997: How This Text-to-Video-Game AI Startup Hit 20M Users

24:50Jon Krohn:One other thing that I love about B2C working when it does is that you, you know, if you lose one user, it doesn't matter. You have tens of millions of them. So, you know, these kind of overall trends, you've got to keep an eye on them, But you can get great data on what's working, what's sticky. You can A-B test really well. And so you can figure out. You can iterate over time and build that more and more and more and more. And it's hard for people to take that away from you. Whereas you can end up in a situation as a B2B startup, you could end up getting one or a couple of these big landmark enterprise clients initially that account for so much of your ARR, of your revenue, that losing one of them is like this huge deal.

25:37Jon Krohn:So then even worrying about losing them can end up being a big deal in your mind. Yeah, and then you might be secretly a ChatGPT rapper and then ChatGPT just goes up and eats you up, as I'm sure has happened to many AI startups. Yeah, for sure. The other thing I'll say on B2C is, I suppose, a subset of B2C, which is kind of UGC side of things where we have a platform. UGC? UGC, user-generated content. So we are a platform, people make things on it, and people then experience and consume those things. And that is fun in a way where we do have a moat technically where we have advanced AI and all the kind of magic and benchmarking and so on.

26:27But the ultimate moat is having users. And with UGC, the ultimate moat is having people making good things. Right now, we have a community Discord. Discord is where people come together and chat. I forget what size it is, 50 ,000 people or something. We have deeply, deeply engaged creators who spend hours. Again, I just want to emphasize that the AI is there to let people accomplish their vision. but it turns out to really do that, you need to be very skilled and you need to put in the work. And one of our modes, as of now, I would say, is that we have a community of creators where we now have people who have been there for months making games and have gone really good and kind of encourage each other, talk to each other, provide feedback.

27:19And it's another reason why it's really fun to work at Astrocade.

27:24Jon Krohn:Really cool. I love that. And one of the ways that you're keeping these tens of thousands of highly engaged people engaged, and this is definitely not burning through VC money, this is very well allocated capital. Strategically, you have a$10 million creator incentive program in the mix. Tell us about that. Yes. So as you imagine, if you have people spending hours and hours and hours and putting in a lot of work, making these games for people to enjoy, the dream for us would be to make it viable for people to do that full time or, you know, actually make it part of their living, as is already the case in Instagram and YouTube and TikTok, all these platforms.

28:08And so as you say, we want to thank our VCs for helping make that possible. We now have this$10 million set aside and already being paid out to users in a very straightforward way. If you make a game that people really like and a lot of people play, we pay you per play.

28:29Jon Krohn:Wow. So it's actually, not only is it free, but you can actually make money. Yes. That's really cool. So the, yeah, in terms of strategy, we've got a B2C, UCG. UGC, yes. I was trying to think of another TLA, three-letter acronym that I could throw in there as like the overall strategy here. Vibe coding, I don't know if he, I guess VC is taken, but you can throw something in there. No, just, I was trying to play with some kind of idea of like what the overall strategy is here for you in terms of, you know, where the platform's going and yeah, how VCs can eventually get a good return on what they're doing, but it's really obvious that this idea of incentivizing creators allows you to get a flywheel of high-quality content being created, get more users in there, and then long-term, yeah, once you have lots of people's attention, there's ways you can monetize that.

29:25So cool. Yeah, we'll get there soon enough. Don't worry, VCs. We'll get you your money.

29:30Jon Krohn:Nice. It's in the bank, guys. Nice. All right. So I recently, in preparation for this episode, I read through recent blog posts on the Astrocade website. And there's a post recommending user-generated games may be the hardest problem in modern REXIS, so recommender systems. And so, you know, that's something I've worked at a bunch of startups where the big machine learning problem is kind of like a matching problem. and we're getting the right information in front of our users at the right time. What is so hard about recommender systems and games? Yes, so it's hard, I think, because it's a phenomenally different type of media from, for instance, video.

30:17And most recommendation system stuff, you know, is either video, like Instagram, Reels, or TikTok, or you can have text if you're doing search, right? So with search, you want to get the user the right thing that they click on. With video, you want to give them something that they watch and like. And so you can have very clear metrics of, okay, this person watched this whole thing end to end, right? And they stuck with it. And so this is correct. With games, a lot of time, there's no endpoint, right? There's no, we don't know if you finished it or not. And maybe there's no way to finish it. if you spent a minute playing the game, we don't know if you spent a minute playing the game because you want to experience it or because you're confused and you're just giving it a chance, but ultimately you dislike it.

31:07Right. So the signals there are much harder to parse, to train an algorithm to do that. I would say in comparison to something like TikTok, where if you're watching it beyond 10 seconds, you're probably interested and you want to see more of that type of content. And beyond that, games are also quite complex in the sense of they have mechanics and they have loops and they have, like, how do you say two games are similar? You can say the genre is similar, but even very small tweaks make it a different experience that one of these things you like, one of them you don't like. So I think it's difficult for a couple of reasons.

31:52The rexus part is difficult because a lot of what's been done hasn't been done for this domain of games. And the ways you do it have to be different because the data and the ways people engage is different. So we are working hard on trying to figure out how to do that.

32:10Jon Krohn:I bet. It's certainly, you know, tricky things would be, you know, you have a new game, you know, no play history. How do you decide like quality, genre? Like there's so many, yeah, be complicated. I can see why it's a tricky problem. One of the ways that you're helping people make better games, regardless of what genre it is or what style, is you have something called the Astro Academy. Tell us about that. Yes, Astro Academy is our not very formal, but semi-formal schooling curriculum where we actually have a set of classes where you learn the ropes of how to make a game from start to finish.

32:51And this is going back to that note of, yes, the AI is writing the code and making the assets, but there turns out to be a lot more beyond that. And if you were just to go there and enter a prompt, R.I. is not bad, but R.I. is not likely to get you top tier content of a sort that really gets millions of plays on a platform. So Astro Economy is our structured way to let people who want to make games and join Astro and become creators, kind of quickly learn all the tricks of a trade that we found out are tricks of a trade as we've been doing this. It's now about to enter season three is what I think we're calling it.

33:35We've had multiple iterations now where every time we've learned what helps people, how we can make it easier for people to learn how to do this.

33:44Jon Krohn:So is it like when you say seasons, like that basically means that it's like a refresh of the curriculum or does that mean like, do people follow along with the curriculum kind of in real time in a cohort or is it like a video series you can, you can take on anytime? It's, it's a bit of both. So we do, I think we post the content online, but it is a cohort based kind of real time program. So this is the third iteration of it where we are continuing to iterate on the lessons themselves based on people who have gone through it before and how it has gone and so on. So I think this next one is the most mature, the most advanced, hopefully the most effective way to learn how to make games.

34:29Jon Krohn:Until season four. Until season four. For people, it's a great resource. I'll find a link to the Astro Academy to put in the show notes. But for people who would just like right now kind of want a summary of some of the biggest tips. Do you have some instinct yourself? I realize that you're not a full-time game developer. You're developing this ecosystem, but maybe you're aware of some of the tricks as to what makes a great game. Yeah, well, I've tried to make many games over the months and years, so I'd like to think I can say some things. One thing I'll say is it may be easy to kind of miss that ultimately it is a creative endeavor.

35:14So the sorts of things that typically apply in a creative endeavor of, you need to have an idea in mind and you need to sort of pursue that idea, have a hunch of what this should be and then start working towards it and kind of explore and iterate until you get to something that feels good. So that's number one, is create it as a creative endeavor where you want to make something compelling. Then after that, there is a very basic thing, which is give it to other people to try out and see if you might think it's fun, but it's not actually fun in practice. Those are some of the very fundamental things.

35:59And beyond that, play games. Enjoy other stuff and let inspiration strike you and follow that inspiration.

36:07Jon Krohn:Cool, great tips. beyond having these academies for people to learn and having a great setup that allows this vibe coding to be so effective. It seems like, if we haven't already talked about this already to some extent, let's dig into it a bit more, but it's clear that a big part of the success of a platform like this is that like TikTok, like YouTube, Astrocade allows you to follow people, to be followed. And so I think that that creates, you know, when you are a creator, when I create a podcast episode, when I create a post for LinkedIn, I'm hopeful that people are going to like it, that it's going to, you know, get some reactions, get some comments, that it's not just something that they, you know, scroll right past.

36:57Jon Krohn:Like you're hoping to create something that's a value to people. And so it seems like having a followers following setup, allowing that kind of engagement within Astrocade, that must be a big part of the success. So, yeah, what do you think allows Astrocade to now have tens of millions of users? Do you think it's these social elements? Do you think it's the quality of the AI? Or I guess the answer that you're probably going to have is that it's both. It is both. It is both. Okay, moving on. No, to the point of following and followers, I think what that gets at is, again, a core part of what makes AstroKate work.

37:37And, you know, we are one of many Vibe coding platforms, right? And there's other, even game-specific platforms out there. But one of the things we've learned, having been doing this for years, even going back to our pre-alpha days, is you need to nurture the creator community. You need to treat it as something that people get good at and kind of, the creators are not disposable. They're not just there to enter a prompt. They're people who actually make it work and the AI is just there to let them do it. So you do need following, you know, subscribing and then following because there are now dozens, maybe even hundreds of people who have made, you know, 5, 10, 20 games and have a certain style and a certain set of things that they like to do and keep getting better and better.

38:32So I do think that this is something that is very likely to happen, similar to Instagram, similar to TikTok. If you like one game by this person, you will probably like their next game and you'll want to see that next game. And we definitely want to enable that.

38:46Jon Krohn:Yeah, it's interesting to think that as a creator in the Astrocade ecosystem, people must, I mean, I'm sure to some extent people are like, oh, like I created a Tetris style game. Now I'm going to make an RPG style game. And now I'm going to make like a first person shooter. And, you know, to some extent you probably, because it's so easy, you might want to experiment with different kinds. But I bet it also ends up being the case that certain kinds of creators really specialize in a niche and that you kind of, you know, they over time figure out how to create better and better games. When you were last on my show, one of the hottest takes that you had was AI as a product rarely works, but AI as a feature does.

39:27Jon Krohn:And now it seems like Astrocade is about as pure AI as a product as you can get. Yeah, what do you think about what Andre said two years ago? I don't think our product is AI. Our product is games, right? I see. I see. So the social platform is the product. The social platform. Yeah. Most people don't go there to make games. It's actually a very small percent of people who make the games. Similar to TikTok, Instagram. Right, right, right. Of course. So the majority are playing the games. And I still, to an extent, stand by that. So obviously, you know, Anthropic, OpenAI, et cetera, of their product is AI.

40:08They are selling the LLM. But aside from these relatively few giants, and at this point, if you want to compete with Anthropic or OpenAI or Google, I mean, good luck. You don't want to be selling a chatbot. You don't want to be selling like a wrapper, right? This has been one of the things that has been pretty clear over the past few years. But if AI can be part of the solution to something and you can provide extra value on top of AI or LLM, I would say Harvey is maybe one good example, right? One of the breakout successes where, yes, they are using OpenAI or Anthropik or whatever, but there is a domain specificity there where they can configure it, tune it, put it together in the right way to make it useful.

41:01So I think that take holds up.

41:04Jon Krohn:Yeah, Harvey is a legal AI app for people who aren't aware of that. And it is a sensation. VCs love it right now. They are. It does seem like they're, well, I guess we'll see. I can't speak for lawyers, but it seems to have gotten quite a lot of adoptions. For sure. I think, yeah, it is interesting, you know, when you think about a business like Harvey, it seems like they would be more vulnerable to Anthropic or OpenAI themselves than an Astrocade would be. Yes. Because you guys have. like you have the social network like to imagine OpenAI or Anthropic being like okay now we're going to make a social media platform for video gamers it's like whereas just it being something you know some future iteration you know instead of Claude code it's just like Claude law and it's kind of like they're yeah that just launched Claude for legal really?

41:55yeah so I don't want to say how it is safe but I would think and I would hope that Harvey, the reason they're not losing their minds about Cloud4Legal is they have built things that are on top of AI that make sense and provide value to their customers.

42:19Jon Krohn:Yeah, yeah. All right. Well, I think I'm ready to move on from Astrocade. If you don't have anything else there, I could move on to Last Week in AI. I'm happy to move on, yeah. Nice. All right. So Last Week in AI, as I said at the beginning of the podcast, my favorite podcast to listen to for folks who want to stay up to date on the latest news. You and Jeremy are both funny. You're very intelligent. You have different strengths. You have different opinions. And so it's nice to hear these kinds. And you argue very nicely with each other. Usually, yeah. It's very polite. And I learned so much. I have a smile on my face when I listen to it.

42:57Jon Krohn:You're now five-plus years in, over 300 episodes. every weekend when you release it, it is still pretty much two hours, which is like the full recording blog that you and Jeremy have set out. Yeah, lately, multiple times, I had to run off to actually go to work because our stand-up is right as we finish recording and Jeremy kind of finished up last minute. So it definitely is one of our ongoing challenges of taking as much time as we have every time. Yeah, exactly. But there's a lot to say, obviously, in the AI world. And that's part of how you're so up to date on these things going on with Harvey and Claude for Law.

43:36Jon Krohn:And I'm sure when I listen to that episode, I'll be up to date as well. How has that changed over time? In the past two years, even when you were on the show two years ago, we talked a lot about AGI and how it's going to change the world. Do you think that your perspectives have changed over the past two years? Or do you think the way that you deliver the world's most popular, I think, AI news show. Yeah. What's shifted over the past two years? I think the shift has been kind of the, I don't know if I would call it a shift, but it's been growth, right? I think ChatGPT, we started this podcast and originally Jeremy wasn't on it, but I started it in March of 2020.

44:26It was like right before COVID, which is crazy to think about. So over 60 years ago. And so for the first couple of years, we were doing it. I was a PhD student at the time. So we were talking about AI news. And at the time, it was deep learning. And we had GPT-free, but it was like inside baseball, right? We heard of GPT-free. Venture GPT came out and it suddenly became a normal thing that normal people knew about outside of academia or tech. But it's easy to forget that even in 2023, a month or two after JGPT, most people probably hadn't used it. It did have explosive growth, but it didn't reach 900 million users or whatever it is in 2023.

45:14So over the last two years, we've seen the continual expansion of AI into everything. where every single product has AI features. Everyone kind of, it's in the culture now where you can joke about the tone that chat GPT has or kind of the AI slop that sometimes permeates media. So if I had to pinpoint a change, I would say we focus more on kind of things that everyday people can interact with. We start up every episode talking about tools and updates to ChargPT and updates to Claude. Whereas previously it might have been more business focused or tech focused.

46:00Jon Krohn:And academic papers. Yes. Yeah. Yeah. You still do cover those things though. We do. Yeah. You have sections, you make sure to cover it all, which is, yeah, part of why it takes so long. I think, you know, it's so easy in the first section to have some stories that feel so important, so revolutionary that you need to dig into them a lot. And then you're like, wow, we still need to do academic papers and business and art and and all these things. Yeah, that's a nice way to put it is we are very strategic about spending our time where it matters and not just like bad at time management. I think that's fair.

46:29Jon Krohn:Yeah, exactly. It's been pretty funny. The times that I've co-hosted the show, we'll have, you know, you will have created before we start recording a spreadsheet of the topics to cover. And that leads to a Google doc that as the recording goes on, you, without even talking, you're just deleting sections from later on. I'm like, okay, this story we can probably skip. It's okay, yeah. That does happen quite often. I bet, I bet. One of the topics that you guys mentioned on the show, of course, that gets coverage as well, which is very intimately related to AI, but isn't necessarily directly related to AI.

47:10Jon Krohn:You could have robotic embodiments that don't involve AI, But today, I'm sure pretty much every time you would, just in the same way that you probably wouldn't create a software agent without having an LLM in it today. Right. Like, why would you handicap yourself in that way in creating some kind of agentic approach? So robots are a big thing. And last time that you were on the show, you said that residential robotics was, you know, a big problem holding back residential robotics is that the robots are too strong and that they're buggy, so they could be dangerous. and so what's changed in robotics?

47:46Jon Krohn:I know that that's a subject area that you have interest in beyond just a news item and so let's spend a few minutes talking about robots and what's shifted where you see things going. Lots to talk about robots and a lot has shifted I think in the last two years and it's maybe something that people are not as aware of unless you listened to Last Week in AI. there are multiple humanoid robotic startups led by people who have been researchers at OpenAI or Google, whatever, like deeply, deeply skilled people working on the problem of humanoid robots. And if you've seen any clips of robots, you probably know that outside of demos from Boston Dynamics, where they do Kung Fu or parkour or whatever, they can't do even basic chores.

48:37But one thing that honestly surprised me a little bit is the amount of progress in human robotics we've seen over the past two years. The people working at these companies of One X and Figure and several others have taken the basic technologies behind LLMs and adopted them to this embodied setting with things like video action language models, I think, VLA, video language action models. So to me, there's been a surprising amount of progress towards general purpose robotics. And that was really the challenge always is we had AI for robotics. I did machine like traversal planning, even had autonomous driving back in 2024 already.

49:25Waymo was already servicing customers. but to do humanoid robotics you just need to do everything all at once you need to do perception you need to do motor control you need to do common sense reasoning and so I've been impressed by obviously you want to take the demos with a grain of salt but it seemed like humanoid robots are making much more progress than I might have expected and the question of whether you'll have these household servants assistants who do chores for you, who do laundry, who do the dishes. I think it's now quite possible that technology is going to be there within, let's say, two to three years.

50:15Two years ago, I probably would have said a decade. Now I think it's closer to two to three years. Wow, that's a big difference. And it's more of a question of economics at that point of like, does it actually make sense to have a robot that costs$50 ,000 to do dishes for you. I probably would not pay for that. I would do the dishes myself. But yeah, it's a very exciting moment in robotics in a way that unless you're kind of following this area, you might be aware of.

50:42Jon Krohn:Really cool. Thanks for that robotics update. And then in terms of the apocalypse coming, singularity, artificial superintelligence stuff, as we talked about on the show last time, one of my favorite quotes from you that I have recycled in a million different ways, paraphrased, probably plagiarized word for word, is one of the things you said, I don't even know if you'll remember this, maybe something you think about a lot, but you really stuck with me, was this idea that it's not going to be as good as we hope it might be, it's not going to be as bad as we fear it might be. You know, it's just going to be somewhere in the middle, there's still going to be problems in the world, but, you know, we'll be able to solve some of them, we're going to create some new ones.

51:24Jon Krohn:You probably still feel the same way today. I think so. I think in some way it might be a cop-out answer where like, it's probably not either outlier. It's probably somewhere in the middle, but it's, it's simultaneously banal, but also it's an important thing for me to reiterate in my head because it's so easy for our mind to roll forward with when something bad happens politically in the world. This isn't confined to just AI stuff. When something politically bad happens or good happens, it's so easy to be like, wow, things are going to be different forever. But it's more like it's a pendulum swinging back and forth more.

52:09Yeah. On the question of generally AI risk and AI danger and even existential risk, I do still kind of have that general sense of you can never predict the future. And the general question of AI risk is inherently trying to predict the future of like, how much should we be worried? And we should be worried. AI has many negative outcomes that are already being evident and will become more and more evident. And I think what you're pointing out is that it's easy to get overly focused on the extreme cases of complete extinction and miss out on the many, many more mundane, but much more real and kind of like definitely applicable things to think about.

52:59So one example, scamming, right? That was the first one that came to mind for me.

53:04Jon Krohn:Exactly. Like now we've talked about some new stories. I'm sure we can't even imagine the amount of scamming that is now powered by AI. This is not something that if you were to be like, oh, what is the number one risk of AI in the next two years, two years ago, I don't know if you would say scamming, but I would say that's one of the most harmful aspects of AI today, probably. And so I think there's always this tendency, we are all as humans kind of attracted to stories and it's easy to think about the most exciting story of AI, which will be Terminator or it'll be self-replicating nanomachines or super intelligence.

53:49but you do need to really keep in mind all the more mundane realities of scamming, of people becoming overly obsessed with chatbots and that becoming their romantic partner and losing ability to socialize. And yeah, on the positive flip side of like the utopian vision of, oh, we're all going to write poetry and whatever, and it's going to be great. that is the same problem of ignoring the middle ground of all the kind of practical applications and outcomes of AI that are inevitable and that we'll need to contend with and are already contending with.

54:28Jon Krohn:Yeah, really great summary there. And as usual, mirroring my opinions basically exactly, which is nice. Some great confirmation bias for me. Love it. And what do you think about nuclear fusion and how that's progressing? It's our get-out-of-jail-free card for everything, right? I think superintelligent AI is generally, yeah. I'm hopeful we'll solve cancer and we'll solve all the diseases and nuclear fusion and all of our problems. But I am a SAI skeptic. I'm a skeptic on artificial superintelligence as opposed to AGI. ASI. ASI. ASI, I'm a skeptic, I think. And that's part of the reasons I'm not as worried about the extremes.

55:16I think the reality is there are inherent limits in physics and in computation and in science. You can't just come up and know the answer to something. You need experiments, and that is the physical world. So I'm very enthusiastic about the way AI will accelerate science. And we've already seen that happen with DeepMind, obviously. But I also don't think it will kind of magically solve everything once it gets good enough.

55:50Jon Krohn:Sure, exactly. It's this gap. I think we were talking about this actually before we were on air, which is the risk of... I just... Should have just kept quiet? Yeah, we just should have not talked at all before the cameras came on. But, you know, there's different definitions of artificial general intelligence. Chip Hu Yen in the episode that we were just recording, but will come out next week for you listeners. That Chip Hu Yen episode, you know, we get talking about AGI. She's saying that, I don't know, you might know this because you host a new show, that Microsoft is apparently like taking open AI to court because there's like a, They had an agreement where once AGI happens, that triggers some things that are beneficial to Microsoft in the OpenAI contract.

56:40Jon Krohn:And so, yeah, lawyers are saying we have AGI. And to me, you and I were discussing how, to me, AGI is – I can't imagine that it really is general if it can't do plumbing, if it can't do electrical work. it isn't good enough for me if it can do everything that you can do at a computer. And that includes things that, you know, you might not traditionally think about as being at a computer, like being a physician. You know, if you have a camera and, you know, maybe some other tools, a lot of different kinds of medical work, you know, could be fully automated. But if you can't also be automating cleaning someone's bed sores in their medical bed, I don't know.

57:25Jon Krohn:It just doesn't feel to me, it doesn't feel expansive enough to be AGI. Yeah, I'm going to be honest. I kind of, I'm glad to discuss AGI, but I hate the way AGI gets discussed, which is there is no definition of AGI. And when people talk about AGI, they have no definition in mind, almost always. So you can't define it many ways. But the typical definition is just vibes. It's just AI has gone so good that it's beyond us or something, right? And I actually remember at one point, Andre Kapoffi was giving a talk at Stanford about AGI and the potentialities of it. I asked him what AGI is, and his answer was, you know, I know it when I see it, which is a reference to some Supreme Court case.

58:15Yeah, exactly. For most people -

58:17Jon Krohn:On porn, I think. Yeah, so it's like people don't have a definition. They know it when they see it, and they're like, that's AGI. But I do think that, if anything, the more kind of outcome-based definition of AGI is real when the impacts of AI are such that society is fundamentally transformed. I guess probably when you know it, when they GIs here. I guess so. And yeah, and then, you know, like the cloud code stuff, the meter, I mean, it's not just cloud code, it's Codex as well, Gemini, CLI. It's not like one company. And that also is kind of an interesting thing showing how, you know, some of the fears around artificial superintelligence and singularities that like, you know, the first person to get there only has it.

59:08Jon Krohn:But it's like, you know, yeah, it just seems like there's so many players there kind of neck and neck anyway. And it's a jagged frontier where, you know, one frontier lab might be slightly better at washing the dishes and another one is a cleaning bed sores. If I can, I will get on my soapbox about what AGI should be defined as. Okay. And I think that's going to be the last. We need to start wrapping up. We've had AGI since GPT-3, even before GPT-3. AGI is artificial general intelligence. Yes, yes, yes. Right? Which is as opposed to artificial specialized intelligence. So if you can do multiple things, if you're not AlphaGo, if you're not a Go playing AI, you're general.

59:49So it's a spectrum, right? There's a spectrum of generality and capability. And I wish people discussed it as such, but that is not the case.

59:59Jon Krohn:Yeah, I think there's a, I'll try to remember to put a link in the show notes while I'm writing down a note now to do it. So it should be there. That you can refer back to my five levels of AGI podcast episode where I go over the Google DeepMind paper. They do a great job defining it. It's very concrete what the levels are and how broadly it applies. It's a good benchmark. My favorite definition is that greenhand paper. Me too. We're the same person. All right. Andre, it's been great having you on the show again and been great recording with you again in person in San Francisco. Loved doing it.

1:00:35Jon Krohn:Before I let you go, as you know, I always ask my guests for a book recommendation. Yeah. I wonder what my previous one was. this time I'll go with one I'm currently reading I can actually tell you what it was last time because I brought that, it was House of Leaves I was tempted to say that again because I do think that's a great book but I'll go with one I'm currently reading or rather listening to which is Red Rising it's a science fiction book about this somewhat far future Martian society it's fiction it's kind of if you like sci-fi and you like socioeconomic commentary and just good old fashion engaging storytelling I really like that book so far nice alright Red Rising thank you for that recommendation and other than the last week in AI podcast how should people be following you I think it's actually kind of tricky I can't even tag you on LinkedIn oh Well, I wish I probably should be using it more, but I'm on Twitter, Andrey Kurekhov, Twitter slash X.

1:01:47And otherwise, I guess listen to Last Week in AI.

1:01:51Jon Krohn:It's great. For sure. That is the best way, I think, to know what Andrey and Jeremy are up to these days. Thank you so much, Andrey, for taking the time out of what I know is a super busy schedule for you. Really appreciate you doing that. And yeah, hopefully we can catch up again on air sometime soon. Yeah, thank you for having me.

1:02:11Jon Krohn:exceptional episode today with dr andre karenkov in it he covered astrocade of course which he describes as a tick tock of video games where anyone can vibe code a game with ai publish it and have other people play it and which has now grown to over 20 million engaged users he talked about his view that there's no secret sauce in the tech stack just an agent some tools and an llm and that the real difficulty is mixing those standard ingredients well and building so your scaffolding doesn't become obsolete the moment the next model lands. He talked about the surprising pace of humanoid robotics from companies like One X and Figure, leading him to revise his estimate for capable household robots from roughly a decade down to two or three years, with the real holdup now being economics rather than capability.

1:02:55Jon Krohn:And he told us why he's a skeptic on artificial superintelligence, since physics, computation, and the need for real-world experiments mean AI can accelerate science enormously, but can't simply think its way to every answer. All right, as always, you can get all the show notes, including the transcript for this episode, the video recording, any materials mentioned on the show, the URLs for Andre's social media profiles, as well as my own at superdatascience.com slash 997. Thanks, of course, to everyone on the Super Data Science podcast team, our podcast manager, Sonja Brejevic, media editor, Mario Pombo, partnerships manager, Natalie Zajski, our researcher, Serge Massis and founder, Kirill Aromenko.

1:03:34Jon Krohn:Thanks to all of them for producing another stellar astronomical episode for us today. For enabling that super team to create this free podcast for you, we are deeply grateful to our sponsors. You can support this show by checking out our sponsors links, which are in the show notes. And if you ever want to sponsor an episode of the podcast yourself, you can find out how at johnkrone.com slash podcast. Otherwise, please help us out by sharing this episode with someone that would love to learn more about Vibe Coding Video Games. Subscribe if you're not already a subscriber. Review the show. I can't tell you how helpful that is.

1:04:08Jon Krohn:If you do that on your favorite podcasting platform or on the YouTube videos. And most importantly, I just hope you'll keep on tuning in. I'm so grateful to have you listening and I hope I can continue to make episodes you love for years and years to come. Also, very special thing. If you've listened this far into the episode, you might love to know that we are doing a special live recording for episode number 1000. That is coming up. This episode was released on June 2nd. And we're going to be doing this just in two days. So if you're a dedicated listener, listening to episodes all the way to the end, you're getting a special surprise.

1:04:50Jon Krohn:Come join us on June 4th at 5 p.m. Eastern time, 2 p.m. Pacific time to get the sign up details for that. You can go to my LinkedIn page. we'll also have it for you in the show notes to check out to go directly if you don't want to go to my LinkedIn page for some reason I posted about it yeah a week ago on May 28th there so yeah hopefully see you for episode 1000 we're going to be doing a live recording so Kirill the founder of the show Kirill Arimenko and I will be there and we will do a little history of the show and And then we will be available for people to come in into the podcasting platform itself.

1:05:35Jon Krohn:So you can use the chat functionality to chat in real time with so far, you know, many dozens of people have already signed up. And you'll also have the opportunity to kind of knock and come join us. I don't know if we'll be able to do that with everyone that comes join. It seems like a lot of people are going to come. But some people will get to come on live with video and ask us questions or make comments in real time celebrating 10 years of this show. So yeah, looking forward to doing that with you. I always sign off as I'm looking forward to enjoying another round of the Super Data Science Podcast with you very soon.

1:06:11Jon Krohn:And yeah, for episode 1000, you will be able to do that June 4th, 5 p.m. Eastern, 2 p.m. Pacific. Come join us. Again, the details are in the show notes and via my LinkedIn page. All right, catch you soon.

1:06:27Thank you.