Tech Brew Ride Home - (TWTR SPC) The TWTR Whistleblower And The REAL AI Revolution

Episode Date: August 27, 2022

First up, we have @KevinTDugan to give us deeper insight into the whole Twitter whistleblower situation. And then, @KylanGibbs of @inworld_ai convinces me we are maybe closer to the Metaverse than I ever thought possible. TLDR: AI might really be at a tipping point. Learn more about your ad choices. Visit megaphone.fm/adchoices

On April 4th, 2023, around 2 in the morning, a man was found stabbed multiple times on a sidewalk in downtown San Francisco. Hey, who did this to you? What happened next turned the story into a political firestorm. Reports have identified the victim as Bob Lee, the founder of Cash App. From Bloomberg Podcasts, this is Foundering, the Killing of Bob Lee, beginning April 16. Welcome everybody to the TechMeme Ride Home Experience for Thursday, August 25th, 2022. Two days ago, it was the hashtag's 15th anniversary. So I got to do, I don't know, I don't want to call it a victory allowed because that sounds self-congratulatory.

But it was an interesting moment to stop and reflect on, I suppose, my contribution to social media and to this hellscape of a platform, which also is wonderful and lovely and connects us to amazing people. It was also, I guess, an interesting moment just to think about how Twitter has both evolved, devolved, stayed the same, and continues to remain this somewhat, I don't know, so-bop-orific platform where people come together and they do stuff. One of the things that I actually want to, I guess, call out now, since we are, of course, recording on Twitter spaces is what was a new announcement as of today, it's been previewed for some time. But that is the move for Twitter to, I guess, embrace podcasting, which is essentially, you know, what we're doing here. We are using the live platform, of course, to share kind of in real time to bring guests on and to facilitate a different type of, like, an, I don't know, like studio experience. But Twitter's move into podcasts, I think is just so interesting, again, sort of thinking back, you know, 15, 16 years to where it all got started, which, of course, was as audio. and audio was one of the first social podcasting platforms. So it's like we've come full circle.

And I don't know exactly where it's going to go, but I thought that was sort of an interesting moment to sort of just consider in the annals of social media history. You know, we discussed you bringing your OG crew on to talk about the anniversary. We can still do that if you want to do it. You know, so many of them have, I think, gone to the wind. in a sense, or just sort of like dispersed. However, I was actually thinking about this. There's sort of on the one hand, the compression of time element

where the things that took place over the course of, you know, Twitter's history kind of happened actually over a longer and larger sequence than a lot of people think. So, for example, it took a very long time for Twitter to get search. And I think a lot of people kind of misremember search as being part of, you know, Twitter from the beginning because most platforms, of course, I've searched, but that wasn't true here. And Twitter actually acquired a company called Summise, which became Twitter's search, and was one of the ways in which the hashtag kind of was church and hoarse into the Twitter platform itself.

So it'd be super interesting, I think, to get folks like that on who had built, you know, kind of apps around the ecosystem onto the platform. And, you know, maybe what we'll do, because I was having a conversation today with Amir Shabbat who runs the Twitter developer platform now and who I worked with the Google and also headed up Slack's platform and just, you know, around kind of a super developer advocate kind of person. The Twitter developer conference is coming back this fall. And so, you know, Amir is interested in coming on the show. That might be an interesting moment to kind of do a little bit of retrospective as well as like talk about where the platform wants to go. They previewed something today. What do they call it?

Called it something specific. That was kind of interesting. Compiled tweets or something. Let me see. they're called tweet tiles. And this is something that he just shared today, tweeted about. I guess it's in the vein of allowing you to add multiple attachment types to a tweet.

You know, so going beyond just like a single animated, you know, GIF or GIF, for example, and having a full rich multimedia experience, which from a developer ecosystem is interesting, especially if you start to think about Twitter more as a protocol that should be able to move. type of medium or media to any recipient and to contexts that relate to groups or

communities or DMs and so on. Let's let's let's do that. But let's see if we can do that after I'm intending to shit all over Twitter right now in this episode. Okay. Um, yes. Okay, I got two more things on the podcast landscape and then we'll switch over. Sure.

One is that I saw another test of, um, of Modifies native podcasting ads, like in the product itself. And it was actually for the podcast episode that you sent me with Derek Thompson and Chuck P. Closterman. Closterman. Closserman. Yeah, yeah. Thank you, Closerman. Yes. Which, by the way, it was an amazing, brilliant episode, and I'd love to dive in on that at some other point. But what happens now is that as you're essentially listening to a podcast episode, the advertising artwork will appear, as the album art and then you'll have a button that can take you to a website. Mostly, I think they all go to websites. It'll say like Shop Now. And you'll be listening to the ad and the entire thing

kind of changes. And for the episode itself, there's now a listing of advertisers that you heard during the episode. So it's starting to really get there in terms of what the podcast, kind of advertising experience is going to be for podcasts. And in the context of Twitter moving towards podcasts and YouTube just launched a new landing page for podcasts at YouTube.com slash podcasts. I just, like, I feel like there's a maturing of the space and a competitive energy in the space that's just really significant and something that we should just pay attention to. Agreed. I'm always told that I need to do more in terms of putting stuff on YouTube, but that's hard for me to do because I put my head in a box every day. But we do post the

YouTube every day. So, okay. Well, You literally do post here or put your head in a box. Okay, anyways, enough of the preamble and the things I wanted to get off at the top of the show. This week has been another big week for Twitter and for Twitter news. And we have none of than Kevin Dugan from New York Mag to come on and talk about some of his reporting that he's been doing about this topic. Brian, do you want to tee this up? Yeah, Kevin, should we call it Intelligencer or New York? magazine what's I don't I don't know that no conclature either one I mean it's it's

intelligence or is it's one of the that's what I say when I when I quote from you on the show I say it's okay sure try to make it works okay let me frame it this way when this story broke earlier this week I of course immediately framed it in the Elon sort of Twitter trial sort of thing. And over the course of the week, I realized that it's bigger than that. And the way I want to frame it is,

all of the things that I left behind are coming back, which is if you've been listening to the show long enough, you know that Twitter was kind of a joke. We used to joke about, oh, Twitter is going to innovate on product. They never did that. Twitter is known in especially Wall Street circles as being like the most dysfunctional company,

you know, especially among social media. They can't make money off it. Other people can make money off the social media, but Twitter never figured out a way to. A clown car, I think, is what Mark Zuckerberg called them at some point. And so all the sudden, I want to leave the Elon stuff till later, but all the sudden what this whistleblower complaint has brought back to the fore is, oh, Jesus, they were a clown car, allegedly. So I'm going to say that everything we're going to say, all people on the show,

everything we say about this is allegedly. But it sounds like that allegedly the clown car shit show was ongoing, and that's what this whistleblower has sort of brought to the fore. or even beyond anything about Elon, right? Oh, yeah. I mean, this complaint shows so much dysfunction up and down. It seems like, you know, I mean, you're right that Wall Street has been pretty correctly diagnosing that. Twitter doesn't know how to make money.

And more of those details are coming out, you know, not only through the municipal complaint, but through the Musk lawsuit. But the things that are, I mean, it's some pretty incredible allocations in there, especially around national security, the fact that Peter Zaco, Mudge, is saying that they employed spies from the Indian government to, you know, be a part of it that, you know, people essentially didn't care. They didn't care about security. They didn't care about spam. Although he did kind of both of the he did play like both sides on that. It really does show not only that things were extremely dysfunctional, but he also puts the blame on Paragra Grawal, right? The current CEO. Yeah, he says that Jack Dorsey was checked out, but he really is gunning for Parag.

And I think that is a really interesting thing. It hasn't really been covered too much. It hasn't been discussed too much how centrally focused he is on the current CEO. Yeah, that was interesting to me as well. I remember thinking when Parag became the CEO that that was sort of a surprise to those of us watching Twitter. And then essentially what he's saying, I'm looking for your quote in here, was like he's laying most of the company's problems as being in his wheelhouse because before he became the CEO, he was the CTO.

And I don't know, it's sort of painting him as like sort of this striver, duplicitous sort of executive, who apparently a lot of the board wasn't even in favor of him taking the top job? Yeah, he mentions that some board members were not in favor. He doesn't really go into that. And since Mudge was not at the board level, I do wonder how he actually got to know that, although he does have access to some pretty high-level people. So, yeah, there are a number of allegations in the suit, which, you know, and these allegations were, this is a roadmap, right?

This is a roadmap for federal investigators to go in and dig some more. And these, a lot of these allegations are saying that, you know, he wanted, And Perag wanted Mudge to hide information that he was one of the forces behind keeping the board in the dark about some of the security problems.

I don't know how true this is. Some people have called into question how much it could really be true. And it is a difficult thing to prove, I think. But he's trying to, you know, I mean, he's, he's a let it fraud. One of the things that this all kind of, I think, hinges on.

First is the FTC consent decree. Look, I mean, almost every company that does social in Silicon Valley is under some consent decree. Google was under a consent decree. Facebook was, you know, essentially there are, as these companies were figuring out how to build products that people would use in the social space, Of course, none of these concerns around, well, I don't want to say none of these concerns around privacy per se, but the ways in which these companies were built was relatively fast and loose because they needed to be. One of the things that I kind of was struck by in reading through the complaint and trying to think through that era and now where things are, you know, 10 or 11 years later, you know, much comes in late to the game in some ways, still in the Jack Dorsey era. And this was actually sort of an interesting thing that I was trying to piece apart. Because, you know, as you suggest, most of the criticism goes to the now CEO, Prague, Agarwal.

However, Mudge was brought in under the Dorsey era. And even the complaint says that Dorsey was checked out, you know, for a good period of time. Now, what I'm imagining, and again, this is my speculative brain. And so I'm not saying that these things actually occurred. But what I can imagine is that there was time when Jack Dorsey, you know, as someone who came up or at least was familiar with the hackerspace, certainly is big in the Bitcoin and crypto land. You know, it spends a lot of time, but like, you know, developers, hackers.

Like, you know, Twitter came out of IRC. So there is this legacy of being within the hacker space and not the business space. And so much comes in later on, and I'm sure they have somewhat of a respectful, you know, like relationship. Certainly an awareness. No, in fact. In fact, Kevin, to step on your, I'm going to quote,

from you. Mudge said that he spoke with Dorsey about six times over the course of a year with each 30 minute call at his own request and Dorsey speaking perhaps 50 words during those calls. Yeah, I understand that, but what I'm saying is before this period. So I do believe there was a time when Jack Dorsey got

checked out and Jack Dorsey was probably burnt out was probably... But he still he still he still he still lionizes and respects these hacker types or whatever.

But the point is, and this is sort of how I was framing it at the beginning, it was, it's the original sin of Twitter, which is no one actually, no one actually passionately believes in it as a product or a service. It's some sort of accident that people sought over and knifed each other over, but no one actually cared enough to actually make it good. If you're willing to believe what Mudge is saying here, it's that he actually gave a show. shit. And he was going into a culture where no one was giving a shit, and that's why he's doing this thing, which we can talk about Mudge in a second. But Kevin, does that ring true to you

the idea that one of Mudge's grudges is that Jack and other people at the top were not paying attention to the things that he cared about that? He was screaming from the rooftops. guys aren't doing enough. Yeah. And like, let's take a step back here and take and think about like what Jack Dorsey has said about his own company, right? I mean, he said that, you know, he wishes it wasn't a company, that he wishes that it was more of a protocol or it's more of some sort of idea.

He literally said that today on Twitter. Yeah, he said that. Yeah, right, right. But, but as much as he wants that to happen, it's still a $44 billion company, right? He can wish it all he wants. And I think when Mudge was brought in, he was hired to look at it as an operating company, right? Not as a protocol, that has a dream. And so you have those two really clashing ways of looking at Twitter.

Yeah, but one of the reasons why I think the attention existed was in some ways, like the CTO, who should have been responsible, for security, right? And you got to also put this in context. Like the January 6th, you know, attack had occurred. You know, Donald Trump had been kicked off the platform. I mean, Jack Dorsey was not living his best life, like having the most fun running this company.

It was like shit show after shit show. And it wasn't about, you know, these ideals that kind of started the platform as this open and free place for expression and dissidents and, you know, journalists to be able to get the word out. Now suddenly people, I don't mean, people like governments. and big, powerful institutions were starting to really care about Twitter.

And so I imagine that his day-to-day was starting to become a real shit show and that these underlying security issues, which I think to Brian's point, I think is a good one, which is like people within the company either weren't empowered, didn't care, didn't know how, or weren't engaged with actually securing the system

because they were more concerned about growth, which frankly, I mean, it is a concern. And I think this is one of the tensions, at least for this show and for the audience of the show, is like a hard thing to work through and to consider, right? You could have, there's never, there's no such thing as absolute security, but you can, of course, prioritize security to a great degree.

And you can require everything be logged and for users to be very, not users, developers, engineers, for people have access to sensitive systems to go through a series of checks and to have a much more rigorous security footing in terms of the way that they get their work done. That also slows things down. where it can slow things down.

So when it comes to Twitter needing to, well, and I think this also, Brian goes to your point, about Twitter finally getting into a place where it's building and shipping and launching things, it's a tension and its set of tradeoffs for risk mitigation. There's a risk of annihilation on one hand where you don't continue to grow the business and the stock craters and the thing's not worth anything.

And on the other hand, you completely secure the platform, but that it's so onerous that no one uses it. Right. So these things are our tension, and I got to imagine that when Dorsey decided to bring much in, he was looking at like all the things that were starting to become risks and threats to the business was like, you know, now Elliott management is like breathing down my neck. Let's just get Mudge in here to actually like, you know, clean house and sort this shit out. Now what I don't understand is how it's possible for Agarwal who oversaw all these issues to be then appointed to CEO. Why would that happen? I mean, I don't know. My guess is that he was the least controversial choice. He knew the system. Yes. How they. worked and you know I mean he did anybody really think that he was going to be a long-time CEO he it right you know I mean but looking from Wall Street's perspective and I've written about this before the company was not doing well

a share price it had been taken over or it would try to take it over a trillion times before and you have a you know a relatively uncharismatic leader it's after Elliot just came in, you know, and, you know, tried to work their magic to make it more valuable. So it looked like a relatively good takeover target. So if I was the board, you know, and Elon Musk comes along, I mean, they had to have been thinking that, like, this was going to happen eventually. Yeah, we should probably get to the Elon stuff in a second. But, Chris, let me point this out to you. and I know that these people are our friends and they've been on the show or whatever,

but look at all the people that are no longer there that got fired, like Kavon and also Mudge. So the whole era that I'm talking about where you remember me doing those shows. Oh my God, Twitter is shipping a product. And like I had to eat. Yeah, I know. I had to eat crow about that. And that was like a running joke on the show.

those people as soon as Musk's bid came in that's how they cleaned ship they were all gone so if you none of us know but looking at it from that

lens and Agriwal was not someone that any of us expected to take the top job I mean he had no profile before so what if the play here was look this is a shit show someone wants to come in and take us out it's it's it's it's a mercy killing right and let's get everybody out of here that's gonna that's gonna that's gonna complain about that

people that work there i mean it's it's hard to imagine it being like a mercy killing like at that level chris there's a lot of money involved and i mean you know know. I mean, that's what I'm looking at. And that's why like the mind, you know, races and it moves to speculation, right? Like, you can imagine like this end-dimensional chess where like, you know, Alon and Doris. This is like the fanfic of, you know, social media kind of catastrophes where Alon and Jack Dorsey and Mudge are somehow conspiring to like, you know, basically sink and tank Twitter because Jack Dorsey is angry at Elliott management and wants to get back at like, but I'm not saying they're going to sink it. I think the plan was

is like, this is a really good deal. We're all going to walk away with nice bonuses. The shareholders will be happy, like, let's just get this out the door, which is where we should probably... Are you talking about the Musk acquisition or which part are you talking about? Yeah. Yeah. Or anybody that would take it. Oh, for sure. Yeah.

That's what I'm saying. Yeah. Yeah. Yeah. Let's like Uncle Elon take it. And then we're going to get off. Right. Exactly.

So, okay. Kevin, let's talk about the Musk angle about this. on the one hand does it it sort of plays into to Elon and his lawyer's argument which is that this is a shit show but then at the same time

it's almost like what the whistleblower is saying kind of confirms that the monetizable daily active users which is all that really matters is kind of on the up and up and credible So where do you come down on this? Does this whistleblower, obviously they're going to depose him?

Obviously, they're going to use this. I don't think they're not going to use this. But does it make it more likely that Elon's case is stronger now or not, in your opinion? Yeah, I mean, they're already using this. You know, there was a hearing yesterday and some filings made that already referenced the whistleer suit. But, so a couple of things. On the monetizable daily active users, the MDOW, the suit is framed like, oh, this is like, it's the big thing. It's going to, you know, expose the problems around MDOW.

But there's, much is essentially saying that the numbers are right. And he's saying that the incentives are right for people to keep spam out of the MDOW numbers. Now again, the way that this works is there's a cross section of users that Twitter looks at, and they remove, you know, spam or bots from those cross sections. And that is essentially how they get their number, that there's fewer than 5% of spam in, you know, the number of people, users that advertisers can make money off. So much is actually saying that that is basically right.

And that is the hinge of Elon's argument, right? Elon, of course, could have dove into this before he tried to take over the company. And so even, I'm skeptical that even if the number is higher than that, and, and you know, I can prove it unless it is significantly, significantly higher, like over 50% or something like that, then he really had the case. Because that would be a material adverse effect. And that could blow up the whole thing. Are we able to speculate or just think about what the reaction from the chance

record was today and what that leads to? I mean, because I thought that the reaction was like pretty measured and it might give some indication as to like what they're willing to allow in terms of, you know, the discovery process and the arguments on both sides. And they like, I mean, you know, we're a couple weeks away from like this, this, you know, case happening. It seems like they might want to just get it done quickly. They do.

And so what Musk's side wanted was basically every piece of data that Twitter had over the last two and a half years. And Twitter said, no, that's crazy. And the judge in the case

said that it was absurd, the amount of information that they were asking for. And what they settled on was the 9,000 or so user data profiles that Twitter actually relies on for the last quarter for the end up and all the data around those 9,000 versions.

So fine, you know, like that is a reasonable thing. And they had said that they would be able to put that together in, you know, 10 days or something like that. So, yeah, I mean, so far the judge is. trying to keep a lid on all of the data requests from the Musk's side, which go in all different kinds of directions. And it's, you know, sometimes a little hard to follow how it would actually impact what they're trying to argue. you know it it does seem like it's it's a lot of you know they're they're trying to get to penis for I think close to three dozen people they got one they got Kvon but and the rest I think

are still under seal so it's not really clear who they were asking for but they're they're asking for a lot and they're trying to you know if you were on his side you would do that too you would just ask for a lot of information, see what you can get, and then try to build a case out of it. So I think, and you said this in your piece, I think in the end what this really does is raise the odds that Twitter just settles. Because as I've been positing, I think that this was a mercy killing, and they were all hoping that this was going to be quick and easy and whatever. And as you say, I think at this point, it already was Twitter's big incentive to just get this over with, and now even more so. And also, I think, you know, Elon is going to make enough noise that, now the question is what's going to happen to Twitter after this. But Kevin, thanks for coming on and talking about it, your piece, which we'll link to in the show notes and is.

linked right now in the space kind of got into more detail than I was even able to do this week. So I appreciate it and let's keep an eye on this one and see if we're all right about this. Thanks, Kevin. Yeah, definitely. Thanks so much for

having me on. Chris, I think I said this week that Mudge was a bigger deal than I knew. Were you aware of him and and his sort of fame in the, um, so, you know, I, I, I was aware of the, the cult of the, um, sacred cow or whatever. Um, and I, I, I'm sure that I encountered, you know, him and, you know,

Kevin Mittnick and a lot of these like, 2,600 type hackers, um, in the past. I mean, he's been active since like the 90s and was one of the first hacktivists. So I certainly, you know, when I was in the identity spaces, uh, with open ID and Oath and learning all those things. Um, no doubt, you know, I've sort of adjacent aware of him. But for example, I didn't realize that he'd been like recognized by the CIA, the White House, the U.S. Army, the Office of Secretary of Defense. And he got the exceptional public service award, which is the highest medal of honor available to civilian non-career officials. So I mean, dude's legit. Not only that, go down a rabbit hole, Google around for this guy. You'll see him. He, I believe, is 1998. He testified before

Congress and if you can imagine what a hacker look like in 1998, a hacker dude, he looks like that. I mean, he's sort of like, you know, he's become more like clean cut, whereas like Jack Dorsey sort of like full-on, like hacker code. Because there's pictures of him with like Bill Clinton shaking hands and stuff like that. And it's like he looks like a guy from 1998 that's a hacker dude. But this is this is the point I wanted to make to you. You and I have commented before that that shy guy in the corner that you met at a meetup in 2004, 20 years later, is a billionaire. It's so funny to me that the hacker that is wearing nose rings with long hair and tattoos and shaking hands with Bill Clinton in 1998 is the chief security officer of a major social media

platform 20 years later. Well, you know, people can evolve and, you know, I don't know, like, the sense that I got was like Mudge had options. And coming to Twitter was an opportunity to continue sort of his hacktivist streak that, you know, he believed Jack Dorsey to be genuine in his desire to one support Mudge and to, you know, kind of clean house. And I just feel like there was like this institutional inertia to really not embrace the things that, you know, he was trying to say. I don't know. I, you know, I will say that some of the folks who are heading up or leading up parts of the Twitter security org are people that I worked with at Google before.

And I know them to be people of, you know, strong integrity, and they do a lot of work, and they're very respectable in their own right. So I think it's a mixed picture. I think what this complaint, at least to me, at least the biggest questions are really about Agarwal and what his ultimate role, you know, is historically and going forward. And as a CEO, you know, he just, Twitter needs something. you know, of kind of like an enlightenment to get its on, you know, it's, it's course back. And it feels like it's really kind of, you know, dovetailing into or fish tailing into

oblivion. Yeah. Yeah. But it, it. Oblivion is the right word because they did have their renaissance and it got be kept by all this. And God only knows what's going to happen. 2020 is going to be a shit year for Twitter. That's for sure.

So my segue for you, my segue for you was people that you used to work. with people that you know tangently. And so we're going to shift away from Twitter to, I don't know how well you know our next guest. But I believe you worked with him or people. I didn't work with Tyler. Nope. So this is the first time that I met Kylin. However, Ilya, which is his co-founder for InWorld, is someone who I interacted with before.

And, you know, so I co-founded a conversational artificial intelligence company. And during, you know, this is like 2016 era, I was like super excited about conversational interfaces and bots and that whole thing. And Ilya had started API.AI, I believe, which then was, God, was, you know, Cullen's going to know the history in the back story much better than me. But there was, there was involvement with Facebook and with Google and all the rest. But Kylin, hello, welcome to the show. Great to meet you. Thank you. I would love for you to just say hello and then maybe give us a little of that background.

And then we'll get into it. Sure, yeah, I can fill in a little bit. Yeah, nice to meet. I'm Kylin co-founded in World AI. Yeah, so most of our team does come from a Google background. I was previously at DeepMind, and most of our team sort of comes from similar conversational AI backgrounds. So Ilya and Mike, my two co-founders, they previously were part of founding API.

which was basically one of the first sort of conversation AI building platforms was acquired by Google, became Google Dialogue Flow, and is now like the foundation for Google Assistant and a lot of other sort of related products. And so that's maybe sort of how you encountered him and in that space. And now I think it's the largest conversation AI platform on the market. So, you know, our team sort of got started and got their chops on what I would call like that old school style of conversation AI. And now we've moved on to a bit of a new venture. Am I hearing a bit of Canadianism in your voice? You are, yeah.

I'm from Vancouver, Canada. Oh, Vancouver. Okay, I was going to say Montreal because there's a lot of conversational allies up there. Don't turn it to French, please. Okay, fair. Well, before we actually dig into In World, I did want to just, I guess, get a sense from you on where, like, what is the state of play on the conversational AI space? You know, I feel like there's been a number of pivots this year, specifically from Amazon

and Google, of course, some of the big ones. majors in the space, you know, both towards and away from, like, their conversational platforms. It seems to me that, you know, Amazon continues to push pretty hard in terms of bringing developers into the Alexa ecosystem and utilizing the technologies for conversational experiences. But Google, I don't know, if there's any sense from what my little stupid puck that gives me access to the Google assistant in my shower suggests, is that Google is not focused on that platform anymore. And basically I have, it's not just a puck, it's a brick. So what is your sense for just like the world of conversational AI? And where is that at? And then we'll get into what you

guys are working on within world. Yeah, great, great. Great question. So effectively, I think conversationally I really started, you know, even like before Siri with this really old school rules based approach, which was like when I said, hey, like specifically, hey, with no capitals, I got a specific response because I've hard coded in those rules. And then you sort of had like, you know, early machine learning and supervised learning where you could learn, okay, things like hey, tend to be responded with, oh, hey, how are you doing too? Right? And so that sort of was like the next step. And then you saw sort of like Bert from Google and a lot of these things come out, which was really kind of allowing the machine learning to handle the understanding part. And then

you could attach basically these sort of responses in the back end of that. And that was really what Dialogue Clow did. And then sort of in the last few years, you saw that with GPT3 and a lot of generative models coming out, the actual ability to, for the models not only to understand the language, but actually then construct or like a representative response has exploded. And I think there's frankly been, so GBT3 came out in 2019, kind of really changed the market. And what actually happened there was there's, you know, the machine learning community, especially in natural language, was going around lots of different architectures. And then you had something come out called the transformer.

And then basically what open AI figured out is you could just scale up the compute of this thing and get really better results with every bit of more compute that you threw at it. And frankly, it took a while, I think, for a lot of the large companies to sort of of shifting focus to that and really realize the opportunity there. And I think a lot of the reason for that was that it's frankly very risky. So the way that the generative models work is you basically give it a response and based on the statistics of language and the information that it's trained on, which is usually like a corpus of the entire internet, it gives you sort of a reasonable response.

And you can imagine a company like Google or Amazon is, that's very risky for it. So they basically stuck with the old school approach, which was, you know, this idea of like these pre-canned responses to particular set. And sorry, just to unpack when you say risky, you mean sort of reputationally harmful because the internet is full of shit. Yeah. So there's two interesting things here. It's both reputational risky because it's, yes, they could say unsafe things. Also, if Google or Amazon regurgitate something that is someone else's IP, for example, that is also another form of risk, right? So it's like who owned that information that the model was trained on?

And it's something that Open AI kind of got away with because I think they're like a research institution and they were at the time smaller. Yeah. So there was a lot of reasons that they didn't do that. Google, you know, I know a lot of these companies have explored this. And I think there is going to be stuff in this. Like you did see Google come out earlier as you with Lambda, which is an example of this, you know, Facebook has a letter. That's the conscientious or the, what is it called? Yes, the sentient.

The self-scentia. I'm going to put it in air quotes. I'm going to put it in air quotes with my hands. That's right. But what I think is happening now is that whole area has matured. So that was 2019. We're almost like for three years later, I think roughly from when GPT3 came out. And I think now you're seeing a lot of these companies kind of back away from investments

in that old school tech. So you know, why am I going to invest in Google Assistant when really it's a voice interface on top of Google Search? That's really what a system is. I think what you'll start to see over the next few years is because of the massive investments. Let's be clear that are being made in these large language models. you're probably going to see a bit of a pause now in innovations on some of these old school technologies like the Google Assistant Amazon Alexa and then in the next few years sort of an

explosion of new technologies that are much more natural much more generative and ultimately powered by sort of this new wave of generative tech and I think it's just taken a while for the safety and for the IP issues for the ability to for example reference materials like where did you get that information like those are hard technical problems that research teams are working on that But if you look at the academia, it's really kind of exploded. And so many AI researchers are shifting specifically into this area. So I presume in the next few years, you might get another version of your legal system that's a lot smarter. Well, I hope that stupid little puck, which I probably got for free, you know, ends up actually being useful again.

But it seems to anyways, we'll get into that. So one of the trends that you're just describing sounds a little bit like, or at least, what are the challenges or headwinds has been that there hasn't really been transparent AI, or essentially AI that is able to explain itself, as you just alluded to. I understand, and I think I read this about Lambda, perhaps, or about the sentient bot, that increasingly AI is able to explain

how it sort of arrives at an outcome or a conclusion or whatever it happens to be. And I guess what I want to understand is whether or not that allows for, like, the larger companies to embrace these things more effectively, Because it can explain how it comes to it or the researchers like, no, it explains itself. You ask it. Like, why do you think that? Or maybe think is not the right word.

But, you know. Am I wrong about that? I thought that a lot of the problem recently with AI stuff was that it was working, but we kind of didn't know how. Am I wrong about that? No, that's what I'm saying. So what kind of was saying is that now there's this whole sort of thrust from. research and academia to make more trustworthy, transparent AI that, you know, essentially can explain how it arrives at outcomes and answers. And so you, that allows you to either address the

model or, you know, change the weights such that. Now it becomes, if not more predictable, a little bit safer. Callan, tell me where I have that wrong. Yeah, no, no. So generally, this whole region is called like explainable AI. And it really depends on. Yes, that's what I did. Sorry. Yeah. So, so effectively that the field is, you know, how do you take a neural? So, you know, You know, when you have really old school AI, right, which was rules-based, it was easy to explain. Then you went to neural networks and people are like, oh, these things are really cool, but, like, I have no idea how it works. In the same way, like, I can't explain to you right now how I'm generating speech based on how my brain is firing signals. And so there's been a lot of work that's gone into this question of, like, how do I explain that?

So I would say you're both right in the sense that, so neural networks are still very hard to explain. So when you ask like Lambda or, you know, R models or any of these things, right, that, like, oh, how did you come to that answer? It will give you the reasoning. It will give you like human level logical reasoning as if you ask human. But in the same way that like I can't explain how my, you know, my different cortexes and my brain are firing right now. I could explain to you like why I'm telling you this answer, right? You ask me a question. I have a certain experience.

I like have this like, you know, I worked at deep mind. Those things are why I'm giving you this answer. But I can't give you the technical answer as like what computations are happening. And that's effectively what's happening in the models. So they've got a lot further in terms of like, okay, if the if the things, says, you know, elephants are only on two continents. And you say, oh, why is that the case? And you say, oh, they tend to be in India or they tend to be in Asia and they tend to be in Africa.

And then you say, why is that? And you can keep going down this reasoning. And where do you find that information? Oh, I found it on this Wikipedia link, you know, these types of things. It can do that. And that's basically because they've added on sort of separate architectures and research to these core language models that help with that. But, you know, Brian's probably still right in the sense that, you know, the model is not

can be able to say, oh, you know, I assigned like a weight of 0.3 to this code. I think that there are ways that you can visualize that and get it through now, but I think in a natural language context, it's still very hard to do. Well, honestly, if you ask any teenagers, you're like, why do they run that red light, right? They're going to come up with all sorts of answers anyway. So it's not that much worse than the state of the art. Yeah. But you know what? I want to come to this at the end in terms of like where AI is going. But like I'm still fascinated by the idea that it's a black. box to even the people that are doing it. But so, let me ask you this. So if you guys are successful, as I'm sure you're going to be, essentially what we're

talking about is, I mean, a really reductive way to say it is to turn NPCs into no. You're jumping ahead. You're jumping ahead. Okay, okay. Yeah. Yeah. Yeah, we're going to get there.

But I think we had to go through a little bit more of a journey. Okay, okay. Yeah. Which, and, and, and, and, and, and, and, and, and, and, and, you know, actually, I'll step back a little bit, but what I'm kind of getting to, or what I'm sort of wanting, you know, one for, like, the listener to understand is kind of like, where we've been, what the state of the art is with regards to known voice assistance, because that's how a lot of people are interacting with these things. And then I kind of want to dub tell that, you know, sort of just by pointing to previous conversations that we had with folks like Matthew Ball about the Metaverse and about places. and about places where you're spending all the time, you know, in-game with, you know, characters that you might encounter. And then to sort of start to think about how you can blend aspects of these, you know, kind of procedural, you know, worlds, let's say a game, you know, that sort of has like, you know, a beginning, middle, and an end and sort of an outcome and bosses and challenges and things like that with what are kind of non-deterministic exploratory realms that are created through artificial intelligence. You know, we've also had conversations

recently about generative AR art. Whether that's like, you know, Dolly or, for example, today I launched stable diffusion non-productment or mid-jurdy. Yeah. And you have these amazing kind of other examples that are quite visual and you're able to sort of wrap your mind or I guess your eyes around that. You're like, wow, that's amazing. Like, how's that happening?

Now imagine that same kind of like technology that is so, I mean, it's not even that it passes the uncanny valley. It's just that it's like a new form of expression. being applied to conversation and to character story and to background. Exactly. So this, I'm sorry, what was it? It was an E-WR that you used, not explanatory AI. Oh, explainable AI.

Okay. So that's the thing that I wanted to point at, right? Once the artificial intelligence starts to be able to describe its reasoning, then you can start to actually inquire about itself, like what it cares about, what its goals are, what it wants to achieve, which to me gets me into the mindset of Westworld, which I have to man, I'm behind in a season. But, no, wait, Chris, Chris, now you were jumping. I am. I'm not trying to, like, steal the thunder, but I'm just trying to say, like, so let's just, like, put all those things kind of, you know, on, on the mood board, so to speak.

And then, okay, Kylan, like, take us back maybe, like, to, I say, like, you guys started, I think, July 2020 or something around those times. Talk about maybe, like, leaving, like, deep mind and, you know, dialogue flow and kind of being in the belly of some of these big beasts. and then moving off to work on your own stuff. Like, I'm just curious why you guys went in the direction that you went and what you guys were considering at that time. Yeah, awesome.

So, yeah, I was at DeepMind. Before this, I was basically leading product for generative models and conversation AI. So there was basically a point at which, you know, GPT3 came out and there's kind of a question like, oh, what do we do in response to us? And so I was working across Google, basically taking DeepMind Tech and integrating into Google products. So you can imagine across with language like search assistant Google Cloud.

And so there I met Mike, our current CEO, who was leading AI and ML at Google Cloud Conversational AI. And we just, you know, we met and had a few projects working together. I'm around that same time, Ilya had basically, you know, gone, stepped out of the Google sphere and was, you know, kind of just thinking what to do next. I think he was maybe in Hawaii at that time. And basically, so you can kind of think about Mike and I were deep in the space of this. Like I especially, so Mike was still working on dialogue, moments of this stuff. I was trying to, you know, work on these sort of large language models and think

through like basically what is the next step beyond GPT3 and how we productionize that. And Billy was, I think, just kind of looking around and being, oh, people are spending a lot of time in these virtual worlds. So you have games, immersive realities, these like, you know, these virtual influencers online. This is like this bizarre trend that's going towards this. Let's be clear. There's no term metaverse existed as something from Neil Stevenson, but not as like a thing at that time. And so basically what happened was, you know, Ilya, I was like, you know, we had this conversation I applied for before.

We used it on like smartphones and things. But ultimately, people are not, you know, that was the big trend at that time with the transition of smartphones around like, you know, in the early 2010s, like basically after the iPhone in 2008. And so the new trend was toward these immersive worlds, basically not necessarily just saying VR, but, you know, video games, like how many kids spend their time in Fortnite, all these different areas. And so he was like, okay, if people are spending.

time in the world, it's actually more natural to have a conversation with a non-human character in these worlds than it is on like a smartphone. Like how often do you actually use Siri, right? Yeah, totally. Just use an app on your phone. And so it was like it actually became sort of a more natural place. So like if the goal is to sort of, you know, say, how do we have this relationship with AI? It's actually a more natural place to have it in these immersive worlds. But, you know, Ilya frankly, like invented a lot of stuff that had to do with that like sort of previous generation of these conversations AI. And then tap Mike and was like, hey, you know, I think is this possible?

And then Mike tapped me and was like, you know, we need to build this. And so at that point, we basically, that was actually only only just over a year ago that we technically started. So it's moved pretty quickly. And so a lot of our team came on. They had previously been at Google and helped build a previous company. You know, we've got other folks that have joined us since. And that's really sort of how it's all come together. And it's been, it's also been sort of pulling a lot of other people from, I think, related industries.

So John Gata, our chief creative officer was the Academy Award and producer for The Matrix. And so, like, we've got to, like, bring in a lot of really cool, credible from the creative space as well, who have had us really think through, you know, what will these future experiences look like? How will AI fit into this? And if people spend time in these virtual worlds or even, frankly, if these AI come into the real world, you know, what does that look like? And that's really what sparked, I guess, the whole kind of process. So I'll stop there. Well, all right. So this is what I was going to say.

the reason that Chris turned me on to what you all are doing, and the thing that blew my mind is, again, I don't want to be dismissive of what we've seen already, but we've kind of alluded to it. Like, sure, the first time you use Alexa to dial up a song, you're like, oh, my God, that works great. But then, like, there's been limits, and sure, it works great in call center settings.

and sure, even GPT3 and things like that, it's cool and mind-blowing that you can, you know, ask for a certain picture in a certain style, and it's better than you can even imagine. I'm not, believe me, I'm not saying that that's not amazing. The thing that you guys are getting at, that it's the first time that I was like, oh, I can see how this will transform something.

And it could transform gaming in this way. gaming is still on rails in the sense that even in a giant open world where you can go anywhere and do anything and the graphics are amazing and you look over here and there's this thing and that thing you're still on rails in terms of what the game wants you to do what you're capable of doing and that's because again most of the game is dumb and you're the only smart thing in the game. Yeah. So the thing that really blew my mind when Chris turned me on to this was like, oh, I can see how gaming could be completely revolutionized if the game is not dumb. If in these open worlds, the other characters that you're interacting with are, I mean, to what degree they can be totally smarter or whatever, but smarter than they are now, that would blow open the sort of storytelling you could do, the sort of experience you could have. like this is let me let me summarize for you Brian like so and and Kylie and I'd love to hear you talk a little about the Disney accelerator and what that experience is like because to Brian's point what he's really trying to say is when he watches an MCU film he would love to instead not just watch it

but after the credits roll and of course there's always some after scene to you know walk up to you know some of the characters and have a conversation with them because the whole point of what your tool does, as I understand it. Now, again, this is where it gets maybe a little hand-wavy, but yet you guys, I think, have the credentials to actually, you know, produce something that's reasonable in the space, is you walk up to them and they actually know or have some sense of their own story. They have some sense of the vectors that are determining their goals and their desires and their outcomes, which is, you know, again, why I mentioned Westworld.

It's like the whole point is to understand what is the ultimate kind of intention of this character, this, you know, non-player character that you're interacting with. What are they trying to achieve? What are they getting to do to get there? What do they already know? What do they don't know? You know, what was the last conversation you had with them? Are they able to maintain state of some sort?

Can they, you know, refer to that state in a subsequent conversation to give you some sense of, like, recall and also personalization? And that's where you understand why Disney would want to invest in this. Because this is how you bring the metaverse of these spaces and these characters to life. Now, I don't, you know, yes, I'm jumping ahead and I'm sort of, you know, accelerating the conversation. But, like, I would love to just understand, like, obviously you can't share anything that's not, you know, like out there, but directionally, does this start to become a new type of creative discipline, a new type of world in which, you know, storytelling and, you know, Brian's background is actually in film, that the way in which we, you know,

bring characters to life, not on, you know, screen per se, but on screens in these immersive environments, starts to change because you're no longer just kind of scripting scenes, but you're scripting backstories and I guess prompts, if you will, that give these characters, like, reasons to behave in certain ways that are somewhat more predictable. Yeah, so honestly, I think you kind of have to, like, the way we tell the story is, if you think back to like the way that we've engaged with content for a long time, if you go way back, right, you have like this oral tradition of storytelling, which is ultimately, you know, people telling stories to each other.

And if, you know, the kid asked the mom, you know, and then what did Achilles do on the ship? You know, she probably could make something up. And the whole story sort of adapted to that kid because she knew what he liked. But basically since, you know, we had old school theater, basically starting, you know, a few thousand years ago, especially starting with the printing press, content became a sort of one-way stream, right? So you had creators and then you had consumers. And those consumers basically didn't get to decide. Maybe except with the attempt at some goosebumps, books that I did love. You know, they didn't really get to decide, like, how to engage with that content.

It was basically pre-riff for them. And that basically continued through film, you know, even like early games, excited. scrollers, there's not a whole lot that you could do there, except for, you know, the jumping in this. It's ultimately this determinist experience that you as the consumer of that content don't feel in control of your experience. And I also have the side thought, which is, you know, what does that, what does that give you in terms of your thoughts or your conception of the world? Like, if I don't feel like an agency in the world, you know, how does that change the way that I think that the world exists? And what does it mean as content starts to adapt to you as a person

and actually opens up that agency and makes you realize, oh, wait a second, like, when I look out of this thing, these are all things that I can change. So I think that you started to see this in like the end of the last century with like the very first MMO games, right, where you had user-generated content and your RPGs sort of exploding where you had more, you know, tire and things that you could choose. You know, you got to choose which Pokemon you had in your party and all these things. And then sort of that continued. And I think over the last few years, what's happened is it's been this. I think it's tied to a lot of other social trends around, like, you know, self-identification, choices that we're able to make, you know, the lifestyles are elite. And now all of some people are like, I want to choose who I am in this game. I want to choose how I get.

And look how big these open world games have been. You know, Grand Theft Auto was obviously one of the early ones, but so many games now have adapted to this sort of open world sort of user-driven way. And so I generally think you see this as well just with the, you know, the events of things like Roblox and Minecraft where kids are creating their own experiences. even Fortnite, right, added on the sort of creation opportunity. And now, though, for the first time, those characters that forever, you know, our favorite characters, I love all my Pixar movies. I also freaking love Marvel. But for the first time being able to have that actual interaction two-way versus me just consuming it,

however the creator gets, you know, wanted to create it, is a completely different experiences and changes how I engage with that concept. And so that's like ultimately, I think, what excites me. Sorry, Brian. Oh, I was going to say, but, but, you know, You're even describing it. You continue to describe it in terms of the user's experience. So if I am in a game and I talk to Buzz Light Ear in a game or whatever,

and he'll behave like Buzz Light Year, or at least how I expect Buzz Light Year to behave, and he'll react to me and change and adapt to how I am. But forget the user. This creates a different sort of storytelling because, essentially we're creating a platform where you where the story itself is outside of the control of what even the creators intended at some point yeah yeah I mean there's it's basically this emergent storytelling so it's you know you you know we think about this a lot like you can imagine a game which is eventually filled with a

population of characters that also have have their relationship their emotions you know our characters for example are able to remember your interactions so if you go up to a shopkeeper and your brood to them and you're mean to them on time. And they're angry. They're actually going to give you like a worst price maybe the next time, right? And so then also they might tell another character. Now all of a sudden all these characters in the world sort of hate you. And that sort of is a very weird dynamic, right?

Where now like this is kind of the point of agency though, right? It's it goes two ways. It's like I have to do it wherever I want. But now there's these consequences. Like you said, we have to now consider what is the character's experience and like, how is that going to feed back into my experience as a player? Um, and so yeah, there's, there's so many things that like, you know, you can have like these branching paths for characters, our characters have an internal emotional state.

So if you get mad at them, you know, their facial expressions and gestures and voice and everything change. Um, and so, you know, you, it's like there, there is still that consideration of like, what is that experience for the user, but yeah, definitely there is this constant evolving internal state of a character that, you know, especially in certain games, we can support a way where like that character can just continue going on, right? And, and we've also thought about experiences where you could effectively just, just have two characters talking to each other for a while. And, you know, they could, they could, they could evolve in all these different types of directions of conversation and emotion.

Right. Forget about players. You could sort of like the, the clockwork universe God theory where it's just like you, you set up all the players in this universe and you wind it up and let it go and see where the story goes. Exactly. Exactly. And like, you know, we, we need to find the right outlet for that specific type of experience.

But that's, that's it, right? I mean, and that's sort of the vision we had of the future is, you know, if you hop into any of the so-called metaverses today, right? Whether it's Roblox, Meta Horizon sandbox, they're pretty empty. And so you can imagine, you know, if these characters are populated in that world, it kind of, that world goes on without you. So you go in. Oh, shit. Yeah. So, wait, wait, wait.

That's great. So it's like clubhouse, but you're not there. And you come back and the conversations have actually advanced. jokes that I've made about how dumb the meta-metaverse has been so far, because it's so empty and so vanilla. But if you walked into it and all of a sudden there was energy and activity going on, even though no one was there. Yeah. Yeah. Yeah.

I mean, and the fact that you could change it, right? So it's kind of like, you know, if you walk into, you know, a party and you just throw like some paint into the room, right? and it splatters everyone. You just walk out. That's going to change the rest of that night for those people. If you came into that party or to that house, like, met those people later on, they're going to remember that. And so you can imagine, like, also just being able to have those types of experiences where, like, you have this ongoing, this ongoing world that lives beyond you, you influence it and then you walk away. That's kind of what I mean of this, like, it changes this concept of agency and also changes that concept of, I guess, like, the, uh, the persistence

of the experience, right? Like, people often talk about, like, persistence at, like, this thing in these immersive realities. But mainly what I think they mean is, you know, that it exists across time and you can go back and see the same chair. But like, what if the dynamics, what if the actual life and the evolution of that world continued beyond, you know, when you were in there?

And I think that's sort of the magic of this. Well, I mean, persistence in that state, or in that sense, I think refers perhaps to statefulness in the sense that, you know, you're not sort of, you know, booting up the cartridge, like every time you go to play the game or into the space, it's that there's a continuation. So even if the chair

has moved, there could be a reason or a backstory for why the chair had moved because things were happening while you were not there. So it's essentially kind of a, like, and that's, it seems to be like the metaverse concept is that there is this other world that is constant. It's like, you know, I was in New York last week. I'm not in New York anymore, but at least I have some idea that New York still exists because I'm pretty sure Brian is there. And as long as he's there, I don't know that. Look, I can be in my own simulation right now. I can be talking to the simulation Brian, which has statefulness. And So who the hell? I don't know.

You're a disembodied voice. You're coming through this weird black mirror device that I have. Anyways, point being, you know, the simulation of New York is ongoing, even while I'm not there. And there will probably be construction projects that will have finished that were underway when I was there last week, that are actually the completion of the projects that I saw, you know, then. So that happening in the Metaverse is a pretty fascinating and, like amazing idea that there could actually be progress in things that are occurring at a normal rate of change. and that that could be, if not personalized to you, certainly part of the entire metaversial experience,

which, I mean, frankly, from a conversation perspective, it's kind of fascinating, right? Because, you know, you can come in and start a whole dialogue or conversation, perhaps with one of these NPCs about the changes that have occurred,

and the NPCs, having been there, having seen the change, could actually tell you about it in theory. Yeah, that's exactly. And I mean, even if you think about just, like, scoping it down to even a one-to-one interaction is, you know, when we catch up with a friend, how you often would spend,

Like if you said that you had a 30 minute chat with a friend, you can spend the first 10 minutes of that just being like, what did you do today? You know, what did you do this weekend? And if you currently talk to a chat bot, it's like they might be able to make that up. But imagine if you talk to that character, you know, fairly often. And there's actually this consistent story and there's this background. And even if like, even if that doesn't actually exist, right, it's, it fills in sort of reality for you. Because it's sort of, otherwise this thing exists in this like disembodied vacuum. And so you're right. And like all our characters do, they have personalities and backgrounds and all. and knowledge and that knowledge is updated. And so it's, it is like a, it's, it's like creation.

So it's a bizarre experience when you actually start experiencing it. Yeah, I have a couple, like, I have several different questions and directions to go in. I also want to be respectful all your time. But that said, while I have you here, I did pin a tweet. And in that tweet, which is in the so-called nest, which if you, you know, see that, you know, has an interface from Westworld where the, you know, for dramatic, In fact, of course, while whatever, you're sort of encountering one of the hosts, the, whatever, the system administrators, let's call them, are like, you know, tweaking and moving,

you know, the character aspects or attributes of the hosts. And Inworld has a very similar type of interface. So my question is really about, and I'm not trying to again cause you to reveal anything that should be known, but to what degree does Westworld get this wrong? Like, I think Westworld is so useful. as a kind of direction of a sci-fi fiction, you know, sort of future. And oftentimes science fiction can lead the way in helping us to understand and predict how we might solve some of these problems.

When you watch that, what do you see and think, oh, man, that is so wrong? And what do you see? And you're like, oh, actually, I'm going to put that in the next version of the app. You know, I, there's nothing I would say is wrong in the sense. I think that at least the, like, frankly, everything I see in Westworld is, like, scarily maybe possible. I'm so afraid and also excited that you want to say that. Not now, but, you know, I think that honestly what I would say today is some of the hardest parts are like locomotion and robotics and these kinds of things, right?

So, you know, actually the brain is like what we focus on, right? Oh, sorry. And I'm like, I'm taking away like the, like the aspects of moving the body and like the muscular structures and things like that. Sure. Like, you know, I'm thinking more if this were, if Westworld were actually conducted in a, you know, in sort of a visual simulation. like the internal personal narratives yeah like Roblox if Roblox were the platform or unity or something like that

to what degree could you achieve a Westworld style experience oh we could oh I got that one of my laptop right now not the same but like very close to it today yeah yeah so that's I guess what I mean is like frankly I'm not I can't say oh it's not possible at these I can't poke too many holes I you know maybe the way that they have it is probably too simple in the way that the tech works, and frankly, their interface is kind of ugly, is what I would say.

But in general, so like for an example, even for the tweet that you pin, like literally what's happening in the emotion engine is you actually have a real-time state of the character that's like basically just a bunch of numbers that are consistently being updated in real time. And the emotion side is basically like the default state of the characters. You can kind of think about this as like, you know, some people are between like a seven and a ten. They tend to be very happy. some people like a three to a five they can be sort of sadder and that's basically setting the like default state of the character and then you can think about the personality as basically how that default state is updated based on what's experienced and we all have this right it's like we all respond it's like if you get mad at the rock versus you get mad at you know um ariana grande they're gonna respond very differently based on the personality you to regardless of where they're you know you have like different triggers and then you also have things like so what you see on you have like fluidity so you know some people's emotion emotions are very mercurial, they're changing all the time. Or so not, and like the emotion engine really does contain all that and is really able

to hold all that. And then we drive, for example, animations and gestures and changes in voice and changes in content of speech, all based on that. So it really is sort of like, you know, this, I don't want to say one to one, but it's very much inspired by how the human brain works. And frankly, when I look at the Westworld interface and like, maybe they have too many knobs there, like that, like, clearly that's like the advanced version that most people would never be able to use but uh yeah i don't think we're all that far off okay so i've i've one more

thing and i know brian's got a question um yeah you know i think i'll like well i'll just make a statement and i'll ask my question like this this to me could be like such a revolution in terms of education you know if you would imagine actually like going back in time and history you know i watched all the john adams movies that that were on i forget what was on maybe hb o and it's just like so interesting right and to be able to go in and then add this layer you know where you have accuracy, or these, you know, told by certain people from history, you know, built into the characters and now you can ask them, like, why did you do this? Or what did you see or whatever? Then you can actually inspect, you know, those moments in time from a very different perspective.

So my question, though, is actually kind of about the competitive market. You know, I sort of asked you, and I think you might have dodged a little bit, the question about Disney. My question is actually about Apple and Pixar, because in 2019, they acquired a company called Polstring. And I know the founder of Polstring. And what seemed to be the case at the time, and I don't really know what's happened to them since because, you know, it's Apple, you know, is that they could have been used to improve Siri and, you know, its ability to hold conversations. But I would imagine that as, again, I guess going back to the creative side of this, that Pixar could use this technology as well as Disney to actually, you know, investigate characters to, you know, coach the characters to behave in a certain way. And even from a, you know, I don't know, like a film or a movie or whatever kind of perspective, you could. I don't know, conduct an entire environment with these characters like this.

How much does this become a creative tool like that versus one that's just built for, you know, sort of multiplayer or single player games? Oh, we really don't think about it, frankly, as a game development tool solely. It just happens to be the use case that resonates with most people. We really, we call it like a creative suite for basically building these AI virtual or AI personalities. And I'm like things like education, as you mentioned, are sort of like, frankly, top of mind. And we can dig into that. But, you know, when you think about Disney and Pixar, like, Disney is, and like, I will come back to that, right? Disney is a character of first company.

Almost everything that you do. I was actually at Disneyland last week. You know, and you go in there, and it's all characters, you know, and it's all characters. You watch their shows. It's like, it's all really just driven by the Marvel shows, the Star Wars shows, your classic Disney movies. And, you know, it's still that one way interaction. And so being able to sort of, you know, one, bring those to life is definitely something, you know, of interest. And then there's also this general question of, you know, how can you per. for example, extend that to allow either consumers to influence their own characters or create their own experiences. And, you know, there is actually applications, you know, related to learning and education and these areas as well that you can imagine, you know, how could you extend something like your favorite character and actually turn it into a functional thing that might serve as like a tutor, for example, or, you know, fulfill something like that. Way's voice that you can like, you know, turn on for your driving directions. Exactly, exactly. And so that's sort of the interest there. And there's, and there's,

definitely like it is a creative tool. That is ultimately what we're creating here. And you know, in the same way that you know, you can go into the, you know, the Dali or stable diffusion and create an image. Really what we're creating is a way for anyone. We're kind of lowering that bar to create amazing characters ultimately, right? And so like if anybody goes on to our studio, like literally, you know, anybody with, that can write, can write a character, right? It is that easy. And so it kind of lowers that ability to actually create these amazing characters. And

we think that also opens up the door for a lot of new applications. Because game developers have been doing this for a long time, but let's be clear, educators, you know, brand ambassadors, all these kinds of people haven't been doing this. And so it really is ultimately moving toward the suite to create these characters and then integrate them into any experience that is as easy as possible. And, you know, Polstring was actually related, the company that you mentioned that was acquired by Apple,

it was actually very similar to like Dialogue Flow. and the previous platform. And so, you know, those things obviously did make their way into a lot of other applications. But I see that, you know, what we're building hopefully sort of is just like the 10x or the next evolution on top of that. And hopefully, you know, empowers a lot more amazing experiences versus what Syria and these things are able to do just because it's that much more powerful. And those real world applications, especially education, are the ones that really kind of motivate me the most because I think that's what we'll have. I think about it, you know, we're not that far off and, frankly, having like, really good tutors. But I think, you know, in a few years you could imagine a place, like, why can't everybody have, like, a Harvard professor actually at their fingertips on their phone who's actually able to chat with them, remember, you know, what they just chatted about.

And I think that's frankly very possible and, like, one of the most exciting things here. Yeah, three more in them will let you go. All right. The first one is just really basic. So right now, and to the degree you can talk about this, and the same way that I can. can grab the Unity engine or whatever. I can grab this as a platform and I can put it into my game or I can make a better Mavis Bacon typing tutor or something.

Like this is a thing that like this beacon. I want this bacon. What did I say? Mattis Bacon? It wasn't a beacon? Anyway, but that's the point, right? This is something that I grab in the same way that I grab unity to create my world. and then I do all the stuff on top.

Is that what you're doing right now? For right now, that's the business model, right? That is right, yeah. So it's basically, what I think about it is like environment or this kind of 3D engines, and then you have this intelligence engine. And we offer also the interface on top of that as well to make it really easy to create and integrate.

Yeah. Okay, so the final tour are where everyone take your gummies right now. So in the same way that currently with all of these fun tools, Like you put in a couple lines and it creates a really compelling, really believable picture based on the lines that you put into it. I could put in a couple lines describing Chris. And then I don't need Chris anymore because the Chris character that I could create would be as good as Chris. And no one would know the difference? Thanks.

Great. Um, you can answer that honestly. That's fine. Uh, no. I mean, the answer is like you actually, like, it can sometimes be creepily similar if you put in enough information. I'll be clear about that. If you put in like the knowledge, the background, the story, you tune it all.

I, like, ask people to try it because in some cases, I've been shocked, especially when, let's say you have more unique backgrounds that like a lot of people don't share. Like, wow, this thing, like, it knows me. Um, so definitely it's possible. And I think in, let's say, as we continue to evolve this in a few months, definitely in a few years, I think we actually have this aspiration for this general idea that why don't you just take, well, Chris, you know, maybe Chris is, you know, well-known enough, but you can take, like, The Rock or one of these famous characters, like, I want this character, right? That's all you should be able to say, and, like, you should be able to basically take that on.

Or, you know, I want a, you know, a funny character that's kind of, like, has the humor of Chris Rock. but you know the like the old school banter of Dane Cook and like I want to see that character right so taking that inspiration so definitely this is something that we're thinking about is how we can draw inspiration from these like you know these well-known characters I think we're also though I will say try it out today and I'm what you get but definitely in the future we're working on ways for people to create likenesses of themselves as well because I think that opens up a lot of really interesting you know it's funny I don't know Kyle and if you know this I

did mention it earlier but the the startup that I co-founded was called Molly. And the idea was that you would actually talk to it about yourself, like indefinitely. And it would actually ingest all of your social media content. And it would then create sort of an index or version of yourself that would be able to automatically answer questions. The inspiration for this was essentially looking at like Gary Vaynerchuk's tweets. And the number of times he has like answer questions, you know, that are the same question over and over again. So the idea was essentially to create kind of like the modern AI powered version of the answering machine, but personalized to yourself, you know, based on,

looking at all the, you know, media that each of us creates. So definitely aligned with that idea. And there was actually for a time, the Messina bot, which you could access over Messenger. It might be dead now. But it would actually pull up my favorite cocktail recipes and it would give you tips based on my four square tips. Like it was actually like pretty great. So, so Brian, if you want to replace me, I'm all on board. I don't want to replace you, but I'm thinking about like we could bring on a Humphrey Bogart at some point. We could bring on a Steve Jobs, which makes me think. Oh, yeah. Yeah. you guys have to be thinking

about the deep this is taking deep fakes to a different level yeah because deep fakes right now it's like well it's in the person's voice it's in their cadence it's using words for sure this is a deeper fake where it feels like it's

yeah I mean definitely there's a I mean first of all I would love to try and create one of these famous characters and bring them on here and just people notice that would be awesome there I think I'm going to request I have a request, Steve Bomber.

I want, if I could have anyone be our co-host, it would be Steve Bomber. What would just be like developers, developers, developers? Well, that's what I want. Yeah. Okay. I mean, I'm all in. Um, with chat. But so, so to the general question though of like the deep figs, I mean,

definitely there's a risk there, right? There's a risk with anything enemies creative technology is the same way like Dolly. Now, you know, with you or Dolly or stable diffusion, you could recreate art from someone else. I think there's slightly less risk when it comes to people because there might be a point, frankly, maybe in the next few years where this becomes more common and all of a sudden you don't know if it's actually, you know, you're your partner messaging you back or the bot that they set up, you know, but I think for now that we're not there and the, you know, there's

probably like these little things that you might notice at least for people you know well. Now, when you talk specifically about like Gary Vee and you have like, you know, people messaging on Instagram and things, there are already, for example, like those people will hire people to respond for them. Like that's very, like those people are not really responding. And so I think where we would end up seeing this used, there are probably cases where you're already maybe not having those people respond or, you know, the Instagram influencers or the, you know, the Twitch streamers.

Those are definitely of interest or what we've actually looked at, for example, and this is like, you know, to put the positive spin on deep fake. So, you know, think about professional athletes and you're, you want to have a personal relationship with professional athlete, but it's pretty. hard, you know, to get LeBron James in a room and have a chat. But you want to ask him, like, questions about his history, and he's willing to share those, but he just doesn't have a time. So he creates a copy of himself. And now you can go and ask those questions.

And it's like cameo.a.i. Yeah. Yeah. Well, exactly. Well, yeah, but I guess live and. Sorry. Sorry. I mean, like, yeah, that was a joke because, you know, obviously camios. The real thing. Yeah. But now we're in the, what was that movie with Michael Keaton? Was it multiplicity or whatever where it's like, he created multiple versions of himself because to deal with his wife, to deal with his kids, to deal with his boss. If I could just clone myself. It's like, oh, right, what if you can?

Not to cut you off, but let me end with this. This is the super macro thing, and I've been searching the whole time we've been talking. It came across my transom today that there were a couple stories about AI researchers. I think it was in the Washington Post or something, but I couldn't find it. Anyway, talking about how they thought. that there's a 15% chance by the end of the decade that AI was going to do something that would be hugely transformative to society on the level of the Industrial Revolution. And then if you go out another 10 years, then it's 40%. If you go out to 2050, they're like, yeah, it's for sure going to. So someone that works in this field, it's one of those things that I always say that, like, oh, I've been hearing that VR is on the, you know,

about to happen, self-driving cars is about to happen, and it never happens, it never happens. Do you feel like we are on the cusp of AI finally happening in a way that it's not just sort of gimmicks and, you know, cute tricks? But like, holy crap, society is different because this thing exists in the world. Short answer, yes. Absolutely. I've been working in a space for a while. I am, especially when it comes to these generative architecture. So there's been a big shift literally in the last three years. And I think every, like I started, for example, doing generative art at the beginning of this year in January.

And I was using these like old school models and things. And it took me so long to create a blob that looked kind of cool. And three months later, Dolly came out. And I can create whatever I want. Now a few months later, you can do 3D generations. Like, it is shocking. And it's simply because like it's tapping in. I think, to this part of like what it makes, like what we can do valuably.

And so there is like this economic incentive as well. And so you just see tons of, tons of resources pouring into it. And I will also make a comparison. I'm sure this will maybe insult some people. But I think crypto is one of these areas, which is frankly, for where it's at is really overhyped. And that's because it's sort of, you know, there's an incentive to sort of over, over, over emphasize the value of it at a given time. And the barrier to entry to get involved with it is slightly lower.

Whereas with AI research, it's really hard to get the AI research or like really strong AI research. So there's very few people who actually understand like what's going on at that like end degree in the state of the art and are reading papers, for example. And I don't think people really recognize like where it's where it's at. And that's why like, you know, even when we go and talk to people, they're like, oh, is that really possible? And then they say, oh, it's possible. So I'm not saying specifically with AI characters or AI personalities, although I think there's a lot of things that are going to happen there. But to this general question of like, will there be something that is hugely transformative, I'd put it above 15%. But I can't, I could try and guess what that would be.

But we won't hold you to it. We won't hold you to it. My last question is just like, how much time would it take to program like the Brian bot or the latest Messina bot? We could create a simple version in like 20 minutes or 10, like, if we could get a really simple version in two minutes. Can I ask you? Because, like, I've said this before. Like, there are literally thousands of hours of my voice out there. Like, what's the input that you need?

So we do lots of input. So a lot of it's like voice. So we like just text. Do you describe basically, we, the paradigm we use is describing to a human actor. So think about the amount of information you'd have to provide. to like Brad Pitt, a really smart Brad Pitt who has access to all of Google's knowledge. Sort of like that's how you describe. So you describe almost a natural language, like how you would describe to the actor who they're supposed to be.

And then we also allow you to, for example, we do train custom voices. So we can train a custom voice with about 50 minutes of footage and we can do emotional and intonated voices with a bit more. We also do like, for example, like you mentioned sort of Brian, right? You could do like example dialogue. So you could imagine actually taking a transcript of a conversation with you and then copying it over and that inspires the AI to speak like you, for example. And then there's other things, of course, like those sliders that you showed. So to be able to tweak the emotional state. And then we also allow you to inject knowledge.

So you can connect it, for example, to knowledge bases. Right now we do natural effect input. We're working on ways to be able to ingest web pages and Wiki and these kinds of things. So all of that, frankly, we're trying to make it super easy. And we're trying to make it easier every day. So, you know, within a few months, I think within, you know, 20 minutes. You could have a pretty representative version. Of course, the more time he's spent on it, the better.

And a lot of it's just thinking through sort of what you want to describe. But not only do I have 2,000 podcast episodes out there in the wild, but I have all the scripts for everything that I've ever written. I've got it coming and going if you need any sort of input. But maybe I would regret that if you take me over and no one notices. But amazing. Yeah, thank you so much. We'll bring on an AI version of you one time and nobody will have to be any

I am 100% down. Let's test it. People will fucking flip out. All right. Let's do it. Yeah. Well, you know, Kylan, it's funny. You know, we started this show today.

It's sort of talking about the hashtag anniversary, which is now, you know, the hashtag's 15 years old as of two days ago. And I think to Brian's point, when you think about or if I think about kind of how these, you know, subtle, and in your case, it's not so simple. technologies, but yet they get to the point where so much of the machine is doing the heavy lifting that allows for this new type of creative expression. It does feel like we are sort of, there's this like tsunami wave that is like building up, you know, momentum. And by, you know, the end of this year, next year, I mean, certainly by 2024, the types of immersive game or just, I don't, we won't even really call them games. I'm not quite sure what we're going to call them. But those environments, those spaces are going to be so different.

I mean, it's going to take, it's going to be like a whiplash, I think. But I think about how like there wasn't social media, you know, before sort of like 2006 to a degree on a widespread basis. And now, you know, as we were discussing at the top of the show, there is huge amounts of money. There's like all this controversy. There's governments. There's spies. There's like all those old, you know, statecraft, you know, literal like physical world building aspects. You can imagine that the same thing is going to be true for these.

virtual experiences. Can I, let me interrupt real quick. Chris, remember the first time you used Alexa and it worked and obviously it blew your mind and that's why you did a startup like that, right? Yeah. And then also remember the first time you played either whatever the Zelda was on N64 or the first time you played Grand Theft Auto, right, where it's like I can do anything. So like it's those sorts of moments in technology that always stick with me where it's like,

oh my God, I can do anything. I think it was like the Matrix, the Matrix Unity demo that came out earlier last year or this year. Max still plays that all the time. Right. So you take that, right, and then you take these, you know, AI character engine. Put all of these pieces together. You take the Dolly type, you know, generative art, sort of environmental generators. That's the kind of world that you can imagine people are going to be like living in.

It's going to be this crazy psychedelic soup. And yet it's going to allow for an entirely different experience of, you know, in conversation or about reality. this is I mean this is what excites me right it's I think that technology is catching up to creativity right so you we used to imagine these amazing worlds and now actually we're able to create them almost like you can imagine if you fast forward maybe a hundred years it's like could you start creating at the speed that you're able to imagine right that's ultimately I think where we're going and I think a lot of these tools are basically just saying how quickly can you translate what's in your head into reality and I think you know we've done that like with Dolly and GPD you know in terms of text and images and I think how quickly can you translate what's in your head into reality and I think you know we've done that like with Dolly and GPDD you know in terms of text and images and I think now seeing this with personalities and people, you know, I think it's just going to keep happening with worlds and environments. And I think that's that's an amazing experience, right? Because I think just allowing people that more capacity for creative expression and ultimately that intersection of creativity and tech, I think is just where like the magic of humanity exists. Can I just say that's the best summation of the promise of this that I've heard from somebody? It's like the thing that you think of and make it happen as opposed to

As long as we get neuralink, you know, it's going to be yeah yeah All right, Kyle, where can people learn more and find out about inworld? Yeah, head to inworld.a.i. If you want to actually create a character, you can just sign in right there and start doing it. We've got a few example ones there. Otherwise, you know, reach out to us if you have any other questions or you have any other thoughts on custom integrations or you're working on a project that's related. Yeah, I'm super excited to kind of see what people build.

You know, ultimately we are a creative platform. We want to see what you can do, push its limits, and build something fun. Awesome. Well, this is super great. We really appreciate your time with us. And congrats on the launch on Product Hunt and you guys raised some money recently. So congrats on all those things. Yeah, thank you so much, Chris.

Thank you, Brian. Really appreciate it. Awesome. Thank you, sir. I love everybody. I love giving our AI overlords that are going to replace me. Got it. Be good, everyone. Thanks, guys.

Good night.