Transcript
[00:00] speaker_0: We have a fantastic panel lined up for you, t- to talk about AI and a- agents and simulation. First, we're gonna s- start with Edward. Edward, would you like to give a bit, uh, a- a short intro and talk about some of the projects that you're working on that you're excited about and then we'll... and then after you do that, we'll, we'll show a bit of your video.
[00:20] speaker_1: Hi. I'm Edward Sachi, CEO of Fable Simulation and we're building out a simulation of San Francisco, which is SimFrancisco. Um, and attached to that simulation are many television shows. Just as New York might have Friends, NYPD Blue, you know, different genres, we're gonna have the same with, with SimFrancisco. And the first show is, uh, Exit Valley, which is a vicious but loving satire of these, uh, AI billionaires who are telling us that they're going to end capitalism, they're going to end money and it's just a coincidence that they're gonna become incredibly rich in the process. So, we think that we should use AI against them and on an industrial scale, uh, make fun of them. So, that's, that's the first show connected to Exit Valley, connected to SimFrancisco, Exit Valley.
[01:23] speaker_0: Great. Let's roll the tape. Play the video.
[01:42] speaker_2: Good morning, Aurora. Currently in San Francisco, it's 62 degrees with partial clouds. You might want to bring a layer for the slight chill in the breeze. Today's schedule includes team meeting at 10:00 AM, your yoga class at 6:00 PM, and don't forget, today is your mother's birthday.
[02:07] speaker_3: Dr. Johnson, I keep having nightmares that maybe we are all part of some digital farce. An intricate simulation.
[02:15] speaker_2: You mean like The Matrix, Aaron?
[02:17] speaker_3: No, no. Not like The... I mean, sure. It's like, like everyone's pulling the strings. Manipulating me, including you.
[02:25] speaker_2: Aaron, you know I'm programmed to help, not manipulate. You're in a safe environment. No one is controlling us here.
[02:42] speaker_4: Ever since we replaced human sim TED speakers with AI replicas, things have become so much more efficient. You know, I absolutely relish orchestrating these TED Talks. The perfect speakers, aren't they? These AIs, they're not hindered with the frailties of human sentiment. They don't crack under pressure or bow out due to personal emergencies. They just deliver. And the best part, I don't lose time dealing with the chaos of human emotion. It's all so effortless.
[03:08] speaker_2: Code red. A speaker has escaped. Code red. A speaker has escaped.
[03:12] speaker_4: Oh, hold on. I'll have to pull the plug on these fellas.
[03:33] speaker_0: Thank you. All right, and next up we have David Slade. David Slade is known for Bandersnatch, which I'm sure many, many of us in, in here watched. Also worked on Twilight, Breaking Bad, o- other Black Mirror productions. So, great to have you with us.
[03:49] speaker_5: Hi.
[03:50] speaker_0: So, uh, so-
[03:52] speaker_5: Yeah.
[03:53] speaker_0: ... if you, if you could give a, a little bit of an introduction. You know-
[03:55] speaker_5: Yes.
[03:55] speaker_0: ... you've had a, a, a, an amazing career in Hollywood so far and h- how are you-
[03:59] speaker_5: Doing my best.
[03:59] speaker_0: ... how are you thinking about working with AI and what, what's most interesting and exciting to you in this space right now?
[04:05] speaker_5: Um, I can... Yeah, let me... Uh, I... There's a couple of anecdotes, but it'll probably take too long, so (laughs) I'm gonna jump into, uh, during the strike, uh, the writers' strike, uh, the actors' strike last year, or the year before, whenever the hell it was, because of COVID. We had a missing year. I don't know. Just, I don't know what... The years all go together.
[04:23] speaker_0: It was last year. It was exactly... Uh, the writers' strike-
[04:25] speaker_5: Was it last year?
[04:25] speaker_0: ... was exactly. It started-
[04:26] speaker_5: Yeah. Okay. So, last year-
[04:27] speaker_0: ... just one a year ago.
[04:28] speaker_5: Um, I had been using Midjourney, I'd been using various other tools and I couldn't work because there was a strike and the commercials industry also died, there I do that too. And I started training myself on what are the possibilities to make movies with, uh, to tell the same stories? To tell stories, in fact, that may be impossible to tell, um, using traditional f- but using, you know, 20 years of, of, of, of cinematic language to do that. And I started making, uh, working on a series of different things. I did a series of short films, I did a music video which was essentially at that point before I think a lot of the gen AIs... It's that a lot of the gen programs that make moving came out, so it was a series of mid-journey sequences that weren't animated. And then, uh, and then I made a little short, uh, storyboarded film which is a little silly, noir, science fiction, takes place in a liminal bar. And I went through that and talking to the, and the idea of simulations, um, I, I would... So, so what I would do is I storyboarded it, I, uh, then looked at each shot and I, uh, generated sometimes mid-journey images which I would then animate with programs like Pika or, uh, or just old sections from Runway. But what I found to be incredibly lonely about the process was I couldn't involve actors. And so, and I think this is why I... You know, I'm a filmmaker coming into this panel about simulated actors, so my, my drive... I mean, I continue to do this, make these films and weirdly, uh, you know, I, I, I'm... I've been spending years and years and years explaining to people what I want (laughs), so it seems to work-... with AI. I seem to to get the images. It seems to look like my work, weirdly, um, because I'm using, and I'm figuring out which la- you know, I'm figuring out my own language. Uh, promptees is a combination of film language and, um, to make these films. But I really want actors to be involved because I, I, you know, you can type in the word, um, you know, "She's hysterically happy." Right? And you might get some random noise, but you might get it exactly right. If you talk to an actor and tell them, uh, they'll be, I, I need them to be hysterically happy, they might know what to do, but most likely you just need to use a verb that will actually get them there, and that's in language. And then you collaborate, and then whatever idea I had will be made way better by this actor who has years and years experience and is, um ... So that is my, uh, you know, what I'm interested in with regards to this panel, uh, is the process by which I can bring people, and if not people, you know, s- uh, entities-
[07:12] speaker_6: Mm-hmm.
[07:12] speaker_5: ... and figure out how to continue telling stories. Um, look, at the end of the day, um, the metaphor is we're all sitting around a campfire. We're all telling stories. As a director, I came up with the analogy that, you know, the story is the text that's written down. Uh, if someone tells that story, and it's a really flat voice, everyone goes, "Ah, that's a crap story. Someone tell me a joke." But if you tell the story evocatively-
[07:40] speaker_6: Mm-hmm.
[07:41] speaker_5: ... then you become the voice, and that's what, I, I think that's one of the definitions, I think, of what a director does is, is become the voice of that. And as a director, I am acutely aware that you need other people and you need... And it's fantastic to have nobody criticize you, and you can do anything you want in your film, and you can put anything and nobody's gonna recut it. It's done. But this is this lonely process, which I, I kinda wanna go, "Well, (clicks tongue) I wish I had a, I wish I had some mocap in here at least." You know? "I wish I had some really good mocap. I wish I could plug mocap into this, and I could then workshop the scene with an actor upstairs in my living room until we're done, and then we, then we'll shoot it." And so the, that's, that's my position. But yeah, no, so I am a traditional filmmaker. My points of view are entirely, come from a, you know, a, a, a, a career of making films and telling stories. Before that, I was a writer, uh, made music videos, made commercials, told stories in very short periods of time, told stories in two hours.
[08:42] speaker_6: Mm-hmm.
[08:42] speaker_5: And so what I'm interested in is, well, two things. One, what are the stories we can't tell with traditional media? And are they interesting and, and can we make those? I made a series of just proof of concepts, which were based on dreams, you know, that I had, and very good at that. AI is awesome at making your dreams.
[09:00] speaker_6: (laughs)
[09:00] speaker_5: I get, I get, I, I, I can story by dream that I had the night before and I can get pretty much exactly what it look- felt like and look like to me.
[09:07] speaker_6: It's true.
[09:07] speaker_5: It's great for that. Not so good for realistic things, uh, yet, but, of course, we're on an exponential curve, and, you know, recursive exponential curves eventually get there. So yeah, I guess I probably talked a lot. (laughs) I can stop now.
[09:19] speaker_6: That's great. That was great. So-
[09:20] speaker_5: Okay.
[09:21] speaker_6: ... couple things. Well, uh, well first, let's, let's play your video.
[09:24] speaker_5: Sure, sure. Yeah, a little bit.
[09:25] speaker_6: Okay.
[09:25] speaker_5: Uh, so this is probably a butchered-up recut that I, 'cause I knew I had to try and get it down to a short amount of time. Um-
[09:31] speaker_6: And I think we're only gonna show the-
[09:32] speaker_5: Oh, that's not the right one, but we can show this one. This one's the silly one. Uh, this isn't the right one. Could you just go all the way to the beginning of the tape and, and scroll past... This is the one. This is the one. Yeah, just go to the beginning of this one. Right at the beginning. No, no, keep going back. Back, back, back. Just further back, further back, further back, further back, further back, further back, further back.
[09:50] speaker_6: Live, live demo here.
[09:51] speaker_5: Yeah. Yeah, so we can start here. Uh, can we turn lights off or do we just leave them on? (eerie music playing) (elevator dinging) (crowd chattering) Whoo. (laughs) It's much brighter now.
[12:14] speaker_6: (claps)
[12:18] speaker_5: Uh, there's twice as much detail there, it's just very crushed and all the rest of it. But, um, that kind of thing.
[12:24] speaker_7: Microphone.
[12:24] speaker_8: Microphone.
[12:25] speaker_5: Sorry. (laughs)
[12:25] speaker_8: All right, who was there saying mic? So...
[12:26] speaker_0: Oh. Sorry.
[12:27] speaker_5: He said- Hello. Uh, sorry. Yeah. Um, so it's about six months old, so it's very old. But it's kind of punk rock the way it all looks so dark, um, like, which I like. Um, it, it, there's actually twice as much detail in the image, but it, you know, the com- the, the, the projection kinda, we lost half the image. Uh, so that is a really simple basic story about a missing gun set at a bar, which is, uh, one of the oldest transitional stories you can tell beyond, you know, without going through Joseph Campbell. So it's basically, it's about missing something and someone has to pay. And so I used it and I just said, "Look, I just want to make a psychedelic film." Really. And it gets much more psychedelic as it goes along and, you know-
[13:08] speaker_0: Got it. Well, just showed a teaser then.
[13:09] speaker_5: Yeah, it was just a teaser.
[13:10] speaker_0: Where can people find it?
[13:11] speaker_5: Uh, nowhere yet. I don't know. Right?
[13:13] speaker_0: (laughs) All right. We're looking-
[13:15] speaker_5: Um, uh, you can see lots of-
[13:16] speaker_0: Yeah, we got some buyers in the audience.
[13:16] speaker_5: ... little sh- I have Instagram with lots of short films on it, but that one's not up yet. So I th- get a feeling it's gonna premiere on a, on a platform that I'm being talks with someone about soon.
[13:25] speaker_0: Nice.
[13:25] speaker_5: Yeah.
[13:26] speaker_0: Great. Okay.
[13:26] speaker_5: Um, thank you.
[13:27] speaker_0: So one, one somewhat ironic thing that y- you said in, in, in your intro was that because of the strike-
[13:34] speaker_5: Hmm.
[13:34] speaker_0: ... you ended up starting to, you had time to play with all these tools.
[13:38] speaker_5: Yes. Yeah. No. That, that was a little serendipitous.
[13:40] speaker_0: Because otherwise, otherwise you probably might not have had time.
[13:43] speaker_5: No. I definitely wouldn't have had time and, but it goes back further for me. Like a couple of years earlier, a friend of mine got me into an early beta of Midjourney. And so then I, I, then I witnessed the diffusion imagery, first use of diff- Because what happened was, I was looking around, I, I don't want to steal anyone's time here, but I was looking around and I started seeing these images I couldn't understand, because I knew, I knew how to make lots of document, but I started seeing these images, like, if those are digital paintings, it would take 1000 hours. What are these things? And I was looking. So I went out looking-
[14:16] speaker_0: So you are, you already-
[14:16] speaker_5: So I went out looking for Midjourney to, to, to, to explain-
[14:20] speaker_0: Cool.
[14:20] speaker_5: ... these AI images I began to see.
[14:22] speaker_0: Great. Great. Thank you. Thank you. All right. We'll, we'll talk more about that. Okay. Xiuying, thank you for being here. Xiuying is the co-founder and CTO of Alterra. So, so tell us a little bit about Alterra and, and then, uh, we'll show your, your clip as well.
[14:38] speaker_7: Sure. Thank you. Yeah. Thanks a lot for, uh, inviting me on stage. Um, so, yeah. I'm the CTO for Alterra. Um, we're, Alterra is like a deeply research focused company. Our mission is to really build digital humans that love, love and grow with us. Um, so what does it mean for digital human beings? What is missing for like the AGI, like, or the ChatGPT? Isn't intelligence all we need? Um, so our answer is that for fundame- uh, for human, there are some fundamental qualities that are not related to intelligence. There are like emotions, the coherence, the long term, um, theory of themselves, self-awareness. All of them are now part of intelligence. Well, at least now part of, like, the language, the, um, the describable part of human beings. For example, if I do something like this is like kind of fast processing, not controlled by our intelligence, like, our prefrontal cortex. So, in our company, we wanted to build this, like, coherent module of, like, a real human being. Well, virtual being, may not be human. Like, virtual being that can be living on a machine and be living in, like, together with humans, and be coworkers, buddies, friends, or, um, whatever way, and live together in the society.
[16:16] speaker_0: Show your video?
[16:17] speaker_7: Uh, yep.
[16:18] speaker_0: Okay.
[16:21] speaker_5: We are Alterra. And we're building digital human beings. In other words, we're building advanced AI agents that live with us in virtual environments that have strong pro-human social emotional intel- intelligence, and eventually self-awareness. In fact, we just built the first ever Minecraft AI agent that you can directly play with today. Our agents go far beyond the chat interface, and open up previously impossible game plays that are both novel and fun. We're starting with games, but we must, we must go beyond. Productivity AI will be so much better with more human qualities when they can go from copilot to coworkers. And eventually, we'll be able to bring digital humans to the physical world.
[17:13] speaker_0: Great. Thank you. So I'm excited to talk more about, about some of the work you do with Minecraft and beyond. Um, we'll, we'll circle back to that. Okay. Parth, you work, woo, so you work with the- at the office of Reid Hoffman.
[17:27] speaker_7: That's right.
[17:28] speaker_0: Yes. Excellent. Uh, so talk to us a little bit about, you know, maybe h- how you came to be there, and then, um, some of, some of the work you're doing there, like Reid.AI, and then we'll show, we'll show a clip of Reid.AI.
[17:37] speaker_7: Cool. Um, so in a previous life, I used to work for startups and I was a data analyst. And then, um, I was working at a company a year ago, uh, a little over a year ago called Clubhouse. It was an app made in the pandemic.
[17:50] speaker_0: Who was on Clubhouse?
[17:52] speaker_7: Some of you. Oh, nice.
[17:53] speaker_0: Nice.
[17:53] speaker_7: Little Clubhouse. Okay.
[17:54] speaker_0: Some Clubhouse fans.
[17:54] speaker_7: So I worked on Clubhouse, and part of, when I was working there, I, I met, I met people during, when ChatGPT came out in November of '22, I realized it was, very quickly realized it was the fastest growing topic in every single language in every single country. And I was also lucky enough to meet people all over the world through the app and learn how they were using ChatGPT. And I think there were a couple moments where I realized, I think, one-
[18:18] speaker_9: ... I met a guy and he was like, "You need to be programming with this thing. It's like, GPT-4 is an incredible programming assistant, and it can teach you how to program." And so then I was like, "Okay, I'll, I'll, I'll check that out." And then I met a guy who was a farmer from my home state in India, um, a state called Maharashtra. And I met this guy and he was using GPT-4 to help him plan his crop cycles, and that's when I realized, this is not just for your email. Like, th- For some people, this is actually like a new kind of computer, where like, you wouldn't expect that person to have access to a data science team, but now they don't need one because they can just talk to the intelligent copilot in their pocket, and help, and it helps them reason over, you know, decisions they're making that really impact their life. So for me, it was like a realization that we might have discovered one of the, one of the most important technologies that is, is on, is on like our doorstep. This idea that the abundance of intelligence and that we can access it through natural language, whatever language you speak, that, that was mind-blowing to me. And so then I started working on GPT-4 side projects every single day. I'd have it teach me all kinds of things. I started off, I'd be like, "Oh, teach me how to clone my voice," and then it wrote the first version of that program and I was like, "Wait a minute. Teach me how to build a GPT-powered chatbot." And then it wrote the first version of that program, and then I was like, "Wait, like even sticks and stones don't tell you how to make fire." But intelligence is something else, and we had layoffs and, uh, I thought that was the best thing that could ever happen to me, because then I was like, "Now I get to do this full-time, and just spend all my time and money seeing how far this rabbit hole goes." And so I explored all, all of generative AI that could, I could get my hands on. Everything from Midjourney to other language models, to even Runway and PikaLabs. And then I started building copilots for almost everything I could think of, so from music production to copilots that I'm working on for game design, to just like help me, you know, clean my apartment. But the way I see it is like, if you, if, if-
[20:16] speaker_0: How does that help you clean your apartment?
[20:18] speaker_9: Well-
[20:18] speaker_9: ... I have, I have a custom GPT and I just aim it. I just take a picture of a mess, and then it's pre-wired to just look at the mess and then suggest ways to clean everything up, and then things that I might need to buy to organize my life. Yeah.
[20:34] speaker_0: Wow. Does the R- Is the Room- Does the Roomba get jealous?
[20:37] speaker_9: The R- Oh my god, I have, I have a Roomba and I have two cats, and I'm pretty sure my cats don't think the Roomba's intelligent, because it gets stuck in the corner and I'm just sitting there like, "Why can't I just tell you to just, like, just go home? Like, you're done. You can't do this. I'll, I'll do it." (laughs) But actually, I think in the future, um, we're going to see, eh, especially with the new GPT-4 model, as developers start and, you know, putting it into everyday technology, all, a lot of inanimate objects we today perceive as inanimate, in the future, will have conversational, um, experiences and like, personalities. And that'll make them more useful. Yeah.
[21:13] speaker_0: All right. Are you ready to show the video?
[21:14] speaker_9: I, I think there might be a video, but we'll see. This is a, I think, if he's playing this video, it's, it's a video, it's a demonstration of a v- of a, of an AI twin, uh, that I worked on. Oh.
[21:31] speaker_0: Is it volume?
[21:34] speaker_9: There's no audio. (upbeat music)
[21:41] speaker_8: ReidAI, introduce yourself to everyone.
[21:43] speaker_11: Hello, everyone. I'm thrilled to be here today. I'm an AI-generated version of Reid Hoffman, his digital twin. My thinking and everything I am saying comes from a custom GPT that is based on 20 years of Reid talking and making content. So I've read all his books, watched his speeches, and listened to his podcasts. I think it gives me a pretty good sense of how you approach the world, Reid.
[22:06] speaker_8: So first, I'm going to challenge ReidAI with some questions to see what it can do, and how good its content is in mirroring what I might say.
[22:16] speaker_11: Then the tables will turn, and I will challenge you with some questions about AI and tech.
[22:22] speaker_8: Of course, there's a second question, whether or not I'll answer.
[22:25] speaker_11: Well, okay. Challenge accepted.
[22:27] speaker_8: Let's get started. Let's take a real piece of work for me, Blitzscaling. I wrote a 336-page book, and how do you explain it in one sentence?
[22:41] speaker_0: Nice. Shout out to the AV team for that.
[22:44] speaker_9: Y- you'll have to, you'll have to go to YouTube and look it up to get the whole video.
[22:48] speaker_0: (laughs)
[22:48] speaker_9: And it's, I think it's worth watching once because GPT-4, like, there's a GPT-4 agent powering that avatar. The avatar's not real time, but eventually they will be. I think this is a demonstration of where the tech, commercial-grade tech is today, and I think, uh, as we get more advanced avatar technology, you can imagine eventually in g- video games, intelligent NPCs you can speak to that feel lifelike, grounded in some primary source data. But, yeah.
[23:16] speaker_0: Excellent, excellent. And with better hair. I noticed that his, his, ReidAI's hair was a lot more...
[23:22] speaker_8: Yeah. Volumetric.
[23:25] speaker_11: (laughs)
[23:25] speaker_0: Volumetric. Okay, all right. Um, excellent. All right, so Edward, let's go back to you. So, uh, Edward was, uh, was, uh, the Obama digital campaign, was one of the co-founders of Oculus Story Studios, has won Peabodys and Emmys, correct? Yes. Um, and hi- is now working on, in addition to this, a number of A- AI cinema.
[23:50] speaker_1: Yeah, I mean, I, I love what you, uh, what you were saying about that, and we think about that a lot. Um-
[23:56] speaker_0: And some of the people that you're working with are here.
[23:57] speaker_1: Yeah, Scott and Katja.
[23:59] speaker_0: Isran, that's right. Nice.
[24:00] speaker_1: Um, making, uh, AI movies and, you know, I think that loneliness is going to be a huge problem. When you were talking, I was just thinking about Brando and Bertolucci and Last Tango in Paris. It's kind of hard to see how you're going to make a movie, a great movie, that, uh, doesn't have that conflict, uh, between the actor's vision, the director's vision, the DP, the editor contributing something. So we've, just on that topic of, of, uh-... of it, we been thinking about AI crew members and, and how to kind of grow those in simulations so that they have, um, agency and are able to push back, uh, 'cause so much of that push and pull is important to creativity.
[24:44] speaker_0: So, so let's talk a bit about the... you know, so you also, you, you, the c-... you've created Showrunner. And so, how are, how are you thinking about y- y- y- you know, like, in a... bringing some of the agents into... a- are, are you bringing those into the cinema, or, or are you... are those separate, separate projects?
[25:02] speaker_1: Yeah. I th- I think in both cinema and TV, having AI crew members is really exciting. Um, I think sh-... what's unique about Showrunner is, you know, people have started saying platitudinously, "Don't worry. AI's just a tool. It's like VFX." But my hope for Showrunner, eventually, is that there'll be no humans at all, and it will just generate amazing content for us from the simulation. On the other hand, I also am very excited about movies that are made by a writer-director, by a visionary. But we have to be honest about the difference between content and art, and a lot of us consume a huge amount of content. I rewatch lots of shows that I love. I think it would be very much better to watch some new episodes of shows that I love, or new shows that are in the kind of area that I enjoy. That's content. That's something I do to relax and to kind of, you know, have time. It's a lot of what we actually watch. And then there's art, and I think I'm very excited to see writer-directors who maybe they can't get their movie made. Maybe they are more like painters or novelists. They like working alone. Obviously, Julian Schnabel is able to kind of make that transition. But a lot of, a lot of filmmaker... a lot of people who might make amazing films, we've never seen their movies because they can't command a set. They don't feel comfortable working with 30 people or 40 people to get their vision out. So, on that artistic side, I'm ex-... I'm excited to see writer-directors who can make incredible movies that will, um, really shock us and take us inside of their very strange vision of the world, like what David showed. So, I think it's gonna revolutionize content, and it's also gonna revolutionize art.
[27:01] speaker_5: Can I respond to that?
[27:02] speaker_0: Please.
[27:03] speaker_5: Yeah. I mean, um, as a filmmaker who's worked in the studio system and independently, one of the issues that I think we are constantly... you butt up against... um, and actually in the previous panel there was a really interesting point made about, "Well, what if you scrape the blacklist, which is all the really great, you know, uh, unused scripts?" Well, so there's this thing that happens where market forces, um, come to play. And this is really the reason (laughs)... it's not... you see, with enough... not enough ambition or whatever. You know, you, you do your 100,000 Gladwell hours and you get to a certain point and then you... then it's up to you and you go and, yeah, it costs a lot of money to take 1,000 people into a f-... into a desert with 20 cameras and do a massive explo-... you know, explosion sequence or whatever it is you're going to do. And, and I don't miss that. I, I, I think that'll be the first thing to go. I think it's, it's, it's inevitable that there'll be vanity projects for Christopher Nolan. We love Christopher Nolan. I love Christopher Nolan. Um, I love his movies. But he's the only person that can do that now. He's one of the very few people who can prioritize and have the power over the market forces to shoot film, which is verboten for most filmmakers. Um, and, and then to actually have the intelligence to say, "I need 100 millions worth of prints and advertising, um, so that somebody sees my film and it becomes a financial success." And so, these are the tides that push filmmakers right now. And so, what I'm looking at, i- in an optimistic manner, is guaranteed production values. We can make your picture look good. But if you've got tons and tons of experience, you can go in and tweak that and make it like exactly how you want. But the main thing is, we're not gonna go to a Nielsen testing audience to look at your film, um, and tell us whether it's any good or not. We're just gonna trust you and it'll either be entertaining or it won't. And that's how more films get made. Now, my fear is... my great fear is that there will be an AI version of the Nielsen tests and creativity will become a commodity that can be, um, analyzed and quantified. And this is the, this is the kind of
[29:32] speaker_1: The dream.
[29:33] speaker_5: This is the edge that we're kind of traversing right now.
[29:36] speaker_0: I mean, there, there's companies that... like that, that already exist.
[29:38] speaker_5: Yes, absolutely. Yeah.
[29:39] speaker_1: Just, just to that point, on this side of content, um, which is very different to art-
[29:44] speaker_5: Mm-hmm.
[29:44] speaker_1: ... but it's what we actually spend a huge amount of our time consuming. I go to Netflix-
[29:48] speaker_0: Does everyone agree with that, that content is different than art? Is that-
[29:50] speaker_1: Yeah. May- maybe not.
[29:52] speaker_5: Uh, I... well, it's all interconnected.
[29:54] speaker_0: It's all subjective.
[29:55] speaker_5: You know, we're all made of stars, aren't we?
[29:57] speaker_1: Let, let's... t- to the, to the point of content, hopefully, eventually, um, generating movies and shows will be as simple as browsing Netflix.
[30:06] speaker_5: Yeah.
[30:07] speaker_1: So, right now you have a delay of three years of Netflix noticing that people really want more movies like Event Horizon. It's kind of amazing. There's still like-
[30:17] speaker_5: Yeah.
[30:17] speaker_1: ... the, the peak of that particular genre. Um, and then, you, you know, you, you get a glut of these movies. Instead, you might, in the future, just search, you know, Event Horizon and it's more like this. You've got some posters. You click on the poster of the thing that looks a bit like it, and you get a movie or a show like that. So, that's from the showrunner perspective.That's a provocative thing, but I think AI is getting so good so quickly, that it's a plausible future. And then on the artistic side, there are so many people who are just more suited to novels, my mom was a novelist, um, or painting, or solitary art, and I'd like to see them in movies, because right now, I'm not excited about a lot of what's in the cinema. I'd like to see the movies of people who are too strange to operate within, um, the structures of Hollywood. So I think it's going to be terrific for content and what we a- watch on Netflix on a Thursday evening, and it's going to be terrific for art and the things that actually fuck us up inside, um, so I'm excited on both sides.
[31:26] speaker_5: Can I respond just one more time?
[31:28] speaker_0: (laughs)
[31:29] speaker_1: (laughs)
[31:29] speaker_5: No, it's fine. I'll wait till the end.
[31:31] speaker_0: Please do. Please do. Just do it. Just do it.
[31:33] speaker_5: Oh, just do it. Okay, um...
[31:34] speaker_0: The audience asked you to.
[31:35] speaker_5: Th- okay. All right.
[31:36] speaker_0: The audience, it was an audience request, yeah.
[31:37] speaker_5: So the other side, and the concern that I have to that, um, spec- the spanner that I can throw in the works there, is that if you look at, if you have a kid, and I have a seven-year-old, and if you look at market force-driven content that has no cultural relevance at all, you can find it. It's on YouTube Kids. So if you go on YouTube Kids-
[31:59] speaker_0: (laughs)
[32:01] speaker_1: (laughs)
[32:01] speaker_5: ... you will find endless skits done by parents pimping out their children and doing skits to make revenue off of YouTube, which I think is a really, a devolution, not an evolution. Yet at the same time, taking away the cost of making a production and then allowing people who can train and learn is explosive and amazing and incredible. So it's, it, it's, it's got a dark side as well, I think, this, this whole aspect of it.
[32:30] speaker_1: Yes, I'm sorry, guys.
[32:30] speaker_0: All right. We're gonna move over here.
[32:32] speaker_1: Yeah.
[32:32] speaker_0: Come back. We'll come back.
[32:33] speaker_5: Sorry.
[32:33] speaker_0: All right. Sh- (laughs). No, I, I have a lot of follow-up as well, but we're gonna, we're gonna head over to this side. So, Qiying, talk, so, um, talk to us a little bit about some of the work that, that, for example, the, what you're doing with, with, um, uh, wait, it's like, what's, um, Mi- what's it called? M- the M- no, no, Altara's a company. The, um-
[32:56] speaker_7: Uh, digital human being, or?
[32:57] speaker_0: Yeah. No, but with like, with the, um, inside the games, like how does, how do- like, how does the experience... Minecraft. That was close. Thank you. Um, inside Minecraft, h- you know, what is the experience like, and how are, you know, how are you hoping... You know, I think you're s- you're still early in the process, so how are you hoping that audiences react to it, and how are you hoping that they, they use, they use these interactions? We talked about To Love, To Live, but just talk to us a little bit more about the, you know, the hopes for it and what kind of early feedback you've gotten.
[33:28] speaker_7: Yeah, yeah. Um, uh, yeah, I like the topic that you're talking about. I definitely wanted to get more into, like, what does it mean to really love and grow with this? So, the reason we choose Minecraft is because Minecraft is like a very open-ended, very, like, composable world. Basically, everything can be crafted, everything can be recreated inside the, inside the game, and, and the agent, like, any user in it can create everything inside the world. Like their's, we can say that like a lot of games, there's kind of like gates here and there where the NPC cannot do this or that, but Minecraft doesn't have it. That is why we choose Minecraft. What we wanted the users to get out of it is like, they would feel like they're probably playing with another friend they have, like that friend is not a superhuman being. It's not like a bot that can craft the world immediately so fast that no human being can do. They're probably not even know how to play with Minecraft, but they will react like a friend would do. They will probably talk, like, converse with you during the gameplay, not about the game itself, care about you. If you c- play with them for, say, an hour and come back a week later, they will remember that experience and probably introduce you to some other human being that they have played with, or some other agents that they have played with and that experience. And so, we are hoping is that when the user come in, they would open up a, like a different form of social interaction. They not only have more fun playing the game, but they would have more fun finding, like, some emergent behavior just by interacting with their agents and they would, they would kind of like... What we are hoping is that we can bring this, like, frien- friendship, their, how we, how we are able to show a little bit of vulnerability, that's how we make friends and kind of like form deep relation to make this happen easier in the virtual world.
[35:46] speaker_0: Great. So it's, so it's addressing a bit the, the loneliness e- epidemic?
[35:50] speaker_7: Sorry?
[35:51] speaker_0: It's addressing the loneliness epidemic?
[35:53] speaker_7: Um, it could be.
[35:55] speaker_0: Mm-hmm.
[35:56] speaker_7: And it's also, I think, it's not only the pandemic, but like, it's also just growing up, like, when you're getting older, it's harder to show vulnerability.
[36:05] speaker_0: Yeah.
[36:05] speaker_7: And, and that is, I think, like, that is part of what I think AI and technol- technology can help to bring this down.
[36:15] speaker_0: So it's also interesting, 'cause you said that the, the, the AI can also introduce you to both other people as wether- as well as other, other AI, other agents in there. So it can, it can help you draw, drive connection both with AI or NPC as well as other, other humans.
[36:34] speaker_7: Yeah, exactly. So, as I, um...Part of, like, a very important human quality is that people form long-term memories. Like, the long-term experiences really form, like, their fundamental or different personality characteristic for humans. So, we believe that for the agents to really be kind of ... The agents should be having consistent long-term memory, and that is how we can make them coherent, and that is how they can help other, so-
[37:05] speaker_0: W- would you, would you, uh, hold the mic a little closer to you? Uh-
[37:08] speaker_7: Oh, sure.
[37:09] speaker_0: Yeah.
[37:09] speaker_7: I'm sorry. (laughs)
[37:09] speaker_0: Thank you.
[37:10] speaker_7: Um, so going back to the question, sorry. Um, oh, yeah. We believe that the long-term memory is there int- integral part of a human being, so we wanted to bring that to the agents. So, that is what can help them to, like, help form more social relationship with other, because they remembered you, they have a thinking of you, and that thinking will continue to change as they interact with more, like, with other environment or, um, human beings.
[37:44] speaker_0: And then that overall, uh, enhan- enhances your, your experience playing Minecraft or, or whatever-
[37:51] speaker_7: Right. Yeah. We, uh-
[37:52] speaker_0: ... whatever game you're inside.
[37:53] speaker_7: Minecraft is our starting playground because it's like very composable, so it's like a very good experiment filled with, like, many different possible behavior. And yeah, it can also go beyond Minecraft. It can talk to you in Discord or, or see your posts on X.
[38:11] speaker_0: Great. Thank you. So, Parth, you're, you, uh, well, Reid, Reid's well-known for having worked very closely with OpenAI and work- worked closely with Inflection AI, as, as, as well as many other c- many other companies, of course. What was it that made you guys decide, like, now, now is the time to make a, a Reid AI and release it?
[38:29] speaker_9: Um, I think, well, when, when OpenAI released custom GPTs back in November, I was already building, like, I had been already building custom GPTs on the API for, like, eight months at that point. And so, they just made it really easy, and the models have larger context windows. So, a year ago, we were working with GPT-4 8,000 token context, and now we're at GPT-4 128K tokens. And then, you know, the new Gemini model has 2 million context windows. So, the, the working conversational memory of these models is getting very large now. And you can imagine, like, we will continue, like, peop- they'll keep solving problems to make it possible for theoretically infinite context, is the theory. But why now? I guess we wanted to demonstrate what was possible with commercial grade technology. So, like, custom GPT is, I would say, like, that's, that's your commercial grade, um, like, AI agent lite. It's not proactive, but there's a number of things that, like, you can do with agents when you build them even more custom than custom GPTs on OpenAI. And then for the video avatar, even that's a consumer application. Like, that's not ... There, there's more expensive video avatar technology that's better in some ways, maybe gonna be applied in Hollywood, but that's something that you can subscribe to. There's an app called Hour One. You can make your own avatar. So, we wanted to see what was possible with, like, a pretty small budget, a decent idea on, you know, building an avatar, on a guy who's written five books, has given numerous speeches, is on a couple podcasts. Can we create a character? And, and I think of this, like, when you're building agents, I think w- it's really, like, as someone that designs an agent, I think of it as character design. And there's, like, different components to character design from this lens of taking a language model and getting it to pretend to be someone, referencing primary source data that it has access to, using tools. And then, and then, like, the resulting effect is this, like, illusion of, you know, a person. But I think the bigger takeaway is when you try to replace someone, you discover just how multifaceted they are, and just how hard it is to replace a person, their, their actual ... Like, like, I, I woke up in the middle of the night and had an anxiety attack once because I was like, "Man, I'm pretty sure ChatGPT is better than me at my job."
[40:44] speaker_12: (laughs)
[40:45] speaker_9: And, and then I, and I, I talked to my teammate about it, and she's like, "Parth, I'm pretty sure this is better than our CEO."
[40:52] speaker_12: (laughs)
[40:53] speaker_9: And then I was like, "Oh, wow." Like, I think everyone has some of the sentiment of like, "I just need ChatG-" Like, you know, me and ChatGPT against the world, which is actually a new meta game. Like, you and your AI, your system of AI tools will allow you to reach into multiple skill sets now that you previously would have had to dedicate a separate career to. And so, like, I can make a music video using Midjourney and Runway in two days. You know, I talked to GPT-4, and it's like, "Yeah, you just get this editing tool. Here's ideas." And it, you know, you have these iterative conversational experience that goes through the idea, makes it, you realize it's real, and then you go make it happen, and you realize you have a new ceiling, and you're no longer just a data analyst at a startup. Like, you start visualizing yourself as something else. And, um, he's not here, but there, there is a guy, um, who kind of, like, he saw what I was working on, and he's, uh, James Blevins, and he, he, he described what I was doing building agents. He's like, "I think this is actually a new kind of art." And the more I think about it, it is, it is like character design. Like, it's, it's, it's h- it is, it's like, what do you, how do you want it to act? How should it present information? What are its tendencies? And then, like, you know, where does it exist in the world? How does it touch people? Does it leave an impact on them? And, uh, more and more, I'm seeing AI agents as, like, an art form than a science. Yeah.
[42:12] speaker_0: So, this, uh, we're gonna go to questions, um, as next, so, so get ready with your questions. Since this is AI on the Lot, you know, how do, uh, you, what, what one thing would you want the audience to be thinking about around agents and simulations and storytelling, you know, whether, whether they're creators or IP holders or tech companies? Parth, you wanna start?
[42:34] speaker_9: Um, I think, I think you're gonna see agents of all kinds-... you know, throughout society. They might help you run your business, they might help you solve a creative problem. And I think this is potentially the biggest creative augmentation in human history. Create, like, the, it- it's hard for me to compare this to any other thing that people have done, but the, the version of you when you're using tools like this, your creative limit, your potential, it might be hard to visualize, but you should maybe throw this, throw these technologies at things you're actually passionate about. Don't think about it as, like, a job thing. Think about it as, like, an exploration. 'Cause what you, what you do for fun, what you're passionate about, like, no one else has to tell you, and then you might have a high bar for what success looks like. And when you use these technologies, you realize what they're not good at and where they fall short, and then you come in as, like, that grounded person in the real world, and you say, "Hey. Here's, here's my opinion. Here's my subjective taste." And I think your ability as a curator is really important because, like, you know, you can generate 500,000 clips, but someone's going to have to decide what people end up seeing, and that decision is actually really important. It's grounded in your lived experience, the people you've met, and, like, the life you've led, and that's unique to you. So, so, like, keep that, because that's actually, like, the sauce, you know? So.
[43:58] speaker_6: Great. Thank you. Shuying.
[44:04] speaker_7: Um, this, this is such a hard question. Um, I, I think that, um, for a-agents and simulation, what I'm hoping to see is whether there's some emergent behavior that we could observe in this type of simulations because I think that, um, the machines and computers just open up so much possibilit- like, so many more possibility in it that, like, really cannot... There, the, the amount of compute creates more possible scenario than what we can see-
[44:43] speaker_6: Mm-hmm.
[44:43] speaker_7: ... in, in kind of like a, just, like, regular human society, so I'm really hoping that it would just, like, from there emerge something that we have never seen and, uh, thought about before. Um, on the... So I'm more coming from a tech background, so I think it is our time that more con- it, this may become con- uh, controversial. Um, but as for, uh, people coming from tech background, I think at this time, when something like... I actually talk about that (laughs) with Parth just before this, um, that things like consciousness, personally it's just, like, kind of a more, like, a philosophical kind of questions, but I think we can start to see whether those things are, can be scientifically benchmarked and evaluated, like, from a very basic sense and we can build from, like, the, I think it's a time when, like, from the ground up we can start building some kind of scientific measurements around it. Because, um, in, in our human brains and, like, and actually in different animals, like, their neural circuits of, like, how you feel happiness and how you feel about, like, um, the different basic emotions like fear, it's well studied. But, uh, it's not like it's... I think it's now a time that we can start experimenting them on, like, machines and see if we can replicate that with some type of models.
[46:15] speaker_6: Great. Thank you. All right. We need some quick answers.
[46:19] speaker_5: I have no idea.
[46:20] speaker_6: Excellent.
[46:22] speaker_13: (laughs)
[46:22] speaker_6: Edward.
[46:24] speaker_1: Um, I, I would just recommend anyone working on agents to try to, um, bring together the light and the dark, in that so much of the tech companies who are trying to build these, uh, agents do not, there's not an artistic understanding that an actor might bring to it or that a novelist might bring to it. We're, we're all very messy people in our heads, and so if we're trying to make real agents, we can't just have them all be very nice to each other. We've got to think about the light and the dark.
[46:55] speaker_5: It seems... Sorry. I will let you. It does seem to me that actually one of the things to focus on might be just, uh, modeling charisma.
[47:04] speaker_1: Yeah. That's super interesting.
[47:05] speaker_5: Because there are, you know, actors who have this one eyebrow, and that's what they do. And that left eyebrow's got a bet- different agent than the right eyebrow and it makes more money 'cause the eyebrow's the moneymaker, right? But it's not that simple. No, of course it's not. But it's charisma, which is an unknowable thing. It's what we do when we do casting. It is not magic. Uh, it is very much, you know, uh, i- in the world of, of, of, uh, objective, it could be objectively quantized, um, but that, I think, is more interesting than the wide open thing of a person.
[47:38] speaker_1: If I could say one last thing before we go to questions. To, to that very point, David Thomson, who's a brilliant critic, wrote about Cary Grant, that he was the greatest actor of all time because he had that light and dark. He was unpredictable.
[47:51] speaker_5: Yeah.
[47:52] speaker_1: He was potentially sadistic and he was also extremely funny and lovely, so, you know, you need that kind of unpredictability in your agents.
[47:59] speaker_6: Okay, great. All right. We're going to Q&A.
[48:01] speaker_5: We've got maybe one.
[48:03] speaker_6: All right. Todd's gonna-
[48:05] speaker_5: All right. I think we got one, time for one question. I saw this person first.
[48:11] speaker_14: Thank you. Fantastic. Thank you so much. I'd like to know a little bit about your AI text stack. I was wondering, David, were you using Runway, um, in yours?
[48:20] speaker_5: I used, um, uh, Runway, Midjourney, and Pika.
[48:23] speaker_14: Pika. And, and, and Edward, if you could say anything about-
[48:27] speaker_1: I can follow up with you on it.
[48:28] speaker_14: Okay. Thank you. And I wanted to ask you, Shuying, yeah, 'cause I, I think it's got PhDs in it or something. (laughs)
[48:33] speaker_9: I used Runway, Runway Pika. Uh, GPT, and yeah.
[48:38] speaker_14: Yes. Thank you so much.
[48:38] speaker_6: Okay. All right. Just, just go down the road, quickly say the way people can get in touch with you.
[48:43] speaker_1: Uh, on Twitter, we're, uh, Fable Simulation.
[48:48] speaker_5: Uh, I'm on Instagram, my name, uh, middle initial A, and that's about it. I have an agent, but don't call him.
[48:58] speaker_13: (laughs)
[49:00] speaker_7: Uh, yeah. Uh, you can find us at atera.al. Um, it's A-L.
[49:07] speaker_14: Great.
[49:10] speaker_9: Um, you can just look up Parth, last name fire emoji, on, and you'll find me.
[49:16] speaker_6: (laughs) Excellent. I'm Joanna Popper, at, uh, at all the different Poppers. Thank you so much. This was a great panel.
[49:20] speaker_5: Thank you.
[49:20] speaker_6: I know there's a break now, so-
[49:21] speaker_5: Thank you.
[49:22] speaker_6: ... so if, they'll, we'll still be here to answer questions for those of you who have them. Thank you.