**Moderator:**

Michael Gold - [X](https://twitter.com/michaelgold) | [Linkedin](https://www.linkedin.com/in/themichaelgold/)

**Panelists:**

Amber Cook - [X](https://twitter.com/100gamesproject) | [Linkedin](https://www.linkedin.com/in/ambermcook/)

Bilawal Sidhu - [X](https://twitter.com/bilawalsidhu) | [Linkedin](https://www.linkedin.com/in/bilawalsidhu/)

Kayla Comalli - [X](https://twitter.com/kay_coms) | [Linkedin](https://www.linkedin.com/in/kalyco/)

Vatsal Bhardwaj - [X](https://twitter.com/vatsal) | [Linkedin](linkedin.com/in/vatsal/)

[00:36.000 --> 00:41.000] **[Michael]** All right, thanks everyone.

[00:41.000 --> 00:51.000] So, we're here to talk about the impact of AI in terms of gameplay.

[00:51.000 --> 00:58.000] And specifically with this panel, what I want to do for all of you is give you kind of a

[00:58.000 --> 01:08.000] presentation of everyone's take on the technology and how that impacts the go-to-market for

[01:08.000 --> 01:10.000] implementing AI and games.

[01:20.000 --> 01:36.000] So, we want to give you a sense of what the go-to-market could look like for a game that

[01:36.000 --> 01:45.000] leverages heavily AI and how people are thinking about it both from the marketing side, from

[01:45.000 --> 01:49.000] the technology side, from the content side.

[01:49.000 --> 01:56.000] There's enabling technology here that's been around that AI empowers and there's also brand

[01:56.000 --> 02:04.000] new stuff like large language models and generative content like stable diffusion and even 3D

[02:04.000 --> 02:06.000] generative content that's on the horizon.

[02:06.000 --> 02:07.000] So, we'll talk about that.

[02:07.000 --> 02:11.000] We're going to keep it pretty non-technical.

[02:11.000 --> 02:15.000] How many of you in the audience want us to get technical?

[02:15.000 --> 02:17.000] Raise your hand.

[02:17.000 --> 02:20.000] We can geek out with you.

[02:20.000 --> 02:25.000] But we'll save that to the end of the panel.

[02:25.000 --> 02:33.000] And what I'm going to do is I'm going to put together a Google Doc with all of the panelists'

[02:33.000 --> 02:38.000] content info and all of our thoughts in written form.

[02:38.000 --> 02:40.000] I'll share that on LinkedIn.

[02:40.000 --> 02:45.000] So, if you follow any of us on LinkedIn, you'll be able to get that Google Doc with everything

[02:45.000 --> 02:46.000] we spoke about here.

[02:54.000 --> 03:02.000] So, with that, I'll let my panelists introduce themselves and we'll kick things off.

[03:07.000 --> 03:10.000] **[Vatsal]** My name is Vatsal Bhagwaj.

[03:10.000 --> 03:17.000] I've been in games for the last 15 years, building games or building platforms and technologies

[03:17.000 --> 03:18.000] for games.

[03:18.000 --> 03:21.000] Most recently, I was Chief Product Officer at Stills.

[03:21.000 --> 03:26.000] Before that, I was Head of Games Tech and Simulation Tech at AWS.

[03:26.000 --> 03:30.000] I had a similar role at Meta/Oculus before that.

[03:30.000 --> 03:36.000] Prior to that, I used to make games, mobile, free-to-play games, and Zynga and Storm 8.

[03:36.000 --> 03:39.000] So, yeah, I'm very excited to be here.

[03:48.000 --> 03:49.000] **[Amber]** I'm Amber Cppl.

[03:49.000 --> 03:52.000] I work in marketing for a company called Hidden Door.

[03:52.000 --> 03:57.000] We're building a platform that allows you to connect with your favorite universes and

[03:57.000 --> 04:01.000] get into interactive fiction very quickly based on these traditional worlds.

[04:01.000 --> 04:04.000] My background is mostly in tabletop gaming.

[04:04.000 --> 04:09.000] I moved over to digital to work as the Head of Marketing for Roll20, which is sort of

[04:09.000 --> 04:15.000] a platform for playing pen and paper role-playing games about five years ago.

[04:15.000 --> 04:20.000] I just switched over to this new AI platform a year ago.

[04:20.000 --> 04:21.000] **[Bilawal]** Hey, everyone.

[04:21.000 --> 04:22.000] My name is Bilawal Sidhu.

[04:22.000 --> 04:24.000] I'm an AI and 3D creator.

[04:24.000 --> 04:31.000] That means I make educational and hopefully entertaining content on YouTube, TikTok, and

[04:31.000 --> 04:34.000] I guess X is what we're calling it these days.

[04:34.000 --> 04:39.000] Prior to that, I spent a decade in tech predominantly at Google where I was a senior product manager

[04:39.000 --> 04:45.000] working on immersive capture, AR/VR platforms, and geospatial 3D mapping.

[04:45.000 --> 04:49.000] I fell in love with all of this good stuff at the age of 11 when I discovered VFX and

[04:49.000 --> 04:50.000] computer graphics.

[04:50.000 --> 04:54.000] And let me tell you with this current wave of AI, that's exactly how I feel, and I'm

[04:54.000 --> 04:56.000] super excited to be here.

[04:56.000 --> 04:57.000] **[Kayla]** Hi, everybody.

[04:57.000 --> 04:58.000] Hi, guys.

[04:58.000 --> 04:59.000] I'm Kayla Comalli.

[04:59.000 --> 05:01.000] I'm the CEO of Lovelace Studio.

[05:01.000 --> 05:06.000] We're building a user-generated content platform that leverages AI tools in a comprehensive

[05:06.000 --> 05:07.000] way.

[05:07.000 --> 05:14.000] So from language models with characters to general election networks for growing mechanics

[05:14.000 --> 05:19.000] and modular components and entire world building using prompt engineering.

[05:19.000 --> 05:27.000] And we're currently just a startup right now, but my experience is in robotics and AI perception,

[05:27.000 --> 05:34.000] working with tools and systems and have pivoted as a lifelong lover of game into how we can

[05:34.000 --> 05:37.000] build that in a big place.

[05:37.000 --> 05:39.000] **[Michael]** Thank you, everyone.

[05:39.000 --> 05:49.000] So there's this holy grail, if you will, about the infinitely replayable game that people

[05:49.000 --> 05:56.000] think AI might enable, procedural content generation might enable as well.

[05:57.000 --> 06:02.000] Is that the most exciting opportunity as you see it?

[06:02.000 --> 06:04.000] This is for the entire panel.

[06:04.000 --> 06:08.000] What opportunities really excite you?

[06:08.000 --> 06:11.000] What technologies are behind it?

[06:23.000 --> 06:26.000] **[Amber]** I think that enables infinitely replayable stories, but I don't really think that's the

[06:26.000 --> 06:27.000] most exciting thing about it.

[06:27.000 --> 06:31.000] I think the scalability is really interesting, but I think the ability to take these fictional

[06:31.000 --> 06:36.000] universes and bring fans really close to them and let them put their fingerprints on those

[06:36.000 --> 06:40.000] stories and write their friends into those worlds and then sort of say, all right, I

[06:40.000 --> 06:44.000] didn't want the TV show to go this way last night when I watched my favorite episode.

[06:44.000 --> 06:49.000] I want to see what happens if we have been able to finish the story out this way and

[06:49.000 --> 06:53.000] create their own campaign and take it a different direction than the writers do.

[06:53.000 --> 06:55.000] That's exciting.

[06:55.000 --> 07:00.000] **[Michael]** Is everyone here familiar with Showrunner?

[07:00.000 --> 07:03.000] We've got a lot of head nods.

[07:08.000 --> 07:19.000] AI generated content that kind of took a South Park episode and made it automated with bots,

[07:19.000 --> 07:26.000] and it was kind of good, but also needed a lot of work to be really good what you would

[07:26.000 --> 07:28.000] expect for AAA.

[07:28.000 --> 07:33.000] I'm wondering if anyone on the panel has thoughts and ideas of what they would do to improve

[07:33.000 --> 07:37.000] that with today's topic.

[07:37.000 --> 07:40.000] **[Bilawal]** The way I think about it is it gives you the scaffolding, right?

[07:40.000 --> 07:45.000] As mentioned in the prior panel, people were talking about world building

[07:45.000 --> 07:50.000] the ability to distill down the attributes of a world you’re building into something and then scale

[07:50.000 --> 07:54.000] that across a bunch of different episodes, perhaps in this case, and the fact that you

[07:54.000 --> 07:59.000] can go and see if something kind of makes sense, whether that's using, you know, the

[07:59.000 --> 08:03.000] Showrunner example that you gave, or taking a bunch of mid-journey art and running it

[08:03.000 --> 08:07.000] through Runaway and seeing motion added to it, I think it's just profound, right?

[08:07.000 --> 08:11.000] Like, it may not be the final product, but it gets you so much closer to it that you

[08:11.000 --> 08:16.000] can make more informed decisions about how you use that limited, you know, kind of human

[08:16.000 --> 08:19.000] resource that you have to achieve it.

[08:19.000 --> 08:21.000] **[Vatsal]** I think just reflecting on both those things of how do you make a forever game and this

[08:28.000 --> 08:33.000] South Park episode which came out, at least the way I think of it

[08:33.000 --> 08:37.000] is, forever games, by nature they are emergent.

[08:37.000 --> 08:43.000] And, you know, and what I mean by that is, can you set up a construct where the simple

[08:43.000 --> 08:47.000] rules, the players and the community can continue to evolve the game

[09:03.000 --> 09:09.000] Now, everything which we have seen in AI today, especially like the most recent trends

[09:09.000 --> 09:16.000] around LLMs and Gen AI, potentially it could put it on steroids, like potentially.

[09:16.000 --> 09:20.000] And the reason I say potentially is that's still the theory.

[09:20.000 --> 09:23.000] I think there are a lot of problems to be worked out.

[09:23.000 --> 09:30.000] I think the challenge is, can you harness these models in a way where you have a scaffolding

[09:30.000 --> 09:33.000] of control so that things don't go off range

[09:33.000 --> 09:39.000] So, again, just to tie it together, I think we can master sort of the hallucination of

[09:39.000 --> 09:46.000] these AI systems in a way that it offers delight and not misery.

[09:46.000 --> 09:51.000] I think we'll see a lot of invention in video games.

[09:51.000 --> 09:57.000] **[Michael]** So, I think one of the things that we should outline for the audience is, what is the state

[09:57.000 --> 09:59.000] of the art now?

[09:59.000 --> 10:01.000] What can it do?

[10:01.000 --> 10:03.000] What can you do with generative AI?

[10:03.000 --> 10:06.000] And also, what can't you do?

[10:06.000 --> 10:11.000] And based on your experiments, maybe Kayla, you want to take that one?

[10:11.000 --> 10:12.000] **[Kayla]** Sure, yeah.

[10:12.000 --> 10:16.000] And this actually goes off of the infinite question as well, because I think that what

[10:16.000 --> 10:21.000] you can do now is exceedingly powerful if you can tie it all together.

[10:21.000 --> 10:27.000] From a consumer side, things get very expensive if you're talking about, oh, I want a system

[10:27.000 --> 10:34.000] that can move from text to 3D comprehensive with meshes and hundreds of thousands of tries.

[10:34.000 --> 10:43.000] But if you want something that has the ability to create a mesh, you could use a parametric

[10:43.000 --> 10:46.000] base engine and just get that in less than 33 milliseconds.

[10:46.000 --> 10:50.000] And it might not be the highest rendering capabilities, but the technology is evolving

[10:50.000 --> 10:55.000] so quickly that if you tie it all together and expose it to players early on, you can

[10:55.000 --> 11:03.000] start leveraging this applied tool where the entire world has, 66% of the United States

[11:03.000 --> 11:06.000] is just gamers, 68% of adults are gamers.

[11:06.000 --> 11:11.000] And this is this growing underlying understanding that this is a skill set.

[11:11.000 --> 11:12.000] Everybody knows it has.

[11:12.000 --> 11:17.000] And if you could just give them the tools and use data-driven research to refine and

[11:17.000 --> 11:19.000] adjust it, then it's unstoppable, I think.

[11:19.000 --> 11:21.000] Yeah, so it's a good word for it.

[11:21.000 --> 11:28.000] **[Michael]** Yeah, the interplay between AI generation and parametric or procedural generation like

[11:28.000 --> 11:33.000] we see in Minecraft and No Man's Sky is a really big opportunity.

[11:33.000 --> 11:36.000] And it's kind of straightforward to chain these things together.

[11:36.000 --> 11:39.000] So I think we'll see a lot of that as well.

[11:39.000 --> 11:45.000] So Bilawal, I wanted to ask you, you have a unique insight to this.

[11:45.000 --> 11:52.000] What topics get the most responses and the most feedback when you make content around

[11:52.000 --> 11:53.000] them?

[11:53.000 --> 11:57.000] What are the general gaming audiences most excited about, most interested in?

[11:57.000 --> 12:00.000] What are they kind of scared about?

[12:00.000 --> 12:06.000] **[Bilawal]** If you have any posts go viral on AI, there's going to be a lot of hate and fear that you

[12:06.000 --> 12:07.000] get, for sure.

[12:07.000 --> 12:12.000] But on the positive side, I think people are very excited about the unbundling of these

[12:12.000 --> 12:13.000] vertical studios.

[12:13.000 --> 12:18.000] Like all these capabilities that have historically been very, very expensive.

[12:18.000 --> 12:23.000] Let's take reality capture and performance capture as two examples, perhaps.

[12:23.000 --> 12:24.000] So reality capture, right?

[12:24.000 --> 12:25.000] Like photogrammetry isn't new.

[12:25.000 --> 12:28.000] It's been around since Time immemorial, right?

[12:28.000 --> 12:30.000] Even before we even had digital cameras.

[12:30.000 --> 12:36.000] But what took data centers just a decade ago, you could do on like a $1500 Nvidia GPU

[12:36.000 --> 12:41.000] using apps like Reality Capture, which Epic Games now owns.

[12:41.000 --> 12:47.000] And then just two years ago, we've seen this rise of neural representations, these sort

[12:47.000 --> 12:52.000] of volumetric light fieldy, not exactly a light field, but close enough representations

[12:52.000 --> 12:58.000] to take just commodity imagery of the real world and create these immaculate 3D renditions

[12:58.000 --> 12:59.000] of it, right?

[12:59.000 --> 13:00.000] So that's magical.

[13:00.000 --> 13:05.000] The fact that you can capture these spaces, bring it into Unity or Unreal Engine and use

[13:05.000 --> 13:09.000] that as your game world, not only that, but like, you know, one of the other things I

[13:09.000 --> 13:12.000] had the opportunity to work at Google is 3d mapping, go get photorealistic 3D titles.

[13:12.000 --> 13:18.000] You have all of Earth now in a game engine, plus all the micro scale data that you are

[13:18.000 --> 13:21.000] capturing at ground level can be layered and anchored to that.

[13:21.000 --> 13:22.000] That's magical.

[13:22.000 --> 13:24.000] But that's just the reality capture side, right?

[13:24.000 --> 13:29.000] Like performance capture, you needed these crazy optical tracking systems, like, you

[13:29.000 --> 13:31.000] know, go rent out a Vicon studio, whatnot.

[13:31.000 --> 13:38.000] And this tech that had been almost incubated for, you know, kind of making whimsical Snapchat

[13:38.000 --> 13:42.000] filters are now finally being applied to content at a much higher fidelity, right?

[13:42.000 --> 13:46.000] So when you look at move AI, what they're doing, or if you look at what Wonder Dynamics

[13:46.000 --> 13:52.000] is doing, taking all these perception AI capabilities and being able to use that as a very small

[13:52.000 --> 13:58.000] team to just use your iPhone or like a Sony camera or small array of cameras to digitize

[13:58.000 --> 14:00.000] the performance and bring it into game engines.

[14:00.000 --> 14:05.000] Like, I think that has that 10x multiplier effect where it levels the playing field for

[14:05.000 --> 14:09.000] indies, but even people in bigger studios can kind of iterate faster because you don't

[14:09.000 --> 14:13.000] have to go get studio time to try out or validate a concept.

[14:13.000 --> 14:19.000] **[Michael]** Just from the audience here, have any of you already replaced motion capture stages for

[14:19.000 --> 14:27.000] capturing player animations with tools like move.ai or class or Rococo?

[14:27.000 --> 14:29.000] **[Bilawal]** Or augmenting it at least.

[14:29.000 --> 14:30.000] **[Michael]** Yeah.

[14:30.000 --> 14:39.000] So that is a big thing to look into because you can use iPhones and really inexpensive

[14:39.000 --> 14:48.000] mocap for game animations instead of renting studio time.

[14:48.000 --> 14:49.000] Go ahead, yeah.

[14:50.000 --> 14:59.000] **[Vatsal]** Just to jump in, just as a former game maker, I think where there are real gains, if you're

[14:59.000 --> 15:05.000] al game maker, is you can prototype and test things really,

[15:05.000 --> 15:06.000] really quickly.

[15:08.000 --> 15:12.000] Instead of three months, you can bring it down to really short timelines.

[15:12.000 --> 15:18.000] But when it comes to how do you take, whether it's diffusion models for 2D or NERF-based

[15:18.000 --> 15:26.000] models for 3D, how do you take these, integrate them into your current workflows at a quality

[15:26.000 --> 15:28.000] which your players expect?

[15:28.000 --> 15:31.000] I think there is lots of work to be done there.

[15:31.000 --> 15:37.000] These are amazing toys, and they'll get to production very, very soon.

[15:37.000 --> 15:45.000] But there is a lot of improvements around building security, how data is handled, integration

[15:45.000 --> 15:51.000] in workflows, which will need to get built out, and is getting built out over the next

[15:51.000 --> 15:53.000] couple of years, which will make it compelling.

[15:53.000 --> 15:58.000] I don't think we are necessarily there yet in many of these cases.

[15:58.000 --> 16:04.000] **[Michael]** I want to know more about the AI bar exam that Johnny mentioned.

[16:06.000 --> 16:13.000] There are so many legal references today, but I think we all need to have these types

[16:13.000 --> 16:18.000] of pass-fail exams, whether they use reinforcement learning or just traditional approaches to

[16:18.000 --> 16:23.000] make sure that the content quality is good.

[16:23.000 --> 16:28.000] I'm curious, Amber, when you have conversations, because you're on the marketing side, when

[16:28.000 --> 16:36.000] you have conversations with brands about licensing content, how do you explain content quality

[16:36.000 --> 16:39.000] with what you're doing with AI, and how do you get them comfortable with it?

[16:39.000 --> 16:41.000] What does the conversation look like?

[16:41.000 --> 16:46.000] **[Amber]** We have a real advantage in that our CEO is somebody who wrote a book on the ethics

[16:46.000 --> 16:47.000] of AI.

[16:47.000 --> 16:50.000] Obama is chief data scientist, so we have a little bit of an advantage, just because

[16:50.000 --> 16:52.000] she has some credibility in the field.

[16:52.000 --> 16:55.000] But what we're finding is that it just depends on who we're talking to.

[16:55.000 --> 16:59.000] Originally, our idea was that we were going to work with literary universes.

[16:59.000 --> 17:03.000] Because everything that we're doing, where a lot of people are talking about motion capture

[17:03.000 --> 17:06.000] and all this stuff, we're building basically an interactive graphic novel.

[17:06.000 --> 17:10.000] It sort of fits into the cozy game category more than anything else, more interactive

[17:10.000 --> 17:11.000] fiction.

[17:11.000 --> 17:15.000] What we need to make people comfortable with is that we're going to take their world, their

[17:15.000 --> 17:17.000] IP, and we are not going to exploit it in any way.

[17:17.000 --> 17:19.000] Batman's not going to die.

[17:19.000 --> 17:21.000] Batman's not going to start kissing random people in the street.

[17:21.000 --> 17:25.000] How do we make those rails there so that we don't violate any IP standards?

[17:25.000 --> 17:28.000] What are the laws of physics of those universes?

[17:28.000 --> 17:33.000] When we started approaching authors, it's just a difficult time to approach authors

[17:33.000 --> 17:34.000] right now.

[17:34.000 --> 17:38.000] We started getting replies from literary agents saying we don't want to exploit so-and-so's

[17:38.000 --> 17:39.000] work this way.

[17:39.000 --> 17:44.000] But what I'm finding is that the studios, the entertainment studios, realize that they're

[17:44.000 --> 17:45.000] moving forward in these ways.

[17:45.000 --> 17:50.000] They're more open to collaborating with us, and they understand that we'll put the work

[17:50.000 --> 17:54.000] in, and we'll have to figure out how do you get brand approvals and put AI in that term.

[17:54.000 --> 17:55.000] How do you...

[17:55.000 --> 17:58.000] You can't approve every single thing that is generated.

[17:58.000 --> 18:00.000] What is that going to look like?

[18:00.000 --> 18:03.000] That's the whole new world of licensing that I think we're going to be playing around in.

[18:03.000 --> 18:06.000] A lot of those bigger studios aren't afraid of it, but they're using...

[18:06.000 --> 18:10.000] They're almost using us as consultants in some ways because they're coming in and they're

[18:10.000 --> 18:14.000] having AI summits internally and asking us to come talk about what it's going to look

[18:14.000 --> 18:18.000] like to partner with us as one of their first AI licensees.

[18:18.000 --> 18:21.000] Some of the more independent studios are...

[18:21.000 --> 18:25.000] That have really strong brands and are very sensitive.

[18:25.000 --> 18:30.000] For example, the tabletop RPG community is very sensitive to AI who send it to the placement

[18:30.000 --> 18:31.000] of authors or writers.

[18:31.000 --> 18:34.000] Those brands are being a little bit more like...

[18:34.000 --> 18:37.000] We don't want to be the first ones out of the gate to announce a license.

[18:37.000 --> 18:39.000] We don't want to be the ones going to bat for AI.

[18:39.000 --> 18:44.000] Once everybody's over those first conversations, then we're happy to partner with you and license

[18:44.000 --> 18:46.000] our content to you.

[18:46.000 --> 18:48.000] **[Michael]** That makes a lot of sense.

[18:48.000 --> 18:49.000] It's tough.

[18:49.000 --> 18:53.000] Nobody wants to be the pioneer with the arrows in their backs.

[18:53.000 --> 18:55.000] That's always true.

[18:55.000 --> 19:02.000] I was expecting us...

[19:02.000 --> 19:07.000] Well, part of me thought that we might have picketers out here.

[19:07.000 --> 19:08.000] I didn't know.

[19:08.000 --> 19:09.000] I'm happy we are.

[19:09.000 --> 19:13.000] It seems to be a pretty AI-positive crowd here.

[19:13.000 --> 19:17.000] We're self-selected to come to a conference.

[19:17.000 --> 19:21.000] There are some tools, though, that are a little bit scary.

[19:21.000 --> 19:29.000] I was chatting with the co-founder of a tool called CoquiAI, which does voice model cloning.

[19:29.000 --> 19:37.000] He was telling me he thought that in games in the next year or two, everything except

[19:37.000 --> 19:41.000] for cutscenes would be AI voice-generated.

[19:41.000 --> 19:43.000] That was his prediction.

[19:43.000 --> 19:46.000] Of course, he's got much skin in the game there.

[19:46.000 --> 19:49.000] The cost is expensive.

[19:49.000 --> 19:56.000] It costs $10 to generate an hour of audio.

[19:56.000 --> 19:57.000] I'm curious.

[19:57.000 --> 20:06.000] If those are the parameters, what do you all think will be the use of that type of AI in

[20:06.000 --> 20:07.000] games?

[20:07.000 --> 20:10.000] Do you think it's going to blow up, or do you think there's going to be pushback to

[20:10.000 --> 20:11.000] it?

[20:11.000 --> 20:14.000] You can all jump in..

[20:15.000 --> 20:16.000] **[Amber]** I had a marketing agency.

[20:16.000 --> 20:19.000] I was looking to hire for short-form video content recently.

[20:19.000 --> 20:25.000] They captured my CEO's voice from a podcast she had done, and then they sent me short-form

[20:25.000 --> 20:29.000] video they created with her likeness, without any permission or anything, just as a pitch

[20:29.000 --> 20:30.000] to get us to work with them.

[20:30.000 --> 20:31.000] I hope it's not that.

[20:31.000 --> 20:34.000] I think content creation is going to be really interesting, though.

[20:34.000 --> 20:39.000] There's going to be a lot of weird toxicity that we're going to have to deal with because

[20:39.000 --> 20:40.000] of it.

[20:40.000 --> 20:44.000] **[Bilawal]** But I think there's also going to be some amazing personalities that come out of it

[20:44.000 --> 20:46.000] because of that.

[20:46.000 --> 20:49.000] I think the costs are going to go down.

[20:49.000 --> 20:54.000] The rate at which we're going from R&D research publication to actual products, now it is

[20:54.000 --> 20:59.000] an interesting situation where it's the independents that are far more inclined to play with these

[20:59.000 --> 21:04.000] capabilities based on all the training data provenance conversations we've had.

[21:04.000 --> 21:09.000] But it's an inevitability that the optimization will happen if we just take the reality capture

[21:09.000 --> 21:10.000] space.

[21:10.000 --> 21:14.000] I mean, it took 12 hours on a TPU to train like a NeRF.

[21:14.000 --> 21:22.000] Last January, and then Instant NGP came out, courtesy NVIDIA, and took a minute on a $1,500

[21:22.000 --> 21:23.000] consumer grade GPU.

[21:23.000 --> 21:27.000] But hey, still playing back and rendering neural volumes in real-time was still hard.

[21:27.000 --> 21:32.000] Now you've got a new technique, gaussian splatting, that's 240 frames per second.

[21:32.000 --> 21:37.000] I think that rate of innovation is insane, and I think it was the Unity gentlemen and

[21:37.000 --> 21:40.000] a couple of the other folks talking on-device at the edge.

[21:40.000 --> 21:46.000] I think right now, I'm doing a content partnership with a semiconductor company

[21:47.000 --> 21:48.000] They're running ControlNet on-device.

[21:48.000 --> 21:50.000] This content will be coming out next month.

[21:50.000 --> 21:51.000] That's wild.

[21:51.000 --> 21:56.000] I needed a crazy machine to play with this tech three or four months ago.

[21:56.000 --> 22:00.000] And then you look at what Apple's ecosystem is going to enable in all of this, I think

[22:00.000 --> 22:02.000] it's going to be absolutely magical.

[22:03.000 --> 22:06.000] You'll have your massive, behemoth, foundational models.

[22:06.000 --> 22:11.000] You'll distill them down, and a bunch of those workloads will be running at the edge, or

[22:11.000 --> 22:13.000] some sort of hybrid approach.

[22:13.000 --> 22:16.000] **[Michael]** Who here in the audience is familiar with ControlNet?

[22:16.000 --> 22:19.000] Raise your hand if you're familiar with it.

[22:20.000 --> 22:23.000] Why don't you explain ControlNet for everybody?

[22:23.000 --> 22:27.000] **[Bilawal]** I think this goes back to your earlier question of what is the thing that people have the

[22:27.000 --> 22:28.000] most negative reaction to.

[22:28.000 --> 22:35.000] I think the moment text-to-image generators came out, text, essentially words or tokens

[22:35.000 --> 22:39.000] that you describe, would condition an image.

[22:39.000 --> 22:46.000] That was really, really dissatisfying for somebody who perhaps has learned how to use a Wacom

[22:46.000 --> 22:49.000] tablet, they're proficient in Procreate or Photoshop, whatnot.

[22:49.000 --> 22:55.000] ControlNet is basically a way to take the old-school AI world, like these discriminative

[22:55.000 --> 23:00.000] models, these task-specific models, and use them to condition the image generation process.

[23:00.000 --> 23:05.000] Basically, it's two models working in concert where you could, let's say, just draw the

[23:05.000 --> 23:10.000] contours of line work, and then have this model fill in the colors, or you could give

[23:10.000 --> 23:16.000] it an existing image, extract edge maps or a depth map, which approximates the geometric

[23:16.000 --> 23:22.000] structure in a scene, and use that to create, to condition the image generation process.

[23:22.000 --> 23:29.000] It's a way of artists exerting control in this higher bandwidth fashion than just frickin'

[23:29.000 --> 23:34.000] text, because text is such a fickle way to describe the intricate composition of a scene.

[23:34.000 --> 23:40.000] That's what ControlNet is, and it's become fundamental to a lot of the AI video work, too.

[23:41.000 --> 23:52.000] **[Michael]** Yeah, and it really gives the artist so much more control over how the image looks, because

[23:52.000 --> 24:00.000] they're able to take some of their prior art, feed it through, and get stable diffusion

[24:00.000 --> 24:08.000] or whatever AI content system they're using, to look at that as prior art and combine it

[24:08.000 --> 24:14.000] and chain it together with the model that they're using.

[24:14.000 --> 24:17.000] It's become one of the fundamental tools.

[24:17.000 --> 24:25.000] What I found really exciting about it was to see the open source movement over the past

[24:25.000 --> 24:32.000] six months, introduce that in real time, and then see all of the different artists play

[24:32.000 --> 24:38.000] with that, and make really cool stuff, and post their workflows on Reddit, and then see

[24:38.000 --> 24:47.000] the evolution of what we call temporal consistency from frame to frame with AI video, like having

[24:47.000 --> 24:54.000] something that looks like an actual video instead of just a fever dream.

[24:55.000 --> 25:03.000] I taught a course in AI content at School of Visual Arts, and I had my students, basically

[25:03.000 --> 25:10.000] we would look at the latest innovations on Reddit, and try to implement that in class

[25:10.000 --> 25:15.000] and see how we did it, could we do better work than we actually reproduced that.

[25:15.000 --> 25:23.000] I'm curious, that was the process I took with my students, but what process do you all undertake

[25:23.000 --> 25:32.000] with looking at the never-ending stream of innovations in the AI space to bring that

[25:32.000 --> 25:40.000] in-house and decide whether you want to try to implement it or not, is it worth implementing

[25:40.000 --> 25:46.000] in your studio, do you want to experiment with it, because I think a lot of us here

[25:46.000 --> 25:50.000] might feel kind of overwhelmed with the pace of innovation.

[25:53.000 --> 25:54.000] Kayla, you want to start?

[25:54.000 --> 25:55.000] **[Kayla]** Yeah, sure.

[25:55.000 --> 26:03.000] So, our paradigm has always been to expose it to the players and let them identify what

[26:03.000 --> 26:08.000] is the most appealing first and foremost, so that first of all means approachability,

[26:08.000 --> 26:13.000] that means that anybody could be creator, so we're a PC-based game, we're not Web3,

[26:13.000 --> 26:20.000] we want no-code mechanics for the players, and so when people think about creators or

[26:20.000 --> 26:26.000] art, they typically think of visuals, as we discussed, computer vision and how you can

[26:26.000 --> 26:33.000] adjust image models, but with language models, players can actually be storytellers, if you

[26:33.000 --> 26:38.000] create a world that gets auto-generated, you can adjust the themes and the styles, if you

[26:38.000 --> 26:45.000] have NPCs that are there in day zero in that world, and beyond that, when you expose more

[26:45.000 --> 26:51.000] game mechanics and experiences, there's these lesser-known tools and frameworks that

[26:51.000 --> 26:57.000] you can use that actually help the players identify these mechanics that they know, that

[26:57.000 --> 27:03.000] they're all familiar with, as they're lifelong gamers, and that kind of gives them the sense

[27:03.000 --> 27:10.000] of being an architect, more so than just a single asset or piece is actually this entire

[27:10.000 --> 27:16.000] world, there's the bar rules for different player types, there's the achiever, the socializer,

[27:16.000 --> 27:22.000] the explorers, the killers, and if you, and everybody, people are typically some combination

[27:22.000 --> 27:29.000] of multiple, and it's just this evolving living space where if you can give them the tools

[27:29.000 --> 27:34.000] and give them the social environment, then it's frozen to this social community and eventually

[27:34.000 --> 27:39.000] these digital nations eventually, that's I believe what the true metaverse could evolve

[27:39.000 --> 27:45.000] into, but it has to start with the players building it out and getting exposed to the

[27:45.000 --> 27:48.000] systems quickly.

[27:48.000 --> 27:54.000] **[Bilawal]** I could briefly add to that and just say it is very overwhelming, every one to three months

[27:54.000 --> 27:59.000] there's the new thing dujour, and everyone forgets about the last thing, and along with

[27:59.000 --> 28:04.000] it come a bunch of these trends manifested in content and apps and startups too, I would

[28:04.000 --> 28:09.000] say you have to have a playground mode and an architect mode is what I've been calling

[28:09.000 --> 28:15.000] it, where just play with this stuff, don't be like, well, this isn't compute, it's so

[28:15.000 --> 28:19.000] compute-intensive, this would never work in production, you're going to miss the wave

[28:19.000 --> 28:24.000] completely if you don't just play with this stuff, open yourself to serendipity so that

[28:24.000 --> 28:29.000] happy accidents happen, and then you go into a far more structured, regimented mode where

[28:29.000 --> 28:35.000] you apply it, and I think given the compression of development life cycles, that's the only

[28:35.000 --> 28:42.000] way perhaps to stay afloat, we were making the rising all tides, let's all boat analogy,

[28:42.000 --> 28:46.000] so get on a boat first.

[28:46.000 --> 28:52.000] **[Michael]** It definitely works for me as someone, like, skipping ads as someone who works on a small

[28:52.000 --> 29:04.000] team, but I'm curious also, in larger organizations, you were with Amazon, how do you

[29:04.000 --> 29:10.000] implement this type of thing in larger, or with lots of, even 50 developers?

[29:10.000 --> 29:14.000] **[Bilawal]** Where do you get the legal approval to play with this stuff?

[29:14.000 --> 29:22.000] **[Vatsal]** I think the way I look at this is, going back to the fundamentals, if you're making a game,

[29:22.000 --> 29:28.000] the game actually has to be fun, and the story has to be compelling, and so all of these

[29:28.000 --> 29:36.000] innovations and GitHub projects, you have to filter them out by, will this actually be

[29:36.000 --> 29:43.000] a compelling experience for your users and players, and kind of talk about how to think

[29:43.000 --> 29:50.000] about users, at least gamers, so I've always started with that filter of like, who are

[29:50.000 --> 29:55.000] my customers, will they be delighted by it, or is this a distraction, because nothing

[29:55.000 --> 30:03.000] is going to replace people making really compelling core games and building this tech around it,

[30:03.000 --> 30:09.000] so yeah, that's how I sort of tend to filter out all of this noise.

[30:09.000 --> 30:14.000]

[30:43.000 --> 30:44.000] **[Michael]** Make it a fun game first.

[30:44.000 --> 30:46.000] **[Vatsal]** Make it a fun game first, yeah.

[30:46.000 --> 30:55.000] **[Michael]** I'd love to dive into some top content examples that come to mind for the rest of the panel,

[30:55.000 --> 30:59.000] because I think that might be super helpful for the audience, but before we go to that,

[30:59.000 --> 31:01.000] how are we on time?

[31:01.000 --> 31:02.000] We've got five minutes.

[31:02.000 --> 31:09.000] **[Michael]** All right, so real quick, ten second content example definition from the panel, and then

[31:09.000 --> 31:14.000] maybe we have time for like, one or two questions from the audience.

[31:14.000 --> 31:21.000] Does anybody have a great content example they want to share with the audience?

[31:21.000 --> 31:23.000] **[Amber]** Like AI generated content?

[31:23.000 --> 31:28.000] **[Michael]** Yeah, like just implementations of the things that you spoke about.

[31:28.000 --> 31:32.000] **[Amber]** One of my favorite things right now is just using like a web browser that's like, sort

[31:32.000 --> 31:35.000] of instead of Google, so I'm just getting, I mean that's one of my favorite product experiences

[31:35.000 --> 31:36.000] in AI right now.

[31:36.000 --> 31:40.000] But I haven't seen a lot in gaming that's really, what I'm looking for is for people

[31:40.000 --> 31:45.000] to be making things that feel really new, and I feel like that's, right now I see a

[31:45.000 --> 31:50.000] lot about efficiency, and I think it's when people are able to start connecting the kind

[31:50.000 --> 31:55.000] of information that like, one game developer couldn't have possessed, and they can create

[31:55.000 --> 31:57.000] a new kind of gaming experience.

[31:57.000 --> 31:59.000] **[Audience]** What browser was that?

[31:59.000 --> 32:04.000] **[Amber]** You, it's not the top, I guess it's just like my, instead of Google, it's my search engine

[32:04.000 --> 32:07.000] in my browser, my Chrome browser.

[32:07.000 --> 32:12.000] **[Kayla]** One of the first features we released was world generation, from like prompts to world

[32:12.000 --> 32:18.000] generation within ten seconds, and that uses a series of breaking down the worlds into

[32:18.000 --> 32:24.000] constituent components, hierarchy systems, relevancy scores, and defining the world

[32:24.000 --> 32:29.000] by those attributes, and because we wanted to build that first, that creates the layers

[32:29.000 --> 32:35.000] and the metadata that can inform exactly how that world gets built, and that's, you know,

[32:35.000 --> 32:39.000] that is somebody creating something that is in a 3D environment in a very quick amount

[32:39.000 --> 32:43.000] of time, and then just building off of that, like with the NPCs based on those world types

[32:43.000 --> 32:49.000] and such, so I thought, you know, that's at least our dream and vision in how that,

[32:49.000 --> 32:53.000] how content can be exposed to the players in new issues.

[32:53.000 --> 32:57.000] **[Michael]** So that's a good, that's a good practical approach to take home, start with the world,

[32:57.000 --> 32:59.000] and go from there.

[32:59.000 --> 33:01.000] Bilawal anything?

[33:01.000 --> 33:05.000] **[Bilawal]** Yeah, I would say one example that sort of weaves together a bunch of stuff that's interactive,

[33:05.000 --> 33:10.000] a little biased because I worked on this, which is Immersive View inside of Google Maps.

[33:10.000 --> 33:15.000] It's the next generation of Google Maps, and basically it's like a triple A game engine,

[33:15.000 --> 33:18.000] streaming pixels from the cloud to any device.

[33:18.000 --> 33:22.000] It's the full, like, you know, uncapped 3D model of the world.

[33:22.000 --> 33:26.000] It has NeRFs, so if you want to take a look inside, you can see the volumetric representation,

[33:26.000 --> 33:29.000] and then layered on top of it is all the simulation data.

[33:29.000 --> 33:32.000] What is the weather like right now, and what will it be in the future?

[33:32.000 --> 33:36.000] And it's just like such a utilitarian scaffolding, but if you take that,

[33:36.000 --> 33:41.000] Imagine the games that can be told on top of that, especially if you have the AR mirror to it.

[33:41.000 --> 33:44.000] Like, I think what people don't realize, like generative AI, again,

[33:44.000 --> 33:47.000] is going to be so crucial to populating these canvases,

[33:47.000 --> 33:53.000] but there's so many other primitives that are sitting around from, like, VPS to, like, 3D tiles

[33:53.000 --> 33:57.000] that people aren't, appling or leveraging to the full ability

[33:57.000 --> 34:01.000] that are there as, like, foundations to then put generative on top of.

[34:01.000 --> 34:05.000] **[Michael]** Absolutely. Any questions from the audience?

[34:05.000 --> 34:07.000] Yes, in the back.

[34:07.000 --> 34:11.000] **[Vish]** Hi. My name is Vish, first of all. Great job.

[34:11.000 --> 34:13.000] All of you are always smiling at me.

[34:14.000 --> 34:19.000] I am the legal that Bilawal was just joking about.

[34:19.000 --> 34:26.000] My job, as an attorney, is to help figure out, you know, gaming companies,

[34:26.000 --> 34:33.000] to navigate these amazing and exciting developments based on a legal landscape

[34:33.000 --> 34:40.000] that was set years ago and really hasn't been streaming related content yet.

[34:41.000 --> 34:46.000] A big issue that always comes up, and especially with clients

[34:46.000 --> 34:50.000] who are very, very excited to exploit this and obviously want to get into it,

[34:50.000 --> 34:54.000] like you said, want to get on the boat, see where all of these things go,

[34:54.000 --> 35:01.000] is ownership rights and how ownership gets allocated

[35:01.000 --> 35:04.000] and the rights associated with that. I'm sure you see that in licensing.

[35:04.000 --> 35:07.000] I'm sure that's why a lot of the authors that you were seeing

[35:07.000 --> 35:11.000] about Emily Hart's painting and all of that are having an issue.

[35:11.000 --> 35:18.000] How do you guys see ownership and where, how much exclusive rights you have

[35:18.000 --> 35:22.000] to the content that you will be able to generate from all of this?

[35:22.000 --> 35:27.000] How do you really feel that is affecting your current state of learning

[35:27.000 --> 35:29.000] as an attorney or whatever you want to go?

[35:32.000 --> 35:36.000] **[Amber]** I think we're just preparing for ending up in court one day.

[35:37.000 --> 35:43.000] I don't know. I mean, licensing just isn't going to be protected by this, right?

[35:43.000 --> 35:46.000] So you just have to prepare for somebody to be the one to go to court.

[35:46.000 --> 35:49.000] And also, do the players own the content that's generated?

[35:49.000 --> 35:54.000] Like if we augment a Batman storyline and curl up with a new thing, that's wonderful.

[35:54.000 --> 35:57.000] But does the player own it because they prompted for it?

[35:57.000 --> 35:59.000] Do we own it because our system generated it?

[35:59.000 --> 36:02.000] Or does Warner Brothers own it because they own Batman?

[36:02.000 --> 36:06.000] I just don't think we're going to know that until we go to court.

[36:11.000 --> 36:12.000] **[Michael]** That's dark.

[36:15.000 --> 36:18.000] **[Amber]** The chest of his legal fees is waiting for these.

[36:18.000 --> 36:22.000] We've been interviewing lawyers and like, whose kids are we sending to college?

[36:25.000 --> 36:31.000] The real truth is what you said too, because the law really hasn't been set right.

[36:31.000 --> 36:35.000] We really have to go to court to establish certain things.

[36:35.000 --> 36:42.000] And again, I think the next, earlier, I think in the next 10, 15 years, there will be a push.

[36:42.000 --> 36:50.000] Again, you know, that's based on historical data, what Disney has done to, you know, keep extending its copyright for the living of stuff.

[36:50.000 --> 36:56.000] If enough people, enough entertainment and media companies, especially in the human world,

[36:56.000 --> 37:00.000] you know, contributing hundreds of billions of dollars to the legal economy,

[37:00.000 --> 37:08.000] when there is more of a lobby push, based on the fact that this is where things are at, you know, what happened with YouTube,

[37:08.000 --> 37:14.000] I think there is going to be a lot more of legislative change if that's going to happen.

[37:14.000 --> 37:19.000] So I think sooner rather than later, we will also see some sort of progress.

[37:21.000 --> 37:27.000] **[Michael]** Thank you. We're unfortunately out of time for this panel, but connect with us on LinkedIn.

[37:27.000 --> 37:41.000] Look for us on Twitter. We'll share the notes.

[37:41.000 --> 37:47.000] I hope you got some good details to take back for your go-to-market in this space.