Inworld AI launches consumer AI runtime after 4 years in stealth, backed by Nvidia, Xbox, Niantic, and Disney

Yes. Yes. Yes. C. Enough caffeine will take you to the promised land of delusions. Am I one of the best newsletter writers on China? Absolutely. Probably the goat. Goat. Absolute goat. Goat. So, thank you for tuning in, Bill. big fan of your work. Without further ado, our next guest, Kylin from InWorld. How you doing?

You look fantastic. I was just watching your video and you look exactly the same. How are you doing? Awesome. Thanks, John. I also laughing because the the caffeine comments just coming before. I mean, we we just had our launch night, so you can imagine I'm heavily caffeinated now. What are you running?

Are you Red Bull, Celsius, Diet Coke, Matina from Andrew Huberman? What are you? All the above. We love to see it. Anyway, kick us off. We're running late today. Uh we kept you waiting. We're keeping the next person waiting. Kick us off with an introduction.

Explain what the company does and whether or not we should ring the gong for you. All right. Uh so yeah, we were founded four years ago now. Um we're basically solving a technical problems in the way of consumer AI adoption. So our team came from Google and DeepMind worked on LLM there.

Basically got very tired of kind of everything flowing into enterprise applications, professional facing applications as we see.

So we basically set off to solve all the technical problems to see how we can actually drive consumer AI adoption which is of course a huge business problem but also I don't know if you know making sure the benefits of AI reach everyone.

Uh we raised $120 million so far and today we uh hey congratulations love it and uh yeah so today we had our biggest launch to date. So we've, you know, it took us four years to get here working with groups like Nvidia, Xbox, Niantic, um, Disney, and now we have the first AI runtime for power consumer applications.

So that's fun. Okay, let's make this super concrete. To the degree that you can talk about it, Xbox, uh, consumer AI, what does that actually mean? How is generative AI instantiate or LLM's instantiating itself in like the Xbox world? What's even the goal there?

So he started off largely working on things like basically talking NPCs. So, we don't basically Hey, let's give it up for NPCs. Let's give it up for They don't They're about to go on a run. They get a ton of hate. The NPCs, they're about They get a ton of hate. They're about to look and feel like real player characters.

Oh, you're an NPC. You're an NPC. Not for long. They're going to get better because of you. Explain it. So, yeah, B, we started out because conversational AI LMS are great at that. Um, you know, games are pretty boring. Anybody who's played a game, um, you know, has recognized that.

So we started out there and then basically we realized that you know we don't just want these characters and basically agentic experiences in games. You can think about every consumer application. You know your language learning apps, your fitness apps, you know they all suck.

I love actually a coach that actually did something effective. Yep. And um and so what we found over the last few years is we worked a lot on the kind of games applications. So these bringing characters to life, you know, the the types of experiences there.

Um and then now we've started working with a lot broader categories.

So, so on on on Xbox, it feels like uh you you could be almost like an API vendor within the Xbox ecosystem that a game developer could harness and run that on the device as opposed to going to Ubisoft and EA and Activision and saying, "Hey, for the next release of Call of Duty or Battlefield 6, uh pay us to train your LLM.

" You want to be able to run it on the Xbox hardware. So, where in the stack is it more like you want to fine-tune it so that it's uh on Xbox's terms and conditions versus you just want to optimize it to actually run on the Xbox hardware? Like where where are the key uh key trees to chop down? Yeah.

So, think about any of the applications. So, an Xbox for example, you're going to have a game. It's going to be a build with Unreal. In a mobile scenario, you might have it built with node. You have that at the application layer.

And a lot of the infrastructure we've built to date has basically been optimized for that type of experience. But now, we're introducing AI. So now you're having a bunch of LLMs or different model calls that are happening. And so think about that as kind of just a second infrastructure layer that needs to exist.

So you've got your core application, you've got your AI, and then you've got all your hardware in the back end.

So we basically sit in that middle layer of not just powering kind of actual, you know, the u the user interface and the specific, you know, uh gameplay elements or app elements, but actually driving the actual generative part of it.

So it could be characters responding, could be mission generation, um all of those different aspects as well as actually generating on the-fly content. So yeah, we've got That's fascinating. Yeah.

So, so it it while you're building the game, even if it's a single player game, you could be in the loop designing or or or generating all the dialogue, but then in theory, it could also make an internet call if you're connected to the web and and and get upto-date uh uh text.

How are are game developers getting comfortable with the unpredictability of AI? I mean, we've everybody's It can be a feature like hallucinations can be awesome because it takes you in this If you have a game for like Roblox for example, their average user is probably 12 years old, right?

They don't want, you know, some some LLM going off the rails and and naming itself something that maybe also just imagine you talk to NBC's like uh the the goal on this mission is to slay the dragon. You go slay the dragon, you come back and it hallucinates and say like, "No, I I wanted you to save the dragonlay.

I didn't want you to slay the dragon. " Also, my name is Mecca. So, yeah, talk about talking. Oh, this this actually happened though. So we we did a lot of really tests around this and it was pretty hilarious. I mean the characters literally make up anything. Yeah, of course.

But then there's been a lot of applications that took advantage of this. So one of our bigger clients, they're called Status. It's a crazy game huge with like Gen Alpha Gen Z. They basically created a game where you could roleplay as a character in another universe is Twitter. So imagine you're like Harry Potter.

I can role play as Harry Potter in Harry Potter universe is Twitter. And then you have like Ronald Weasley, Draco Malfoy. people don't really exist, but then the AI can actually take advantage of the fact that it's hallucinating and making things up to actually come back with that.

And where we see that is like with the interesting with consumer apps is they've kind of gotten to this point where if you for example have to design manually content and it takes you 30 days to design that content and then your users consume that in 20 minutes. You have to do a lot of work to create months of content.

So AI kind of smooths out as well that content creation curve so that you always have this kind of implant loop which is you know key for things like retention. It's awesome. Um what's uh what's next for the business?

Is it just like uh expansion within your current I feel like the pool of value that you can create in any of the companies that you listed is pretty significant if you just keep delivering better and better products, better value, ramping those up versus going broader. Is there going to be like an SDK at some point?

Are you going to go general availability and some some kid who's building an iPhone app will be able to vend this in? Yeah. So, uh overall today we're actually launching that next that next phase. There you go. Good timing. Exactly what I described.

And uh yeah, so what we realized was that as we were working with those gaming and media partners, it wasn't just them but broader consumer applications.

Basically anybody who is dealing with multi-million user scale that has to be you know consumer cost, consumer latency, consumer quality which is more about entertainment than the factuality people are used to chatbt which we also heard about a lot about last week. Um and so people actually want to be engaged by it.

Um so for us it's been kind of expanding to the broader consumer space across like outside of just games and media. And then the other part is releasing the runtime that we're releasing today. Um, and that is kind of the big push that we've been making for the last 4 years and basically allows things to autoscale.

We had a developer today who called it vibe scaling. So, you know, people can vibe code a lot of applications, but then it takes them freaking 6 months to be able to actually productionize it. And so, everybody comes to me, I've heard like a bunch of, you know, sea level executives be like, I coded an app in four hours.

Why does it take my team six months to productionize it? We're basically like automating a lot of that scale as well. Exactly. Um, and then automating also the ML operations. So most teams don't have an infrastructure team.

So we're basically taking over a lot of that, automating it, and also allowing people to do experiments so they can just launch tons of experiments and find what works. So that's basically it is moving to broader consumer, moving deeper in the stack of the infrastructure layer with the runtime.

Um, and we think it solves a lot of the problems that we're seeing. Well, congratulations. Thank you for hopping on the stream. We will talk to you later. Have a great rest of your day.