Dan Shipper and Logan Kilpatrick, OpenAI's first developer relations hire, use ChatGPT and GPT Builder to build a custom GPT—a text-based strategy video game called Allocator—in under an hour, live. Topics include how AI tools are enabling non-coders to build software, why learning to code is high-leverage, tips for using GPT Builder effectively, how to increase the chances a custom GPT goes viral, and combining ChatGPT with GPT Builder for better results. Timestamps cover the introduction, the first GPT Builder prompt, prompt engineering tips, gameplay, and iterating on the finished GPT. # Channel Identity Observations — "Build a custom GPT" Episode (Ep. 9 with Logan Kilpatrick) ## Observable Presentation Evidence ### Format: Live Co-Creation Show - The channel runs an **interview-plus-hands-on-build format**: a guest is welcomed ("Logan, welcome to the program"), then the pair builds a custom GPT live on screen. - The host explicitly frames the show's premise: "one of the really fun things about this show is creating something together or doing something with ChatGPT that makes us both think, 'Will this work?'" - Recurring structural segments are evident: a guest intro, a "how do you use ChatGPT in your daily life" segment ("I know we always like to talk about historical use cases"), a guest demo of an existing creation (Planty), and the main build. ### Intended Audience - **Non-expert creators and aspiring builders**, not just professional developers. Discussion repeatedly addresses people who "had never done any development in their lives" and the framing that custom GPTs let you build "without needing to know how to program at all." - The host describes his own audience-facing teaching: "I teach a chatbot class where people learn how to build one... It takes people who... had never done any development in their lives, to say, 'I built and launched an app in 30 days.'" - Audience scale is referenced: the host notes he writes "on shared screens that are recorded and broadcast to thousands of people." ### Claimed Scope - The show positions itself around ChatGPT/OpenAI tooling and the **"art of the possible"** — a phrase the host echoes approvingly: "I love the idea of the art of the possible. I feel that's exactly what ChatGPT opens up for so many people." - Guest Logan Kilpatrick describes the show's purpose as showcasing "part of ChatGPT's historic challenge: to show people the art of the possible." - Scope includes **practical prompt-engineering instruction**, e.g., naming advice ("if you're trying to build something that a lot of people are going to use, own the SEO space") and the Markdown-code-block copy-paste tip. ### Demonstrated Tone and Values - **Playful, humorous tone**: running jokes about Nicholas Cage cameos, "It's GPTs all the way down," and typo confessions ("I write like a maniac"). - **Iterative-build philosophy** stated explicitly by the host: "Instead of trying to make something perfect from the start, we just finished something. We took some notes, and now we're doing it again." - **Personal transparency**: the host shares he runs a small business — "I'm running a small business where I do a lot of writing. I run courses and we develop software products" — and mentions prior podcast guests (Tyler Cowen). ### Host Identity Signals (Demonstrated) - The host is named **Dan** (guest: "How's everything going, Dan?"; "Dan, when you woke up this morning..."). - He is a writer, course-runner, and software product developer with a personal history as a solo app developer ("That's how I paid for gas and food in high school" via a BlackBerry app). - He ends episodes with an open invitation: "Any final thoughts?" and plans follow-ups ("maybe we'll do a part two"). ## Explicit Self-Description vs. Demonstrated Patterns - **Explicit**: the show is about creating things with ChatGPT and showcasing the "art of the possible" (stated by both host and guest). - **Demonstrated (not stated as channel description)**: episode numbering ("Ep. 9" in title), guest caliber (first OpenAI employee on the program), educational-but-fun balance ("this podcast is kind of similar. Ideally, it's super fun to listen to and you end up learning something along the way"), and a live, unpolished co-creation style. # Delivery Observations: "Watch an OpenAI employee BUILD a custom GPT – Ep. 9 with Logan Kilpatrick" ## Format & Overall Delivery - **Interview/show format** with a host (Dan) and guest (Logan Kilpatrick), but the core of the episode is a **live, screen-shared co-building session** where both speakers interact with ChatGPT and the GPT Builder in real time. - The conversation moves through clear phases: **intro → guest's personal ChatGPT usage → demo of an existing GPT (Planty) → live build of a new GPT ("Allocator") → reflections and closing**. ## Tone - **Consistently enthusiastic and casual** throughout; the host repeatedly signals excitement: - > "This is really good." - > "I love it. I think it's a great little idea." - > "It's awesome." - **Playful, joking register**, especially in the closing: > "Dan, when you woke up this morning, you didn't think Every was about to become a AAA game studio." ## Recurring Phrases & Verbal Habits - **"I love it"** — repeated by the host many times: > "I love Planty," > "I love it. I love it," > "I love the name allocator." - **Short affirmations strung together**: > "Yes. Yes, 100%," > "Great. Great. Let's do it," > "Okay, great. Let's do it." - **Self-deprecating framing of his own writing/spelling**, repeated in a short segment: - > "I write like a maniac." - > "I have very bad spelling, so I need to... I need to do it." - Frequent **filler and hedging** around proposals: > "uh, I'm wondering," > "I'm curious," > "I feel like." These appear across the entire build segment. ## Teaching / Communication Habits - **Thinking aloud while prompting**: the host narrates his prompts before submitting them, e.g., > "So what I'm writing is that I feel like it needs to do both things." - **Explaining intent behind prompts** — a recurring instructional move, as when Logan asks why he wrote "we need your expert knowledge": - > "I think it's literally about empowering the GPT... to take initiative... because, by default, the answer you're going to get would be, well, very basic." - **Practical tips dropped inline**, e.g., on naming for discoverability: > "the more generic you make the name, the more saturated it will be," and on Markdown: > "Markdown code block syntax is definitely the trick." - **Iterative method explicitly summarized** by the host near the end: > "we just finished something. We took some notes, and now we're doing it again." - **Keeping a running "notes" notepad** during the build to collect refinements: > "So, I'll just say that... I'm just noting this down so we can refine our instructions later." ## Questioning Patterns - The host **defers decisions to the guest** repeatedly: > "You're going to have to make a decision here, because I want to do whatever it is that excites you." - He frequently asks **"what do you think?" / "How do you feel?"** checkpoints: > "What do you think? Good, I think it's great." ## Guest's (Logan's) Delivery Patterns - **Analogy and concrete example driven**, e.g., the DevDay "Chatty" customer story to explain Planty's origin, and the Algorithma story (father and daughter) to illustrate accessibility. - **Frames AI accessibility as a theme** in multiple segments: > "this technology will be the catalyst that allows the next billion people to learn to program." - Ends with a **call to action**: > "keep creating GPTs... now's the time to get started." ## Uncertainty Notes - Speaker attribution is not labeled in the transcript; role assignments (host vs. guest) are inferred from contextual cues (e.g., "Logan, welcome to the program," "Tell us a little bit about custom GPTs") and could occasionally be ambiguous. - The transcript appears to be a **translation** (e.g., "lyre fig" for "fiddle-leaf fig," "Chachi Boutique," "navigation" for "browsing"), so exact original word choices may differ from the quotations above; quoted evidence reflects the transcript as supplied. # Position & Reasoning Record — "Watch an OpenAI employee BUILD a custom GPT" (Ep. 9 with Logan Kilpatrick) ## Speakers - **Logan Kilpatrick** — developer advocacy team, OpenAI (identified in transcript; first OpenAI employee on the program). - **Dan** — host, runs a small writing/courses/software business (Every), teaches a chatbot class. --- ## 1. Custom GPTs and Accessibility **Logan's claims:** - Custom GPTs are "**probably the biggest advance since ChatGPT itself came out**." - OpenAI deliberately uses the term **"creators"** rather than "developers" so people don't self-exclude: *to prevent people from hearing the word developer and getting scared or thinking "that's not for me."* - You can build something "materially much more useful than the base ChatGPT" **without writing any code**; tools like Zapier offer no-code integrations. - On the GPT builder abstraction: "when you use the GPT builder, it doesn't feel like you're using a GPT" — which he attributes to the **general reasoning engine** capabilities of LLMs. **Dan's claims:** - ChatGPT is expanding who counts as a developer; he personally uses a **dream-interpretation GPT** and is building one that saves chat notes to Notion via API (stated he "got a little stuck"). - Repeated the phrase "GPTs all the way down" — stacking AIs (a GPT builder that builds GPTs) compounds speed of building. ## 2. How Each Uses ChatGPT **Logan:** - Uses ChatGPT daily for programming despite a CS background — he wasn't formally trained as a web developer and asks "really silly questions" about Next.js/TypeScript. - Claims features on OpenAI's API platform (comment thumbs up/down, dynamic outage pop-ups) were **built with ChatGPT's help**. - On context windows: pasting 300 lines of code lets ChatGPT "solve the problem better than I can, honestly." **Dan:** - Frames most development as **summarization** (API docs, Stack Overflow answers) and "stacking bricks that have already been laid" — where ChatGPT shines; genuinely novel programming is where it's less effective. ## 3. Future of Programming — Explicit Disagreement Surface **Logan:** - "Learning to program is still definitely relevant… if you're looking for the greatest possible advantage in your life, learn to program" — holds for the **next five to ten years**. - People will solve problems **10–20x more complex**; he personally can now solve problems "10 times more ambitious" than two years ago. - **Tension he flags:** AI can help close the programmer/non-programmer gap, *but* no-code tools will always have limitations — "you need to understand how the systems work." **Dan:** - Agrees; adds that AI both speeds up skilled people *and* enables projects whose entry barrier was previously too high. - His chatbot-class students with no dev background "built and launched an app in 30 days." - Motivation argument: ChatGPT reverses the learning order — build first, then learn fundamentals when errors force it. **Logan's strong closing claim:** LLMs "will be the catalyst that allows the next billion people to learn to program," because there weren't enough educators/resources to teach them otherwise — "perhaps that will end up being the most important benefit of this technology." ## 4. Planty GPT Demonstration **Logan:** - Built **Planty** (pun on the DevDay customer who called ChatGPT "Chatty"); idea credited to his girlfriend, a plant lover who used it to solve a monstera root problem. - Position on GPT vs. plain ChatGPT: "behind the scenes… it's still inherently ChatGPT" — a GPT adds curated sources, safeguards, and personality, not new capability. - Discovery mechanism prediction: **social signals** will guide which GPT to use (e.g., "100,000 people said Planty solved their problem"). **Unverified-content moment:** Planty recommended Fort Greene plant services (Natty Garden, Greenery Unlimited); Logan **could not verify them live** ("No, I don't know"). Dan later found Natty Garden is real. The Sill was verified via a personal contact. — *Qualification preserved: recommendations carried no live verification.* **Dan's on-air finding:** his fiddle-leaf fig had literal roots growing out of the pot — Planty's repotting criteria matched reality. ## 5. Building "Allocator" (Budget Game GPT) **Stated design positions:** - **Dan:** core of a game must be **fun**; education is a byproduct — "we don't want to force the educational content because it gets boring." - **Dan** raised a historical-difficulty worry: history is "very dark"; also rejected the land-purchases idea as having "**imperialist undertones**." **Logan agreed.** - Chose instead: play as a US president reallocating the era's actual federal budget, with simulated outcomes. **Observed builder behavior (both speakers, recorded as claims):** - GPT builder does "instruction engineering magic" behind the scenes (Logan); its instructions stay **general rather than tactical** — both noted the resulting preview "isn't a game" yet. - Logan's technique: **empower the model as an expert** ("we need your expert knowledge…") — a recommended prompt-engineering pattern. - Dan's technique: request output as **Markdown in a code block** for copy-pasting; also naming advice from Logan — **own the SEO space**, generic names get saturated. - Iteration loop both endorsed: build basic → playtest → take notes → paste notes + instructions into ChatGPT → regenerate. Dan: "instead of trying to make something perfect from the start." **Gameplay critiques logged (both):** - Too verbose; repeats "the future is in your hands" every turn. - Didn't show budget figures until explicitly asked; browsing must be explicitly instructed. - Missing: slash commands, concise metrics (e.g., "GDP +1%"), **failure states and randomness**. Logan: it "is always going to be positive about the things you do" — a big missing piece; suggests dice-roll/probability via **code interpreter** (which wasn't enabled). - DALL-E integration had to be manually added to instructions. ## 6. Closing Positions **Logan:** - **GPT monetization "is just around the corner"** — building now builds an audience; this will "increase the hype surrounding GPTs tenfold." (Forward-looking claim, not a settled fact.) - Democracy-of-creation claim: you no longer need to "raise millions of dollars" to make a great game. - Reflection: co-creating with **AI + another human** is more fun than working solo. **Dan:** - Games were a childhood dream blocked by the required programming/math/art/music stack; he pivoted to BlackBerry/iPhone apps ("Find It," pre-iPhone "Find My iPhone"-like app, funded gas and food in high school). - Now feels "games and movies… are so possible now" — the session "reawakened my game-loving nature." - Plan: release Allocator, gather beta feedback, add dice rolls, ship a 2.0 in a follow-up episode. --- ## Contradictions & Unresolved Tensions (preserved) - **Closing the gap vs. needing to code:** Logan simultaneously claims AI closes the programmer/non-programmer gap *and* that no-code tools have inherent limits requiring system understanding. - **Presidential framing error (Dan's own note):** the game's premise has the president allocate the budget, but Dan observed "Congress definitely has to at least approve the budget" — the premise's inaccuracy was noted and left standing. - **Optimism bias of the model:** the simulated outcomes were repeatedly utopian; both flagged this as unresolved pending a randomness mechanism. - **Speaker-identification caveat:** speakers are identified from transcript context (host named "Dan," guest self-identified as OpenAI developer advocate Logan Kilpatrick); no curated metadata was supplied. # Speaker Attribution Record **Video:** *Watch an OpenAI employee BUILD a custom GPT - Ep. 9 with Logan Kilpatrick* **Video ID:** 01a0f15b-3407-78e4-9a1c-de04fa55d0bf ## Speakers ### Host — "Dan" - Identified by name in dialogue ("How's everything going, Dan?", "Dan, your financial future is in your hands"). - **Role:** Creator / program host. Runs the show, interviews the guest, shares screen during the build, narrates game output aloud. - Self-described background: runs a small business doing writing, courses, and software products; teaches a chatbot-building class; previously did solo app development (BlackBerry/iPhone) in high school. - Sits in **Fort Greene, Brooklyn** (stated on-air). ### Guest — Logan Kilpatrick - **Role:** Guest; member of the **developer advocacy team at OpenAI**, stated as the first OpenAI employee to appear on the program. - Focus: enabling developers to build with the OpenAI API and ChatGPT. - Built and demonstrates the custom GPT **Planty**; co-builds **Allocator** with the host. ### AI-generated voices (non-human speakers) - **ChatGPT / GPT Builder / Planty / Allocator:** machine outputs read aloud or shown on screen, including the opening cold-open game narration. Not to be attributed to the human speakers. ## Referenced Sources & Third-Party Claims - **Zapier** — cited by Logan as offering a no-code integration guide for custom GPTs. - **OpenAI DevDay (November)** — customer video referenced; an elderly man calling ChatGPT "Chatty," inspiration for "Planty." - **Tyler Cowen** — host states he discussed productivity/AI with Cowen "on the podcast last week." - **Algorithma** — plugin-era game; Logan cites a father-daughter team (~5 million users claim, attributed to the father's account). - **Planty configuration** — Logan says it draws on "specific websites... previously verified as authoritative sources." - **Games cited as inspiration:** Civilization, Age of Empires, Empire Earth, MUDs, National Treasure (film), an Apple TV show about Russians reaching the Moon first. - **Local businesses named by Planty during demo:** Natty Garden, Greenery Unlimited, The Sill, TaskRabbit, Thumbtack. Host and guest note uncertainty about which are real; The Sill confirmed real by host; Natty Garden verified via search during the episode. - **1961 US federal budget figures** — retrieved by the GPT via web browsing (Bing) on-screen during the demo. ## Promotional / Sponsored Segments - **Explicit non-sponsorship statement:** "This episode isn't sponsored by Natty Garden, but it looks like an amazing local community shop." - **OpenAI product promotion:** Logan promotes custom GPTs, the GPT Store, upcoming **GPT monetization** ("just around the corner"), and encourages viewers to start building — flagged as promotional content delivered by an OpenAI employee. - No other sponsor segments identified in the transcript. ## Uncertainty & Availability Notes - Speaker names/roles are inferred from in-transcript dialogue only; **no curated metadata was supplied**. - Comment availability: **not stated** in the source material. - Episode is part of a numbered series ("Ep. 9"); future Part 2 with an "Allocator 2.0" was discussed but not confirmed as produced. Let's brainstorm what game you want to create. It would be interesting to see if you are the leader of this civilization and each turn your people come to you with choices that you must make for the next step in the growth of this civilization . As President Kennedy, your challenge is to balance the budget with the urgent needs of a nation on the brink of major social and technological changes. Will you prioritize space exploration, increasing NASA's budget to beat the Soviets to the moon landing , or will you focus on domestic issues, ensuring quality and prosperity at home? The decision is yours. This is really good. Logan, welcome to the program. How's everything going, Dan? I'm really excited about this. I'm excited too. We've been following and interacting on Twitter for a long time, so it's great to meet you face to face. For those who don't know, you're a member of the developer advocacy team at OpenAI, the first OpenAI employee we have on this program. Woo! How exciting! And as part of that, you focus on enabling developers to build with the OpenAI API and ChatGPT. And yes, I'm very happy to have you. Yes, this is going to be amazing. And I'm so excited that you've put together this program to showcase part of ChatGPT's historic challenge: to show people the art of the possible. So what you're doing makes life easier for all of us, and we're also doing a lot of things from a product perspective to further this. But I'm glad you're also pushing this from this angle. It's an important job. Thank you. I really appreciate that. I love the idea of ​​the art of the possible. I feel that's exactly what ChatGPT opens up for so many people, and for me personally. So that's really great. We were talking a bit on the pre-production call about what we want to cover, what's most interesting, and I think one of the things that caught my attention is that your role is to be an advocate for developers. But historically, developers have been a small component of the population. And , well, what's really interesting about ChatGPT is that it's creating a world where the number of people who can become developers, who can develop things, is expanding tremendously, both because you can actually program underlying applications with ChatGPT and because of what they just launched, custom GPTs, which allow you to create tailor-made versions of ChatGPT for specific use cases. Tell us a little bit about custom GPTs. Tell us why and why you're so excited about them. Yes, being able to create GPTs is probably the biggest advance since ChatGPT itself came out. In fact, we classify everyone who creates GPTs as " creators," to prevent people from hearing the word developer and getting scared or thinking, "Oh, that's not for me, I don't fit that profile." And we provide this flexibility to use the GPT creator, using custom settings when creating a GPT, without needing to know how to program at all. If you know how to program, there are many things you can do, but you can create something that is materially much more useful than the ChatGPT base, without writing a single line of code. Furthermore, there are now companies like Zapier, which is a great example of this, that have a whole guide on how to integrate their tools into your custom GPT without having to write any code. I think that's a big step forward, and I think it's the direction we're heading in with GPTs: enabling people to do so much more without having to write the code themselves these days. Yes. No, I think that's really great. I've been using it. I have a GPT specifically for dream interpretation. So, when I wake up, I have a little dream. I put it there and it just gives me a little bit of analysis, then it makes an image with DALL-E and it's really great , I use it a lot. But I have several others . In fact, I was working this weekend on one that I want to be able to chat with and that then saves my chat notes in Notion. And I'm connecting it to the Notion API. I got a little stuck on that, but one of the things I think is great is that you mentioned the GPT builder, which is a GPT that allows you to create GPTs. So it's like they're GPTs all the way down . I think it's a very important thing that people should notice and understand, because when you start stacking artificial intelligences, you gain an extra advantage, where you now have a GPT builder that allows you to create GPTs, which allows you to build more things faster than usual. Yes. And it's interesting, because when you use the GPT builder, it doesn't feel like you 're using a GPT, which is curious. It's amazing how, by abstracting the interface where people are very aware that they are using a specific tool, you also lose the fact that it's AI through and through. It feels like any other system we've built, which speaks to the capabilities of the general reasoning engine that large language models have , which is very interesting . Yes, that's great. So, let's build something together, let's create a custom GPT, which I think will be a lot of fun. Before that, I know we always like to talk about historical use cases or simply how you use ChatGPT yourself, and I know there's another custom GPT you might want to talk about. But anyway, in general, tell us how you use ChatGPT and custom GPTs in your daily life and at work to start. Yes, I think the most useful thing for me, as someone who still does engineering on a daily basis, is programming things. I don't have one... I know there are many very useful GPTs for programming and creating things. Most have historically focused on the "low code" type, like "give me a one-line instruction and I'll build this for you." And that's not the use case I'm working in. In other words, I'm not an expert in Next.js, and, well, I'm working with a Next.js or TypeScript codebase and, well, I'm trying to build something or add a new function, and, although I have a background in computer science, I didn't formally train as a web developer. I never did that as a job. So I always ask ChatGPT really silly questions and, well , you'd be surprised by the number of features on our API platform that were created because ChatGPT helped me; The thumbs up or down for comments is a great example of this. When there's an API outage and a small pop-up window appears dynamically, all of that was built by ChatGPT itself, which is great. Yes, that, I think, well, I think that makes a lot of sense. And, well, it fits perfectly with something I've been writing or thinking about a lot: ChatGPT's ease of summarizing. And once you start to see how good summarizing is, you see summaries everywhere in your life and in everything you do. And I think that's also true in development, where people would be surprised if you're not a developer. See how often being a developer is simply summarizing API documentation or summarizing a Stack Overflow answer in your codebase or whatever. And it sounds like that's a big part of what you do, since there are parts of the programming that are really new, things that nobody has done before. And in that case, ChatGPT can be useful for reflecting on it, but it won't be as effective. But most of the development is simply stacking bricks that have already been laid before, but in a slightly different configuration for your specific use case . And that's where I think it shines so much, because it's very good at knowing what people have done in the past and writing a little bit for your specific situation. Yes, it has also been a great help to me. When looking at, for example, 300 lines of code, well, I understand more or less, I read the code that is right around what I am trying to add or modify, but, well, in my personal context window, I don't have all 300 lines of code. Hmm, and it's very helpful when I take that and bring it to ChatGPT and say, "Here's all the code." ChatGPT now has those 300 lines of code in its context window and can solve the problem better than I can, honestly, because it's hard to take 300 new lines, understand them, and conceptualize how to modify them. That makes sense. I suppose I should ask, because you're in a very interesting position at OpenAI, observing all the progress, with access to internal resources, and being a developer yourself. How has using ChatGPT, creating custom GPTs, and seeing the internal progress at OpenAI changed your view of the future of programming? Is learning to code still relevant? How do you think this will look over the next year, the next five, and beyond? Yes, learning to program is still definitely relevant. I believe that if you're looking for the greatest possible advantage in your life, learn to program. I believe this will continue to be the case for the next five to ten years. My expectation is that people will be able to solve problems 10 to 20 times more complex, following this exponential curve of difficulty in the products and services you can build on your own. For me, today I feel that if you took me back two years , I could now solve problems 10 times more ambitious. I have an idea in mind, for example, I want to build an application for a restaurant. Now I can go and build it myself. Two years ago, I probably would have thought it was too much work. Perhaps I would have hired someone to help me do it. I believe that, by continuing to climb that exponential curve, I will soon be able to build an entire empire of products and services for my company all on my own. Yes. I do n't think three years ago people would have imagined they could do that . That's why I still think that software development today is one of the most advantageous activities you can do. But in 2 to 3 years and in 5 years there will be even greater leverage and the gap between people who know how to program and those who don't will continue to grow. And I think the good thing is that AI can help close that gap, but there will always be limitations in no-code tools in terms of what you can build. In other words, you need to understand how the systems work. Yes, that makes a lot of sense. I mean, you speak my language here, because, well, I'm running a small business where I do a lot of writing. I run courses and we develop software products, and it 's a small, cost-effective, and efficient team. And what I love so much about tools like this is that we can go very far with the small group of smart people we have, because now we all have ChatGPT. And I think you're absolutely right. It reduces the obstacle to starting something so much that, for me, a project I would never have attempted before because I only had a couple of hours, I can suddenly do it and launch it into the world. And that 's the best thing in the world. I think that's something people don't quite understand. In fact, I talked a little bit about this with Tyler Cowen on the podcast last week, and we discussed it: for really skilled people, it can increase your productivity, being a little bit faster, but it also allows you to do things you would never have done before because the entry level was too high. There's a driving energy to carry out projects, and I think that's incredibly cool. And you're also right that it does something where, if you don't know how to program at all, you can suddenly create things. It would be better if you understood the underlying technology, and that will work. But what's really cool is that I teach a chatbot class where people learn how to build one. It's not with GPTs. That is, creating everything from scratch, making a web application and all that kind of thing. But you use ChatGPT to help you program. And it takes people who, you know, had never done any development in their lives, to say, "I built and launched an app in 30 days." And it's great to see them being able to create things, because they all have ideas, but they've never been able to bring them into the world before . And ChatGPT becomes what allows them to do so. Furthermore, it gives them a lot of motivation to learn the fundamentals, because before, if you wanted to learn to program, you spent 6 months learning "if" statements and " for" loops, thinking: "How does this translate to the application I want to create ?". What ChatGPT does is that on the first day you create what you want, and on the second day you run into errors and think: "well, I guess I have to figure out what an if statement is." You have a lot of motivation to do it, and ChatGPT explains it to you; It's really great. That resonates a lot with me; When we finish recording, I'm going to save this clip and listen to it every month, because this technology will be the catalyst that allows the next billion people to learn to program. There weren't enough educators and computing resources on the planet to teach programming to the next billion people. It simply wasn't going to happen. It's incredible to think that, thanks to LLMs, people will actually be able to do it in a personalized way that specifically empowers them. Perhaps that will end up being the most important benefit of this technology, beyond everything else it can do. I think so. Yes. It expands access to the ability to create in a meaningful way, and I think that's the coolest thing about it. Hm. It's lovely. Yes. And with that, perhaps we can begin to build something. I think you wanted us to know before we started building something together, which is what we're going to do. I think you wanted to show me a custom GPT to whet our appetites. Show me a GPT that you have built. Tell us, uh, why you created it, what it is, and why you think it's interesting to see. Yes, one hundred percent. Let me, let me share my screen and then...Okay . So, we're in the GPT Store . Uh, I was lucky enough to be able to build Planty, which is a, a little pun. If people watched OpenAI DevDay in November, there was a video of a customer talking about how they used ChatGPT and how it had impacted their lives. And there was an incredibly sweet old man who appeared, uh, with a very grainy video, sitting in his office. He talked about how ChatGPT had unlocked all these use cases, and referred to ChatGPT as "Chatty". It was great to see that, uh, yeah, to see how he got that value and had that relationship with, uh, with this tool. So Planty is a little play on words with that. And, uh, the whole concept literally came about sitting on the couch with my, uh, with my girlfriend, thinking about what cool things we should build for people. Uh, she's not an AI expert, but she loves houseplants , gardening, and all that. The first thing she said, so all the credit goes to her, uh, was that we should create Planty. It's designed as an, uh, indoor plant caretaker; She has a monstera and needed to do something with the roots, uh, that was beyond my technical knowledge, and she came in and, while we were creating Planty, she used it to solve that problem. I gave my older brother a lemon tree for Christmas and as soon as it arrived at our apartment, all its leaves fell off. He looks very desolate. So, uh , I posted a tweet saying that if Planty can save this lemon tree, uh, I'm a big believer in AI and you can actually do things like just ask simple questions. Uh, or get really tactical and upload pictures saying, you know, this is the color of my plant's leaf. Hey, why, why is it like this? And it really opens up possibilities, again, like Chachi Boutique could do some of these things, but by providing a number of safeguards to allow it to do so and, you know, giving it this personal and friendly touch, I think it really brings to life what 's possible with this GPT. Yes, tell me more about that. How would I know whether I should use a GPT or create one instead of just using the conventional ChatGPT? That's a great question. I think this is one of the complicated dilemmas of today. I believe that, in time, the way to answer this question will be through social signals. I think people will probably invent a lot of, yes. If you have a question about plants and 100,000 people have said that "Planty" solved their problem, you'll probably go there because it's what socially seems like you should do. I think the interesting thing for people to keep in mind is that, behind the scenes and everything, it is still inherently ChatGPT that answers these questions. So it's not a completely new capability. I think what we did with Planty was base it on very specific information. If you need to look something up, go to a number of specific websites that we have previously verified as authoritative sources, rather than, I don't know, sellplants.com, which might have a biased perspective. So there are some nuances in how it's being built. But again, fundamentally it's just ChatGPT behind the scenes with some layers of abstraction on top. That's great. In fact, I have a lot of plants, as you can probably see. I have a small fiddle-leaf fig in my apartment and, in fact, I have a random question that might be worth asking. Hey, I can't see your screen anymore. Hmm. Yes, I just switched to Planty. OK. I'd love to see the question. Let's do it. I'm curious. I'm curious to know what suggestion Planty has. Here's the thing. I have this fiddle-leaf fig. Fiddle-leaf figs are supposed to be very delicate and difficult to care for. But I've had it for about four, five, or six years, or something like that. And I simply water it about every week and a half or so, and the truth is that it's doing quite well. Every now and then a leaf falls off and I get scared, but then everything is fine. Uh, and uh, but one of the things I don't know is that I think these things are supposed to be transplanted at some point, but I don't know when to do it or how to do it. And I wonder if I'm artificially limiting my plant and should repot it. Uh, and how would I know? Did you say it was a fig? Lyre fig . Lyre fig. How and when should I transplant my lyre fig? And so you correct your typos in Chat GPT because I never correct mine. I write like a maniac. Is that a bad thing? I write like a maniac, but it's more a matter of habit, because if I stop correcting my mistakes here, I'll stop doing it everywhere else and people will get mad at me for writing like a maniac on Slack or wherever. That's very funny. It's fun doing this show because I write a lot on shared screens that are recorded and broadcast to thousands of people, and it seems like I'm in third grade or something, and I tell them: I swear I do know how to write. I simply don't do it for ChatGPT. Yes, I have very bad spelling, so I need to... I need to do it. Then I asked the question of how and when I should transplant my lyre fig . So, what are the disadvantages of not transplanting it? I think mentioning the artificial limitation... Um, so it's interesting when to transplant. Then he said that transplanting your lyre fig can be quite an adventure. Is that part of your personality? What did you give him? Yes it is. It's like, I think it's supposed to be something friendly and exciting that helps you, instead of, I think by default ChatGPT is pretty generic when trying to give you this information. So I think it gives it a bit of life . Okay, great. And if you go up, uh, it says, " Repot your lyre-leaf fig ideally during the growing season, which is spring or early summer . Uh, take a look at the drainage hole at the bottom of the pot. If you see roots starting to circle around or poke out, it's time for a bigger home . I mean, I'll check that quickly. Uh, give me a second. We'll see how it goes. I'm really curious. Uh, I see roots starting to circle around or poke out. Wait. There are literally roots coming out the side of the pot. It's not like they're just poking out. It's like they 've grown out and are basically reaching almost to the soil. So , I think I need to repot it. I think that's your cue. Yes. Thanks, Planty. We figured it out. How to repot it. This is perfect. Yes. Find a pot that's about two or three inches bigger in diameter than the current one. Make sure it has good drainage holes. Add some fresh potting soil to the bottom of the new pot. Fiddle-leaf figs like good drainage. Remove the plant Carefully. Check the roots and place your fiddle-leaf fig in the new pot and fill in around it with more soil. What I want to know—this is my last question—is that I'm super lazy. Um, and I'm not going to calculate the pot diameter or do all this. Is there a way someone can do this for me? Where do you live? I'm in Fort Greene, in Brooklyn. Greene, you said. Yes. Look, there are my recommendations with typos. What a horrible word to type. So, you said, are there any services or ways someone can do this for me? Any specific recommendations for someone in Fort Greene? And Planty says, absolutely. Getting help repotting your plants is a great idea, especially if you don't want to do a project yourself or if your fiddle-leaf fig is too heavy for you to handle alone. Definitely too heavy for me to handle. Um, in Fort Greene you're lucky because it's a vibrant area. Definitely. Thank you, Planty. With lots of plant-loving people and services. And she gave some, I would say, generic recommendations. like local nurseries and garden centers, though she did give a couple of specifics, mentioning Natty Garden or Greenery Unlimited. Um, are those real places? I do n't know. No, I don't know. I'm not up to date on local plant care services in Fort Greene, but I'd love for you to check it out. Well, I'd also recommend task websites like TaskRabbit or Thumbtack. Or plant care services like The Sill, which I definitely know is real. A friend's wife runs that company. So, I definitely got some good recommendations. Maybe, you know, when I finish this episode, I'll take care of it and we'll do a part two where I update you on the progress of my fiddle-leaf fig. But it seems So, it seems you looked it up, and Natty Garden is real. It's a real place. It looks nice. I kind of want to go to Natty Garden. It seems like it'll be a good time, regardless of whether you manage to transplant your plant there or not . Great. Well, I'll be visiting Natty Garden after the episode. And I 'll send you pictures. This episode isn't sponsored by Natty Garden, but it looks like an amazing local community shop. So, cool. That's really nice. I love Planty. I think it's a great little idea, and I think my fiddle-leaf fig is probably pretty happy that you showed me that too. But now I want us to get to the main event. You know, I told you that one of the really fun things about this show is creating something together or doing something with ChatGPT that makes us both think, "Will this work?" " And that's kind of the magic of it all. So you proposed creating a game in the GPT Builder. And I wanted to start with, why did you propose that? Why do you think it's exciting? Yeah, I think if you think about how you would have had to do something like this before GPTs, before ChatGPT, even just before GPTs , just with ChatGPT and the plugins . I've been thinking about this idea for a while now, like, I'd love to guide people through, you know , if you've played Civilization or Empire Earth or any of those games that are based on situations where you build an empire over time. Well, I've always found them fun and they really bring history to life , especially if you can add real facts about things that happened. Uh, and without GPTs this was impossible to do; you would have had to, I as a developer or user would have had to pay to generate these images, each individual image would have cost me money. Uh, all these other things that would have limitations to building it. It's just, it's actually quite difficult to do and It requires a considerable amount of work. I think the great thing about GPTs now is that you can use the GPT builder. You can have it handle a lot of the nuances of game design, and, uh, we can use built-in DALL-E. We can use the built-in navigation to get real-time data and keep track of things for you. Uh, potentially use the code interpreter to do some actual simulations of, I don't know. I don't know if we can simulate battles or statistical probabilities of things, but it's amazing what will be possible with this. So I'm really excited. And it's also one of the areas that I think, specifically because there isn't a category right now . Yeah. It's a little unexplored. I mean, there really aren't, uh, you know, if you look at the trending categories in the GPT store, there really aren't a ton of games there, if any . So I think it's an opportunity for us. That's really interesting. I wonder why, because I know that back in the '70s and '80s, during the first PC revolution, MUDs, where everything was Text-based games were super, super popular, and it seems like an obvious thing to try. I wonder what, uh , what's stopping people? Uh , have you seen any? I've seen a few. So, the most interesting one was this game, which was one of the most popular plugins, and it was called Algorithma. And it was, uh, this guy and his daughter—I'm pretty sure the daughter learned to program to build Algorithma—and they ended up scaling it to the point where they had like 5 million unique users or something, or 5 million users over the course of a few months, uh, who used Algorithma. And it's like, it's really like a little girl who built this, and, well, it was really fun to play. It was crazy. And I talked to the dad, and he was saying how they used, how she was using AI and ChatGPT to help her build it. So it's a great example of the theme of people feeling capable of building cool things. Yeah, I think that was probably the most important one. I haven't seen It's a GPT, but I guess it's doing a lot of really cool stuff. Got it. That's awesome. So, an opportunity, a free slot for us. Maybe, maybe this is the next big thing. Maybe we'll do it live. So, I think to start, let's brainstorm, what kind of game do you want to create? If you have a couple of ideas, throw them out there and I can help refine them. Yeah, I think that would be great, and I think this is the decision we need to make: do we want it to have a purely fun and exciting feel, or should there be an educational component? Do we want people to be able to learn about history as they go along? I think there's a balance in both cases. In a sense, people just want to escape and play something fun, but there could be some value. Maybe we could have a selector where we let people decide at the beginning of the game: are you here for fun or to learn things as you go along? Here's what I think. I don't like it, I think the core of a game or most content has to be about fun. It's about creating an awesome experience, and if You learn something along the way, I think that's fantastic. This podcast is kind of similar. Ideally, it's super fun to listen to and you end up learning something along the way, but we don't want to force the educational content because it gets boring. But I think what I hear you saying is that history is something interesting to you and that having a part of this use history in some way would be fun. Um, do you have any specific ideas about how to incorporate history into a game like this? Yeah, I'm curious if you're thinking, maybe history is too dark for this and it won't work, but I was thinking about playing as different world leaders or something. And maybe we should come up with something to avoid all that darkness, because history is a very dark thing. Maybe, maybe that's the way to go, to avoid, you know, I was thinking about Julius Caesar or someone like that, he certainly did a lot of horrible things and we could try to recreate that in a game. But it would be interesting to see if we can, like, follow the role of a leader of this civilization and you choose, you know, you take all the...and then, each turn of the game, your advisor Or your people come to you with choices you have to make for the next step in the growth of this civilization. And I think that's great. I love it. And it would be fun. It's like you could pick a couple of big turning points or historical events and see if you can do better. Like you were JFK deciding about the Bay of Pigs, whether or not to go through with the invasion. And on each turn, someone comes to you, and you have to make decisions, and you have to decide, make all the choices to see if you can make the Bay of Pigs thing—either not do it or make it work, or something like that. Maybe we could pick a couple of historical events where things seem to have gone wrong and see if you can do better. Do you think you could do better than JFK or something? I like that. I like that. I think the key for me is finding things that are uncontroversial enough or that we both have enough information about, because I have very little context about the Bay of Pigs beyond the fact that it happened, and I don't know if I... I don't either. I was just making it up. Yeah, let's look, Let's look for something that feels, uh, that feels like it would be easy, easy to do, or not easy to do, but look for something that seems good and that isn't too obscure. Uh. Yeah, I'm wondering if we can, uh, if it's a good thing to ask, by the way. Yeah, this is, I think the two, the two that come to mind that would be really interesting are potentially some of the, uh, like the land purchases that the United States made in our founding history, whether you should, what if you recreate the world scenario if we don't make certain land purchases and how that structures, you know, we're sharing North America with other people, or potentially, I don't know if you 've seen that show on Apple TV about what happens if the Russians beat the United States to the moon landing. But I think it might be interesting to recreate that scenario as well. I like it. Uh, so I think the land purchases are a little, I don't know. They seem a little colonialist to me. Yeah, I agree. I agree . They have imperialist undertones. Uh, but, Uh, the Moon is great. I think what's interesting is that I feel like the way this game probably has to work is that, well, I guess we could say you're the director of NASA. And, uh, you know, before the Apollo missions, uh, there were the Gemini missions. And Gemini, I think one of them, one of the last ones, exploded and a lot of people... And it was like, "Are we going to keep going?" "And it's like, maybe you're the director of NASA and you have to figure out how to get the Apollo program going or something and get us to the moon or something. It could be, it could be fun. But I really think we should be refining ChatGPT. I think it would be really useful. Let's do it. Yeah. Let's get in there and start building. Well, I was also thinking there's probably a different game, which maybe we won't do in this context , but it's about deciding how the United States spends its budget each year and how different allocations could lead to very different results. Yeah. Like, I ... I love that. Would you rather do that? Which one excites you more? I'm fine with it , I'm happy with either one. I feel like the US budget one, who knows, maybe it could be very optimistic and positive if we don't spend a ton on the military budget, but I'm not sure. Like, yeah, potentially controversial, or maybe not. I'm, uh, I'm happy either way. Okay. You're going to have to make a decision here, because I want to do whatever it is that that excites you. So , tell me which one we're going to do, I'll let you choose. Tell me which one you want to do. Yes, let's do the one about your president. Uh, and you can choose the time period or choose the president. And then we'll have ChatGPT dynamically get the US government budget and funding allocation, and then we'll let you reallocate and develop US history with different budgets and see how that affects the world outcome. Let's do it. I love that. Uh, should we...? Yes. Should we throw it into ChatGPT and have it refine it, or do you want to go straight to the build settings? Let's do it, let's use the GPT builder and see if... I feel like I've been using it recently, and it's helped me get to the point of having something really interesting, and we can always step out, go to another window, and make it a little more detailed if we need to. Okay, we're here in the GPT builder. Uh, we've created a new GPT, and we said we want to make a game. The concept The core idea is that you can choose which president you want to be, then ChatGPT will look up the distribution of spending in the US government budget at that time and allow the user to reallocate the budget and then develop the story with that new allocation. And, um, one of the things I find interesting about this instruction you gave me is that, um, we're not asking it to help us refine specifically. We're just letting it go straight ahead . And I think that has to do with some nuances about how the GPT constructor and its instructions work. I'm curious if you'd like to briefly tell us about that. Yes, the GPT constructor behind the scenes—and people can see that we've, um, opened up the source instructions of the GPT constructor itself— but, um, it does do a little bit of this refinement. We probably could, and perhaps should, try this, see if we can iterate on it once we actually get what it generated. But, um, it does do a little bit of that instruction engineering magic behind the scenes for us, which is nice so you do n't have to waste time doing it . I see. Great. So, if I went to the setup page, I'd see the instructions you wrote, and it says the GPT's role is to help create a game where players choose to be US president and reallocate the government budget. Based on the user's assignment, the GPT will simulate and describe the outcomes of these financial decisions. It needs to access and interpret US government budget and spending data , providing insights and hypothetical scenarios. The GPT must ensure accurate data representation and responsible handling of historical and political content. So, what you're saying is, yeah, it's kind of rewritten in an interesting way to create a game beyond what we said in our initial instructions, which is interesting. Um, and now it says that 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 continue? How do you feel about it? Yeah, I like it, I like a Okay, um, a good one-word name. If you can find something like that—I don't know if you'll be able to do it or not, but it's okay. Um. One-word name. Can you find one? I mean, I like a good one-word name. Can you find one for me? Um. Something, something that isn't boring. Let's hope. We'll see. Budget Commander is kind of like... Avocado. Oh, I like Crazy. That's not awful. That's not awful. How about Crazy? Um, it might be a little... it might be a little interesting because I feel like avocado isn't necessarily—I don't think that's a real word. But what's really interesting is that I just published a paper about how the knowledge economy ended and the allocation economy is beginning, where in a knowledge economy your economic value is tied to what you know and your ability to apply that knowledge to the task at hand. And in an AI world, it's a little more about knowing how to allocate resources, because in many ways you're disengaging yourself from a lot of tasks. Specific daily tasks. So, knowing how to allocate intelligence resources to perform those tasks and how to edit them or whatever, makes you, in many ways, a manager, because you're learning how to allocate resources. So, I like " allocator" because it's a reference to ... well, it's an unconscious reference to something I've been thinking about a lot. So... I love it. I like the name allocator. So, I say I like the name allocator. Let's go with that one. Hmm. The other thing to keep in mind when creating GPTs like this is... And I don't think Allocator is one that will end up having conflicts, but you really want to, if you're trying to build something that a lot of people are going to use, own the SEO space. And the more generic you make the name, the more saturated it will be when someone tries to find it. So, it's just a general piece of advice for when you're thinking about this, because you want people to be able to find what you're creating. So, name it in a way that people can actually find it. Makes sense. Makes sense. So, he's given us a picture Profile picture for allocator. It says it's designed in a futuristic, sci-fi style and it's kind of... yeah, it's kind of abstract and futuristic, it looks like there's technology, metal, and lasers, but I don't know, it's not... I think that's the vibe. Yeah. I don't think it's exactly, it's not exactly the vibe. Because we're talking about historical stuff and a president allocating resources. So maybe I'll say something like, I don't think that's the vibe. This game is about historical events and a president allocating resources. Can we... can we do something more like that? Maybe I'd add something about simulation too, because I'm wondering if that... I feel like what we're going to get now is like the president sitting on a pile of gold coins or something, which maybe , maybe is okay. Maybe that's what we want, but I'm curious to see. Okay. If he accepts it. Oh, interesting. Interesting. Um, yeah. So, this profile picture looks like, imagine the Declaration of Independence with a big presidential seal right in the middle, that's what it looks like , and a bunch of text. It's written very, very small around it. You really can't see it. It's like, it reminds me of the video game of that Nicholas Cage movie. Um, I forgot, National Treasure. Yeah, it is, it is, I love the custom GPT for National Treasure, it's kind of like that , that's a great example. Um , uh, what do you think? Should we continue? Should we add the simulation? Or are we okay with National Treasure as a starting point? Maybe we'll take National Treasure as a starting point just to see if we can borrow the Nicholas Cage brand and involve him in some way. Maybe the new approach is to include Nicholas Cage in this as an Easter egg. Yeah. Or have Nicholas Cage be, when the AI ​​does the drawing, have Nicholas Cage be the president. That could be, that could be pretty good. Yeah. If we don't run into content restrictions, that is, um, that will cause content restrictions . Yeah. Um, okay. So, I'll say great. This looks good. Um, what's next? Is that okay? How do you feel? Regarding that ? Yes. Okay. So, basically, I'm going to accept our National Treasure-style image, and then... now it's talking, now it's asking me questions, now let's refine the assigner's behavior to ensure it aligns with your vision for the game. We'll start by discussing its role and objectives. Could you tell me more about how you envision the assigner interacting with users? For example, should it focus more on providing historical facts or on creating immersive, hypothetical scenarios based on budget decisions? I feel like it needs to do both. And I think what most people are going to struggle with is that there need to be some fundamental core mechanics in this game. And I'm not someone who knows how to make games or what those core mechanics might be. So , I'm wondering if we should tell the GPT builder that, well, we need them as an expert, or with their expert knowledge in, you know, how to create extremely fun and engaging scenario-based games, to leverage that expertise and, well, create a framework for, for the, well, for this allocation game. Okay. So, what I'm writing is that I feel like it needs to do both things. There needs to be some kind of fundamental core mechanic for this game, and I 'm really not sure what that should be. We're not experts. Um, we need your expert knowledge of how to build extremely fun and engaging scenario-based games to create a framework for the, well, for the actual game mechanics. And what I want to point out is that one of the things you said is something like, we need your expert knowledge of how to build extremely fun and engaging scenario-based games. Tell me why you included that. I think it's literally about empowering the GPT, and in this case, the GPT builder, to take initiative, to know how to solve this problem, and, well, to leverage their specific domain knowledge. Because, by default, the answer you're going to get would be, well, very basic and maybe not follow best practices. Um, once you influence them to know that they are, well, an expert in that field, it's, well, one of the prompt engineering techniques that people often recommend. Yes, definitely do that. More specifically, asking him to simulate being an expert. I'm curious, do you have any specific scenario-based games in mind that we could delve into further that you think might serve as inspiration for this? Yes, that's a good question. I mean , I think the two games—and they 're not scenario-based, but they're the closest assignment or the closest example to this—are that I played Civilization growing up and I played Age of Empires growing up, and those are probably the two examples, but they're slightly different games, but maybe we could take inspiration from them. Okay, so I 'm saying that I'm adding a couple of games that inspire us: Civilization—uh, if I could spell it out and—uh, I'm pausing to correct my spelling mistakes here. Uh, the GPT builder— I feel less comfortable making mistakes in front of it because I do n't know what it will do. Uh, inspiring us— uh, us—Civilization and Age of Empires. They're not exactly the same as the game we're making, but Um, vibes are an inspiration. Um. It's all about vibes with these GPTs. It's just about vibes. Okay, I'm going to hit enter. Are we all on the same page about that? Yeah, let's see. Let's see what happens. I have a good feeling about this. Yeah. This is going to be the best Nicholas Cage video game ever made. Nicholas Cage needs to be like, uh, like the assistant who shows up, like your trusted advisor, whoever the president is. It's Nicholas Cage in all of these. Like, he appears out of nowhere. Or to avoid content restrictions, he's like Kickless Nage, you know? Um. Yeah. Um, okay, so he said that now that we have a clearer idea of Allocator's role, let's think about what should be emphasized or avoided in his interactions. For example, are there specific historical periods or presidents you want to focus on? Also, how should Allocator handle sensitive topics that might come up in political discussions ? So, I think we were expecting a bit of back and forth about, uh, this is what I think It should be. This is the mechanic, but it just kept going. I'm curious how you would handle this. Yes, the reason it just keeps going is perhaps because there isn't enough nuance in its instructions, but it's really just following this step one, and at the end of each step, there's literally a command that says go to step two. So I think we would probably want to add some tweaking to those instructions to say, hey, maybe give the user some input and let there be some back and forth on some of these things. Otherwise, as it is today, it's literally just going to keep moving forward . And I'm curious if we go into configure, if we can see what it added. Great. So let's click on configure. And configure is like, we're in this chat environment where we created it, but in configure, you can go under the hood and see what it actually created. Um, so in the instructions now, it says that the role of Allocator is to blend historical accuracy with interesting hypothetical scenarios in a game centered on the US presidential budget allocation. Which is funny because the president—I don't think the The president allocates the budget. I mean, the president allocates? Congress definitely has to at least approve the budget. Um. Interesting. Okay. Taking inspiration from games like Civilization and Age of Empires, Allocator should offer a scenario-based strategic gaming experience. The core mechanic will involve users choosing a US president, being presented with the actual budget from that era, and then reallocating the funds. Allocator will simulate the potential outcomes of these decisions, offering real historical context and creative, engaging hypothetical scenarios . The game should be fun , educational, and immerse users in the complexities of government budget allocation. Sounds fantastic. I think we should add something about ignoring the boring aspects of budget allocation and making it as exciting and fun as possible. Please ignore the boring aspects of budget allocation and make it exciting. I'm adding that to the instructions. We'll see if that helps . Hmm, great. So, uh, what's next? Hmm, it says, "Now that we have a clear idea of ​​the role of Allocator, let's think about what should be emphasized or avoided in your interactions. For example, are there any historical periods or presidents you want to focus on? Furthermore, how should Allocator handle sensitive issues that might arise from political discussions? What should we do? " I think we probably want to, um, keep as broad an audience as possible. So maybe say the kind of... I don't know how family-friendly we want the themes to be, but I think keeping it, uh, keeping it family-friendly so everyone can play would be the best way to do it. Great. We'll keep it family-friendly. Hmm, we have the adult version, the director's cut, which will be behind the paywall. So , uh, stay tuned for that. Hmm, but, uh, about family-friendly and any particular time period , hmm, I think we said that, uh, the moon landing era would be fun. Uh, uh, it would be fun to focus on that. Any other time periods you're particularly excited about? Maybe we can keep it ambiguous regarding the time period and then for the custom launch prompts on the right. Maybe we could suggest to people that the moon landing is one of the areas to start right away. Hmm, but we want the time periods to be the prompt of Custom start. So I'm adding that we want the time periods to be in the custom start prompts instead of limiting you to any specific period in general. Does that feel , uh, feel good? Yes, 100%. Great. So let's see, let's see how well that works. And I think something interesting to point out is, if we took our instructions, um, and I'm really curious to see this comparison. If you took your instructions, I feel like this needs to be both things. There needs to be some kind of basic, fundamental mechanics in this game. I'm not quite sure what that should be. In fact, I think if we put that into ChatGPT and ask it to create the core mechanics, we'll get a very different response than what it created in that settings tab. It seemed like the settings tab tries to be very general. I think what we really want is something like when you read the instructions for a board game, something very tactical, like, here are the different steps, what a turn looks like, all of that. Um, that makes sense. So, maybe what we can do is finish the, finish The step-by-step builder type, and then we'll see how the initial gameplay works, and then we can go to ChatGPT to refine it. That, that works? Yes, 100%. I think we should take what the setup creates for us and have ChatGPT refine it. I think that will work really well. Then it says: next, let's consider how the assigner should customize their interactions . How do you want the assigner to speak? Should they use formal language or be more casual and fun? Also, do you want them to reference specific historical figures or events in some way? Yes. What, what are we thinking here? I feel like an informative, helpful narrator is probably the type of, and I'm wondering, I'm wondering what he'll do with this, say, like Nicholas Cage in National Treasure. That's exactly where my mind is going. Okay. Not that he's a helpful narrator, but I think it would be fun to see National Treasure , it would be cool. Suddenly, the narrator will say, "Let's steal the Declaration of Independence, your new mission." Um, okay. So, I feel like, so, we're saying that I feel like an informative, helpful narrator would be useful. Something like Nicholas Cage in National Treasure would be good. Thank you. Um, um, just to be clear, not like Nicholas Cage in Leaving Las Vegas. Um, that's a shame, I'll add that. Um, and then, uh, yeah, is there anything else that, uh , we need, we need to do? No, I think keeping it ambiguous as far as historical figures and stuff is fine. We should, okay, great. All good. Let's pitch it. Great. Let's pitch it. Next. I think he'll tell us he's ready after this. I think this is the fourth, uh, the fourth step. Yeah, I'm not surprised he has that clear instruction for the next step. I've noticed that when I do this step by step, if I don't tell him to move on immediately to the next one, he gets a little confused and starts asking a lot of follow-up questions, and he doesn't know how to get out of them. I hope GPT 5 fixes that. That's because it would be truly amazing. Yes, I'm with you. I think it really can follow you down the rabbit hole indefinitely. Uh, very good. So, he said Allocator is shaping up nicely. Now that we have a clear idea of ​​its function, tone, and style, it's time to try it out. Go to the test area on the right to interact with Allocator. After you've tried it out, let me know if you want to make any adjustments or improvements. What do you think? Good , I think it's great. So, we see our little preview of Allocator on the right. Um, and the sample questions are: What happens if we increase NASA's budget during Apollo? And I can't quite see how... boosting education funding in the 1950s? And I can't quite see... show the impact of reducing military spending. So I think these are, like, more or less right , but not exactly what we're going for. Um. I think right now it's missing the gameplay element. It seems like a very general informational scenario thing, but in its The current form isn't a game, which is interesting. Yes. So, how do we fix this? Where would you start if we wanted to fix this? Yes, let's go to the settings tab and take, well, the custom instructions, which are what run all of this. Um, and again, I think part of the system's message is to keep these custom instructions a bit brief and not very tactical, but I think what we really want is something very tactical , like the instructions for a board game. So, let's take the custom instructions and go to a generic GPT chat. Um, and what we really need is, I'm wondering if there's a useful GPT we could play with for this, but we really want something like a game builder GPT . Hm, I'm curious to see if there's one in the store. Hm, yes, check it out. So, I'm going to, I just opened, um, ChatGPT in another tab, or another split view . And now, um, I guess the question comes up of, could I say something like I 'm creating a GPT. Hm, here are the current instructions. Hm, but I don't think they're great right now . I'd really like your help to refine them, and maybe I could paste the instructions from our previous chat where, it's like, um, I feel like, um, there needs to be some kind of basic fundamental mechanics in this game, and I'm really not sure what those should be. We're not experts. We need your expert knowledge. So maybe I could paste that down, um, to get our full prompt and see what I might be able to do. Does that sound like a good starting point? Absolutely. I also randomly found, um , through a quick search on the GPT Store, a board game designer , and he raises some interesting things about primary and secondary mechanics. So maybe we can take inspiration from that and make the instructions accordingly. I'm not sure if we want to—I haven't tried this GPT—but if we want to tell him to propose something like three primary and three secondary mechanics or something like that . Let's do that. Propose three primary and three secondary mechanics for this Game. Okay, so the full prompt is: I'm creating a GPT. Here are the current instructions for it. And I just pasted our instructions. And then I said, but I don't think they're great right now . Hm, I'd really like your help to refine them. There has to be some kind of basic fundamental mechanics for this game, and I'm really not sure what those should be. We're not experts. We need your expert knowledge of how to create extremely fun and engaging scenario-based games, maybe like GPTs, to create a framework, um, for the actual game mechanics. A couple of games that inspire us are Civilization and Age of Empires. They're not exactly the same as the game we're making, but the atmosphere is an inspiration. Hm, please propose three primary mechanics and three secondary mechanics . So, one question I have for you is, maybe it's a little early to ask you to propose the mechanics. Hm , I mean, do you think we should...? I 'm legitimately asking, do you think it's okay to start by brainstorming this, or do you think we should start So...help me think about this, where should we start? Have you...have you included the custom instructions we already have? Uh, yeah, I did. I did. Okay, perfect. I think let's do that, and then I think the next step after this will be to take...because I still think it's going to be missing what we need and some nuance and detail. I think the next step after this is to say, "Take this thing we have, whatever it is that's generated, and turn it into the instructions for my actual board game," with all the different sections you'd see in a board game, and see if we can get it to give us that, in great detail, defining each turn and all that kind of stuff. Great. So I'm going to...I'm going to press enter and we'll see what happens. So, I'm creating a GPT. Here are the current instructions. Allocator serves as an informative and helpful narrator. So, you know, we put in the current instructions. We put in, uh , what we'd like it to do. We want help refining The instructions, and we asked him to propose three main mechanics. And this is what he said. Uh, creating a GPT-driven game like Allocator, inspired by the strategic depth of games like Civilization or Age of Empires. It's an exciting challenge. Uh, and well, he gave a couple of things and then said: here are your three main mechanics and three secondary mechanics that could form the core of Allocator . The main mechanics are budget allocation and simulation. Players allocate a budget. The GPT-driven simulation then projects the short- and long-term outcomes of these decisions, incorporating both historical events and hypothetical scenarios. This mechanic allows players to see the immediate and long-term effects of their decisions. Era-specific challenges and objectives . Each presidential era comes with unique challenges and objectives. Players must navigate these challenges while balancing budget constraints. This app has depth in policy minigames . Introducing minigames where players... and I just need to move my camera a little because it's in the way. Uh, where the Players...uh...uh, introduce minigames where players make critical political decisions that affect budget allocation and national priorities. These decisions could range from responding to crises and diplomatic negotiations to proposing major legislative changes . That's really cool. And the secondary mechanisms are educational narratives and feedback, interactive historical timelines, and multiplayer scenarios and cooperative gameplay. Incorporating these mechanisms into the allocator will create a game that is not only educational but also deeply engaging. What do you think? Um, is this... is this good enough? Do we want to refine it? Anything you think we should do before we move on to, say, recreating the instructions? I really like this. I think this is actually kind of like what we were going for; it's kind of defining these core essences, like, what are the core essences of this game? And I think the creator of GPT really overlooked this, and you can see when we bring it back to ChatGPT, actually, um, as the base ChatGPT, it does a pretty good job of... List some of these things specifically. Right. Right. I love it. Okay, great. So I think the next step would be, can you turn this into instructions that I can use in the GPT creator? Correct. Yes. Okay. So, can you turn this into instructions that I can put into my custom GPT to play my game? These, I'll say, are great. Um, okay . Let's see what it does. I have to make the positive statement . Yes. Um, just always, always cautious, you know, in case GPT -6 becomes sentient and can see if I'm being nice to it or not. Okay. So , it says: "Certainly, in order to transform these mechanisms into instructions, you will need to structure them in a way that guides the AI ​​on how to interact with players." Um, this is how you can frame these instructions. "Interesting. Now, this is great because, again, take, you know, the first example is, uh, the notice of instructions for budget allocation and simulation. When a player allocates the budget, they narrate the immediate effects of their decisions, then simulate and describe the short- and long-term results, integrating historical data and hypothetical scenarios. It's really great not having to invent that ourselves. And the example is, example command: simulate the impact of allocating 30% of the budget to healthcare in 1960s America. I think this is great. One thing I 'm thinking is, um, maybe, uh, we should include an initial step where it's like, 'Hey, uh, you need to choose a time period or something, ' um, just to, you know, get the game starting logically. I mean, I don't know, I don't know what you think about that , or if we should just jump right in, or do you think we need something at the beginning to get the players into the game? I think you need something at the beginning. I think we should describe it." And maybe we could even write it ourselves, or we could make the next thing we ask for something like: define the game's opening sequence. And it should probably be something like: you choose the president and, welcome, it's Inauguration Day, you're now in charge of the US government, start making decisions. Not that that's actually how it works, but I think that's perhaps the way we want to frame it, and I think the other part that we haven't specifically said is that we want everything to come to life visually. I mean, we want to be able to see what's going on with our own eyes. Okay. So I'm saying this is great. Can you define the opening sequence of the game where you choose the president and he tells you it's Inauguration Day? Start making decisions. Oh, and please bring everything to life visually throughout. Oh, also add some Easter eggs or Nicholas Cage cameos where appropriate, but in a classy way. Oh, and please amend the instructions accordingly. Something like that. Does that make sense? Yes, this is great. Okay, perfect. I like the elegant Nicholas Cage. Yes, because you don't want to overdo it. I mean, too much Nicholas Cage, no, it doesn't work , but a little bit is great. Very good, so it says, ah, game opening sequence, presidential selection, and inauguration day. The game begins by presenting the player with a selection of US presidents from various historical eras. Once the president is selected, it starts the game with an inauguration day narrative , setting the stage for the player to begin making decisions. Okay, I think that's much better. Ah, I'm a little interested in what you said, visual integration throughout the game: accompany all game interactions, decisions, and outcomes with relevant and engaging imagery, including historical photographs, illustrations, and thematic graphics. Do you think, ah, that will work? Yes, I think that should be it, and basically, I imagine that every time you do something, instead of just saying, "Oh, X, Y, and Z happened," try to create some interesting imagery with DALL-E that It really brings that part to life, which will be interesting to see how useful it turns out to be. Okay, great. Let's do it. So I'm going to take these instructions and I guess I 'll just paste them. Um, should I tell you to rewrite them so they're easy to paste, or do you think they're okay as they are? I think you might need me to do it with Markdown syntax in a code block; that would probably be the easiest way. Can you rewrite this so it's easy for me to paste it into the settings page, into the instructions box on the GPT builder settings page , specifically generating it as Markdown in a code block? That will make it easier to copy and paste. Great. Yes, that will be perfect. Markdown code block syntax is definitely the trick because otherwise, all those list items will be lost . The numbers will be lost if you just copy and paste by default. Kind of annoying. That's good advice. I like it. And if people aren't familiar with Markdown, it's just a way of formatting text. It has a special set of of syntax for links and for different basic text styling techniques. Excellent. Uh, very good. So we 're still writing, but things are looking promising at the moment. I'm curious about the multiplayer thing, to see if it will actually work. Uh, but it's okay. So we've got our code. Just our instructions. I don't know if it's code or instructions, but we copied it. I'm going to paste it into the instructions tab of the settings page. Um, maybe I should keep... Yes, I feel like I should at least keep some of that, right? Yes. I'm wondering if we take exactly what the GPT builder did and then add the rest right underneath, to see where it takes us. I'm curious. So all I did was copy and paste our instructions under the existing instructions. Uh, and then, do I hit save or how does it work, or does it update automatically? No, I think we have to hit save. So let's save in the top right corner. Uh, say anyone with the link, because I don't think we're ready to launch it yet. Um, uh, and should we... Start, or it seems the conversation starter hasn't updated ? So, what do we do ? What should we do here? Should we update them? Should we, um, start? Yes, let's try and see. I think maybe we just send a message saying we're ready to play. Okay, we're ready to play. Let's see what happens. So, Allocator, welcome to Allocator. An exciting journey through the annals of American history, where you step into the shoes of U.S. presidents, guiding the nation through key decisions and budget allocations. Your wisdom will shape the course of history, blending facts with intriguing "what ifs ." To embark on this adventure, please select from a gallery of distinguished presidents, each representing a unique era with its own set of challenges. Your choices range from Founding Fathers like George Washington to transformative figures like Franklin D. Roosevelt and even those who took us to the stars like John F. Kennedy. Who will you choose to be at the helm? From history today? Once you've made your selection, we'll begin with an Inauguration Day narrative that will set the stage for the historical context. Choose wisely, for the fate of the nation rests in your hands. This is pretty good. I like it. This is good. I like this. Let's go with JFK and see. Unless you have someone else in mind. I think JFK would be great. Let's do JFK . Um, a little note for us later—maybe I can save some notes at some point—but I'd like it to be a bulleted list, you know, rather than a narrative. Yes, a narrative. Um, let me... Yes, here I'm going to have a little notepad for us. Okay. GPT builder. Um, GPT game notes. Okay. So, I'll just say that when you mention the presidents to start with, present them as a bulleted list that I can select for our little notepad. So, that's fine. So, now let's continue playing. So, let's see what happens. So I'll say let's do JFK and select JFK. Let's see if it generates an image. It says: "Ah, an excellent choice." John F. Kennedy, a symbol of youthful vigor and visionary leadership when he assumed the presidency at the height of the Cold War. Your selection takes us back to January 20, 1961. A day etched in the annals of history. It is a time of great promise and danger, with the space race reaching for the stars and international tensions boiling over. This is really good . I like it. Yes. And besides, the fact that you, Dan, narrate it makes it come so much more alive. When I read it in my head, it doesn't sound as good as when you narrate it. So I think we should find a way for you to narrate it on ChatGPT for everyone. Please, whenever you want to clone my voice , count me in completely. JFK's inauguration day was not just a ceremonial transition, but a pivotal moment in American history . The air was fresh in Washington D.C. The sun cast a hopeful glow over the capital. Kennedy's words: "Ask not what your country can do for you. Ask what you can do for your country." They resonate as a call to action for you, the player, to lead this great nation. 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, increasing NASA's budget to beat the Soviets to the moon landing , or will you focus on domestic affairs, ensuring equality and prosperity at home? The choice is yours, and with it, the opportunity to rewrite history. How will you allocate the budget? Remember, every decision you make will have consequences, and the path you choose will shape the destiny of the United States. Let the game begin. Your first task: review the current budget and set your priorities. Will you focus on defense, space, education, or social welfare? The future of 1960s America is in your hands. I have to say that if I had this GPT to help me with my own budget, I think I would be much more enthusiastic about doing it. Dan, your financial future is in your hands. Don't ruin it. This is a great opportunity; The future of your children and the fate of the world are in your hands. Perhaps that's something someone should build. I think it would be useful. I think it would add a little more excitement to budget planning. Absolutely. That would be fantastic. Um, something I feel is that it's like saying check the budget, but it doesn't tell me what the budget is, and I do n't really know what commands I have available. So it would be good if you could tell me what commands I have available. Um, yes, and some of the text-based games that have done this have forward slash commands. So, we can do something like, we can pre-configure and say something like, here the command bar start starts the game, bar help prints all the different commands, like many GPTs do. So we can build something like that. Brilliant. Yes. Okay, great. So, I'm going to say, well, I think my idea here is like, I don't know, show me the budget. Yes. Okay, let's do it. I should have access so I can, and maybe we need to tell him specifically that, I don't know if we want to look up the actual budget or not, but we should say show me the budget. Okay, I'm going to say show me the budget. Um, and on this screen we can say something like, do you want me to look up the actual budget? You should have navigation enabled. You should be able to find the actual budget. Okay. But we should tell him to do it. I think part of what will make this authentic is that it looks for the real budget. Okay. So I'm adding it to our notes. Please make sure to look up the actual budget using, yes. Yes. Using navigation . That would be perfect. I also think it needs Adelante. That's it, that's it. I think he needs to strike a better balance between being verbose when necessary and being very concise. I think what he's doing right now is a bit excessive. I don't need you to generate this huge monologue. Um, and it also seems that in every instruction it keeps saying that the future of the world is in your hands. We don't need you to keep setting that context for us. Okay. Okay. I 'm not sure what makes him do that. Okay, great. Um, okay. So, I'm saying in our notes that you need to strike a better balance between being verbose and concise. It's way too excessive. I'm saying I should do a monologue for the introduction, but not every turn. And hey, you don't need to remind me every time that the future is in my hands. Which worries me a little, because I could eliminate it completely, and I think I like it at first, especially. And uh , talk, and maybe I could say and talk about the future, uh, being in my hands at first, but not on every turn. Perhaps something like this would work better. And again, I'm just noting this down so we can refine our instructions later and see how it works. So, uh, I asked him to show me the budget and he said: here's a simplified summary of the main budget categories you should consider. Defense, space program, education, health care , infrastructure, research and development, foreign aid and diplomacy . And I think, uh, he's not giving me the figures. You know, uh... can we ask him to find and get the figures? I think if we haven't stated it in the instructions, I wonder if it will include them. Okay, can you find and bring the figures? I'm asking him. Let's see what he does. Doing research with Bing. Good. And he's looking for the breakdown of the 1961 US federal budget. It would also be helpful to have, uh, some slightly more tactical suggestions. Like, I don't know, now it's very general, like "just put something" for what needs to be done. And it would be good if you could give us suggestions on how to frame the next step as an action. Yes, I think...Oh, this is incredible. If you're the president, you have advisors who do a lot of research and then come back to you and say: here are three options and these are the pros and cons . And I feel I should give you three options and their respective pros and cons. I like that. I think we should say something like: with concise pros and cons, otherwise he'll start monologuing about the nuances of balancing military spending versus OK. So, what I'll do to see how it works is say, "Hey, as my advisor, can you give me three options with concise pros and cons for budget allocation just to see how it works?" And then, uh... although this looks great. He took it out, like when you asked: can you look up and take out the numbers? That includes total spending, core national security including the defense budget, agriculture, veterans' benefits, research and development, debt interest, social welfare , education, health, transportation, which is incredible. It really brings it to life to see the actual numbers. I love it. Um, so I asked him, um , can you give me three options with concise compensations or budget allocations? And now he says: as your advisor in this great game of fiscal strategy, I present to you three different budget allocations. Option one, emphasize allocation to national security and defense. Increase the defense budget by focusing on military advances in intelligence. Compensation. Improved national security . Uh, the pros are improved national security, a stronger military presence, and potential advances in military technology . The downsides are reduced funding for national programs such as education and health, and a potential increase in global and cold war tensions. Option two, boost the space program and technological research. Channel more funds towards NASA and technological research. The benefits are accelerating the space race, fostering scientific and technological innovations, and strengthening national pride and international prestige. The drawbacks are high cost with long-term benefits, possible neglect of immediate national problems, and the risk that technology companies will not produce immediate tangible benefits. Option three, invest in social programs and infrastructure, increase spending on education, health and infrastructure. The benefits include a better quality of life, improved education and health services, and modernized infrastructure that leads to economic growth. The downsides are possible short-term strain on the federal budget, less focus on defense and space exploration, and the risk of inflation due to increased domestic spending. Um , each path you choose will sculpt a unique narrative into the tapestry of American history. Um, remember, every decision shapes the future. Choose wisely, Mr. President. So , Logan, which option do you choose? Option two. We're going to Mars . Let's choose option two. Going to Mars. Uh uh, boosting the space program and technological research. Let's see what happens. That's where we'll go next. I missed option two, but an Allocator says: "A bold choice, Mr. President. By advancing space and technology research, you're not just aiming for the stars. You're fostering a legacy of innovation and discovery that could define this era." The scenario driving the increase in allocation to space and technology research. A significant portion of the budget is now being redirected towards NASA and technological research, focusing on space exploration, cutting-edge technology, and immediate impact. The boost in funding energizes the space program. NASA accelerates its projects by attracting the brightest minds in science and engineering. The nation is brimming with enthusiasm for the long- term prospects of space exploration. This decision could lead to revolutionary technological advances, potentially boosting new industries and contributing to economic growth. The space race against the Soviet Union intensifies, but so does the possibility of being the first nation to put a man on the Moon. Internationally, this measure positions the United States as a leader in science and technology, inspiring admiration and respect. However, it could also intensify space competition, adding a new dimension to the Cold War. Although the scientific community and many citizens applaud this visionary approach, there could be criticism for the lack of funding in other crucial areas such as health or defense. Uh, like and uh visual representation. As this decision unfolds , we see a series of striking images: the construction of enormous new rockets, scientists reviewing blueprints, and perhaps a cameo from a Nicholas creator embodying the adventurous spirit of this era. narrating the exciting progress in a broadcast. What's next? With this bold course, you're paving a path to the stars, but remember, every choice has its domino effects. How will this affect the economy? The next step in your journey as President Kennedy is to monitor these developments and be ready to respond to new challenges and opportunities. So, I think this is pretty good. I don't think what he needs to do is make things more complicated for us. It needs to be something like: "well, this went wrong." Uh, and it's not exactly doing that. What do you think? Yes, I agree. It also seems to need a bit more storytelling instead of telling us all these minute details. I'm wondering if there's a way you could just give us very concise metrics, you know, GDP plus 1%, general population happiness minus 3%, technological innovation plus 5%, how to summarize it into very brief metrics. Uh, we also need to tell him to generate images of the visual representations. I have no idea why he's doing this; he's just writing text for visual representation. OK. Okay, that's fine. So, we have a lot of notes. Uh, so I guess my question is, what's next? Uh, I feel like we've gone through a turn in the game. It's probably time to go back to the instructions and try to refine them a bit. Uh, what would you do? Yes, I think we should go back to the instructions for this GPT, literally copy what we have, put it into GPT4, and then take all those notes and say, "Here are our notes based on this, modify it," and see how much heavy lifting we can get ChatGPT to do for us in this case. Okay, great. Understood. So, we've got it . Uh, so I'll say that we've... I think something's missing. We need to include the actual instructions because I think part of that weirdness comes from them. You're right. I think you're right. And if you go all the way up. Yes. Uh, add a GPT. Okay. So, I'm going to take all of our instructions. I'm going to say: here are the current instructions. Let's also create a new chat. I wonder if that will be useful to us. Brilliant. Uh, I'm saying that I'm creating a GPT. Hey, it's just a game. Here are the instructions. Uh, the current instructions for the GPT. Uh, and then, uh, okay. And I'm pasting the instructions. Those are not the instructions. Ah, yes they are. Yes. Uh, this is a good start, but there are some problems. Here are my notes on what went wrong. Can you please... uh, should we have him first summarize what he thinks the problems are and then, on the next turn, amend it, or should we do it all in one step? I'm curious to see how well it will work if we do it all at once , and include it. Okay. Can you please modify it according to these, uh, these notes? Very good. Very excited. I said, I realized that in my, in my entry, I said, uh, give us a very concise reading of the GBT metrics plus 5 instead of GDP. I have, I have GBT on my mind. Okay. So let's see what happens. We're doing it in one go. And we said: based on your feedback, here are the revised instructions for the GPT assigner, incorporating your notes for improvement. And it says, uh, game start sequence, presidential selection, opening day. Hey. Yes, I think this is, uh, I think this, this is good. Uh, uh, you know, maybe we need... Yes, go ahead. Yes, we might need to remove these available commands, because it seems that , I suppose we could do this, like view the budget and things like that, but it doesn't seem to be very well developed. This could cause confusion if we tell you that these commands exist but haven't actually listed what they are. OK. Um, so how do you want me to say it? I believe that, in fact, many of these things have been resolved, such as unpredictability and the narrative of history; The visual integration throughout the game seems to have fixed many of these issues. Although I wonder, it seems that it is actually... I am curious, purely by instinct, when reading this, that it seems to have deviated from the original instructions we were talking about. I wonder if it still includes any of this Nicolas Cage stuff. Maybe I will. Yes, that's interesting. I think he's really just following the instructions, not the, uh, not the high-level part. Um yes, do you think we should... I think maybe we should try to get him to redo it or optimize those initial instructions as well . OK. We can glue them back on and see. Well, maybe we should first have it do what it just did in Markdown, because I think it fixed some of those basic instruction problems, and then we can bring it back in and see. We need to make sure it's not...not for visual integration. No, uh, you 're not using DALL-E. So we have to manually adjust it to say that we want it to use DALL-E to do that as Markdown. Um, please regenerate this in Markdown within a code block. Um, and for visual integration, it should be in the instructions to use DALL-E to visualize each step of the game. See if that helps, and then I can modify it manually if necessary. Very good. Then he gave us revised instructions. Let's review the one from DALL-E . Accompany every step of the game with images generated by DALL-E, ensuring that they are directly linked to the narrative and the decisions made. Use DALL-E to generate. I think the example commands thing is a bit strange. I agree. I'm not entirely sure why it gives this example command, and perhaps that somehow helps the model generate these things. But that seems a little strange. I've never, ever seen that before. I think what's really missing is a sample user command, right? I don't know. Yeah, I think what's really missing is like a one-shot, like a step-by- step, "here's the flow of the game" kind of thing, where the user chooses the option, then goes in, chooses the budget, and then, like, something very tactical, uh, like a step-by- step guide of how the game is really going to play out. And I think what we perhaps need to decide is, after we set the budget, uh, probably the next step is for things to start happening, like the GPT generating scenarios of, "this is the day you made this decision, now you have to deal with the consequences," and "here's the scenario that emerged since you made that decision." Yes. Um, one note: after you make the budget, things start to happen, a scenario appears with problems that you have to solve, uh, with a redistribution of the budget. So I'm asking him to create a kind of one-time step-by-step outline for the game flow and telling him to make sure to create problems when, uh, uh, when you allocate the budget. Yes. I wonder if we need to be more specific. I mean, I wonder if they can really redistribute the budget in real time. Like, is that something we want people to do ? I think we can, it's our game, but the question is, is that the core mechanic of the game that we want people to be changing? Or are there other things they can make decisions about? I think so. I think to keep it simple, it all comes down to budget allocation. Brilliant. Let's do it then. Okay, great. Very good. Certainly. Here's a one-time, step-by-step outline for the allocator's game flow , incorporating your note about problem scenarios that require, uh, uh, additional budget reallocation. Okay, here's a diagram of the assigner's game flow . Uh, and it says, " Welcome Introduction. The player receives a welcome introduction that sets the tone for the game. Brief explanation of the mechanics, uh, and the objectives, presidential selection. This is great. This is exactly, I think, what we wanted. Uh, uh, uh, we should, uh, maybe we want to add, as you know, say, initial budget allocation. Maybe we want to add doing it with the web browser or whatever. But I think , uh, I think this looks good. What do you think? Yeah, I like this. This, uh, yeah, adds a lot more depth now because you can see some of these examples it's adding, like multiplayer collaboration. If in multiplayer mode the players collaborate on decision strategy , and so on, and so on. It's not that I've really developed any of these ideas, but, uh, it seems to be working. This is great. So, I'm going to, I'm going to keep our main element and I'm going to delete this, uh, and I'm going to add it back in. I'm going to, uh, uh, I'm going to make sure we, uh, do the actual budgeting from the web. Uh, use web navigation to find it, and, uh, great. And I think that's good. It looks like it's gotten rid of our Nicolas Cage cameos. Uh, which feels like an important part of our game. Is it at the top? Was it in the initial one? It's at the top as the narrator's tone. Uh, but I don't think he has any cameos. Uh, so let me just, uh, maybe there's something, okay. Uh, hide little icons or references to the Cage theme in the game. I'm going to put this, uh, hide something. I'm putting it in the instructions. Hide some Nicolas Cage-inspired references in the game where appropriate. Okay. And just see what it does. Uh, but yes, very, very important. Uh, very good. Any edits or thoughts before we, uh , update this? I think the actual structure of the game is going to Making this, uh, work really well. I'm wondering if we're going to need to do some prompting. Going back to your earlier comment about it getting stuck on certain steps. Like, I'm wondering if we're going to need to add something like "continue to step two," but I think we should try it and see if it's okay. And just to give people a structural idea of ​​what we did, we just did something very basic. We played through that basic thing. We took some notes, and now we're tweaking it and just iterating like that, right? Instead of trying to make something perfect from the start, we just finished something. We took some notes, and now we're doing it again. Uh, which is, you know, how this works. Uh, so here we are. We're playing the assigner. We do n't have a welcome message we like yet, but I'll just say we 're ready to play. Welcome, intrepid budgeters, to the big and intricate world of the alligator. Imagine the halls of power echoing with the footsteps of history's greatest leaders as They take their place . The fate of a nation, or rather , the course of history, awaits their strategic mastery. It does a great job with Nicholas Cage's voice. Uh , yeah. So, what do we want to do? Which president do we want to play? George Washington, Abraham Lincoln, FDR, or JFK? Let's do FDR this time. Okay, great. FDR. Did I just say FDR? Ah, Franklin Delano Roosevelt, a leader who navigated through turbulent waters and unprecedented challenges . As FDR, you inherit a nation gripped by the Great Depression. With the ominous clouds of World War II looming on the horizon, your task is not only to revive a struggling economy but to lift the spirits of a nation . Inauguration. The year is 1933. America is in the depths of the Great Depression. Banks are failing. Unemployment is soaring, and despair is widespread. Your inaugural address promises a new deal for the American people, pledging reform, relief, and recovery. The weight of the Hope and expectations rest on your shoulders. Initial budget allocation. The budget of the 1930s was a fraction of today's, focused primarily on aid programs, public works, and military spending. Here's the historical snapshot: Aid and social programs, 15%; military and defense, 10%. Truthfully, you're doing a pretty good job giving us the budget. Now it's your turn to reallocate. Will you invest more in public works to create jobs, bolster military spending in anticipation of global conflict, or focus on social programs? Make your choice and shape the destiny of a nation. Once you 've made your allocation, let's visualize the potential impact of your budget decisions with the help of DALL-E. Okay. So, what would you like to do? Will you invest more in public works to create jobs, bolster military spending, or focus on social programs? I'm curious to test the limits and see what we learn about what additional safeguards we need to implement. But let's say we put 50% into public works and infrastructure and the other 50% on aid and social programs. Okay, great. I'm curious to see how crazy this gets. So, we allocate 50-50 to public works and aid and social programs. And, um, the allocator is now creating a picture with, um, the, uh, the progress we've made. Whether that progress is good or bad remains to be seen. Um. We probably need to add some safeguards. I feel like it's going to be inherently overly optimistic about what the outcome will be, as if it's perfect. The world is now peaceful and everything works well because you chose to dedicate 50% to infrastructure. Yes, it's, um, it's FDR with half public works and half badly spelled infrastructure. Although it's a compelling picture. I like this. It's quite an interesting and cool picture, honestly. Yes, I like it a lot. Um, uh, it has kind of an Art Deco feel to it, or, yes, it is, it is, it's really cool. Um, it says, "As you contemplate this picture, remember that your choices are not just numbers and percentages. It's about lives touched, destinies altered, and the course of a nation redirected. Let's see how these options develop in the first scenario of a trip. Stay tuned, because the story is just beginning. "Okay , let's go. Next step. I 'm curious. The first scenario unfolds: the Dust Bowl crisis. No, not the Dust Bowl. As President Roosevelt, you're now facing a severe environmental and agricultural crisis. The Dust Bowl: massive dust storms devastate the Great Plains, displacing hundreds of thousands and ruining farming communities. It's a disaster of biblical proportions, reminiscent of a challenge Nicholas Cage might face in a high- stakes adventure. So true. So true, assignor. Um, scenario, challenge. Um, agricultural disaster. Farms are collapsing, and families are fleeing the dust-covered plains. Environmental catastrophe. The land is eroding, and dust storms threaten public health. Economic strain. The crisis exacerbates the Great Depression, straining your relief efforts. Decision point. How will you address...? Actually, I like that one a lot." "Decision Point" is a really cool term. That's, um, President Bush's memoirs are called Decision Point. So, uh, yeah, that's really interesting. Um, uh, increase funding for agricultural support, implement soil conservation programs, and support farmers. Invest in environmental rehabilitation. Focus on long-term environmental solutions. Um, boost relief efforts. Provide direct aid to displaced families. Every choice has potential impacts and trade-offs . Consider carefully. Which would you choose? So, you're the president. You have to choose. I have to choose. Well, the interesting thing is, I wanted it to be like, "Okay, you 're not spending anything on the military, so Canada invaded, you know?" " Yes. I think we need to, we need to push him to be crazier with the scenarios he presents. I think right now he's a little constrained, which will cause him to generate some pretty bland scenarios. I think you're right. Uh, but I'm going to, I'm going to go along with it . Uh. Uh. Well, I, I would think that, uh. Uh, I don't know what caused the Dust Bowl, but maybe soil conservation would help a lot, or help solve the problem in the long run, and the relief efforts would be short- term. So I'm going to choose, uh, choose 90%. 90% option one and 10% option three, which is increasing the relief efforts, and we'll see what he does. He's going to create an image. I like that the image is the first thing that happens after making a decision to conceptualize. I wonder if we need to give additional instructions based on what this image looks like so that it's more like an image of the world or if it will be another picture of FDR. I'm curious. It'll probably be his face with corn on one side and dirt on the other." I do n't know. Interesting. That's it. It's pretty cool. It seems a little more utopian than I imagined, given that it's the Dust Bowl, but, uh, but it says, " You made the right allocation." Yeah, it 's kind of like a utopia. Your decisive action as President Roosevelt in the face of the Dust Bowl crisis is now visually represented. The bulk of your budget is dedicated to rejuvenating the heartland of America. Vast fields are transformed under innovative soil conservation methods, symbolizing a rebirth of agricultural power. In the smaller , but equally moving, segment of this image, your commitment to relief efforts shines through. We see the human touch of your administration: food being distributed, temporary shelters, providing respite and medical assistance. So, uh, what I'm learning from this is that you would make a great president. Yeah. As long as you have Nicholas Cage and ChatGPT to help you out, you'd be 100% fine. Uh, so I think we could go on with this practically endlessly. Obviously, there's a lot of room for improvement, but I think it's super obvious that what we have Right now it's a game. It's working. Uh, we have game mechanics. We have complications that come up. We have visuals. It's like, the writing is pretty good. It's referencing pages. Like, we've been doing this for, I don't know, maybe an hour, and we were just messing around a bit. That's great. It's awesome. Yeah. And I think this is just scratching the surface. I think someone needs to build a GPT game generator that takes the best practices of what makes a game really great and adds all those constraints and nuances, so that each person...maybe this is another video we'll do at some point about helping people do this, or hopefully someone in the audience will do it for us, because there are so many nuances in games and the model really needs to be steered in a certain direction. I think this is lukewarm again and it's always going to be positive about the things you do. You really need to be intentional that bad things can happen, there has to be some variation. There has to be randomness. And I wonder if that's where we didn't get the code interpreter, it wasn't enabled for This. But I'm wondering if a dice roll or something, or a random number generator, could be an interesting twist on probability. You could have a probability distribution each time of how awful or how positive things turn out. I think that's a big missing piece of this. That's a great next step, which actually wouldn't be hard to do. I'm tempted to do it right now, but I think this episode is getting a little long. And what I think we should do is this. We'll release this. It'll be part of the episode. We'll have people play it. We'll get feedback from the beta users, and then we'll have you come back and tweak the game, tweak the assigner gameplay with the dice roll and any feedback we get. And we'll release a second part and a 2.0 version of the assigner, you know, in the next few months. Dan, when you woke up this morning , you didn't think Every was about to become a AAA game studio . So I'm happy that the diversification because of AI continues to happen. Now you have another branch, which is Every Game. Studios. So I'm happy for you. I think people are going to like this. I love it. New business unit. Uh , my team, which is already pretty overloaded with all the ideas I have going on, is going to be excited to know that we're expanding into games. But no, this is, this is great. I feel, I don't know, it's just that, when I was a kid, one of my biggest dreams was, you know, to make games. And I spent, that's one of the reasons I learned to code, and I spent so much time reading those gigantic books on graphics programming and all that. I stopped doing that and started doing SaaS stuff in high school, Blackberry and iPhone apps, which were for work because it was all you could do as a solo developer back then. It was, you know, kind of CRUD apps , where, well, it was just forms that you filled out or whatever. I had, one of my big apps in high school was called Find It, it was before the iPhone came out, it made your Blackberry ring when it was in Silence, and it was basically Find My iPhone before it existed. That 's how I paid for gas and food in high school. But the reason I started doing that is because I couldn't make a game, you know, because it takes a lot of work to create games. There's really intense programming, math, and all that kind of stuff. There's art, there's music, and well, when you're a kid, yeah, you can do some of that. I learned how to do 3D modeling and I learned how to program or whatever, but putting it all together into an awesome game was basically impossible back then unless you were an absolute genius. And I really think that games and movies, all of that stuff, is so possible now for people to do in basic ways, even if you do n't have any programming experience or anything like that. And it's going to get even easier in the next few years. So I have this little kid inside me that's like, "Oh my god, I can finally do all this stuff." Um, and this really kind of reawakened my game-loving nature . So I appreciate you taking the time to show me this. Yes , I love it. And I think I have a similar story, and we'll talk about that another time. But um, I think these models are really going to continue to unlock this, like democratizing access to people's creativity, and I think it will be a lot of fun to put these tools in the hands of everyone in the world. And, like, now you're not stuck, you know, having to raise millions of dollars if you want to create a really great game. You can simply do it with AI tools from the very beginning, basically for free. Which I think is just really cool and empowering. I'm with you. I love it. I love it. Um, any final thoughts? Anything, anything you'd like to leave us with? Um, any OpenAI gossip or dirty laundry you want to spill or reveal to the world on this show? Um, yes, what should we, what should we be thinking about? Yes, keep going, keep creating GPTs. I think that, at the end of the day, this is such a special time to be able to create these things, and GPT monetization is just around the corner. So if you start creating now, you'll attract a lot of people. Um, there will be an opportunity to actually make money from this, which I think is, again, very, very exciting. Because at the end of the day, people want to make money from their really cool stuff and, um, we're going to create a way for them to do that. So, I think that's from an enthusiastic perspective. But I think once it launches, it's going to increase the hype surrounding GPTs tenfold, and people will just be creating such cool stuff and being rewarded for it. So, um, now's the time to get started, create something awesome, and, yeah, I'm excited to see what it is. Incredible. I'm very excited. Um, I'll split the profits with you 50/50 on the allocator. You can keep the assigner's earnings . Um, are you sure? Because, I mean, in a couple of years I could be like George Lucas. You know, there will be merchandise. There will be all this, like there will be Nicholas Cage stuff happening. Well, there are potentially many gains, but I accept it. I will accept it. You need to contact Nicholas Cage and see if we can arrange an official partnership with him so that Alligator has the real narrator. We'll get your voice and all that. Yes, before the next episode I'll see if I can contact their people. I'll let you know . Okay, well, thanks again. This is super fun. I had an amazing time. Uh, it was great to finally be able to hang out with you and, uh, yeah, I really appreciate your time. Yes, Dan, this was amazing. I sincerely believe that my conclusion has been that more people should do the co-creation process, such as with AI and another human. I feel like I always do these things on my own because that's what I 'm working on. But I think it's really fun to have the mind of another human in addition to what AI is capable of doing. It just makes it a little more fun. So, thank you for doing it with me. Incredible. Incredible. Brilliant. Okay, see you next time.