Latest / Newcomer Pod / Someday That NPC Could Be More Alive Than You Are (w/ Amy Wu & Keith Kawahata)
Transcript
- 0:10Hey, welcome back. This is Cerebral Valley.
- 0:13I'm Eric Newcomer. My guest on this episode will be
- 0:16Amy Wu of Menlo Ventures, who was an investor at FTX and deep
- 0:21in the crypto world and is now hardcore into AI Gaming
- 0:25Investments, and Keith Kawahata, who's the CEO of a stealth AI
- 0:30gaming startup and has been a mobile gaming exec at a bunch of
- 0:34different. Companies.
- 0:35So we need to talk to him. I need to hear more about that.
- 0:37Yeah, exactly. I have Max Child and James
- 0:40Wilsterman, my Cerebral Valley Co hosts and Volley Co founders
- 0:45here. This episode is very much in
- 0:48their domain. We're going to talk about
- 0:50artificial intelligence and its impact on entertainment, and
- 0:54some of the time, particularly their field gaming.
- 0:57Welcome back. Happy to be here.
- 0:59Explain Volley and the how sort of the AI games world as you
- 1:05guys experience it. Sure.
- 1:08Well, we're definitely in a very specific corner of AI games, and
- 1:13we're built around voice control.
- 1:14So we build games that you talk to rather than play with a
- 1:18controller or, you know, using a touch screen like your
- 1:21smartphone. So we have games that you can
- 1:23play on. You know, an AI device like an
- 1:25Amazon Alexa as well as that's where the majority of them are.
- 1:28That is where the majority of our users are today.
- 1:30But actually the fastest growing part of our company is on Smart
- 1:32TV's, Roku and Fire TV. And we have different types of
- 1:36games you can play with your voice.
- 1:37We have game shows like Jeopardy, so we have like
- 1:40Jeopardy on Roku and you get a big board in front of you like
- 1:43you're actually on Jeopardy and you can buzz in with your remote
- 1:46and you know say give me US Capitals for 200, Alex.
- 1:49And you know what is Sacramento and?
- 1:52Yeah, we have a number of other kind of voice control gaming
- 1:54experiences, including storytelling, kind of fiction,
- 1:57Choose your own adventure that you control with your voice, as
- 1:59well as some music stuff. Yes, Sir.
- 2:02Is that still? Are people still into that or is
- 2:04it? Yeah, I mean, it's actually
- 2:06pretty popular. That's our Choose Your own
- 2:07adventure fiction game to it gives you like two options and
- 2:11you go one way or the other. It's been out for six years.
- 2:14It's still going, but it's not. I would say it's not the cutting
- 2:17edge of all. Yeah, stop talking about.
- 2:18That but no. But I think it's an interesting
- 2:20tie to where we are today with AI because the reason we built
- 2:24yes sire at the time was to really simplify what you could
- 2:28say to the device, really dumb it down a lot so that it never
- 2:32misunderstood you or you know, couldn't process your language
- 2:36input. So it was 2 choices, you could
- 2:38either say yes or no and that would affect the game.
- 2:41I think where we are today with AII think it's much, you know,
- 2:44the technology has improved drastically where it, you know,
- 2:47it recognizes things better. But also now with LLMS, we can
- 2:51take in a lot wider sort of inputs around natural language.
- 2:57So you know, obviously part of the volley bet is which has a
- 3:02huge AI component because there's sort of the voice
- 3:04recognition and all that and we could probably do a whole
- 3:07episode on and you guys are obviously big advocates of voice
- 3:10being sort of one of the main ways we'll talk to AIS.
- 3:14But I'm curious like when you think of gaming and AI
- 3:17generally, I mean I think of like creation or what are the
- 3:20pieces where you actually think artificial intelligence and
- 3:23generative AII mean assets? You know, like what are the
- 3:27parts of games you see AI fitting into?
- 3:30I think there's a lot of different layers of the stack as
- 3:32we would call it, right? And I think that, you know, our
- 3:35company's built on some stuff that really started working.
- 3:38Five or six years ago in the sort of last generation of you
- 3:41know, deep learning, machine learning, AI, whatever you want
- 3:43to call it at the time. And that was just speech
- 3:45recognition getting pretty good all of a sudden and you know
- 3:48Alexa being one of the big steps forward there.
- 3:50And then secondly like what we call like natural language
- 3:52processing, which is like turning human phrasing into
- 3:56meaning, right. And both of those got pretty
- 3:58good like 7-8 ten years ago, however you want to measure it
- 4:01and that was sort of what enabled our company to exist was
- 4:04like. You could build a game on an
- 4:06Alexa device and when you said you know something to a a
- 4:10storytelling game guest sire, it would understand it and we could
- 4:12do the right thing, right. But to your point, what's
- 4:15exciting about, you know, the last year and a half and the
- 4:18reason we have a cerebral valley conference is the sort of the
- 4:21generative element of AI has really taken this hugely
- 4:24forward. The idea that we can create new
- 4:26things that that you know, AI can create things that.
- 4:29Humans either couldn't create as quickly or they can do remixes
- 4:32that have never been done before.
- 4:33And there's a few different things that AI can generate.
- 4:36One is just text, right? It can generate scripts for a
- 4:40storytelling game or a movie, right?
- 4:42It can generate images like you were talking about assets within
- 4:46a game, so like art that you could use within a game, you
- 4:49could have an AI generate it, like a stable diffusion or a mid
- 4:52journey. And then one thing that's
- 4:54relevant for us is also it can generate.
- 4:56Fake voices, right? It can create synthesized voices
- 4:59very easily. You know 10 years ago.
- 5:01Does Amazon let you do that right now?
- 5:03Yeah. So Alexa was one of the first,
- 5:04you know, high quality synthesized voices.
- 5:07But ten years ago it cost like a quarter $1,000,000 to make a
- 5:11good synthesized voice. And then today with the sort of
- 5:14generative transformer wave, it costs 5 bucks now.
- 5:17So it's come down by, I think with the script we're going to,
- 5:20you know, you know, if I'm too busy, they're going to slice in.
- 5:23Yeah, right. Scott's going to fill in my
- 5:25voice as needed. So yeah, so that's what I
- 5:28actually said on this. Yeah, exactly right.
- 5:31You know, some people call it like deep fake voices where I,
- 5:33you know, I think it's like synthesize voices is kind of an
- 5:35amazing. I mean, it's literally like
- 5:37100,000 times better than it was 10 years ago, which is like
- 5:40really big. Yeah, one of our most.
- 5:42It's recent games on Roku where you can play Jeopardy.
- 5:45It's actually hosted by one of our employees.
- 5:48We synthesize his voice and it sounds amazing.
- 5:51It sounds really realistic, but we were able to do that because
- 5:53of the advancement in the technology.
- 5:56Right. And so in in all kinds of
- 5:58gaming, whether or not you believe voice is the input or
- 6:00controllers of the input or it's mobile gaming or whatever.
- 6:03I think the output stuff, the content of the game, the art and
- 6:07the assets and the voices and the way you hear the experience
- 6:11and the music as well we haven't talked about as well.
- 6:13Like you could synthesize music and games.
- 6:15Those are all like completely going to change or changing as
- 6:19we speak across the whole world of gaming from like AAA, you
- 6:23know, crazy console games all the way down to like more casual
- 6:26experiences and and more sort of family game night type stuff,
- 6:29which is a lot of what we built. James, which one do you think
- 6:32will be the actual most revolutionary?
- 6:34Or, you know, there's a lot there, but which piece of it are
- 6:38you most excited about? So Max and I really like this
- 6:42scene in her, which people probably don't remember very
- 6:46well, but it is a scene in HER where the main character is
- 6:51playing a video game and it's really interesting because the
- 6:57character in the game is like a foul mouthed alien.
- 7:01Sort of child. And the player has to kind of
- 7:06interact with it, with natural language and convince it to help
- 7:08him on a quest. But you know, it gets the
- 7:13character is not really helpful. It's like very adversarial and I
- 7:19think that opens up some new kind of ideas and gaming and an
- 7:23AI. Right now, every assistant is
- 7:26trying to be extremely helpful. Whereas once you kind of put
- 7:29this in an entertainment context, you can have
- 7:31adversarial AI that are, you know, leading you astray.
- 7:35What what do you think feels like I, you know, we see I
- 7:39haven't played them, but you know, you hear about the No
- 7:41Man's Sky, like they're all these space games.
- 7:43I don't know if they're not using LLMS necessarily, but
- 7:45they're trying to generate sort of.
- 7:47Audit generation. Yeah, right.
- 7:49There's a lot like, let's throw technology at this stuff at, you
- 7:53know, human storytelling is still, you know, the best thing
- 7:56out there, and a lot of this stuff can end up feeling
- 8:00gimmicky. What do you which feels close
- 8:03that that would actually be useful?
- 8:05I think there's, I think there's three layers of like complexity
- 8:08of the content generation you can do in gaming and in sort of
- 8:11entertainment more broadly. I think the first layer is you
- 8:13can just generate some content. Let's say it's, you know, a
- 8:17script in a game or in a movie, or in our case like a trivia
- 8:22question, let's say. Right?
- 8:24And. You can generate it from an LLM.
- 8:26You can ask an LLM for a Jeopardy question right now.
- 8:29You'll get something pretty good.
- 8:31You know, we would look it over and have a human in the loop
- 8:34doing some editing, but then we we put it into one of our trivia
- 8:38games, right? And so that would be like, hey,
- 8:40the humans who work for Volley or whatever game company now
- 8:43have to do a lot a little better.
- 8:44Yeah, they just got a little better, right?
- 8:46They're they're faster writers. Or let's say they're an artist,
- 8:48right? And they can generate artwork
- 8:50way faster, or generate a bunch of ideas really quickly and then
- 8:54discard some of them. And iterate on some of them, but
- 8:56in the end there's a review process and then these things
- 8:58ship into the games, right? That I think is like happening
- 9:02as we speak, right? I mean within our company, like
- 9:04that's happening as we speak. Within other gaming companies.
- 9:07I'm aware of people already doing generative art for in this
- 9:11kind of pipeline where they use a mid journey or stable
- 9:14diffusion or whatever or some specialized tool and then they
- 9:16review it and then they ship it out to to their players, right?
- 9:19So. That's like Level 1 and that's
- 9:21happening like for sure. But it is basically like
- 9:24augmenting the human in the loop, basically.
- 9:26Then there's like the second layer, which we talked a little
- 9:29bit on the first episode about, but it's like you start
- 9:32personalizing the content within a gaming experience in real time
- 9:35or you were talking about a TikTok feed being personalized
- 9:38for each person. But a gaming content is often
- 9:40easier to sort of think about where it's OK, let's use the
- 9:42Jeopardy example again. Let's say that.
- 9:44Every question within the Jeopardy game that I'm playing
- 9:46is is literally decided for me by a large language model idiot
- 9:51and he needs to win enough for him to get claims.
- 9:54Well, not just difficulty adjustments though.
- 9:57But what if you were able to say, you know, if you're really
- 10:00like obsessed with Harry Potter, you could just make the whole
- 10:03board about Harry Potter. It would generate instantly,
- 10:05right? Or if you're studying for.
- 10:08You know an AP physics class? Like, why not put that into our
- 10:12game and play a Jeopardy board of AP physics classes?
- 10:15Right. I think those will be shipping
- 10:16out more and more in the next three, 612 months where you're
- 10:19creating personalized experiences and or you're
- 10:22enabling user generated kind of content experiences in games.
- 10:25And then the last level was sort of what James was talking about
- 10:28where AI are characters in games, right.
- 10:31Let's just say you're playing Zelda and you run into some
- 10:33dwarf or whatever and like. It just like talks to you right?
- 10:37And you're talking to the dwarf and like you it has some mission
- 10:41in mind which is to give you a potion or whatever and but you
- 10:44can ask you like anything. You could be like hey like where
- 10:46should I go next? Like why is my sword like you
- 10:48know wearing down like how many you know coins is it going to
- 10:52cost me to upgrade my shield or whatever right?
- 10:54You can ask me questions but in the end there's some object of
- 10:56that conversation right? And I think non player
- 10:59characters like that. Generated by large language
- 11:02models is, you know, it's really expensive is honestly the
- 11:05biggest thing holding it back right now.
- 11:07If you want to run a really good large language model like at GBT
- 11:10at at that kind of scale, having millions of gamers playing it,
- 11:13it's like really expensive. But I think that in a couple
- 11:17years or a year or two or whatever, there will be ways to
- 11:20sort of minify these models or focus these models or just the
- 11:24hardware will get cheaper or whatever, and then you'll start
- 11:25to see the stuff in. World of Warcraft or Zelda or
- 11:28whatever, you know, and and just to zoom out, like why are we
- 11:31talking about gaming besides obviously you guys being
- 11:33experts. I do.
- 11:34I I mean gaming in so much of technology is sort of the
- 11:38bleeding edge of what consumers experience.
- 11:40Like obviously in AI right now there's a ton of infrastructure,
- 11:44wonky stuff that the regular person's never going to see.
- 11:46But I do think gaming and entertainment will be where a
- 11:51lot of this stuff emerges. You know, we we've talked about
- 11:54before, you know sort of the chat apps which are sort of.
- 11:57Interactive entertainment, Yeah, right.
- 11:59Well, I mean, NVIDIA, which is the biggest company in AI right
- 12:01now, literally spent decades selling GPUs to gaming
- 12:05companies, right? I mean, NVIDIA exists because of
- 12:07gaming, right? I mean, to your point, gaming is
- 12:09always it's, you know, it's the bleeding edge of entertainment
- 12:12because it is fundamentally just like interactive entertainment.
- 12:15And you know, when you add technology to entertainment, you
- 12:18get gaming basically, right. So I think that.
- 12:20Yeah, I and I also just think that if you look at industry
- 12:23dynamics, gaming now makes more money than movies and TV and
- 12:28music like all put together. It's it is the biggest
- 12:30entertainment industry in the world by a huge margin.
- 12:33So it is like fundamentally how people actually really
- 12:36experience entertainment these days.
- 12:38I wanted to pull be the journalist and annoy my friends.
- 12:42Would you use at Volley, mid Journey or Dolly assets or what?
- 12:47Yeah. What are your lawyers telling
- 12:48you in terms of like actually putting these things in the
- 12:52game? Because some of this is going to
- 12:53be like, which big company? You're not as big, but which big
- 12:56company is going to take the risk on this sort of thing,
- 12:58right? Yeah, I think that's a really
- 13:00interesting question and I honestly I think we are
- 13:03discussing this a lot internally to give a real answer.
- 13:06It's it is a topic of debate inside our company to say the
- 13:09least. And I think that you know, I
- 13:11think the maybe we'll get to the whole copyright discussion later
- 13:13in in the the podcast, but I think that the question of
- 13:17whether or not. Training a huge image AI on a
- 13:21bunch of like images you scraped off websites is legal or not is
- 13:25like an open question. And so I think we would try to
- 13:30use something where we had some good information about the
- 13:34provenance of the training data. And I don't know, James, you've
- 13:37obviously been thinking about this even more than I have.
- 13:39Yeah. What?
- 13:39What do you think? Yeah, I think that right now the
- 13:42answer is no. We don't use AI generated images
- 13:45in our products and I think we are internally.
- 13:49Debating that and trying to understand where this image data
- 13:52came from and how do we feel about the legal implications of
- 13:56that. I think that there's so Dolly
- 13:58Three was announced recently. I think they specifically
- 14:02mentioned there would be like robust opt out features for
- 14:05artists and their training data. And I think they will start to
- 14:09see more and more, you know, like Max said, tools that
- 14:13actually have to delineate the provenance of the data and then
- 14:16maybe that opens up. You know a legal path for it,
- 14:20but it's really an open question.
- 14:22Did you guys see the Getty Images just launched a image
- 14:25generation tool? Yeah.
- 14:26So that's pretty exciting. I think the progress in 100%
- 14:29licensed image generation is very fast, right.
- 14:33And so I think if you are concerned about that side of
- 14:36things, I think that you now have Adobe Firefly, which is all
- 14:39licensed. You have the Getty generation
- 14:42tool, which is all licensed. So I guess I just sort of think
- 14:44this isn't. There's going to be that much of
- 14:46a problem even in a year from now, because you're going to
- 14:48have pretty robust, you know, license generative tools.
- 14:52I do think like an interesting subsequent question here though
- 14:56is these are these types of questions are being litigated in
- 14:59the courts, right? I mean, there are a bunch of
- 15:01lawsuits occurring. Sarah Silverman and people.
- 15:04Sarah Silverman and other artists, and I don't know, What
- 15:07do you guys think? Do we think that there's a
- 15:09clear? Outcome of that based on current
- 15:12law or do you think they're you know that these it seems very
- 15:16serious like it I I do think there's got to be some copyright
- 15:19I mean not being a lawyer but just.
- 15:22Do you, do you think that applies to LLMS to then?
- 15:24I mean, that's essentially what Sarah Silverman is saying,
- 15:27right? Yeah, Yeah.
- 15:28Yeah, I hope. I mean.
- 15:30Would you have an issue if I could create a newcomer article
- 15:32based on scraping all the newcomer articles?
- 15:34I mean, you're a journalist, so a lot of your original content
- 15:37is from reporting, not from your writing.
- 15:39But yeah, I mean definitely, yeah.
- 15:42So you you do have it. I mean that's part of, I mean so
- 15:45yeah, this is exactly I the writers strike, we want to talk
- 15:48about it, let's talk about it because it's directly relevant
- 15:51to this question. I think you know even me sort of
- 15:55being very bullish on AI and like around it, you can sort of
- 15:58laugh a little bit at the writers making like prioritizing
- 16:02AI sort of like a top priority. It feels paranoia.
- 16:05On the other hand, all the things we've said in this
- 16:08conversation, it feels very current that, you know, they
- 16:10basically won protections that you know, you know, chat product
- 16:14products can't replace their work.
- 16:16Like the writers get to choose if they use it and sort of a
- 16:19bunch of different. And they got that basically.
- 16:22Yeah, that was the resolution, right.
- 16:23I mean, do you think they needed it?
- 16:24I don't know. I guess that was that a real and
- 16:26present danger. So I have a good story on this
- 16:29front, which is I was meeting a big like Hollywood studio honcho
- 16:33like six months ago and. He was like, oh, you know, what
- 16:36do you think about all this AI stuff?
- 16:38And I was like, oh, I think it's pretty cool.
- 16:39You know, you prayed on with chat, JPEG, whatever.
- 16:41And he's like, yeah, like, I asked it to come up with some,
- 16:44like, new movie ideas for our studio.
- 16:46And he's came out with seven ideas.
- 16:47And he's honestly like five of them were pretty good.
- 16:50He was like, he was like, I think I might use like five of
- 16:52them, honestly. And just like a real Hollywood
- 16:55studio guy. Yeah, exactly.
- 16:57I never know what good ideas of you in the face.
- 17:01I mean, I'm just saying, I was in the room where the guy was
- 17:03like, Oh yeah, like these are. These ideas were pretty good my
- 17:06friend and and it didn't produce screenplays for him for the
- 17:11record, but it had some concepts that were going to work for him,
- 17:14so. Yeah, well, I think the
- 17:16agreement that was struck here, you know, a lot of it had to do
- 17:19with can you take, can the studios take these old scripts
- 17:22that they own that the writers wrote but, you know, gave away
- 17:26those copyrights essentially to the studios.
- 17:29And they use them to train future scripts.
- 17:31And I think what is, you know, going kind of unsaid there is
- 17:36that, you know, everyone believes that in the current
- 17:40legal system, legal frameworks like the studios have every
- 17:43right to go use those scripts for training and write future
- 17:46scripts with. However, you know, the unions
- 17:48were able to negotiate sort of like.
- 17:51Retroactive rights for their writers on scripts that they
- 17:55already sold to this they make a broad point on this that is
- 17:58eating up at me. I mean, first of all, it seems
- 18:00right now like foundation models are becoming pretty
- 18:03commoditized. There are several of them.
- 18:05Like maybe open AI stays far enough ahead and finds other
- 18:09modes, but like in some ways like the creative expression is
- 18:15more unique, right. Lord of the Rings is like more
- 18:18singular. And yeah, I just don't think
- 18:22like somebody should be able to just steal.
- 18:24That was a great creative work. And just because people didn't
- 18:27foresee this technology being around at the time, like, why
- 18:31should just like these technologists who found who are
- 18:34the first ones to figure out how to steal the idea in a way they
- 18:36could get away with, be the ones to profit off of it.
- 18:40I'm on the creative side here. I don't think that.
- 18:44You know, I said Volley could create a game based on Middle
- 18:48Earth, and that's just because of existing IP protections that
- 18:52prevent us from doing that whether or not we write the
- 18:54content with AI or not, right? And likewise, nobody's can go
- 18:58and write a whole, you know, a bunch of newcomer articles and
- 19:02claim they're part of newcomer because you have you own the IP.
- 19:06So the question is more like can I, you know, write a brand new
- 19:10IP script that you know has some?
- 19:13Training data of the original Lord of the Rings series.
- 19:16And and it gets to the one of the core idea is an LLM.
- 19:19Because obviously, like a human being, you wouldn't say you
- 19:21can't go and read Lord of the Rings and then go write another
- 19:25book having like been changed, right.
- 19:26It's like a point that's basically what all art is, you
- 19:29know? But with the machine you're
- 19:30like, well, you basically just pull little bits and pieces from
- 19:34everything and that's stealing, right.
- 19:36So yeah, I mean, they're big, like philosophical, like how,
- 19:39you know, how different is it from how a human being sort of
- 19:42operates? Yeah, I mean I think you know we
- 19:45one thing we think about on copyright side, but also I think
- 19:48on the sort of like conceptual side of you know stealing versus
- 19:52remixing versus original content is like.
- 19:55Like in the US, we actually have this pretty robust framework on
- 19:57copyright. We call it, you know the the
- 19:58fair use test of whether or not you can reuse some portion of
- 20:01copyrighted work, right? And when you can, it's called
- 20:05fair use, and when you can't, you're infringing and and you
- 20:07can be sued or whatever, right? And there's a bunch of key
- 20:10elements of the fair use test that I actually think are all
- 20:12really interesting and relevant in the AI discussion and the LLM
- 20:15discussion, And the image and music generation is also really
- 20:19hot topics here, like the key pieces are like.
- 20:23How transformative is the use of the original work?
- 20:26You know, is the new work compared to the original work?
- 20:28So how much have you changed? Basically, like how different it
- 20:30is actually fairly important. How much of the original work
- 20:33you are using or replicating is also fairly important.
- 20:37Whether or not the use is commercial is also like a big
- 20:39part. Are you selling this thing for
- 20:41money or is it just like a fun project or is it educational?
- 20:45And then the last piece which we've all kind of been talking
- 20:47about is are you impinging upon the market for the original
- 20:53work, which is like really important, right.
- 20:56So yeah, so to the Lord of the Rings discussion, like if I
- 20:59created a, you know, a fantasy novel that said in a sort of,
- 21:03you know, vaguely related universe of the Lord of the
- 21:06Rings, but I didn't call it the Lord of the Rings.
- 21:08And, you know, I wasn't stealing any of the characters.
- 21:11And, you know, Jr. Tolkien's estate wasn't going to
- 21:14sell less Lord of the Rings books because I came out with
- 21:16any fantasy novel then that's a pretty good use.
- 21:19You know, example, like fair use for.
- 21:20I'm not really impinging upon the original.
- 21:22And it's also quite transformative, right?
- 21:24I think music is the one where this is like the most stark in
- 21:27some cases because there's these AI tik tokers that are making
- 21:31like fake Drake music, right. And like fake like Taylor Swift
- 21:35music and stuff, right. And the music industry obviously
- 21:38doesn't like this. They're having them trying to
- 21:39have them all taken down, right. And it's an interesting
- 21:42question. If I make a fake Drake song,
- 21:45like, obviously that's not very transformative, right?
- 21:49But am I impinging upon the market for Drake, right?
- 21:52Am I like, am I capturing some portion of Drake love in the
- 21:56universe, you know, with my fake Drake song?
- 21:59And I kind of think so, yeah, I sort of agree.
- 22:01Like, I I think that the fake Drake tracks are probably
- 22:03copyright violations, right? But then the second question is
- 22:06like, what if I make a completely new piece of music,
- 22:09right, that is trained on every pop music, you know, piece of
- 22:13pop music ever, but doesn't sound like any existing artist
- 22:16in a meaningful way, right? I'm not pretending it's Taylor
- 22:19Swift, I'm not pretending it's straight, whatever.
- 22:21It's just a new band that you know, it's the Max James and
- 22:24Eric Band that puts out awesome pop music because it's a really
- 22:27cool AI creation based on existing pop music.
- 22:30Is that fair use? If you train it on all the pop
- 22:32music, I'm sure this will be litigated quite heavily, but I
- 22:36think you can make an argument that it's fair use.
- 22:37You know there's also a. Question that AI is new and
- 22:40there should be new regulations. I I'm very torn on what
- 22:44independent law what ought to be the case.
- 22:46On the one hand it's it would be nice to have a system where
- 22:49artists got paid proportionally to how much their work you know
- 22:53sort of created the circumstances for the new AI
- 22:56based invention. On the other hand, just, you
- 22:58know, all those sort of barriers, you know?
- 23:01Just create lots of hurdles to new ideas and you know, it's
- 23:04like the same type of thinking that prevents like, you know,
- 23:07remixes and sampling and all that stuff when that's what the
- 23:10consumer ultimately wants and enjoys.
- 23:13And also in the context of an LM or a transformer model like it
- 23:17seems pretty hard to. Tie this new output of the song
- 23:20like that. Yeah, attribute it back to the
- 23:23thousand artists, right, who had some factor who deserve some
- 23:27fraction of a penny, right. I just think it's not practical,
- 23:34but if but if I make fake Van Gogh art on mid journey, should
- 23:37the Van Gogh estate get a cut of?
- 23:39It feels like in some just world.
- 23:41Any, any. I mean, yeah, maybe.
- 23:42It's just interesting, right? You can do it.
- 23:44I mean, so, but yeah, Van Gogh, if it's like Calvin and Hobbes.
- 23:48Yeah, definitely. You know, Yeah.
- 23:50I mean, if you call it Calvin and Hobbes for.
- 23:51Sure, but I think this is just we don't even need to know
- 23:55whether it was Air I or not. We already have these laws in
- 23:57place. If IA Court looks at it and it
- 23:59looks like Calvin and Hobbes or it is called.
- 24:02You're already in trouble. Like, why does it matter if it
- 24:05was generated by AI? SO that's kind of why I push
- 24:07back against the idea that we do need new laws, because I've
- 24:10never heard a better framework than this one.
- 24:12I guess I'm not saying that don't exist, but at least this
- 24:14one covers like a lot of the relevant considerations here,
- 24:17right? And the big one being, are you
- 24:20impacting the original art? You know the value of the
- 24:23original work in a negative way, right?
- 24:25I think that's super important. And it's hard to figure out,
- 24:28even in an AI world, unless you go whole hog and you're just
- 24:31like, yeah, scraping any copyrighted content is illegal
- 24:33for an AI. It's OK Well, yeah, that's a new
- 24:35law. We.
- 24:36Can make right? Isn't that you know?
- 24:38There's certain humans you know with the fan fiction have, you
- 24:40know, dump, you know, played around with stuff.
- 24:42But I mean there, it's not commercial, right?
- 24:45It's not commercial. It's not commercial.
- 24:46Yeah, right. Let's say we agree that this
- 24:49there don't need to be many new laws here and kind of the
- 24:51existing fair use framework is sufficient I guess.
- 24:55That does seem to lead to a lot of potential challenges in, you
- 25:00know, the creative fields and the workforce that generates art
- 25:04and for games and scripts, for films and, you know, songs going
- 25:09forward, right? So do we think that's
- 25:11inevitable, essentially that the like creative industries are
- 25:15going to need less people to generate this type of content?
- 25:21I mean, it's hard to say. I mean, part of what Max was
- 25:23saying in the beginning is sort of the democratizing effect that
- 25:26it could allow. Or sorry, Jamie, you know, you
- 25:29guys run together. No.
- 25:32Yeah. Yeah, exactly.
- 25:33It's giving, sort of. People with less capabilities,
- 25:37the tools to to do more, which could just just big companies,
- 25:41right? I I guess I just think there is
- 25:42a different discussion to be had around people who own copyright
- 25:46and how that affects the the, you know, future works versus
- 25:48people who don't own copyright but who work for companies
- 25:51generating creative output, you know, and how that what their
- 25:56lives will look like going forward, what their work streams
- 25:58will look like. I mean, we don't want a world
- 26:02where all future creative content is just echoes of what
- 26:07came before and humans stop. But that that also kind of gets
- 26:11to whether LLMS can be creative. If you believe that LLMS are
- 26:15capable of original output, then that won't be the case, right?
- 26:19Yeah, you're right. There's an argument that we
- 26:21could be in a renaissance of quickly all these new art types
- 26:24we've never experienced, It doesn't feel like.
- 26:26Like Dolly. And stuff are producing that
- 26:28yet, right, do you think? This, I mean, I don't know.
- 26:31I think that if you prompt these LMS correctly and spend some
- 26:35time on it, it's not like 100% of the time you'll get creative
- 26:38output, but I believe it is possible to get it to generate
- 26:42new ideas. And you want to be in a museum.
- 26:45You can put up a green painting on the wall and call it art, but
- 26:50you need to have a really pretentious explanation for what
- 26:52the point was and where It's in conversation with everything
- 26:55else and these machines cannot produce.
- 26:57Any sort of like narrative for why they're doing what they're
- 27:00doing. So it is not.
- 27:01It's not art. This is here the original.
- 27:05Green painting man. But Max, to Max's point, if the
- 27:08studio exec can just use this tool to generate commercially
- 27:14viable ideas for movies that are better than anyone on the staff
- 27:18has generated, I I think that. Alone is an interesting.
- 27:22Right. And commercial generation is oh,
- 27:24we should do Star Wars plus. But that's what that person is
- 27:27optimizing for. Theoretically, I mean, but yeah.
- 27:30All right. I wanted to ask if you could
- 27:34like you know dream up, you know using generative AI some like
- 27:41sequel to a favorite work. Like what would you like to see
- 27:44or what would most excite you that feels like in reach because
- 27:49of? Some of the stuff we've talked
- 27:51about. I guess I can answer that.
- 28:01I feel like, you know TV shows that maybe ended for their prime
- 28:07or something, right? I'm trying to think what a good
- 28:09example of that is. But or.
- 28:11Here's a name. Firefly is a famous.
- 28:13Yeah, Firefly, everyone says. I liked it.
- 28:15It wasn't life changing. But let's talk about lost or
- 28:17something, right? If I could, I thought the ending
- 28:20was not great, but not as terrible as the Internet
- 28:24believes. But I would be happy to go rerun
- 28:26the last season, right? Wouldn't that be cool to see if
- 28:28you could create? You know 100 different versions
- 28:31of the ending of Lost. I love that idea.
- 28:33I love the idea that, you know, like, first of all, I quit Lost
- 28:37at the end of Season 2 being like they can never pay this
- 28:39off. So I'm very proud of that.
- 28:41But the idea that like you know with Game of Thrones and stuff
- 28:43where like clearly the author really can like puzzle out the
- 28:48like solution that they need, but it feels like there should
- 28:51be some right answer that you like.
- 28:53Throw an LLM at it. Maybe there's a right answer
- 28:55that exists. I think if they could redo the
- 28:57last season in Game of Thrones, that would not be a terrible
- 28:59idea. Whether or not, yeah, like
- 29:02whether or not the writer, whether or not RR Martin is
- 29:04giving them guidance or someone else.
- 29:06But I think that would be a good one to rerun.
- 29:09Use the generative AI as the excuse.
- 29:11Who cares if it's actually better than, like some random
- 29:14writer? It's Oh well, we have this new
- 29:15technique. I mean, it's not really a sequel
- 29:17for me, but I think it would be possible to create just like
- 29:21really awesome music that I love, right?
- 29:23That's personalized, right? I think that again, I'm not like
- 29:27if it's trained on a bunch of copyrighted material and artist
- 29:29should be paid for it. And I'm cool with that.
- 29:30I'm, I'm figuring out how to pay this AI company.
- 29:33But just like the pure product itself, it would be really cool
- 29:36to have personalized music, you know, on demand that really just
- 29:40hit the sweet spot for me or whatever, right?
- 29:42I mean, I just think that would be so cool.
- 29:45And I think not that hard from a, I mean conceptual
- 29:49perspective, just train an LL or not an LLM.
- 29:52But as music generation model, everything Max loves, everything
- 29:55he's listened to more than 10 times in his life or whatever in
- 29:57his Spotify library, right? I just think that would be
- 29:59awesome. I don't know.
- 30:00That'd be really cool. Well.
- 30:01You could also imagine a world. I mean, you're talking, I think
- 30:04about. Novel sounding art, right?
- 30:07Not the Drake example in that. Yeah, right.
- 30:09I don't need another Kanye album.
- 30:11We kind of work out that well. But another Beatles album?
- 30:15I think a lot of people there would be a market for that,
- 30:17right? But yeah, I do think maybe this
- 30:20ends up looking a lot like the fan fiction community is where.
- 30:23Anyone can create a new Beatles album.
- 30:25They just can't sell it, right? You can't.
- 30:27I mean, that would be pretty sweet too.
- 30:29I know, right? Shareware Beatles albums like
- 30:31you just have a ranking of the best ones basically, I mean.
- 30:35Yeah, Do you forecasting cultural, do you think?
- 30:38I mean, you can imagine, yeah, people get to like refresh you,
- 30:42you know, you got a personalized version of every episode or you
- 30:44don't like the take. It's like a Magic 8 ball and
- 30:47you're like, Oh my God, take it again.
- 30:49Like I that seems so culturally, like, hollow to me.
- 30:52I guess it's very exciting to me where you could create new art.
- 30:56Like, I think my answer to this would just be, I mean, I just
- 30:58think the NPC like creative NPCS, like where you're actually
- 31:02engaging with them like any game, just like any open world,
- 31:05you know, you can just imagine playing.
- 31:07I don't know. Any Horizon 0 Dawn where like
- 31:10they're real characters and they have sort of.
- 31:13Their own agendas that are different every game.
- 31:15And to bring this back to the last fake tech trend, which was
- 31:18like the Metaverse, right, The Metaverse populated by a bunch
- 31:21of characters that you know weren't that flexible or
- 31:24interesting, I don't think is that exciting.
- 31:26But the Metaverse in a world where you can run into an
- 31:28infinite number of digital characters that have interesting
- 31:31stories that you can go on interesting quests and missions
- 31:34with and and compete and collaborate with.
- 31:36I just think it's. I mean I want to hang out with
- 31:37other people in a in an ARVR world as well.
- 31:40But I think hanging out with really cool, interesting non
- 31:43player characters that are powered by LLM is just a really
- 31:46fun idea too. And so I just think Ready Player
- 31:491:00 is way more fun if half the people in Ready Player One are
- 31:52like interesting, conversational, you know, AI
- 31:54driven characters. Also because AI will be much
- 31:57happier to spend tons of time in the Metaverse than regular
- 32:01people, and so you could actually populate it with them
- 32:03versus forcing strap on. There you go.
- 32:06They're always on. We haven't talked about movies
- 32:09as much as I thought we would. I'm curious, I guess I'm
- 32:13personally skeptical like the movie business is anywhere is
- 32:16near affected in the near term like.
- 32:19Maybe sort of I what they sort of you know development has
- 32:23changed but it feels like. I do think like just that seems
- 32:27to be just driven it seemed, yeah, I mean the generative
- 32:32video space is like very early right from the current
- 32:35technology, but it's still amazing and so they.
- 32:38My prediction on the first episode was that Tik Tok's being
- 32:42revolutionary. So yeah, right.
- 32:44I mean I just don't like, I think this sort of bums me out
- 32:47as someone who watched a lot of movies growing up.
- 32:49But I just sort of think the movie business is in like a
- 32:51secular decline. And I think all these other
- 32:53types of content are like eating it up, whether it's like short
- 32:57form video or television series or TikTok or AI generated
- 33:02TikTok. And I mean as we discussed at
- 33:03the beginning, gaming is bigger than all these things put
- 33:05together because it turns out like having some interactivity
- 33:08and some form of, you know, play as part of your entertainment
- 33:11experience makes it more fun and makes it more engaging.
- 33:13So I just think that movies are a sort of almost like
- 33:17anachronistic type of content. I know that's a really hot take.
- 33:19But this idea that you have a 2 hour thing that like you don't
- 33:23interact with, that has if a sequel comes out, if ever, it
- 33:27comes out five years later and it's also only two hours long
- 33:30and it's not personalized in any way.
- 33:32And also the business model is falling apart because nobody
- 33:35wants to go to theaters anymore. So they can only make Marvel
- 33:37movies basically now. So it's just not, It's caught
- 33:41between a lot of factors that are bad for the future of what
- 33:45we think of as films or movies, right.
- 33:47And I think, you know, TV has already been eating movies
- 33:50throughout our lifetime where I think most of the like prestige
- 33:54dramas that have come out in the last, you know, 10 or 20 years
- 33:56of a lot of a moon on TV, which like never happened before 2000.
- 34:00And then I think also I think gaming will start to eat into
- 34:03that. And I think, you know, short
- 34:04form video, whether it's TikTok or YouTube or or some AI
- 34:08generated version of these things, we'll just start to eat
- 34:09into that also. So I guess, yeah, the sort of
- 34:11depressing take is I don't think movies like matter anywhere near
- 34:14as much as they once did, which is why I think the other stuff's
- 34:16more interesting to talk about. Yeah, that's some very
- 34:20depressing on movies. I'm sorry.
- 34:22It feels like areas where you can iterate.
- 34:25I mean, short form video is just so perfect because you can
- 34:28really test it out like the machines can learn from it.
- 34:31Whereas yeah, movies require a ton of intention and sort of
- 34:35broad thinking, which is still the domain of humans.
- 34:38So I don't know if everyone in the movie business would be so
- 34:40depressed by what you're saying. It's like, Oh yeah, the real art
- 34:44sort of the long form thinking is a little more insulated than
- 34:48the let's run an AB test and see how people react to things.
- 34:52And so there there's some good in that too, definitely.
- 34:56I mean, I still think there will be great movies throughout our
- 34:57lifetime, but I think that where the new tools are coming into
- 35:01play is in categories where, yeah, maybe the movie makers
- 35:04aren't that interested in. Or if we really believe in the
- 35:07extension of the gaming discussion we had at the
- 35:09beginning and we can create extremely intelligent NPCS.
- 35:15Can you just put them in simulations that are fun to
- 35:18watch in the same way people watch reality shows or, you
- 35:22know, watch films and TV's? The stories are so compelling
- 35:26that but they're just occurring because there's a simulation
- 35:29running. I don't know.
- 35:30Maybe that's where. We are.
- 35:30That would be a fun The Real Housewives of Open AI here.
- 35:33Yeah, exactly. Patent that, yeah.
- 35:37Yeah, that sounds good. Yeah, little Big Brother, but
- 35:40just with AI running around. Yeah.
- 35:42Or. And then and then, you know,
- 35:44you're essentially able to iterate through thousands of
- 35:47different reality scenarios, right?
- 35:49Throw an island in there, or a rose ceremony, whatever it is.
- 35:54All right. Stay tuned.
- 35:55I have Amy and Keith up next. Keith, Amy, welcome to the show.
- 36:04Thanks so much for joining me. Thanks so much, Eric.
- 36:07Yeah. Thanks for having us.
- 36:08Amy, I I thought, I mean, we'd start off with you just as.
- 36:12Sort of the venture capitalist who sees things broadly, can you
- 36:17talk a bit about your interest in gaming?
- 36:22Yeah, it's it's a it's a lifelong passion really.
- 36:26I haven't worked in the game industry but have been, you
- 36:28know, playing games. I was a gamer growing up and
- 36:31play a lot of, you know, early, like, you know, first person
- 36:35shooters like the Doom and the Wolfenstein.
- 36:39Era games to like CS and then Diablo.
- 36:42Era to more of the modern day and and as an investor you know.
- 36:47What an industry. I love that you get.
- 36:48It's like, oh, we did all these fun things and that's part of
- 36:51the credit of it. Like, I should spend more time
- 36:53in gaming. Anyway, yeah, it's so true.
- 36:56It really is. I mean it's pretty important I
- 36:58think because especially if you decide to invest in content,
- 37:02first of all it's really difficult and and there are
- 37:05people play testing really like hundreds if not thousands of
- 37:07games every year. But it is really important to to
- 37:11play them because then you can get kind of a feel for what
- 37:14really exceptional looks like in different genres.
- 37:17Yeah, in in terms on the investing site, didn't you, you
- 37:19invested in Epic at one point, is that right?
- 37:22That's right. Yeah.
- 37:23Well, do the first foray was actually when I was at Insight.
- 37:27This is like 10 years ago now and Insight was majority owner
- 37:31of Jag X Runescape and and that was like an early experience
- 37:36because I I still remember when. When there was like a big
- 37:40monetization change in the game and everyone sort of the players
- 37:43were like outraged and and they kind of saw that you know
- 37:47Insight was a big shareholder and so then you know there were
- 37:51like bomb threats called into the fund and yeah people were
- 37:55player. I mean, yeah, it was a great
- 37:59example of just how intensely passionate gamers are about
- 38:03their games and and changes to it like.
- 38:06And so then later on you know I joined Livespeed and one of my
- 38:09mandates was actually investing games and games there.
- 38:12And we did early stage investments in companies like
- 38:14you know Faraway which is web three side really experienced
- 38:18game makers from Scopely and then and then also like AAA
- 38:21studios like 1047 and and lastly games and others and also growth
- 38:26stage there was Epic Games and then also triple dot studios
- 38:29such as mobile Studio. And then on on the crypto side,
- 38:32I mean, I'm sure there. Other things, but I mean it's
- 38:35you Go Labs, right, the board ape company, I mean part of the
- 38:38investment thesis. There.
- 38:40What is it you got right. OK yeah that you were gonna that
- 38:45they were are we're going to build out sort of games and sort
- 38:50of in an entertainment world around to that is that right?
- 38:52Or how much was games part of investing in them?
- 38:56In the you go round and this was early last year I think that.
- 39:01First and foremost, the exciting thing was that with board of
- 39:04Yacht Club and then you know they then acquired Crypto Punks
- 39:08and a couple of other brands. It was the I would say the blue
- 39:12chip brand in the NFT space. And so there was a lot of value
- 39:16in that and that they could turn into potential like consumer
- 39:20products that a lot that both NFT holders and aspirational NFT
- 39:24holders could use and love. And and so games was a a big
- 39:28part of that strategy all. Right, Keith.
- 39:31Tell, tell. Just give us a little bit of
- 39:33your overview. You have a stealth start up,
- 39:34which you know, amazing. You know when I'm prying that
- 39:39you're going to be able ducking and dodging on some of that, but
- 39:42give give people. A little bit of your background
- 39:44in the games industry. Yeah, sure.
- 39:48So actually just before, before I got into games, I did five
- 39:53years in the US Army, artillery and and Special Operations and I
- 39:58I bring that up because it it's. It's unexpectedly like kind of
- 40:01come back as as part of my life in game.
- 40:04But for the last 1215 years I've been in the games industry at
- 40:10startups, mostly later stage startups.
- 40:13So you know, past the kind of 0 to one point more at the growth
- 40:16phase as well as like larger conglomerates.
- 40:19So companies like Kabam in the early days of web, Facebook and
- 40:23then mobile gaming, Gree International, the subsidiary
- 40:26of, you know, the large social network Gree in Japan,
- 40:30Wargaming. I'm actually visiting some old
- 40:32friends and colleagues here at the Wargaming office right now,
- 40:36makers of, you know, the phenomenal World of Tanks.
- 40:39And then most recently was Ads app Loving where I led the games
- 40:43business there. To our speaking the language of
- 40:46actual in particular games, because I think you know not
- 40:49everybody follows like the games business.
- 40:51What were the games that ended up defining a lot of your career
- 40:54in terms of spending time? Yeah, so you know I've, I love
- 41:00every and I I mean this very generally.
- 41:02I love every game that I've been a part of in in one way or
- 41:05another. Even those that have like if he
- 41:07was kind of sharing resulted in some funny real world
- 41:10circumstances like getting getting offices firebombed and
- 41:14threatened and stuff like that. Even that is actually just a a
- 41:16symptom of the passion that exists in the industry.
- 41:18And I I share that passion as a gamer as well.
- 41:20Maybe not to the violent ends, but yeah, so.
- 41:23So games. Games is what people would know.
- 41:25So marquee games for me, you know, the first game I ever
- 41:29worked on was a game called Last Chaos, which is a Korean
- 41:33developed MMRPG that we were publishing over in the West.
- 41:37No, not many people know about it.
- 41:39Like if you weren't one of the 100,000 or so people that tried
- 41:41it, like you wouldn't know. But it's it.
- 41:43It speaks to the longevity and some other elements of the
- 41:46industry because that game is still alive today, believe it or
- 41:49not. But game the people would know.
- 41:52So King is a Camelot would have been you know an early mobile
- 41:56actually first Facebook and then early mobile 4X strategy game
- 42:02that I spent some time on and then also at Kabayam, you know
- 42:06as a part of a number of IP tie in games as well.
- 42:10So we did Marvel, contents of Champions.
- 42:13We did a couple of while I was there.
- 42:15We did a couple of Fast and the Furious titles as well.
- 42:20And then let's see next Wargaming.
- 42:24So World of Tanks, World of Warships.
- 42:26You know, there's pretty big titles in the shooter space or
- 42:29in the strategic shooter spaces, as I used to say back then.
- 42:33And then most recently, oh sorry, I almost forgot.
- 42:35Actually, the most played game that I've ever had the pleasure
- 42:39to work on and and with is Subway Surfers with the.
- 42:42Oh my God. Every TikTok video right is
- 42:45exactly. I mean it is the most
- 42:47downloaded. I believe it is the most single
- 42:49downloaded game ever. Wow.
- 42:51It, you know, it counts its installs in the multiple
- 42:54billiards and what's your portion of the human race like
- 42:57has played that game at one time or another?
- 42:59Do you have a typical role in these games or like how would
- 43:01you describe what you tend, what the role you?
- 43:04Yeah, so early, early in my career I was in the product
- 43:08management and design side, so very hands on individual
- 43:11contributor, but then was quickly leading teams.
- 43:14So like I led the team, you know kind of as the general manager
- 43:18on as a Camelot. I'm pretty early in my career
- 43:22and then since then, so that would have been, you know, maybe
- 43:25five years into my 12:12-ish year stint, I've been more on
- 43:29the corporate executive side. So at, you know, war gaming, I
- 43:33built and and ran the mobile division at App Love and I ran
- 43:38the games business. And so as these things happen
- 43:41like I became further away from kind of directly involved in the
- 43:45creation operations and more on the business side.
- 43:48And now you're now you're thrust back what?
- 43:50Like exactly, Exactly. Two person, 1 1/2 person.
- 43:53Yeah. Because I'll tell you, like,
- 43:55there's nothing better in the industry it's than like actually
- 43:58touching a game than actually, you know, touching consumers in
- 44:02that way. You know, that's what I think we
- 44:04all get in this industry for. And so I'm happy to be back.
- 44:07And I I might poke at more what you're doing at the end, but
- 44:10broadly? Fair to say it is like a game
- 44:13development type approach rather than a particular game.
- 44:19What do you mean by that? What you're what you're.
- 44:20Working on now, are you building a game?
- 44:22Like is your company a game, like building a particular game?
- 44:26No. So we're we will build games but
- 44:29our mission is more about ushering forth what we see as
- 44:34this like new generation of games empowered by AI.
- 44:39And so that's that's what our roughly our mission is.
- 44:42It'll involve making games and. Evolve.
- 44:43Yeah. We finally get into the picture
- 44:45lines of technology, yeah. And and by the way, like I'm not
- 44:48that us being stealth and and not sharing is not because I
- 44:51don't want to. It's actually more as we I think
- 44:53talked about or the call or before you started.
- 44:56It's more that I struggle to concisely explain what it is
- 45:00that we're doing. So you can so look, you open the
- 45:02door. I'm gonna try to explain and
- 45:04stumble way, way through. So yeah, nothing's off the
- 45:06table. OK, cool.
- 45:07We'll come back to that. Amy, can you give a sense of why
- 45:11we think something's going to happen with artificial
- 45:13intelligence and gaming or like what what you see, the promises?
- 45:17Like, you know, in investing there can be a sense like, oh,
- 45:20new things happened. It's an inflection point like
- 45:23let's plug it into like our favorite thing.
- 45:25But it does feel like with gaming in particular, there's
- 45:29sort of a lot of optimism. I'm curious like what what you
- 45:32think will make. What's is it what's happening in
- 45:35generative AI with like? The image generation or what?
- 45:38What in particular do you think is important here?
- 45:41Yeah, I can break it down South. And also, you know, Keith can
- 45:44give so many examples where he probably used AI in all of his
- 45:48games. But AI in the gaming industry
- 45:52has kind of been around since the beginning of games Like back
- 45:56in, I don't know, fifties, 60s like starting with arcade,
- 45:58right? Because you know, fundamentally
- 46:01you're playing against. An opponent a lot of times and
- 46:04before there was you're playing against a human you're playing
- 46:07against a bot and and then with bots against other bots and so
- 46:13that now is captured under kind of like you know non playable
- 46:16character AI as definition. But this concept of vision gains
- 46:21for multiple decades of time, and that's.
- 46:24The NPC and probably when the most regular people or like you
- 46:28know as a gamer you'd be like, the AI is like.
- 46:30Sucks or like you, you know, famously I think like Mario Kart
- 46:33where the AI just like is taught to like Slingshot to catch up if
- 46:36you're too far ahead and all that.
- 46:38Yeah. So we've all, we've all.
- 46:40And like if you play a game too much, you sort of learn the mind
- 46:43or whatever the rules are of the NPCS.
- 46:47So yeah, that's an that's an exciting opportunity because
- 46:50you're you interact with them and you sort of get to know how
- 46:52they think. So that that's certainly one.
- 46:54Sorry, God. Yeah, I know exactly and.
- 46:57And I think, well we'll we'll hit the categories first and
- 47:00then we'll go back into sort of the evolution of each of these.
- 47:03The other, the other one is acid generation which also has had a
- 47:08history in games cause back as far as you know there's like
- 47:12Diablo or maybe even before the concept of procedural generation
- 47:16of different levels. So you know, you're playing like
- 47:18an action RPG and you're playing different like, you know,
- 47:21dungeon levels and. And they're actually
- 47:23procedurally generated. So you know, it might actually
- 47:26look different for example, for different people each time you
- 47:29go into a particular dungeon. Has been a concept that's been
- 47:32baked into game engines for a very long time, which which I
- 47:36think like the modern. I'd say like the post like last
- 47:39fall like LLM version of these categories has been how do you
- 47:44use large language models in these categories?
- 47:49Because AI has been actually using games for a very long
- 47:51time. And then the newest category
- 47:53which Keith is, you know, building in is what what about
- 47:57applying large language models to generating code that creates
- 48:01games. So taking that in terms of asset
- 48:03generation, one step further into actually game logic as
- 48:06well. And people haven't done it yet.
- 48:09And so it's like an exciting hypothetical space.
- 48:12I mean, people are starting to build like basic games in that
- 48:14space, but, and I think, you know, we'll get to it, but the
- 48:18questions here are like one there is.
- 48:21You know, a tremendous amount of enthusiasm.
- 48:24I think some people in the games industry with a kind of like a
- 48:28over hype of of the capabilities of AI, you know ahead of where
- 48:33the actual technology is in some cases and in other cases people
- 48:37say Oh yeah, this is the inevitable next step in
- 48:39evolution of that. And that's one of the reasons
- 48:41why you know AI devs are so excited.
- 48:45Because there's no. The concept doesn't need to be
- 48:48explained as existence industry for a long time, but I think
- 48:52what everyone needs to remember is that you know, the bar for
- 48:54creating content and games is so high and and new technology is
- 48:59only like a part of that. It's a space where really
- 49:01content is king. And it's just about like, you
- 49:03know, can you play, can you produce better content?
- 49:05That sort of fits into a big question for me, which is, you
- 49:09know, I think there's this trend of like the democratization of
- 49:12gaming or the sense that like you know.
- 49:16Roblox or, you know whatever, you're trying to like maybe help
- 49:20the user create games you're trying.
- 49:22Or on the other hand, on mobile that there's just sort of a we
- 49:25can make a bunch of games sort of cheaply and see what works
- 49:28and then invest in that. Is, is that a true sort of
- 49:33narrative or it At the end of the day, are most people just
- 49:36flocking towards premium games that are really thought through
- 49:41with like the best techniques, Like what What sort of the
- 49:44optimism? For this sort of democratization
- 49:47story in gaming and how much do you think AI is is going to move
- 49:53sort of us closer one way or the other.
- 49:55One thing that is easy to forget and and I am guilty of this as
- 49:58well, even having been in the industry for all these years is
- 50:02just how big and broad like the industry is, especially on the
- 50:05consumer side, right. It's something like 434 billion
- 50:08people. I don't know what the number is,
- 50:09how play games. So that's like you know, sizable
- 50:11portion of the human race. It is not like, you know, 20
- 50:15years ago you would have said like when we said gamer, you
- 50:19know, you thought of some like 30 or 20 something year old guy
- 50:21in his mom's basement or something like that.
- 50:23That's just not true anymore, right.
- 50:25It's a very diverse market. Of and games are bigger than the
- 50:28movie business and everything. It's like movies, music, TV, you
- 50:32put it all together, games is bigger.
- 50:33So it's like the largest form of entertainment that exists.
- 50:36And so all of those kind of use cases that you mentioned are
- 50:39real and are big. So hyper casual gaming which is
- 50:43something that you kind of touched on as these like high
- 50:46volume kind of mobile launches like that's actually a a large
- 50:51category unto itself, something like 40 to 60% of the installs
- 50:55across the mobile platform which is the biggest platform in
- 50:57gaming or hyper casual style game.
- 51:00But that's not to say like you still have Diablo releases like
- 51:03billion dollar franchises that launch.
- 51:05So all of that exists and all of that is part of game.
- 51:07Nice, nice. When a business is big enough
- 51:09that you're like, oh, we can go after one or the other, they're
- 51:12both promising. I mean I'm curious like which
- 51:15whether you think, yeah, the high end gaming or sort of the
- 51:20sort of cheaper sort of more democratized game creation is,
- 51:24is going to be helped by the trends that we're seeing here.
- 51:28So I think all, all of it'll be helped but not equally, right
- 51:33and not in the same way. So the high end, you know what
- 51:37we think of as AAA but actually there's probably a bunch of
- 51:39titles that we wouldn't consider AAA but meet this what I, what I
- 51:43call it like the high complexity kind of side of the spectrum.
- 51:48So that can, you know AAA is typically like high complexity
- 51:51on the fidelity side, but there are games that are high
- 51:53complexity in the kind of puzzle formation or other things that
- 51:56are behind the scene or the multiplayer side of things.
- 51:59So those those are all in the same bucket.
- 52:02I expect those types of games. I expect for us to get more of
- 52:06those games with AI, more of them produced.
- 52:08So increase in productivity, increase in creativity via the
- 52:13use of AI and then also new kind of features and components like
- 52:19we talked about like a ILLM driven NPCS and then AI also
- 52:25like natural language as an interface for gaming that will
- 52:29be new as well. So I think that that kind of
- 52:31thing is going to happen on that you know high complexity side.
- 52:36I also think that things are going to change a lot on the
- 52:39kind of democratization on the UGC user generated content side
- 52:43of things as well. So some of it'll be similar like
- 52:46natural language as an input, right?
- 52:48Further abstracting and and thereby democratizing like
- 52:51interaction with, you know, platforms like Roblox or even,
- 52:54you know, more complex professional tools like Unity
- 52:57and Unreal that will allow a lot more people to engage with the
- 53:02creation of games. You know, from an intent
- 53:06standpoint. In other words, have a game in
- 53:07your head easier to make it playable.
- 53:11But then I think there's actually, and this is where my
- 53:13company is building tours, we believe that there's a third
- 53:16category that gets born out of all of this, which wasn't
- 53:20possible before and comes from the convergence of not just AI,
- 53:25but also a number of other things, which allows for, yeah,
- 53:30so for for for games and entertainment that look entirely
- 53:34different than what we've seen before.
- 53:37Well, OK. Different in what way Or give us
- 53:39a little bit more there. Yeah, so, so, so here's where
- 53:45all I've had to struggle through.
- 53:45So like, the way I like to to come at this is from the player
- 53:50perspective. So, like, there's been all these
- 53:53things in games. That for my entire life as a
- 53:56gamer and all the customers that that you know, millions of
- 53:58customers that I've interacted with their billions really in in
- 54:01the games that I've made all of these expectations that we've
- 54:03had these unmet expectations and they keep being unmet and it's
- 54:07just so disappointing and sad and so things like the fact that
- 54:12games are largely A finite experience.
- 54:14So even World of Warcraft still around right over that early MMO
- 54:18I talked about the still around yes, it's true.
- 54:20But like if you ask anybody that plays World of Warcraft, like
- 54:23the amount of content that's being added, some people even
- 54:27say it's dead. But like even that which kind of
- 54:29pushes the boundary of finite still feels finite in many very
- 54:32important ways and similar to that would be like unrealized
- 54:36depth. And so you guys touched on this
- 54:39a bit with like procedural generation and I think it was
- 54:41something you said, Eric, where like a player can kind of learn
- 54:45the mind of the NPCS and that gets boring on like this feeling
- 54:50of alive. You know, Another another more
- 54:52contemporary example would be Starfield, right.
- 54:54So everybody is super excited for Starfield.
- 54:56Oh my God, another Bethesda game.
- 54:58Amazing, amazing. And it is amazing game.
- 55:00But what is like, you know, one of the the often repeated
- 55:03critiques is like, yes, there's 1000 worlds that can go explore,
- 55:07but they all feel relatively the same they.
- 55:09Feel right. Except for those people who
- 55:11wanted to bring up Starfield as sort of the red flag for AI
- 55:15gaming. Just because it's like, oh, we
- 55:18can do everything, you know, we it can be infinite or whatever.
- 55:21You know the sense that. I feel like more and more games
- 55:24are, it's like, oh, maybe the way to win an open world is to
- 55:27be even more open in world, you know, and and then and then it.
- 55:32You lose sort of the story, right?
- 55:34You hand it over to sort of assist, you know, some sort of
- 55:37system to generate the world. But then you can't sort of put
- 55:41enough human storytelling in to to power that expanse, I guess.
- 55:47And that's how it's been with kind of you know, historical and
- 55:51contemporary techniques of procedural generation and others
- 55:54things we called AI before but are different from you know the
- 55:58large language models, image models, things that are are
- 56:00being used today and like with this also it's all so to
- 56:04continue on the things, these unmet expectations like the
- 56:08experiences are also relatively static.
- 56:10In other words like there is a necessary one-size-fits-all that
- 56:14goes into when you're developing a game, right.
- 56:17I I can't customize the game for every player.
- 56:19I can create systems so you know, you can choose and you
- 56:21can, you can, you know, design your character, your player
- 56:24character in in Starfield to some extent, right?
- 56:27But you're still limited in your narrative choices.
- 56:29You're still limited in, you know, the in in many other
- 56:31things. And then because of the the
- 56:35natural kind of formation of the industry and how games are
- 56:38developed as well before AI, you end up with these fragmented
- 56:42experiences. So like social is fragmenting
- 56:45and everybody, like one of my favorite examples of this is
- 56:47like you know, try and play a game with a one of your friends
- 56:51on PlayStation, another one on PC or something like that.
- 56:54Like shit, it's getting better, right?
- 56:56But but damn, it is not great. And even beyond that, like if
- 57:00you think about the kind of social interactions you can have
- 57:02inside of a game, those really don't exist across games not
- 57:06even within these you know small ecosystems like the PlayStation
- 57:10Network or others that do try to create a social graph across
- 57:13game. And then finally like what what
- 57:15kind of summarizes all of this up for me and and speaks to you
- 57:17know the direction that that we're going is games are not
- 57:21personal, right? And I think that that that's
- 57:24that's what we view as kind of this new frontier is a is an era
- 57:28of personalized gaming experience.
- 57:31So building games for individual level persons, for segments or
- 57:38so not even I I actually think that on this you know this third
- 57:42category of gaming that I was describing this native AI gaming
- 57:46is not is going to look very different not just from a
- 57:48product and consumer perspective but in a how you build a
- 57:51perspective as well. Because, you know, creating an
- 57:55AI game team, a game director, a personal game director who is
- 57:59learning your preferences as you play.
- 58:02Much like you know, Spotify is learning your preferences as you
- 58:05listen. Or like the.
- 58:08Game is building as you're going through it exactly and that's
- 58:13and so, so things still have to get built in order to make that
- 58:16happen. You must create you know
- 58:18capabilities within a simulation that an AI, game director and
- 58:22game team can work within on. But like the what we would have
- 58:28traditionally called creating the game will be very different.
- 58:31And I actually think that this one of my dreams or my hopes is
- 58:35that this actually brings gaming much closer to what our earliest
- 58:40and most free experiences with kind of interactive
- 58:44entertainment probably were. Which was like hanging out in
- 58:47the sandbox with your friends, right?
- 58:48Where like there was no game director when you're in the
- 58:51sandbox, right? There was no, like designer.
- 58:53It was like, no, I'm, I'm the spaceman, you're the alien.
- 58:56Like who? Who am I?
- 58:57Shoot you? Oh no, I have a force field like
- 58:59that kind of merging or blending or blurring between creator and
- 59:03consumer. I think comes with this native
- 59:06AI gaming future as well. One of my favorite books is
- 59:09Ender's Game and you know in that in that book in that you
- 59:15know he Ender is playing with a computer that creates basically
- 59:19like psychological tests for him that sort of evolved the player
- 59:23where you know the teachers don't even know where the gaming
- 59:25is. I imagine that's still far away,
- 59:28but it is amazing to imagine this world where.
- 59:31The game is sort of responding to the player and producing
- 59:36stuff as it goes. I mean that, you know, I, I I
- 59:40think a lot from the framework of like TikTok, right, where,
- 59:44you know, that's a key sort of AI sort of sorting, you know,
- 59:50showing people what they want, running experiments based on how
- 59:54people engage with stuff to see what which should rise to the
- 59:57top. Yeah, I guess my, the thing that
- 1:00:02I'd been hung up on with like longer form content is it just
- 1:00:05like harder to run those experiments.
- 1:00:07So even if like, you know, we're able to create like videos from
- 1:00:12like whole cloth will, like we really have movies that are like
- 1:00:16AB tested in that way because it's just like too hard to run
- 1:00:20the experiments. What I hadn't really imagined is
- 1:00:22this world you're talking about where like, Oh well, you have
- 1:00:25somebody in the experience and you can sort of run many
- 1:00:27experiments on them while they're playing.
- 1:00:29The movie comparison is an interesting one and I think that
- 1:00:32games in, in this case we actually have an advantage
- 1:00:34because there is a lot more signal data that we get from a
- 1:00:39person playing a game. Then you are probably going to
- 1:00:41be able to get, you know, the person watching the movie.
- 1:00:44Amy, could you take the temperature on where you think
- 1:00:48startups are right now? I imagine you've been sort of
- 1:00:51meeting with them like I don't know, yeah, what's what are you,
- 1:00:56where do you, where are you seeing a lot of energy going and
- 1:00:58sort of how, how, how active is this or like what?
- 1:01:01Yeah. What's your view on sort of the
- 1:01:04AI game space at the moment in Startup World?
- 1:01:07Yeah. So now that, so now you know as
- 1:01:10a partner at Menlo I've been covering games consumer and
- 1:01:13games. But in games obviously you know
- 1:01:15AI is a big part of it and I would say that the that the
- 1:01:21where startups are and also I would say the.
- 1:01:24Game studios of willingness to adopt right these tools and and
- 1:01:30techniques really depends on where which of these kind of
- 1:01:33categories that we we talked about the three you you have
- 1:01:36like kind of MPC, you've got actually I'm going to swap the
- 1:01:39order and we start with like asset generation creation and
- 1:01:44then MPC and then sort of like cogeneration we are kind of in
- 1:01:48that order of like I would say like readiness of adoption
- 1:01:51because. On the on the asset generation
- 1:01:54side, so this is, this is like taking image models and
- 1:01:57generating kind of 2D and 3D images.
- 1:01:59This is kind of like a where there's a lot of I would say
- 1:02:03heat from artists because immediately they're like wow,
- 1:02:06our our works, you know, our our works of art are being kind of
- 1:02:10used to train these models without our permission.
- 1:02:13And also you know are our jobs threatened?
- 1:02:16And for some people, they are actually one of the studios when
- 1:02:19I was at Lifespeed, the founder told.
- 1:02:22Did tell me that, you know, they're actually able to reduce
- 1:02:26the size of their 2D artist team by about 50% because you know
- 1:02:32using a number of the tools out there whether it's like Leonardo
- 1:02:35or Scenario or a number of these tools, they're actually and
- 1:02:38actually the the number one tool that people use are actually
- 1:02:41just mid journey and or building on stability, stable diffusion
- 1:02:45directly. They're able to generate, you
- 1:02:48know, let's say like concept art.
- 1:02:51Everyone's trying out using these tools or concept art, but
- 1:02:54actually getting some of the 2D assets to a point where it's
- 1:02:57like you know 70% there and then taking it from there.
- 1:03:00I think where the where the hardest problem is this 3D model
- 1:03:05generation and the model or asset generation.
- 1:03:09So the models really aren't quite there yet and so you have
- 1:03:14tools out there that are a combination of.
- 1:03:17You know, they'll they'll produce a model that say
- 1:03:20anywhere from like 20 to 5060% of the way there.
- 1:03:24And then they might actually have be mechanical turking like
- 1:03:27the the remainder of the process to get you to like 70%.
- 1:03:29And then your 3D artist takes that mechanical Turk, as in
- 1:03:32there are people trying to flesh it out.
- 1:03:34Is that what you're saying? On the company side, yeah, Yeah.
- 1:03:38And nobody wants to say the word people, yeah.
- 1:03:42No, I mean, I mean right now you cannot replace like a 3D artist
- 1:03:46in terms of generating because the models just aren't there,
- 1:03:49but because that is like the hardest and also the most cost
- 1:03:52intensive, I think that will really unlock this piece of it.
- 1:03:57Big question from game studios would be, well how do I protect
- 1:04:01my IP? Because you know if I.
- 1:04:05Will my IP be used to train this model, you know, not OK with
- 1:04:09that? Or like if I generate using this
- 1:04:11model assets, can I actually use them in my game and have it be
- 1:04:15IP protected? And so I think this is a big
- 1:04:17kind of area debate right now, and which could be helpful for
- 1:04:24indie type developers, right? If if they don't have anything
- 1:04:28to lose, they're willing to risk copyright more.
- 1:04:31You know, it gives you a competitive edge.
- 1:04:32You can be a little bit looser. While, you know, big companies
- 1:04:36are certainly very worried about copyright.
- 1:04:38Both. Both.
- 1:04:40You know, whether they're going to hurt their own material, like
- 1:04:43you're saying, or if they create stuff with these techniques,
- 1:04:47whether they actually have any copyright claim.
- 1:04:49Right, because there's some early signs that you don't.
- 1:04:53Yeah. Anyway, that's a great point.
- 1:04:54Yeah, absolutely. I mean, indie studios.
- 1:04:57It's a big part of the thesis with this AI wave right now is
- 1:05:01that. Cost content is the number one
- 1:05:04cost in game making and so you can actually bridge this gap a
- 1:05:08bit from for indie studios that are early that are like adopting
- 1:05:12tooling earlier. I think over time it converges
- 1:05:16And so then I think like in the end state, these tools will
- 1:05:19actually just make it easier for all creators of they're all
- 1:05:22artists to be creating these assets, which means that the
- 1:05:26the, the artists within these large gaming studios actually
- 1:05:30just be more efficient. And so I think it's still that
- 1:05:35may still happen, but there's probably like an early adopter
- 1:05:37advantage right now. But I think that's kind of
- 1:05:39predicated on how good these models are.
- 1:05:42And I think when it comes to putting assets into production,
- 1:05:45I'm still not hearing almost anybody putting in production.
- 1:05:48People are mostly using tooling on the concept side because in
- 1:05:52terms of animating these these assets, particularly if it's 3D,
- 1:05:57it's still, you know, we're still pretty far from that.
- 1:06:01Yeah. So that's kind of acid
- 1:06:02generation which is for this along adoption and right now
- 1:06:06we're looking at founders that are enterprise ready, you know
- 1:06:09knows how to make enterprise workflows and create them and
- 1:06:12also go to market and and basically make a created
- 1:06:15experience and also security and privacy that really scheme
- 1:06:19studios need particularly on the larger side.
- 1:06:22Do we see a foundation model for gaming?
- 1:06:25Like has it? Have there been companies that
- 1:06:27are trying to like do the Open AI thing specifically for
- 1:06:31gaming? Or are they mostly just trying
- 1:06:33to build off of Dolly Mid Journey and other sort of
- 1:06:38general general tools? The image models are very good,
- 1:06:43you know, in terms of what's what's out there and also like
- 1:06:47you know, kind of developments on the open source front.
- 1:06:50I think where a game specific model becomes interesting is
- 1:06:55around generating code. And and the question there is
- 1:06:59like, well, what are they training the data from?
- 1:07:01Because you know, game code is, is, is basically IP that's held
- 1:07:06within gaming companies, mostly large gaming companies that are
- 1:07:09very incentivized to having that be in kind of like the open
- 1:07:13public property so that other studios can actually use that to
- 1:07:17generate games. And so I think that is like a
- 1:07:20big open question, but current models really aren't trained to
- 1:07:24be able to generate. Well, the neither the
- 1:07:27specificity of code nor kind of like the complexity in terms of
- 1:07:31knowing like well, what makes a fun game versus just like an
- 1:07:34average not fun game. And because it's really like at
- 1:07:38the outlier is where you actually have, you know, where
- 1:07:42the fun is created and that's produces these large outcomes.
- 1:07:46Yeah, all right. The, the non player character,
- 1:07:48the AI in the game. I mean, I think it's such a fun
- 1:07:51thing to think about for the reasons I was talking about
- 1:07:53earlier. You know, people very much
- 1:07:55experienced it. I mean, and and just given the
- 1:07:58charm, you know, the fact that people think they're dealing
- 1:08:01with a real being when you know, you're interacting with a
- 1:08:04character bot or or whatever. Or like, you know, famously the
- 1:08:08Google employee at Lambda like feels like once people are
- 1:08:11interacting with three NPCS that have some of this LLM technology
- 1:08:16behind it, it's going to be mind blowing.
- 1:08:19Yeah, I I'm curious like Keith, like from game design, like what
- 1:08:25are are we going to get like a sort of, I don't know like
- 1:08:29what's it called Grand Theft Auto?
- 1:08:32Is Grand Theft Auto or Red Dead or something going to have just
- 1:08:35like totally free thinking LLMS like running around or do you
- 1:08:39think that would break these games?
- 1:08:41Or like how likely is it that somebody could just plug in like
- 1:08:44sort of much more independent minded MPCS into sort of how we
- 1:08:48think about traditional open world games?
- 1:08:52Yeah, so personally I don't think the NPC, so I think the
- 1:08:58NPC thing will happen. However, I don't think it will
- 1:09:01be as or at least in the narrow way that I imagine most people
- 1:09:05think of it, which is like like you were saying, like a Grand
- 1:09:07Theft Auto and like all of the people walking around or a lot.
- 1:09:11I think that that that will come probably not if you if you know.
- 1:09:15The kind of development timelines at Rockstar and how
- 1:09:18long they've been working on the latest one I would be shocked if
- 1:09:22they kind of shoehorn it in here at the end but we'll see it in
- 1:09:26games for sure. However I think that if you
- 1:09:29broad how I don't think it'll be like that big unto itself
- 1:09:34because the question there's all of these follow on questions.
- 1:09:37So once you have like an LLM able to interact with the
- 1:09:39creativity that we've seen in applications like ChatGPT but
- 1:09:43within a game world the. Just interacting at A at A at a
- 1:09:47language conversational level. Like, I don't know.
- 1:09:51Like what freedom does the LLL have exactly?
- 1:09:53What agency does it have? So I think when you start
- 1:09:55talking about maybe a a more broad like kind of description
- 1:10:01would be like an independent agents acting within the game
- 1:10:05that are able to actually have an impact on the simulation
- 1:10:08itself, on the gameplay experience itself more than just
- 1:10:11like what we've what we're used to with NPCS.
- 1:10:14Where like they might have a set of kind of actions that they
- 1:10:18take, or if it's a bot that you're playing against, you know
- 1:10:20it plays like you beyond that to something that is more material,
- 1:10:23maybe more towards, you know, think of the difference between
- 1:10:27an NPC and an RPG and a Dungeon Master in an in AD and D game,
- 1:10:32right? Well then, well then that's when
- 1:10:33it starts to feel like The Matrix.
- 1:10:35You know, you've got Agent Smith in the game, you're shooting
- 1:10:37against it. And then if the game developer
- 1:10:39is empowered this AI system to like, you know change the rules
- 1:10:43or in some way, that would yeah, be pretty mind alter or an easy
- 1:10:47way to break games, you know, in.
- 1:10:50Exactly. And and what I'd like to do is
- 1:10:52actually link this to what Amy was just saying and and for all
- 1:10:56of the, you know, developer like game developers out there and to
- 1:10:59get a little hint on some of the technical solutions that we're
- 1:11:02exploring right now, I totally agree with what Amy was saying
- 1:11:05about generating code. It's something that we've been
- 1:11:07experiencing as we've been testing out different models to
- 1:11:11to try and generate code. But there is like.
- 1:11:14I think this falls under the definition of code, but it's
- 1:11:16different in an important way, which is data.
- 1:11:18So like often times a game simulation is a set of code
- 1:11:24which describes like which describes all the interactions
- 1:11:26or which is interpreting data. And so a lot of game design
- 1:11:31actually happens at a data level and.
- 1:11:34LLMS are already so your foundational model, pre trained
- 1:11:37LLMS, llama, GPD 4, whatever are already very good at data.
- 1:11:43And So what do I mean by that? I mean like you know take a
- 1:11:46game, Call of Duty right? There is like a whole bunch of
- 1:11:50Call of Duty is code, but then a whole bunch of Call of Duty is
- 1:11:53data and the part that's data is like the the configuration, the
- 1:11:57data that configures the weapon that you have.
- 1:11:59So you have an M4. And it's got like an attack
- 1:12:01power of 50 in a range of like 35 that those numbers are data
- 1:12:07and you so then you could easily say like come up with different
- 1:12:10guns or something, yes. Yeah, exactly.
- 1:12:13So that is already with it very much within the capabilities of
- 1:12:18the models that we see today is the ability to take natural
- 1:12:21language, understand game design, understand is
- 1:12:23anthropomorphizing, but like to take natural language and
- 1:12:26effectively translate it into. Functional contextually relevant
- 1:12:30data inside of an existing simulation right.
- 1:12:33So you could have way more skins like oh it generate us a bunch
- 1:12:36of skins generate and and there where you have the constraints
- 1:12:41you know it's not going to destroy your game.
- 1:12:43Maybe you can you can review it before versus like my sort of
- 1:12:48Agent Smith like wild futurist sort of thing is like OK, let an
- 1:12:52AI rampant in in the game and like mess with it.
- 1:12:57Yeah, Matrix is actually a great.
- 1:13:00Example because like, you know, kind of foundational, at least
- 1:13:02in the original Matrix was this idea of, you know, Neo as the
- 1:13:06One and the Agents could break the simulation, right?
- 1:13:11They worked outside of the simulation.
- 1:13:13What we're saying here explicitly is like, yeah, sure,
- 1:13:16that's cool with movies and all that, but like.
- 1:13:17An agent working within the simulation can actually be very
- 1:13:21powerful depending on how expansive your simulation is and
- 1:13:23so to like wrap it back to the NPC.
- 1:13:26Beyond just having a an interesting conversation, if an
- 1:13:29NBC agent inside of a game has the ability to change data, that
- 1:13:35can actually feel like pretty significant in terms of.
- 1:13:40You know, a world feeling alive and things like that.
- 1:13:42As a matter of fact, if you looked at like Starfield, right?
- 1:13:45And I don't know this to be true 100%, but I would guess it to be
- 1:13:47So a lot of what's going into the design of those worlds that
- 1:13:51don't feel so much alive or the ones that do.
- 1:13:54The big difference is the world's that don't feel alive.
- 1:13:56Where procedurally the data was procedurally generated and the
- 1:13:58world's that do feel alive were hand designed by people to feel
- 1:14:03so, but all within the same simulation and so really just a
- 1:14:06difference in data. Yeah, on the MPC side, part of
- 1:14:11the conversation is that you gotta remember MPCS are pretty
- 1:14:15good today. You know this.
- 1:14:17This is what it players have been interacting with for the
- 1:14:19last, you know, few decades. And so there are certain genres
- 1:14:25of games in which kind of a IMPCS which is moving from like
- 1:14:28a decision tree model for an MPC to a more open like kind of LLM
- 1:14:33model for MPC is actually discernible to justify the
- 1:14:37increased costs right around. Having that kind of deployment
- 1:14:41would be in specific genres I would say like you know open
- 1:14:45world and RPG being some of the most obvious ones.
- 1:14:47And we're we are seeing, you know, new studios being born
- 1:14:51with that approach. They're like, we're going to
- 1:14:53build a game where we're differentiated by our characters
- 1:14:56being LLMS, or they're going to sell LLM techniques to other
- 1:15:00other people. So it's like so there's there's
- 1:15:04a lot in between there and I would say there's yeah there's
- 1:15:06like studios that are saying all right we're going to have these.
- 1:15:11We're going to have a IMPC based characters that interact with
- 1:15:16you that that actually will like enrich like where the the
- 1:15:20stories might actually evolve like in in gameplay that starts
- 1:15:25getting really closely tied to well how does like you know game
- 1:15:29code also. Evolve, you know, dynamically
- 1:15:33to, to Keith's point, and I think a lot of this theoretical
- 1:15:37and what a lot of these game desks need to figure out is at
- 1:15:41what, at what point do you give all those degrees of freedom?
- 1:15:44And where do you actually just codify in the systems, which is,
- 1:15:47you know, how game making is done today?
- 1:15:51Because yeah, 'cause there's a cost consideration and then
- 1:15:54there's also the controlling the player experience.
- 1:15:59Part of it as well, you know if you actually have a a locked in
- 1:16:03decision tree, you know like you are creating player paths for
- 1:16:06these different segments of players and you're really
- 1:16:09optimizing your their experience and you have a lot of control
- 1:16:11there and where is giving up some of that control actually a
- 1:16:16good thing? So I would say like, you know,
- 1:16:18large studios when I talk to them on the MPC side are like
- 1:16:21we're going to do kind of like proof of concepts.
- 1:16:27Maybe with our smaller games, maybe with the new game and see
- 1:16:30whether it actually impacts and improves the player experience
- 1:16:35and then translates to like longer player retention and
- 1:16:38higher kind of lifetime values and how much players are
- 1:16:41spending in the game, etcetera. Before we make the determination
- 1:16:45that it is completely game changing and we're going to
- 1:16:47change all of our MPC systems across the games.
- 1:16:51Yeah, yeah. I think this gives an
- 1:16:52opportunity for a startup that wants to go.
- 1:16:54A lot more further in this and and especially with their core
- 1:16:57IP then where the incumbents are taking it.
- 1:17:00So it's an exciting definitely. I can see the risks for like a
- 1:17:03big developer. One like you, like you know with
- 1:17:06Microsoft and Sydney or whatever, you don't you don't
- 1:17:08know what the LMS will say. If you're just like sending out
- 1:17:11a game and you like free response from NPCS could be like
- 1:17:16terrifying. And two, I imagine it's got to
- 1:17:19be like much riskier for just like creating bugs in a game
- 1:17:22where you don't know the whole decision tree and you introduce
- 1:17:25a part of a system that's basically like trying to come up
- 1:17:28with a bunch of wild answers that you haven't gained out
- 1:17:32before. But I mean, as a consumer, I
- 1:17:34desperately want it. Like, you know, I feel like the
- 1:17:37sense that like this game could come off the rails or that you
- 1:17:40would have like an AI that's like feels Wilder.
- 1:17:43Moving to our third category, I'm glad we have such like a
- 1:17:45good framework like assets NPCS and sort of the the actual games
- 1:17:51being and code generated by with help from AI.
- 1:17:55Do we have like the first like AI game or like what are what is
- 1:18:01the sort of state of play right now in terms of games, sort of
- 1:18:05from whole cloth or components of games that are like human
- 1:18:10fairly out of the loop? Expect us to begin on the the
- 1:18:15game generation side. At the lowest level of
- 1:18:17complexity possible. I mean it just makes sense to
- 1:18:19start there. The funny thing is is actually
- 1:18:22it you don't have to get very high on the complexity spectrum
- 1:18:26before you start to be capable of generating a large portion of
- 1:18:31the games that are actually played.
- 1:18:32So if you think of you know categories of gaming especially
- 1:18:35on mobile categories such as social casino, puzzle, I mean
- 1:18:39puzzles like a multi billion dollar category social casino is
- 1:18:42as well with with billions of players.
- 1:18:46If you think of like 2D right is actually a low complexity game
- 1:18:50on because it is like pixel graphics and Sprite sheets and
- 1:18:54like 2 dimensional and all of that.
- 1:18:56So like RPGs, top down strategy games, like again, like there's
- 1:19:01actually a lot of gameplay that exists under that level and so
- 1:19:06we might see the game revolution in like 2D gaming.
- 1:19:08Like obviously mobile is popular right now, but also if we're
- 1:19:11saying OK 3D assets are really hard, we want to democratize,
- 1:19:16We're closer you know with with sort of yeah lower production
- 1:19:20type games like it could mean a bunch of really cool like well
- 1:19:23and you look at, you look at. Games slash platforms like
- 1:19:26Roblox and Minecraft, right? That again, we're low complexity
- 1:19:31on the fidelity still or largely low complexity on the fidelity
- 1:19:33side, but we're able to go very deep on kind of the game playing
- 1:19:37systems and customization and and creation and all of that.
- 1:19:42Yeah, I think that that's where. A lot of the kind of, well, I'll
- 1:19:46call it visible innovation will happen because of the problems
- 1:19:51that like Amy was talking about earlier, right?
- 1:19:52Creating 3D models, rigging those models like from a
- 1:19:55generative perspective on visual effects, all of that shit.
- 1:19:59I think there's entire teams of people at AAA studios that are
- 1:20:02working on. Like simulating better, more
- 1:20:05realistic like hair or movement and and stuff like that, right?
- 1:20:08That's going to continue AI. Well, that's the classic, but
- 1:20:10that's like every Pixar movie is like we've invented a new way
- 1:20:13for curly hair to move. And like that's the technical
- 1:20:16actually and underpinning and that's something is is
- 1:20:18ultimately meaningful when it's accumulated altogether for sure.
- 1:20:22But it's it that that type of advanced movie less obviously,
- 1:20:25yes, I think on that the the kind of code generation side
- 1:20:28within this within this three-step framework, you know
- 1:20:31it's going to start at the low complexity side.
- 1:20:33And So what we've already what we're seeing is we are seeing
- 1:20:35some early experiments kind of hit around Twitter.
- 1:20:38So you had you know people are using and actually I've done
- 1:20:40this as well. It's one of the things that got
- 1:20:41me really excited and I recommend anybody out there that
- 1:20:44you know wants to dabble in this space, go do it is you can sit
- 1:20:47down with Chachi BT right now and through like not one shot,
- 1:20:51but recursively build the game. Like I've done.
- 1:20:53There's a guy on Twitter with Yeah, I've done that.
- 1:20:56I love that. Yeah.
- 1:20:56Yeah. Yeah, yeah, yeah.
- 1:20:57I played B&D with Chachi BT too. Yeah, exactly, exactly.
- 1:21:01So. So that's already happening,
- 1:21:02right? And then I'm talking even
- 1:21:04generating code I tried to play. Poker with it too, though.
- 1:21:06It's sort of like in in visual or like just chapter, just like.
- 1:21:10You know, in the text format I got with D&D just to Stew on it
- 1:21:15for a second, I the memory problems are what drive me
- 1:21:18crazy. It's just like that it forgets
- 1:21:20it and that you sort of have to try and remind it.
- 1:21:23But it's so fun to have ChatGPT like make itself a character and
- 1:21:27everything. Yeah, what would be fairly
- 1:21:28probably easy is for you to spit out a a really like via ChatGPT
- 1:21:32to spit out a fairly simple Python simulation or, you know,
- 1:21:36pick another code. But Python seems to work
- 1:21:38reasonably well. That sets up like the framework
- 1:21:41for AD and D and then to use like GPT 4 as the kind of
- 1:21:46language model within that simulation.
- 1:21:48And you could do that and you'll have like you know you'll have a
- 1:21:52a a better working like DMD kind of game that you created and and
- 1:21:57so that's that's there. There's a few you know if you
- 1:21:59want to check out FRVR and I know the the founders over
- 1:22:03there, you know they've been working on this thing called Fr
- 1:22:06VR4. They have some demos out some
- 1:22:08videos out. I think they're starting to let
- 1:22:09some people I believe they're a game film and some people were
- 1:22:12able to play with it where it's actually like so natural
- 1:22:14language prompt based you put it in and and you start to generate
- 1:22:17like games playable like experiences and there are
- 1:22:23there's a number of startups that are also like like like us
- 1:22:25working on on this stuff as well and I expect in like the coming
- 1:22:28year my my guess is next year we'll start to see a lot of
- 1:22:31these you know hit more of a a broader production and and see
- 1:22:35like what what they can do. Takes like three to five years
- 1:22:38plus develop a AAA game, right. I think some of the more, yeah,
- 1:22:42casual games will be faster. So like you know maybe they're.
- 1:22:45I'm actually at on Real Fest in New Orleans right now.
- 1:22:48Like seeing like what maps could be created.
- 1:22:50With these kind of tools on UEFN, like that'll be
- 1:22:53interesting cuz that'll take, that'll be faster to market.
- 1:22:56Uni. What was it?
- 1:22:57Uefn the What's the unreal Like Fortnite creator?
- 1:23:01OK, OK. For Fortnite, yeah.
- 1:23:04Do you have a? Do you have a sense just like
- 1:23:07armchair of which? Big sort of game developers will
- 1:23:12be most forward thinking or where where you see the most
- 1:23:15aggression in terms of embracing what's going on.
- 1:23:18I have a straw man. Obviously you know a lot of this
- 1:23:20is like secret a lot of places, but my sense is that like a lot
- 1:23:24of new tech Asia developers were faster to it.
- 1:23:29So you know Netease tends to. I think both of the companies
- 1:23:31are developing their own like large language and image models
- 1:23:36and have been experimenting with all the tool tooling and.
- 1:23:40You know they were some of their they were some of Asia
- 1:23:43developers were some of the for like for a calm to us like we're
- 1:23:47the first to kind of put higher quality I would say like you
- 1:23:52know games watching games out live and and so I think they've
- 1:23:57been they've been they've been moving and they've been
- 1:23:59experimenting kind of for years in the AI space as well.
- 1:24:02Keith, I mean, I wanted to just like come back to your company
- 1:24:04for a second, like. I don't know, in two years, what
- 1:24:08do you think? What is the first like thing
- 1:24:10that you provide or like what's? I know there's a lot of sort of
- 1:24:14ambition there, but just sort of in terms of something concrete
- 1:24:17like what do you think the first move for you, you all is?
- 1:24:22Yeah, so we're the first move might actually be AB to B
- 1:24:26solution. I think there are many like
- 1:24:28areas inside gaming right now where on you know our technology
- 1:24:33can provide a a solve a very real problem that that some some
- 1:24:38studios and publishers have that are creating or publishing games
- 1:24:41of that low complexity threshold.
- 1:24:43The thing that really excites me though and and you said two
- 1:24:45years so, so that gives me more space to work in because I
- 1:24:47expect next year for us to be executing on B2B is this you
- 1:24:51know like lower complexity games that are creating themselves as
- 1:24:57you play. I do think that we are you know
- 1:24:59within two years of of us being able to launch games like that
- 1:25:02and honestly we're not going to be the only ones.
- 1:25:04So I'd expect to see a a number of games like that on the market
- 1:25:07within two years. And you, well, you'll be
- 1:25:09designing some of it. You'll be like, oh, it's a
- 1:25:12Subway surfer game, but the AI is free to do this thing.
- 1:25:16Or you think from sort of the beginning in some way the AI is
- 1:25:20the way that we're doing it is we are designing kind of the the
- 1:25:22foundational capabilities. It's sort of like you know, if
- 1:25:25you think of think of chat JPT right and GPT 4 and and language
- 1:25:31models right. They are attempting to predict
- 1:25:33the next word, right? And and what has been just so
- 1:25:37like so unexpectedly, at least for me like surprising is that
- 1:25:41in trying to predict the next word when you feel like throw
- 1:25:45enough parameters and like training data and all that stuff
- 1:25:47at it, you end up with what feels a lot like intelligence
- 1:25:50and like something that's emergent kind of capability.
- 1:25:53And so for us it's about predicting the next gameplay
- 1:25:56experience, the next moment of gameplay, the next level and I
- 1:26:00would expect similarly so. So like you know we're testing
- 1:26:04our systems with trying to figure out hey can it make a
- 1:26:07match three-game. Can it make a a level of
- 1:26:09something that looks like Mario Brothers.
- 1:26:11Yes and yes. So you have to build the.
- 1:26:13First level and then you're like, what's the next?
- 1:26:18Yes, actually that's part of, but not for.
- 1:26:21We have to build a level. So yeah, you're really fast,
- 1:26:26like technical under the hood. So like we have what's called a
- 1:26:29game descriptor, which is a data structure that describes the
- 1:26:32game in data or describes a game in data.
- 1:26:35And then is is run through an interpreter that actually
- 1:26:37produces the game. And then we use right now Unity
- 1:26:40to to actually render and interact with the the
- 1:26:42application. And so in that like framework or
- 1:26:45schema, like we produced the first game descriptor, the first
- 1:26:51data that describes a game within the simulation.
- 1:26:53But then and and let's say that the first thing that we built
- 1:26:56was like a something that looked like a 2D platformer like a
- 1:26:59Mario or build Prince of Persia's.
- 1:27:01But then we said here's a game descriptor of Mario to the LLM
- 1:27:05now make us a game descriptor for a top down shooting game.
- 1:27:10And it did and it works. And so like that's like it's
- 1:27:14it's like in a way if that if that answers your question
- 1:27:17right, like yes in a way you need to populate it so that it
- 1:27:20understands the simulation and the parameters that it has to
- 1:27:22work with. But it's not like I have to make
- 1:27:25the first level of all the games.
- 1:27:27I just make a game and it functions and works and then all
- 1:27:30the other games can be generated.
- 1:27:32Amy, I just wanted to ask like a lesson or two from like the
- 1:27:36crypto gaming experience or like moving, you know, I I think on a
- 1:27:41super surface level take it's like.
- 1:27:43There was a lot of excitement in crypto and now there's a lot of
- 1:27:46excitement in AI like will you know, some of it overhyped or
- 1:27:49like I don't know what. What are the lessons from having
- 1:27:52gone through sort of the heat of crypto and now sort of and
- 1:27:57thinking a lot about sort of AI stuff now?
- 1:27:59I'll, I'll I'll kind of keep my answer to the crypto blotching
- 1:28:03game side just because I mean the broader crypto space.
- 1:28:06I would say that the the token launches and the abuse of that.
- 1:28:11You know, really kind of dictated a lot of that
- 1:28:12narrative. But on the gaming side, I to be
- 1:28:16honest, I I always thought there was like way too much venture
- 1:28:19dollars within the blockchain gaming space.
- 1:28:22There were a lot of really bright developers that embarked
- 1:28:27and wanted to make games for the first time, which is really
- 1:28:30difficult. I mean, games are really
- 1:28:31difficult to make, right? Not only do you have to know how
- 1:28:34to make a complicated system work that balances the game and
- 1:28:39produces. You know, kind of our thousand
- 1:28:42hours of content for people to have fun with, but also how to
- 1:28:45live operate that afterwards. I mean this is, it's just really
- 1:28:48difficult. And so it took some time before
- 1:28:52I would say experienced game devs really actually came into
- 1:28:56the blockchain industry and was excited as well in terms of
- 1:28:58finding ways to leverage blockchain technology that that,
- 1:29:03you know, could make gameplay feel like something that a
- 1:29:06player hasn't experienced before.
- 1:29:08I mean if it basically feels the same, then.
- 1:29:10I would say anything that adds friction actually it's just bad
- 1:29:13in game design but it has to be it has to be part of like it has
- 1:29:18to create a gaming meta that that is actually totally
- 1:29:23different in in that and is unique And so I would basically
- 1:29:28apply and so if you look at all the VC dollars put like that
- 1:29:32went into the blockchain games industry a lot of it was not to
- 1:29:35fun teams with a clear vision of it was more just like.
- 1:29:39I'm going to fund it because it's in this category, but the
- 1:29:41devil is absolutely the detail, especially with the rate of
- 1:29:44failure in the game's content side.
- 1:29:47And so I would apply the same kind of, you know, view on with
- 1:29:53with with the AI intersection which is 1.
- 1:29:56Like you know, just having the content be AI generated does not
- 1:30:01mean anyone is going to find this font, and so the bar is it
- 1:30:05needs to be really great content.
- 1:30:07And also it needs to be and compared to what the other
- 1:30:13things that I could play in the market, because you know, the
- 1:30:15most valuable thing is actually person's time.
- 1:30:17You know, why would I play some game when I can just play Breath
- 1:30:21of the Wild or Diablo Or, you know, CS?
- 1:30:24Investors may want to invest because it has sort of a new
- 1:30:27technique, but at the end of the day the players don't care that
- 1:30:31much how it was created and want it to end up being a great game
- 1:30:35and. The people who can do that, or
- 1:30:36often the people who've built great games.
- 1:30:39Before players don't care about ideology, they care about and I
- 1:30:43but I'm game to ask you sometimes, but I mean players
- 1:30:45just care about like one of the content is just fun.
- 1:30:48All right. Our last question what looking
- 1:30:50forward five years like with the promise of today, like what what
- 1:30:55is something that you're most excited about what might be
- 1:30:58possible in in gaming and AI. When I look back five years ago
- 1:31:02in gaming what I'm most and then I look forward five years.
- 1:31:06What I'm most excited about is for me personally and I think
- 1:31:10many people within the industry this last five years have been
- 1:31:12kind of boring just straight up like as a as an industry
- 1:31:15participant you know it took five years to make Starfield and
- 1:31:20again I'm gonna play it, I'm gonna enjoy it.
- 1:31:23But you know I I doubt that I'm gonna be blown away by it.
- 1:31:26And that's really been the story just over and over and over.
- 1:31:29And the truth is if I ask myself and I look at the, you know, the
- 1:31:32the best games that were launched in the last five years,
- 1:31:35none of them were, you know, none of them were unlocked or
- 1:31:39made possible by some specific technology that came out in the
- 1:31:42last five years. There's a lot of same same,
- 1:31:44there's a lot of sequels and like that's honestly
- 1:31:47unfortunate. And it's not just limited to
- 1:31:48games. It's been all over, I think
- 1:31:50entertainment and media, when I look forward to five years, what
- 1:31:52excites me is that I think that picture will be different.
- 1:31:56And I also think that a lot of the terminology we use describe
- 1:32:01gaming is going to get so mushed and blurred and blended that
- 1:32:07it's just going to look very, very different.
- 1:32:08I think what what is a consumer or a player, what is a creator,
- 1:32:13what is a a streamer or a passive participant.
- 1:32:17I think all of those lines are going to get blurred in really
- 1:32:21exciting ways as AI enables you know a a a change in how we
- 1:32:26interact with technology in a democratization to kind of the
- 1:32:29interaction with technology and and that unknown is what excites
- 1:32:33me. That's unknown is what I'm I'm
- 1:32:35striving for and and whether I'm right with the company that
- 1:32:38we're building or not like I'm I'm here for the change I I.
- 1:32:42Something new, Yeah, I I totally agree with that sentiment.
- 1:32:45And yeah, the blurring that feels so inevitable.
- 1:32:49That's a exciting vision, Amy. What would what would you offer?
- 1:32:55Yeah, I would agree. So I think in five years most of
- 1:32:59the game comments will still be the same game comments.
- 1:33:02Will there be one or two new entrants that has launched
- 1:33:06something pretty incredible leveraging you know the new
- 1:33:10tech. I I certainly hope so.
- 1:33:13And I think where I hear, where we hear like skepticism,
- 1:33:16incumbents still will both make us like, you know, kind of
- 1:33:22proceed cautiously because you know.
- 1:33:25And that's so much knowledge is back there.
- 1:33:27But also, all right, this is not where a lot of people are
- 1:33:30willing to risk, you know, most $9 billion titles.
- 1:33:32So this gives an opportunity for a startup, an indie studio to
- 1:33:37really break out. So yeah, I I'm hoping there will
- 1:33:40be a couple of of sort of AI native studios that are able to
- 1:33:43create something super awesome. Awesome.
- 1:33:47Keith, Amy, thanks so much for coming on the show.
- 1:33:49This is great. Thanks so much.
- 1:33:51Thanks, Eric. All right, fun talking about
- 1:33:53games. Thanks so much.
- 1:33:55That's our episode. I'm Eric Newcomer, Co host, our
- 1:33:58Max child and James Wilsterman Co founders of Volley and my
- 1:34:02long time friends. Thank you to Scott Brody, our
- 1:34:05producer, Riley Kinsella, my chief of staff, Gabby Caliendo,
- 1:34:09key person of Volley who's helping make this whole
- 1:34:11conference happen. Shout out to young Chomsky for
- 1:34:15the theme music. Please like, comment, subscribe
- 1:34:18on YouTube. Follow us and give us a review
- 1:34:21on Apple Podcasts. And of course, subscribe to
- 1:34:24newcomernewcomer.co, Become a paid subscriber, support
- 1:34:30everything we do, and then go play Volley games on your Alexa
- 1:34:34say, hey Alexa, let's play Song quiz anyway, All right.
- 1:34:38Thanks so much. See you next week.