Hands-On with AI: From Tabletop Redesigns to the Compiler Paradigm

0:00Greetings listeners and welcome to the game design roundtable, a show by game designers, four game designers. I'm Dirk Neymire, joined on the core team by my co-host David Heron and the show's editor and co-producer, Elena Neymire. As you may be aware, AI models such as Clawed and Cursor are having a tremendous impact on the game design field, including in our personal practices. As a result, AI in game design is going to be an ongoing topic here. While we are sharing our own personal experiences and in some cases, even recommending these technologies, your personal creationary tools and choices are entirely your responsibility. In the worst possible case, AI models can damage your files and systems. The advanced levels at which we are using the tools are particularly vulnerable to adverse outcomes in the case of human or machine error or even simple bad luck.

1:03Tread carefully. Listeners, for a decade and a half, our show has been offered ad-free for the comfort and convenience of you, our friends in game design. If you'd like to give us a small thank you, hop over to our Patreon where we have a $5 tier just for you. This humble amount from enough of our listeners will cover our costs and help the show continue into the 2030s and beyond. Just head over to Patreon and search for TGDRT. That's the game design round table. Many thanks to those who support us, it does make a difference. Now, on to the show. David, good morning. Morning. I like seeing you every week. It's great. We often will batch record. This is great. I appreciate it. Something we didn't talk about too much if it all last episode, but has been a thread for us is AI. You've talked in the past about you're sort of sizing it up, looking for some assistance

2:07we haven't jumped in as of our last conversation. Where are you at now, David? I've set up a dev environment. Pretty good. Including a couple of things. What I think is most interesting is since the last time I sort of stumbled through or sort of thought about this, which might have been maybe six months ago, we're now at the point where Clode was able to actually really aid in setting up some of the more complicated parts. Yeah. Which was interesting. Setting up some test servers and stuff like that so we can run headless unit tests, which is something that I know is important in development, but would not have any idea on how

3:12to set it up and was sort of assuming it was going to be like a sort of a back and forth between like asking what to do and then fumbling through. It actually just did it. It's magic, brother. It's amazing. It's interesting. It is magic in that, I think in a very real sense where I don't understand necessarily like the underlying principles, but someone who would be able to do it. It's not like it appears to be magic. It isn't, but that is great. Yeah. And magic is just a trick, of course, too. Right? It just makes me happy to say it is an illusion. And I think where in terms of actually like building a thing, after a few sort of like

4:19trying to figure out like what the workflow would be, what I have sort of come to the realization where I'm going to be most more comfortable is less of a sort of build the plane in mid flight as opposed to figuring out what my proof of concept, like what is the minimum proof of concept going to be and then creating design documents that I can work with, clothe to see what happens. So where I'm at right now is I think of the term of the scope of what I want to do. And it's really just going to be a very simple tower defense game with some sort of like Tetris Mergie things. Right? It's like, and just imagine, like if you can imagine like the Tetromino's of a Tetris thing, well, if you place them down, well, what if they merged?

5:20Yeah. Okay. So like I don't necessarily think it's a particular compelling design. It might be, but more of, I mean, to get some stuff on screen, there's going to be some click and drag. Yeah. There's going to be some, you know, enemy creatures of varying types that spawn off screen and move on screen. Probably. And probably inheriting, you know, characteristics that I would manually outline in a JSON somewhere or some other sort of file. They'll basic attack stuff. It'll, you know, tracking hit points on many different objects and then some like recipe stuff. And like it's four or five different systems that all need to work together to work and you will have components that are exposed to something that needs to be human readable

6:22and human, human, uh, editable. So, so I'm treating it. I'm sort of structuring this initial first pass of, well, what would I do if I was working with a designer or with it with an engineer, right? Like, what would I give them? And we'll see. So, um, nothing, nothing to report as of yet in terms of, of, of anything operating. But it's, it is, it is, it is movement at least and I have a, I have a target. I'm thrilled. And, and are you using it pretty regularly or sporadically? Do you have any sort of a schedule for yourself or deadline? I did. And then things got a little bit derailed. I'm listening to hear like I'm, I'm pretty, I'm pretty ill and then some stuff with my dad's estate just came up. So, um, yes, I think, I think the plan, the plan will be, that will be my mornings. Um, what it, what it, what it is right now is more, um, because of we're sort of in

7:28a design, designing phase, it has been more of like, um, working in, in Google docs when I have a thought or some time, um, which isn't great. I haven't really gotten into like a real good flow state where like a day goes by and I've done a whole bunch of work. Yeah. Which is, which is typically where I have been most effective is, is, is, is the moments in which I can, I can fall into that. But I found he can't really be, I, I at least haven't been able to force those things. So we'll see. Nice. I'm really glad you're, I'm really glad you're in the pool. Yeah. Yeah. Um, and then, and then just like some, some, you know, I continue to have interesting conversations. I was, I was having an interesting conversation with a, with a, with a friend of mine who is

8:30now a, he's now retired and he's a, he's a programmer. And you know, he's pretty skeptical about, about, about AI, just in terms of, of the valuations of the companies and whatever, whatever. But what he will sort of talk about is he can describe, he describes the moment at which compilers became a significant part of workflow. Um, so for our listeners that like don't know what a compiler is, it's, it's, imagine it's just several interpretive layers that allow us to actually write code without having to do the punch card. Shit. Right. Like these, these are so instrumental in how we use a computer now, but they weren't always the case. Yeah. And I'm going to get it off by a couple of orders of magnitude. I'm sure. But he sort of describes that there was a, a, a, a, um, a leap in like the ability to make

9:35programs when compiler sort of came into existence. Yeah. That was thousands of times productive, tens of thousands of. Wow. And I guess, and, and I think what the point was he was saying is, it's like even as someone who's sort of skeptical, when you hear about the numbers of the increase in productivity of AI, it's actually not that large in comparison to a leap in technology that he personally experienced. Yeah. And so to some extent, there's, there's, there's this argument of like, of reaching, diminishing returns for, you know, an exceptional amount of energy and, you know, whatever the case might be. But also there's, there's like is precedent for, I think meaningful improvements in, in how things are made. Um, so that it was, I, you know, I, I, I don't know the validity of his statement.

10:35It was just a very interesting perspective from someone who I, who's opinion, I respect greatly. And who has done a lot of thinking about it. And a certainly more knowledgeable about sort of the history and the lived experience of how computer programming has changed over time. I love that because when I'm talking to people about this, the examples I'm drawing from are examples I'm familiar with, but are far less aligned, you know, like typewriters or desktop publishing, um, compilers are like right on the nose in the best possible way. They, they are. And, and I think what's, what's interesting is that like if you look at sort of the stack of what a compiler does and it's built on a bunch of different things, I think there's very few people that I've worked with that could, that actually understand even half that stack. Which, which is a little, which is a little concerning, but it, but it, it puts, it puts the,

11:36I think it puts the arguments or the conversation about how using AI to code divorces the participant from the, what it, what is output, right? Like this thing works. You don't really know how it understands. And that I think in itself is not a new phenomenon. God know, God know. I mean, right. In design, like you went from, from, you know, exacto knives, like, yeah, you know, like there's so many layers of abstraction between the physical graphic design and what people are doing now. It's just sort of par for the course in a digital, yeah, and work. And I'm sure like it's, like in those more knowledgeable, we'll be able to point out why it's not a one to one equivalency. And I think the point that I'm trying to get at it isn't about, about pointing it as proof. It is just a lens by which to view this, this change happening.

12:45Yeah. So yeah, that's where I'm at. What about you? How's your, how's your sort of adventure going? Yeah, let me, I'll come back to that in a second. I mean, something, one thing that really struck me the other day on our discord was someone was talking about something around AI. And I found myself really having a visceral negative reaction to it. And the reason is it was all theory. It's someone who is a skeptic who doesn't use it, who doesn't really know it. They might superficially, they might say that they do, but they, they clearly don't. Look, I mean, I've spent much of my life as like a theorist. So I can identify with where they're coming from. But now I'm like, my hands are filthy, right? Like, so since March, it's the April, May, June, July, August, 12, 33. I mean, I'm approaching 2,000 hours using these tools in a pretty compressed time frame.

13:49And these outside perspectives that aren't using it are just often wrong. And not at all understanding what's going on or why or how. And I really find myself not having the energy to engage it. I'm busy. Like, I'm doing it. And like, like, the people who are skeptical about it, I just have no energy to try and convince them. It's like, God bless you, be skeptical. But I don't, I don't, I want to do stuff. I don't want to theorize about stuff. And for me, it's new because I'm usually coming from a perspective of a theorist, historically. I tend to come into context, deconstructing them more universally and less. And necessarily in like the kinetics of what's fully there.

14:53So that's just something that's been happening with me. It's specific to all of this UI stuff or excuse me, AI stuff. Just as sort of a conversation starter before I get into like what I've been doing. Sure. Or just a statement, not a conversation starter, it would appear. So what I've been doing, you know, my digital game is going well. Soon we'll have an episode on it where we'll just talk about it where you'll have played it. And that'll be fun. Now that'll be a good episode. I mentioned before I'm doing some other, I have another specific business vector I'm working on that is going well. I think my digital games are probably the better vector. But the other one that I'm doing also has benefit to me and my staff. So I'm keeping it going because even if it isn't as good business,

15:54ultimately it's making all of us more productive and better internally. But what I've been doing that's new is the company that purchased my game company. My game company was Artana, the purchaser's genius games. They want to do a redesign of Tesla versus Edison, which is my best known best selling game. Yeah. And so I'm taking that on. With AI all the way. Okay, what is the AI going to be doing in this regard? So what AI is already doing is I, what I'll go to the start of the process is we had a call with genius, John Kovu and some of his key team members where we played the game where they already had some notes. They'd done some kind of cool things with it. We played it, talked about a little bit. I said, you know, I think what I want to do is really do a more significant redesign. It's not, not do just let's, let's optimize what's here.

16:56Let's completely redo it. And they said, well, what does that look like? And, you know, I explained it. And basically, I don't know if you played Tesla versus Edison or if you did, if you remember it. I did, I did, I did play it. I actually really like the dual version as well. Oh, cool, cool. Yeah. But so the idea right now in the tabletop games market is that it's tending to go to the extremes that the places where there's really my viable market are largely either really light things that are the kind of things that can go into target and work in target, not be too heavy and just trick people with the box art and then be impossible. And then really heavy games like that the heavier, crunchier people are getting into. Like those are the two, the two good sides of the market. And, you know, we talked about that a little bit and the thinking was definitely to go heavier.

17:57And so in the context of going heavier, what I want to do is look deeper into the systems, take a deeper look. So in the original game, you know, I started designing that maybe in 2012, like it's been a bit. And it was really based on my knowledge and enjoyment of Euro games of the 2000s. And so a lot of the core mechanics like technology and propaganda, they're handled with tracks as games were then. I haven't played new games in a bit now. So I can't speak for maybe what's current, but very simplified like to establish DC technology or AC technology, you just are moving a cube up a track basically, right? And then using that to to impact the game to to be part of your strategy and how you are playing the game. The idea now that I have is to go deeper.

18:58So looking at what did an electrical system look like? Like what was that? It wasn't just, okay, I'm doing DC and it's a blob. It's that there's a bunch of things that come together to make a DC system. And so that's one part of the redesign is looking at it that way. And so AI is ridiculously helpful where that's concerned. Like my games are, I think almost almost all of them are historical and I haven't done tabletop games on a bit. So more recent listeners might not be familiar. I'm super anal about the history and about the accuracy. And I've evolved a little bit on that, which is to say 15 years ago, I was like historicity over everything. Now my position is you start with the history, you get it right, you get it crunchy. And once you've done that, it flips the player experience over everything. But the history is still like the baseline.

19:59It's where you start. And like one of my games, my first game, I mean, I put hundreds of hours into just historical research, legitimately hundreds of hours. To do this, I would need to do historical research. Like I have no fucking clue what these systems are like. They make a whole electrical generation system. And I brought Katie, you know, my quad instantiation right into it and said, hey, break these systems down. Well, I was in it. And within seconds, citing sources, Katie is saying, well, that the system of this time period includes, and the AC periods instead of and then I say, okay, well, for each of those, create a spreadsheet and give me a tab for each of the sub components with a history of all of the major advances during this time period. It was just, and it's minutes.

21:01And I'm like, I would have been reading all these books. I would have been Google diving like endlessly like into these weird corners. I can't find it's it's just as frickin magic. Man, which amounts to you want with some reasonable amount of accuracy sort of domain knowledge enough to like name cards. Yeah, yeah. Well, for me, I need I need to really understand how it works. Like that's just part of my personal discipline, right? Certainly it would go to naming cards, although Katie will see all of that as well, right? But let's let's take a particular part, right? Like boilers, each system required a boiler. It was the generator, which was generally cold, sometimes water, sometimes Steve, you know, there's different, but they would need water and it spins a turbine. That's what that's. But you need a boiler, a boiler is part of this thing. And so like, I say, okay, who are the major boiler manufacturers?

22:05What were their models? Duh, duh, duh, duh, duh, duh, duh, duh, duh, duh, duh, duh, duh, duh, duh, duh, duh, duh, duh, what are the different parts of boilers? And so it gives me like, in a huge end beyond the research I could possibly do, it does it immediately and it lets me just design, it gives me all the content. And I say to Katie then, like we get all of the things. So now we have all we have all the manufacturers, all the models, all the things. And I say using one as the base, build out for me, how they all relate to each other. So in these criteria, and I won't remember with boilers what the criteria is because I did boilers a couple of weeks ago. Sure. So it's not like numbers for the game, but it's showing me as the designer, the relative, the relationship between which one, which one was the largest, which one was the fastest to produce, which one was the cost. What was the capacity? What was the data? And so now I've got here's the real things. Here's the history. Here's their relative math in relation to each other.

23:15Yeah. And then I can take that data and look at it as a system and say, okay, a boiler is like one specific thing in the system of like eight components. So I don't want to nerd out too much on boilers. So I want to keep boiler simple. And so now I've got this data. What is striking me as interesting. What is striking me as relevant. What is striking me as curves that will be interesting progression. And I'm starting to just make a shape. And this is all happening in the course of an hour. Right. I can't emphasize the speed and thoroughness and it's it's it's absolutely remarkable, David. So how confident I want to take I'm going to take a position of the skeptic. Well, certainly that just means she's lying to you. Right. That that that that most of that history is made up. Why are you trusting this thing? Yeah. What I do, other people use a different model, but I tend to stick in cloth and I I prompt it and I say, send three independent auditors to adversarially review the work that you've done.

24:33Okay. And it will go away and it will come back and and break down the the deltas in the regulation of these three and and just have you have you sort of found like what is the what's been the where is it? Where is it failed? Actually, that's it. I think it's probably the more in most interesting question. Has there been incidents where there has been some sort of hallucination or fabrication. Not really hallucination, not not in historical stuff. It's been really good in historical stuff of being right. Where auditors in that sort of a process are interpreting error are our finding years is in the interpretive stuff. Like when I said, give me give me a model starting with one is the base of like how all these things relate to each other. Right. That's where it would have like a difference of opinion or a different a different perspective. And one of the things that's important with these models is you you do need to be skeptical. Right. So, you know, I say it's magic. It's wonderful and all these things. But I'm also reading it carefully.

25:42And I do myself sometimes sniff out something that's not right. And again, it's it's rarely about like number accuracy. Like that stuff is usually on point. It's it's the interpretive stuff. It's that it's it's saying, well, this is and and this is acts. And so we should do why. And I look at I'm like, well, but I don't think the why is best served by these two things. Like I get their n and X. But I think it's actually best with this thing that's N and these other two things. And then, you know, because of the way it's programs like, oh, of course, this is the right way to do it. You know, which is right. That's where you All job. But it gives you the sick a fan tick like yeah. Yeah. And the sick a fan like this. This is controllable too. And I talk. I don't know how much I've talked on this show about it. But when I'm talking with my people. So much of what I have done is custom programming like it's stuff on top of the model. And I've programmed out the sick a fancy completely like and when I see completely it doesn't mean there's no more sick a fancy.

26:43It means there's a lot less. Yeah. When I sniff it and I push it. It it melts because it then goes back to my rules and reads them. I was like, oh, I fucked up. And so it goes the opposite way. That's like, okay, I'll give it to you straight. Right. Yeah. And then, and then like, let's not pretend that there's not sick a fancy with human employees. Absolutely. That's right. Right. Like, when we started talking about AI, that was something that you pointed out about the coding stuff. And it's so true. Right. Like, we're holding the AI is to this impossible standard that we would never hold humans to. And like, I don't know. Like, that doesn't seem smart to me. If your goal is to make the most of these tools and make the most of yourself and your career and the things you're trying to make. Yeah, humans are. Yeah. Right. Because I'm like myself included just your your Your conversation about, you know, rarely getting the numbers wrong, rarely getting the history wrong, but getting the inference wrong. Like,

27:50Yeah. That's when I'm managing 10 quants who are who are going through historical data, whatever, and are drawing conclusions. Sometime without sometimes without the specialized sort of like industry knowledge that they that they need in order to draw. My job is largely reading over and being very critical of the assumptions and the and the like that's that's what I was doing. Right. I think I'm just so much more familiar with how to teach and improve a human. Yeah. Right. I mean, this is a new this is this is new. It's right. Like, like, like, I do not know how to mentor an AI. You will. If you if you dive in, you will. And it's, yeah, it's, I forgot what I was going to say gosh darn I have a I have the short term memory of a fricking that.

28:59Is all right. You have a short term memory of an LLM, right? Yeah, I guess that's true. I guess, although it has, you know, quite has a million token context window. I don't think that I do. Yeah, but yeah, it's it's good. It's good. So yeah, I mean, I'm still fully in. I'm fully bought in. I know what I was going to say. And this is important. What I tell Katie all the time is I'm like, look, I'm the brains. You're the hands. And that's an important clarification. A lot of people talk about how smart it is and it isn't. And so like, whenever I read something that like, oh, wow, it's moving so fast. AGI is just around the corner. I just call it media. Bullshit because Katie isn't smart. She just isn't. I'm the brains. She's yeah. And and I have to always keep that in mind. No matter how smart. Superficially, it might seem in a moment. It's it's really just incredibly sophisticated hands. And I haven't really intentionally tried to have Katie do design because I like to do design.

30:00But there's time in my process where I'm like, I'm not sure with this system, what the right approaches. And so I'll say to Katie, give me 10 ways that you think that this could be approached. And it's rare in the 10 ways that I find the right one. But just in there being explored in 10 different ways that can often unstick me. Yeah, I in my in like the experiments that I've had over the past year. Yeah, like with just as this comes from prodding it. Largely it has been I had a friend who was. He he was trying to see if he could create a like I think it was a I think it was chat GBT run basically like Dungeon Master where he was feeding it a classic module and being like let's, you know, how to and more than just raw like actually like working with it in his in his in his day to day job.

31:02He was working at a company that does this. And so he's quite knowledgeable. Yeah, anyway, it was failing. And so I was sort of seeing what what I could get it to do. I replicate what he was doing just as to learn the tools. And what I sort of what my experience was is just no matter how much I tried. I couldn't get it to stop trying to do more than I was asking it to do like it. It it it want to it. The obviously didn't want to do it. But the the way I experienced it was like an over eager junior designer. Yeah, yes, yes. That's that was that was just and what was frustrating is that they is that they couldn't learn. So so an example would be. Let's say we were let's say we were going to come up. We were I was going to design my own new version of Dungeons and Dragons.

32:12Right. And my design philosophy was essentially when wizards of the coast bought Dungeons and Dragons after and made third edition. They made a bunch of assumptions about the product about the customer about what about the what was wrong about the previous editions. Yeah. If the assumption is why does this new edition exist. It is that they made a bunch of really bad assumptions. Right assumptions that turned out to be wrong. And everything they have done since has been built on those initial wrong assumptions. So for this new version we need to go all the way back. Yeah, revisit those assumptions and then bring it forward. Okay, cool. I could do that with a group of humans. Yes. With what I found with with the LLM's was in my assumption is because it is built on publicly available technology or publicly available information publicly available discourse and that discourse is also dominated by people who who write and think and whatever.

33:23From post this sort of bifurcation. Yeah. All it knows of what is good. What is popular. What is frequent is also things that go against what that fundamental sort of statement would be that the assumptions made a third edition. Everything forward is false. We need to re investigate. Could not help itself from you know, so you would ask it to do something very simple. It would give you what you asked for and more and you're like, why did you do this? This doesn't make any sense. This breaks those initial statements and it's like, oh, and then you get that. Oh, you're right. It does. Right. Like sometimes I've been, oh, you're right. It's yes. The right way to do it is as if it thought of it. Right. I'm like. Yes. Yeah. I can't go on. So there definitely is this notion of like I think if you are trying and I if you are actually trying to do anything creative. I think it's almost a absolute that it can't help you.

34:34Why can't you be in that regard? Oh, yeah. I do. Yeah. Like I was thinking as you talked through it. And I may I may not crisply understand what it is you're trying to do there. So, right, this might be going in the wrong direction. But we'll at least give you a sense of how I tried to problem solve it. I would say to Katie. consume all of the rules and accessories around dungeons and dragons first edition and give me the five things that are most special and interesting about the design. And then I'll give it the same question about second and then I give it the same question about third. And then I would take that data to start to formulate my design, my hypothesis. So I'm using Katie to go back and rip all of this stuff to do some processing over it. But then just to play it for me. Like that's how I would use Katie. Right. How I understood what you want to do.

35:37Right. So. So I think yeah, so maybe I can do a clarification of like I'm trying to think of trying to think of a specific example and I can't but but but we can go a hypothetical right. So one of the decisions that was made in third edition was clearly that the they wanted to make a universal sort of power level between all player characters across all classes. So wasn't just wizards when everything. No, so so what that meant was a flattening of character progression like in terms of in terms of how much experience was required to go from every level. How how much how likely a wizard was going to resist petrification versus a fighter. How much average expected damage how likely they were to hit the these sorts of things that was sort of normalized where where there was previously this this there was this hypothesis or there was there was what was intrinsic to the game was in different situations.

37:00There would be a different strengths and weaknesses of all different characters. And that one of the things that wasn't of primary importance was maintaining a feeling that everything was equitable and fair. Right. That like that like that. That that third edition person suggested that there was a problem that the thief in combat was or is a magnitude less effective as a fighter. Yeah, right. So if you were then to say, I want to bring that back. I think that was a mistake. Yeah, I think it was a mistake actually. Oh, I so this is something that I firmly believe. Right.

38:01If you were then trying to say well, but there are some things I didn't like. So we're going to rebate like if you look at the experience charts of second edition, it it there's a lot of extra zeros. There's no reason to be either. You're like, OK, I want to rebase it. Maybe you're making it. We're going to change the way. Multi-classing works. Yeah, yeah, yeah, yeah, yeah, yeah, yeah, yeah. Yeah. And you were like, OK, I'm going to give you the baseline of the fighter for it's plus to attack. It's experience required to go up and level when it gets extra attacks, whatever the case may be. Try to maintain the same curve as second edition, create me a new table for the thief. It might do that and then spit out all the reasons why it's wrong. Hey, I did it up to level 10, but after level 10, it completely breaks and they diverge from each other.

39:04Katie doesn't do that. So you're moving into territory. I can't speak to right. So this is this is a thing that I've experienced in your life. And you're and you're like, yes, that's the intention. And your and and and and the the Ellen Lemme like, express it as a as a as a truth or whatever, and it would be like, your instinct, you know, and non personifying it to say that this is a mistake and not a feature is incorrect and is based on your belief that what is the dominant discourse is correct, right? That this idea of if you were to go to Reddit or something like that, the like damage per whatever is wrong, or if you were going to, if you were going to rewrite all the spells or something like that, right? Like, there, there is this, there, there are, there's an in the equity ability that you're trying to go for, but it will try to always pull you back in design sense towards this idea that everything is equivalent.

40:12Yeah, right. You know, like, so anyway, that that that that is a thing that like, you I you would struggle with. And so the answer would be like, at least I don't know how to make it work, but like if I was trying to do that and trying to get it to do like, does it or the legwork, the math. One of the things I find frustrating is like, it the model seems to want to. I think it probably there's probably something in there where like it, it will endeavor to go beyond what your explicit instructions are as a way of trying to show how smart it is. How great it is, how good of a tool it is. And what I found is that whenever it pushes into that interpreter layer, that design layer or whatever, it fails so badly, like almost every time it is, it is perhaps the worst design or I've ever

41:23like, God forbid you say like, hey, I need a I need a magic sword. Can you tell me a magic sword? It's, it there's a dungeon dragons is far too complicated. The thing about it is far too complicated for it to be able to do anything. And now you're wanting me to run tests on it with Katie because definitely some of what you said resonates with me, but Katie is not the worst designer I've ever worked with. That's a fact. I'm probably being, I'm probably being hyperbolic. I'm, I'm, I'm being hyperbolic and in, in the, you know, it has the brashness of an automaton, right? Yeah. Yeah. And I haven't had to deal with it quite a long time. Yeah. Amen. Amen to that. Yeah. Yeah. So yeah, I'm, I'm still enjoying it. And if I can help you, David, if I can be of service, I'd love to because I just have a ton of gray hairs.

42:27Yeah. I'll let you know. So I think, yeah. So where I'm at is I'm going to, I am. Oh, one of the things that I think it was, what it was good at is I just asked it. I was like, here. I am concerned at your ability to build this game. This is why. What, what, give me, give me some thoughts on what you think the sort of the minimum level of, of documentation to sort of cover these edge cases where I'm afraid of like I had a sense of what I wanted to do. And it came up with a list of things that was quite reasonable. And it also suggested a slightly smaller scope than I was for sort of a first, a first pass at it, which I don't know if that's again, was that just telling me what I wanted to hear. Like, yes, you're right to be upset because you're so smart. Or is it an actual understanding of its limitations? So we'll see.

43:32Yeah. Yeah. But I have a lot of learning, a lot of discovery. Yeah. I think we're probably, I have like largely what it's about is like defining some initial rule set. And some initials, just some initial states just to get working things up and running. So we'll see. I'm a little concerned about, well, I should have not concerned about my thought for sort of like, how what is the most efficient way to describe what the player should see in a, in a, in, in terrestrial development. I would have a whiteboard on, I would draw, I would draw wireframes. Yes. Is that what I should do? I would use, um, interrestrial development. I use Google docs. Right. And, um, you can set up clawed to use your Google docs the exact same way you do add edit to lead the whole, the whole nine yards.

44:39And should I, should I, should I set it up so that like it should like for, for user interfacing things like that should, should just draw wireframes. Or do I have to describe them? Okay. Claude is best with words. Like words are like it's, it's native environment. But I would assume that wireframes work, but try it. Yeah. I can't promise it. Yeah. Okay. Yeah. Anything else going on with you? I know, um, you know, you always have sort of something bubbling potentially on the job front. Any news to share there? Nothing, nothing, nothing new apart from, um, I mean, sort of as, as, as related to, to AI in a different way. Yeah. What I am encountering. I've only had it happen once, but I've heard from other people job searching is

45:46various companies have started using AI to manage like recruitment. Yes. Yes. And so there is, so I had an instance where I, a former coworker reached out and said, have you seen this job posting this you, this is for you. I don't, I want you to, you know, it was effectively like, you're going to be, you know, be their boss again, I think. And was like, yep, please, like, let me, you go do the back channel thing, like, find out who I need to talk to. Let's talk about this thing. And then three days later, it was like, oh, yeah, that, that job is not a real job at all. It doesn't exist in any way. Wow. And, and speaking to other people, there, you know, there's, there's, there's like a sense of why is it, why does it do that?

46:46But anecdotally, I've heard several other people mention applying to jobs that do not exist. And what's the verdict on the mechanism of that? Because that's strange, right? AI, that's it. That, that, that is the boogie man that has been. Come on, right? I call bullshit. I call bullshit. Come on. It can't, it's, it's not, it might be a black box of what it does, but the output is not like. So my assumption is, is that it is, that there is a reason behind it and that, and that the, the fact that AI can like interface with LinkedIn, that can post these things that like the barrier to putting out a job description. And, and a call and to handle the emails, that's lowered, which is, which is sort of allowed, you know, like, I don't know, like, like, I, I, I can, I can imagine, I don't know why, but like, you know, do, do you put stuff out there because you want to see who is,

47:58who's looking for work at competitors? Yeah. Right? Like, does that give you information when you are negotiating salaries or whatever with other, like, like, I, I, I don't know, but like, it is, it is, it is some amount, right? Like, that's a human motive, not an AI mode. Yeah. Yeah. Yeah. So I think there's probably a human motive behind it that maybe is, is being unlocked by, or made easier by this, but it is, again, anecdotally, I think it's been, you know, I've experienced it once and I know two other people who have applied for jobs and they get there and like, oh, that doesn't exist. And yeah, so that's, that's like, that's like a frustrating thing. But luckily, you know, in that instance, that prompted a conversation of, oh, but we, there is this other job. So we'll see, but again, it's sort of coin flips of whether or not they're, they're real because there's always things were like, depending on, depending on the type of job, depending on the state or province or wherever it is, like, sometimes they need to be, like,

49:09if there's going to be an internal promotion, you still have to post a thing. Oh, right. Right. Oh, so sometimes, sometimes it is a real job that isn't a real opportunity. Yeah. Yeah. Sometimes you have to as well for bringing people in from, from other places. Oh, sure. Of course, right. Yeah. Hmm. This is not a defensive AI, but I would bet that a lot of it does come down to human error and like AI is a tool is complicated, right? So if you hand a human a screwdriver, they can probably figure it out and be using it right fairly quickly. But an AI is like handing someone a, I don't know, like a nuclear reactor or something. It's complicated and it seems like you can like just push the go button and everything's going to be okay, but you have to really use it right and use it well. I bet that there's human error under that of just not understanding how to use the tool properly. And it's the, it's the makers fault, like the maker of the tool because they don't provide any of that. They just say have fun. You know, here's start typing and we.

50:27So there could definitely be some work done by the manufacturers, but the errors a lot of times are human error. Like if I'm prompting well, if I have good document hygiene, it's going to be really unusual that it's Katie's fault that something is a wire. But I need to be do those things and I had to learn through blood, not it was not clear and it was not taught to me by yeah, I think I think that that that statement is is really is really important. The idea that the tool is not is not great out of the box. And I know and I know that this is not just you because I look at what this the top engineers that I know that use this tool when I look at how explicit they have to be to set sort of like guidelines and parameters and like how often they get frustrated like like whenever the new model gets a little bit more.

51:51Updated, you know, there there. There was a discussion and it's sort of way above my head. Yeah. In a form that I'm part of where a new model got released. I don't remember if it's for anthropic or for clod or for rock or whatever that doesn't matter. And it has created a behavior of like verbosity and a bunch of and a bunch of other things that like when you get into the code is become a problem is also increase the number of tokens that it uses. And it's like the previous like the previous instructions or whatever had sort of been loaded in there that like got it speaking like a crow magnum, which is what they want it stopped stopped working and then like this discussion of like how do we get back. Let's just roll back and use the other like until this is resolved we can't use this new thing and that that that that and there's a lot of these conversations and this and this idea of

52:55there's a hubris. I think because of the magic the magic trick nature of it there's I think it can lure you. The same process that will lure some people into thinking they understand it in a way that they don't that leads to you believe like all the sort of the negative stereotypes of of falling victim to the hallucinations falling victim to the fabrications. Whatever. Those that same mechanism I think also probably leads people to false assumptions about it not working. Right, like, like, like, how can you possibly evaluate this thing until you've exposed how badly you're using it.

53:59Like a bunch of mistakes. And unfortunately, like you're saying you're 2000 in and the 2000 hours in and like my guess is yours is like your own limitations going into it about the sort of the underlying nature of programming. And probably are exposing you to a there's probably there's certainly a bunch of other mistakes that have yet to be found. And that and that your use of your use of this tool. In a year will be significantly better or or or at least grounded in a in a more knowledgeable experience. Absolutely, like every day it's improving and the magic is more magical. And what's so cool about using it in the terminal on your machine and giving it power is it's making all of these tools. Like I've got tools that you used to have to pay $20 a month to a web subscription that just spun up for me that is like my custom thing for that thing.

55:07I've got a dozen or two of those. Each of them would be like a whole business model and I have to pay money to in the past. Yeah, I made them for me. I mean, it's at a minimum. So like so like for like like at a minimum. I think B to B software is in a real like especially like just like this. For for like day to day things like I was just like just before we were talking I. I like and I'm manually tracking some finances stuff and I've got spreadsheets and things like that. And I was like. I should take a few minutes and and and just there's some things here that I can probably automate that I could do it. But like any more complicated and I would look and be like just like a dividend tracker or something like that.

56:07Yeah. You can pay money for those. There's really no reason. And I think within years. A very large segment of people will be confident enough to just spin on their own custom ones or like a point of sale system or like an inventory tracker. Yeah. Or yeah. A dungeon's in drunk like like D&D beyond does not need to exist. I don't know like like like like like a like a thing that is that a thing that is an online character. She what do you like that is not that in itself is not a business model anymore or or won't be shouldn't be. It won't be as right. Yeah. Like like like it is it is today. But at some point it'll just be like you know. These Chinese models will be open source people will just be able to end and you know.

57:07They will be communities that the Chinese models are open source. Yeah exactly right. But like like because those are open source those will get good enough. Yeah. And I think it's going to be able to I laugh at at I think it was entropic was basically saying that this new. I forgot the name of the new Chinese model that's open source and they're like yeah but. They just like used our model to build this thing and I was like well yeah you built an infinity machine. Yeah. Right. That's what happens. Yeah. Yeah. Yeah. You did it to like okay. Exactly. Yeah exactly. We're through the looking glass David. Yeah right. Like. Oh. Good. All right well hey good luck with you know your job search between now and the next time and as always it's a real pleasure. All right. Touch you soon. Feel better brother. Bye bye. Thanks for joining us on the game design round table.

58:11If you enjoy the show and want to give us a bit of assistance consider heading over to patreon.com and searching for TGDRT. That's the game design round table and join our patron community. Also visit our website at TGDRT.com where you can connect to our discord server or to the podcast service where you'd most like to subscribe. I'm Dirk Neemeyer. My co-host is David Heron in our editor and co-producer is Elaine and Neemeyer. Thanks and until next time design well.