if the Russians beat the U.S. to the moon. But I think that could be an interesting way of playing out that scenario as well.

Dan Shipper (00:32:54)

I like that. So I think land purchases are a little bit colonial for me.

Logan Kilpatrick (00:32:56)

Yeah. I'm with that. I'm with that.

Dan Shipper (00:33:01)

It's giving imperialism. But the moon is kind of cool. I think the interesting thing is I feel like the way this game probably has to work is like you're doing—I mean, I guess we could say like, you're the director of NASA and, before the Apollo missions were the Gemini missions and Gemini, I think one of the later ones blew up and a bunch of people died and it was like, are we going to keep going? And maybe you're the director of NASA and you have to figure out how to get the Apollo program off the ground and get us to the moon or something like that. It could be kind of fun, but I really think we should be refining in ChatGPT.

Logan Kilpatrick (00:34:00)

Yeah, let’s do it. Let’s go in and start building. I was also thinking there's probably a separate game, which maybe we don't make in the context of this, but around deciding how the U.S. spends their budget every year and how potentially the different allocations lead to some very different outcomes.

Dan Shipper (00:34:21)

That’s really good. I love that. Would you rather do that? Which one are you more excited about?

Logan Kilpatrick (00:34:26)

Well, I'm happy for either. I feel like the U.S. budget one, who knows, maybe it could be very optimistic and very positive if we don't spend a ton of stuff on military budget, but I'm not sure. I feel like, yeah, potentially controversial, potentially not. I'm happy either way.

Dan Shipper (00:34:56)

Alright. You're going to need to make a call here. Cause I want to do whatever is exciting to you. So tell me which one—we're going to make you make a choice. Tell me which one you want to do.

Logan Kilpatrick (00:35:08)

Let's do: You’re president and you get to sort of choose the time period or choose the president. And then we'll have ChatGPT dynamically pull up like the budget of the U.S. government and the allocation of funds, and then let you sort of reallocate and then like play out American history potentially, with different allocations of budgets and see how that affects the outcome of the world.

Dan Shipper (00:35:32)

I love it. I love that. Should we throw it into ChatGPT and have ChatGPT refine it? Or do you want to go right into the building configuration-type thing?

Logan Kilpatrick (00:35:50)

Let's use the GPT builder and see if it's—I feel like I've been using this recently and it's been helping me get to the point of having something that's really interesting that we can always jump out and go into a different window and have it do something a little bit more verbose for us if we need to.

Dan Shipper (00:35:58)

So we are here in the GPT builder. We have created a new GPT and we’ve said, “We want to make a game. The core concept is that you get to choose which president you wanna be, and then the GPT will go and search the U.S. government budget slash distribute spending in that budget. Let the user reallocate the budget and then play out the world with that updated allocation.”

And, one of the things that I think it's interesting about this prompt that you gave me is we're not asking it to help us refine specifically. We're just letting it go straight ahead. And I think that that has to do with some nuances about how the GPT builder works and its instructions. I'm curious if you want to tell us really briefly about that.

Logan Kilpatrick (00:36:35)

Yeah, the GPT builder behind the scenes—and folks can see we've sort of opened the source instructions for the GPT builder itself, but it is doing a little bit of this refinement. We could probably and maybe we should try this and see if we can iterate on it once we've actually gotten what it outputted. But it does do a little bit of that prompt-engineering magic behind the scenes for us, which is nice so that you don't have to spend time doing that.

Dan Shipper (00:37:05)

I see. Cool. So if I went into the configure page, I would see the instructions that it wrote and it says, “The GPT's role is to assist in creating a game where players choose to be a U.S. president and reallocate the government budget. Based on the user's allocation, the GPT will simulate and describe the outcomes of these financial decisions. It should access and interpret data from the U.S. government's budget and spending, providing insights and hypothetical scenarios, the GPT must ensure accurate representation of data and responsible handling of historical and political content.” So what you're saying is it sort of rewrote it in an interesting way to create a game beyond what we said in our initial prompt, which is kind of interesting. And so now it says, “To start shaping this GPT, let's give it a name that reflects its unique role. How about Budget Commander?” What do you think? Budget Commander? Should we keep going? How are you feeling about it?

Logan Kilpatrick (00:37:52)

Yeah, I like a good one-word. If it can come up with something that's like, I don't know if it'll be able to do that or not, but—

Dan Shipper (00:38:05)

I'm saying I like a good one-word name. Can you find one for me?

Logan Kilpatrick (00:38:09)

Something not boring, hopefully. We'll see. Budget Commanders a little—

Dan Shipper (00:38:14)

Allocato.

Logan Kilpatrick (00:38:17)

Oh, I like Allocato. That's not horrible. That is not horrible.

Dan Shipper (00:38:21)

What about Allocator? Could be kind of interesting because I feel like Allocato is not necessarily—I don't think that's a real word. But the really interesting thing is, I just published this piece about how the knowledge economy is over and the allocation economy is beginning, where in a knowledge economy, your economic value is sort of pegged to what you know and being able to bring to bear you know to the task at hand. And in an AI world, it's a little bit more about knowing how to allocate resources, because in a lot of ways you're abstracting yourself out of certain day-to-day tasks. And so knowing how to allocate the resources of intelligence to do those tasks and how to edit them, or whatever, you become in a lot of ways a manager. And so you're learning how to allocate resources. So I kind of like Allocator because it's an unknowing reference to something that I've been thinking about a lot. So, I'm saying, “I like the name Allocator. Let's go with that.”

Logan Kilpatrick (00:39:27)