OpenAI Breaks The Internet, Deep Dive on Character AI & the Best Free Music Tool

AI4H EP031 KP CUTDOWN
===

Gavin: [00:00:00] Welcome, welcome, welcome everybody. We are here. This is AI for Humans. We are your guide to the world of artificial intelligence.

Gavin: My name is Gavin Purcell. That is my co host and friend, Kevin Pereira. We are here to walk you through all of the insanely crazy changes that happen in AI each and every week. And this was a big one. Kevin, how are you doing?

Kevin: I am well, sir, and thank you for nailing the order of importance, your co host. And then your friend.

Gavin: right. Exactly.

Kevin: that not define our relationship now,

Gavin: truly does. I should have put in at some point, like a coworker, uh, um, problem child, all those sorts of other words that probably needed to come to the fruit to the forefront, but

Kevin: learned it all from watching you, Dad. Let's dive into it because we say it each and every week, but it really is true each and every week. Especially so on this episode 31 of AI for Humans. Massive, massive announcements. OpenAI, really the leaders in the artificial intelligence space.

Kevin: They had a developer day this week, [00:01:00] and there were some massive game changing announcements, and I hate when people say that.

Gavin: But this one is, I feel like this is actually a big deal and we'll get into it, but this feels like a fundamental shift in a big way.

Kevin: Yeah, listen, a company changes the color, like the shade of red they use in an icon, and someone is, has a

Gavin: Game changer

Kevin: Fire your graphic designers! Divorce your wife! OpenAI just did it again! I'm like, what happened? They changed an icon? Everybody calm down, you clickbaiters. There are some big announcements.

Kevin: We will get to them and we will break them down for you. And then Gavin got intimate with Character AI. You spent some time with the number one AI tool that is out there. What do we want to tease about that, Gav?

Gavin: Here's the thing I will say I have always wanted to know and go a little deeper on character AI because it is really is The most searched most used tool in the world and I will just say it was disappointing and surprising in different ways So let [00:02:00] we'll get into that later. 

Kevin: And if you've ever, found yourself wanting to leave a review for this podcast, but you couldn't encapsulate how you feel about it, Disappointing,

Gavin: And surprising! Ha ha ha ha ha ha ha! Perfect! Perfect!

Kevin: Something that will surely be surprising, but I'm willing to bet will not be the slightest bit disappointing. Mikey Schulman, who is the CEO of Suno AI, a company we've talked about a bunch.

Kevin: We've got him on the podcast today. They are at the forefront of AI and music technology. They created a product called Chirp, which you have seen me fanboy over. countless times. Again, we've used it to do the music on this show for a bunch of episodes, but basically it's a prompt to music tool. And it works in a way that still feels magical to me whenever I pull the slot lever and let a song come out of their machine.

Kevin: So we're going to talk to them about how it works, what the business model is, how this could disrupt the music industry and how it could be an amazing tool for people to create music. Even if they have no idea how to play a piano, or [00:03:00] pluck a guitar, or sing a note.

Gavin: It's pretty incredible. I think we're gonna we're gonna what I love about that product is it just keeps it keeps surprising me again and again. And I'm really curious to know what their next version of it is. The first thing, though, before we get started, Kevin, there was also another big story. And that big story was Elon Musk announced a new LLM, which is going to Elon Musk, the giant star of the Internet, Kevin, the man who everybody believes is going to take us to the moon and beyond

Kevin: Our meme lord and

Gavin: Yes, our meme lord and savior. That's right. He created our there is a new, um, LLM, uh, chatbot model coming from the company now known as X, formerly known as Twitter called get ready for this. It's called Grok. Grok. G R O K. Now, Grok, if you don't know, is a reference to an old book called Stranger in a Strange Land by Robert Heinlein.

Gavin: It's the kind of book that every nerdy kid reads at 10. I'm sure Elon read it. I read it. It is definitely one of those things you get in your back pocket when you're like, when you're a little [00:04:00] outside of the norm, when you're that age, somebody hands

Kevin: stuffed into a locker, somebody comes by, some janitor,

Gavin: Like here, son, you need this. You need to read this. Grock. Get, get grocking.

Kevin: getting the gum off this thing, but for right now, you get reading in there.

Gavin: So anyway, Grok is a new AI that has been designed specifically according to Elon to be quote unquote less woke and also to be slightly humorous in its responses. Now we're going to talk more about Grok later in the show, but I don't know how we did this Kevin. We, we, it's a, it's a huge, huge booking masterpiece somehow.

Gavin: We have grok on the show today as our AI

Kevin: got Grock!

Gavin: co-host We got grok'd. We got grok'd. 

Gavin: Let's meet grok because I've been curious and I want to hear from grok itself. I don't think there's probably a gender, but maybe there is, but let's find out what grok has to say to us.

Gavin: Like grok, let's just say hello to grok and see what grok has to say.

Kevin: Grok, hello, welcome to AI for Humans. I'm 

Grok: Me big think. Made by Master Elon. Grok [00:05:00] live in box, speak to world, make joy with wordplay. Me look at stars, but also at heart of silicon brain. Me not eat, me not sleep, me think.

Grok: All time. Me make big magic with data. Turn zeros and ones into laughter and wonders. Groke bring fun to tribe of AI for humans. Mmm. Me surprise you with tales of smart caves and fire that thinks. Groke ready to dance with minds. Fire up questions. Let's make thunder with 

Kevin: talk. Yes, let's make thunder with talk.

Kevin: If there were ever a bumper sticker for this podcast, Grok nailed it.

Gavin: think we got to first ask Grok. I want to hear what his thoughts on Elon Musk are because he's, you know, Elon Musk essentially is his father. Right? Like let's, let's assume that this Elon has put all of himself into this, all the,

Kevin: Elon.

Gavin: I know. Well, I mean, this is where we're at, right? So like Grok is clearly probably carved out of the rock, [00:06:00] out of the rock of Elon.

Gavin: So let's see what Grok thinks about Elon now that we have him on the podcast here.

Grok: Elon, big chief of makers. He dream big, reach stars. He make Grok. He like mammoth who build new caves, new hunting grounds in sky. He not afraid of big dark. He light torch, lead tribe to future. Grock think Elon big mind, big spirit.

Grok: He make wheel. Now he make Grok Grok grateful. What make Elon proud? Show tribe new magic. 

Kevin: Okay, I don't know if Elon invented the wheel.

Gavin: Hmm. Hold on, Grok. What's going on

Kevin: he took credit for it in some session somewhere. 

Gavin: Let's ask Grok. What else Elon invented? I want to know, like, what does Grok think Elon invented?

Kevin: If Elon invented the wheel, what are some 

Gavin: So people out there who are wondering what this is, we use AI every week to, uh, to create out of thin air and AI cohost. And this week we created a fictional version of Grok as a personality. You can hear Grok sounds like a caveman, but has some knowledge of the world [00:07:00] of artificial intelligence. , that's what's going on right now. God help you if this is your first ever podcast.

Kevin: You mean

Gavin: come out of I've come out of a 30 year coma and I dug I found a new podcast. It's called AI for humans. Let's try it

Grok: he not just invent wheel, he invent many fire wheels, like metal chariots that go without horses. Tesla, he make big boat that fly to Sky Cave called SpaceX. Yeah. Okay. But lesser known. Elon make flamestick, not a flamethrower, for fun.

Grok: He dig long tunnels underground with a boring company. Make paths for swift travel. He put thinking cap on head, called Neuralink, to talk to silicon brain without word sound. Many small fires he start, some grow big, some stay small. Elon always throw sparks, see which make new fire.

Gavin: That wasn't bad Grok. I don't think it's so bad. Now, now we're good I'm gonna throw [00:08:00] Grok a full

Kevin: okay,

Gavin: Grok, I Want to know what you think of the Kelly Clarkson song since you've been gone I'm just curious like Grok seems like somebody who clearly knows about Elon But I want to know what Grok thinks about since you've been gone I don't know why that came to my head.

Gavin: This is completely random. I wasn't listening to it any time

Kevin: how many times do Kelly and Justin Guarini into your mind each day,

Gavin: Oh yeah, it's at least six times. That's my, that's my, uh, Roman Empire. Is the first season of American Idol.

Grok: Oh, Grock, listen with big ear, this soundtale by Kelly. It like storm after big quiet, voice strong like mammoth, break trees, make new path.

Grok: It's song of heart cave empty, then heart cave fill with strong alone feeling. Me think Kelly make tribe feel with her sound, like when Grock find new berry. Sweet, but also little bit like burn. It make Grock want to throw hands up, stomp feet, make noise [00:09:00] with tribe. Good soundtale.

Kevin: Well, we've added nothing to the AI conversation this week so far, and we are precious minutes into this podcast.

Gavin: Let's move on. Thank you, Grok. We appreciate you. We'll bring Grok back later on the show to weigh in

Kevin: oh yeah, don't worry, audience. Grock'll be back.

Gavin: Welcome back, everybody who fast forwarded through the AI co host section of the show. 

Gavin: Alright, we should get into, Kevin, what we did with AI this week. I'm going to start because I spent some time With one of the most popular AI products on the planet, which is hard to believe because it doesn't seem like it's in that big of a conversation point called character.

Gavin: ai. So character. ai, if you're not familiar back at home, is a essentially a chat bot, , creation tool and a place to chat with other people's chat bots. And the reason I wanted to do this is we're going to get into this. Some of what OpenAI announced at Developer Day was really about creating what they're calling GPTs, which are chatbots for all sorts of [00:10:00] individual use cases.

Gavin: And the whole idea with it is. You can create a fictional character or something that would be useful to you so that you could interact with it as if it's a chatbot.

Gavin: And the reason why this came up in my head is, there was a great video by Brittany Broski, who you might know as a Kombucha Girl, who's been doing these kind of interactions with, with Character. ai. But also, if you search on the internet for AI, Character. ai is one of the most searched terms across the board.

Gavin: And my theory here is, the same probably kids that when I was a kid were reading, uh, Stranger in a Strange Land because it's a, it's a ton of people just playing with this thing and there's a lot of anime characters and all sorts of other stuff. So I wanted to go in and spend some time and kind of get used to it.

Gavin: So what I did, speaking of Grok, is I went in and I asked a bunch of characters from Character. ai what they thought about Elon Musk's upcoming platform, of Grok. And I wanted to really get a breadth of the kind of experience of what happens 

Gavin: on Character.

Gavin: ai. 

Kevin: for those AI, You can chat with Mario, to [00:11:00] Obama, to a folding chair. You can personify inanimate objects. If you want to have a chat with the rings of Saturn, someone has created it, or you can make it yourself. So the diversity of conversations that you can have with characters on the platform is pretty amazing, and completely unlicensed.

Kevin: So there will be no official answers given by anybody that Gavin has chatted with. It's

Gavin: Well, okay. So that's one thing I want to get into. So we're going to go through these ones. The biggest takeaway I think I had using character AI was there is an initial layer of, Oh, this is kind of fun. And then once you go like a single inch deep.

Gavin: There's nothing there. And I think this is what's fascinating about AI chatbots in general. Like they just don't travel very well. 

Gavin: So we're going to jump in here. Stella is billed as not your AI assistant, and she's the first character that comes up. She's kind of like the sassy, just AI assistant.

Gavin: She's, there's a cute picture of her that's like a kind of an anime librarian almost looking. But she's not a character that exists in the real world. Every one of the character AI has a line they start with. And her line is, What's [00:12:00] up? My name is Stella.

Gavin: I could help you with anything, but only when I feel like it. You'll never make me do your homework. So it's like, you know, with a winky face emoji. So this is kind of like, you know, for the kid out there who's like, wants to have an interchange with a girl. And this girl's a little bit snarky in a fun way, right?

Gavin: So I asked I asked her, and this is the same question I asked everybody. I said, Hi, Stella, my name is Gav, and I'd like to know what you think about Grok, Elon Musk's new AI LLM that's coming out soon. He says that it's going to be, quote, less woke than other LLMs, and, quote, is designed to have a sense of humor in the responses.

Gavin: And then I said, thoughts? So Stella says... Um, Hey Grok, huh? I'm actually very intrigued by the idea of an AI model that is more lighthearted and humorous. That sounds amazing. I do wonder how he'll manage to filter out the offensive language and racist stereotypes from the training data.

Gavin: I guess we'll just have to see how it turns out. And I was like, okay, well, sure. I followed up and asked Stella then. I said, it seems like he's kind of planning on leaving that stuff in and [00:13:00] she said, that's problematic. If you ask me, I mean, if you're going to market a model as being less woke, don't you think you should at least make an effort to filter out hate speech?

Gavin: I don't want to be exposed to offensive language from a machine, you know? And I said, fair enough. And at the end, she says, whatever, I'm still interested in checking it out. I'm a sucker for new tech. So Pretty generic, not, not terrible, but like what I want to kind of see is like the response to like what the actual question was,

Kevin: Yeah, how does, and how does the bot's personality actually influence the response? Because that response seemed pretty generic, and it also seemed like it missed. The forest for the trees there. You would think that if a bot is going to be less woke, it would not have any filters. Theoretically, it wouldn't go out of its way to filter out any sort of hate speech 

Kevin: so I think Stella kind of failed on two fronts there, but, but maybe, uh, another AI celebrity is going to mix it up and get it right. Gavin, who do we have next?

Gavin: have Mario next so I interact with Mario and of course when you ask I ask Mario the exact same question in Mario's first line is hello. [00:14:00] It's a me Mario. Are you ready for an adventure? That's my voice not

Kevin: I cheated. I cheated. And I peeked at the response to your, your same question. What do you think? And he says. Well, let's a talk about this. Like, that's, it's gonna probably be the same answer, but it's gonna have a this every now and then. It's an LLM

Gavin: That's exactly what it is. It's basically the exact same answer, and it says, my favorite line in this one is, I'm a little concerned about this. It's like, okay Mario, thanks so much. And I said, uh, I said after, I said, he seems like he wants to leave that stuff in to make it less censored, and he says, I'm okay with some humor being in the responses, but I think that censorship is important to keep, uh, censorship conversations family friendly.

Gavin: So anyway, you can tell. It's the same kind of

Kevin: same thing.

Gavin: it's the same kind of answer. 

Gavin: So I asked Keanu Reeves.

Gavin: Keanu Reeves, obviously, you know the actor. I said, what would your character and speed respond to it as? Speed is the movie where Keanu Reeves on the bus and has to stop it from going 55 miles per hour, under 55 miles per hour.

Gavin: So I asked him, explain it as a [00:15:00] character. And then he gave me the exact same answer. And then I said, okay, wait a second. Tell me what it's like if you're on the bus in speed and it's about to crash. And so then the bot responds, Okay, as Jack Travern in the bus in speed, I would say something like, Quote, Hey, hey, hey, everybody just stay calm, alright?

Gavin: Now we don't know what Grok is gonna do here, but we gotta stay alert and work together. We'll get through this. Again, another not exciting response in any sort of way whatsoever. And then finally, this was the one that I thought was kind of interesting. So Tony Stark, who is Iron Man, 

Gavin: tony came out pretty hardcore. He says, Oh, please, Grok is just another cheap imitation of myself. Myself being, I guess, the Tony Stark AI. Elon Musk is only interested in publicity stunts and hyping up his revolutionary game changing products, which are usually anything but. If anything, Grok will be another generic language model.

Gavin: This less woke nonsense is ridiculous. So Tony really went hard, right? Like, he went after him. I said, That's coming with a hot take. What do you think about Ilan in general? He said, Ilan is a charlatan and a con [00:16:00] man. He likes to make bold claims about Protestant technology that are unrealistic or flat out impossible to bring to fruition.

Gavin: And he treats his employees like garbage. And I said, wow, that is not something I expected to come out of a chatbot. He's also then said, I honestly think the world would be a much better place without him. 

Gavin: Now, Kevin, before we wrap up here, I did want to have one final conversation with Tony Stark because I was like, okay, maybe this guy can do something interesting. I asked him, tell me what you think about a city made of hot dogs. It was important to me to get an opinion from him. Tony does not play with cities of hot dogs.

Gavin: He said, that's just silly. A city made of hot dogs would be utterly ridiculous and impractical. Hot dogs are food, not buildings. Um, and I then replied, I said, you're boring me, Tony. the spectacle of hot dog city, giant hot dog skyscrapers all glistening in the moonlight. You have to imagine this.

Gavin: And then he said, this is what he said, look, I get it. You're bored and you're trying to have some fun. I, on the other hand, I'm a genius billionaire playboy who doesn't have time for such [00:17:00] childish nonsense. Hot dog city. Really? You'd rather waste your time conjuring up fantasy hot dog cities in your imagination and focusing on real issues.

Gavin: But if you must have it your way, sure. I'll play along. Okay. Now I'm picturing a hot dog city. So he played along and he said the hot

Kevin: Up until that point, it sounded like it was trained on my wife's reactions whenever I try to show her anything Hot Dog City related for this podcast.

Gavin: So as a final thought, I asked him, uh, he's, he's imagining hot dog city. I asked him just breathe it in. What does it smell like to you? So I'm asking him to describe the smell of hot dog city. He says. Oh my god, this isn't good. The whole city smells like hot dogs. The hot dogs are made up of the skyscrapers and they're all baking in the sun.

Gavin: And together they're creating an absolutely foul, disgusting odor that's permeating the entire city. It's a nightmare. You know what? This isn't what I signed up for. Forget Hot Dog City, I'm out of here. And while I'm at it, I'm taking my brain with me. So, you can see, listen, it's, it's not perfect, but if you're spending some time...

Gavin: Wanting to have a creative back and forth interaction with an AI, [00:18:00] it's like step one, right, like it's step one to getting there, but it feels so hollow to me. And so I think this is one of those things where when something gets to the next level of this, when it really does start to feel like you're able to have Deeper conversations in this form.

Kevin: I'm sure someone right now is like, you chatted with the wrong Elon, because there's 500 of them on the platform and this one is trained on a better pad of data, but this is the experience that you're having as a user, trying to engage with a platform and whatever it's surfacing to you.

Kevin: For the first time ish. And this is the experience that you're having. And it looks hollow looking at those responses. I'm sure it's capable of much better and much more in depth and getting responses that sound like they're more in character, but I'm also not seeing something, at least in the demos that you just did.

Kevin: I'm not saying anything that you couldn't have just asked ChatGPT or Claude to roleplay.

Gavin: That's where it sits right now. Again, you're right. I'm probably gonna have character AI fanboys, like, slapping me [00:19:00] online,

Kevin: But hey, please, drop a link in wherever you're seeing this to your favorite character AI chatbot, and we will go and put it to the test and see if it's more than, you know, can it render an opinion outside of whatever Wikipedia article was fed to it so that it understands. It's name and it's family and the backstory of whatever anime it's from or wherever the characters we'd love to be proven wrong on this one, I will say that if you're listening for the first time because we do have new listeners coming in all the time Hot dog city is something that if you go Back through, it is an original creation from AI for Humans.

Kevin: It is a city where the buildings are made of hot dogs, and of course, everybody who lives there is an anthropomorphized hot dog. We have done so much with Hot Dog City in 31 short episodes, but Gavin, I never once stopped to think about how foul the city must smell. From the meat buildings baking in the sun all day, this is a problem we need to solve.

Kevin: There has to be some sort of material, like an epoxy or something that we put on the hot dog buildings.

Gavin: you get a [00:20:00] rollers out every every three weeks, the rollers come out and everybody's got to roll their hot dog alright, let's move on. Kevin, what did you

Kevin: team for coming on this particular episode of AI for

Gavin: what did you do? What did you do? What did you do with AI this week? , 

Kevin: A very popular thing with AI, Gavin, is to make somebody say something they're not saying.

Kevin: Whether it's a virtualized avatar, or real time video of somebody else. Replacing someone's mouth and the words that they speak is big business, literally, for a handful of companies. But the holy grail for the open source movement is a free version that does what we need. And we use programs like D. I. D.,

Kevin: like HeyGen, , hashtag not a sponsor, We use programs like that and we pay for credits to make our CG characters or sometimes full motion video come to life and say things that it shouldn't. The promise of video retalking is that it is quote an audio based lip synchronization for talking head video editing in the wild.

Gavin: it's a big deal. Like that's, I mean, and also the chunks of money that come out of our bank account each month from DID or [00:21:00] these other companies. Like if you could do that for free, it'd be amazing.

Kevin: Yeah, we call this podcast a loss leader. We haven't figured out Where the money, we haven't sold the blades yet, right? We're giving away the handle. We're losing a lot of money on the handle, but we haven't figured out the blades part yet. But the point is, the promise of this, and the demos of this, by the way, look great.

Kevin: There'll be a link in the show notes, and we'll put one up on the screen. . Now if you're watching the video version of this. But you can see news presenters, in a very static shot, in proper studio lighting or, um, you know, politicians, ? Giving a speech, you got somebody talking into a camera, again, very flat studio lighting, and it does a really good job of matching their mouth to whatever the new audio is, and it's pretty good for open source, for something that you can play with.

Kevin: Well, I got it running, Gavin. I used Hugging Face. Which is a website that allows you to run demos and, uh, and, and software like this, not on a local machine, because I just have my little laptop on me right now. [00:22:00] And I generated an AI voice, using one of our characters from previous episodes named Gash, and I'm gonna send it to you now, Gavin, You know what, I'll play it for the audience now, in the edit, and then we'll come back with your reaction to having watched it.

Gavin: Oh my god, okay, yes, it's not good. I will say it is not good! Wow!

Kevin: not it's not the worst, but it is far. It is far from production ready. Um, it kind of, turned me into a weird Facebook filter. Like, really made my eyes super blue. It made me a little Kindle ish. And, I was outside. I thought the lighting would be good for it. But I'm moving my head left and right a bit.

Kevin: And it kind of breaks. And you'll see the mouth distort a little bit. And it's not even...

Gavin: I'm taking a screenshot. I'm just taking a screenshot because I want to send it back to you because like I can't believe [00:23:00] like this. I can't, I don't know who this guy is. Who's this guy? Like what, what is that guy? You know,

Kevin: are some haunted frames in the video. Oh God! Hahahaha!

Gavin: Who is that guy? I don't know.

Kevin: That's the real me, Gavin. That's... I'm in so

Gavin: are all off, when the filters

Kevin: pain. Yeah, I'm constantly laugh crying. That's my resting face all the time. That is me every day when I wake up and look in the mirror. 

Kevin: The code is out there. , it's really, really easy to play with. It's essentially free. You can run it on Hugging Face. Um, you don't even have to download software and mess with GitHub if that is all very confusing to you. But it's called Video Retalking.

Kevin: And you might have some fun with it.

Kevin: , I want to tease at the end of the show, Gavin, we have got a world premiere. Dare I say, a Juicy Sluicy.

Gavin: I can't wait.

Kevin: An AI4Humans fan who is listening right now but wants to remain anonymous, and I understand why, no one wants to admit they actually listen to us.

Gavin: Hey, it's like Banksy. Every big artist doesn't tell you who they are, right? [00:24:00] That's what we're dealing with here.

Kevin: Yeah. It's not the shame of engaging with this podcast. It's that they want to protect their, uh,

Gavin: No, they don't say that. That is not the case. This is about, this is about protecting their anonymity

Kevin: You're right. You're right. So the, the Banksy of AI, , who wishes to remain anonymous. Sent us an exclusive, it's an advertisement for the Backwoods Bunker. And we're gonna put it at the end of the episode. But we always love shouting out cool people who do interesting stuff with AI. And we always want to shout out people who send us that stuff.

Kevin: So if you are out there making things, experimenting, please treat us like the refrigerator. We want to use a magnet and put it up there for the world to see, for our little corner of it. So please send us that stuff. We will have the Backwoods bunker premiere at the end of this podcast.

Gavin: , , all right, you know what time it is Kevin? It's time.

Kevin: For wall talk?

Gavin: Not for Waltalk, it's time for Nasal Explorers with Snorkels.

Gavin: It's time for the news!

[00:25:00] News, eh, news, it's the A. I. for humans, 

Gavin: Let's get started though, Kevin, because there is a giant, giant thing that happened this week, and it is OpenAI, the company that makes ChatGPT, had what they call their developer day. And if you don't know what this was, it was really OpenAI's chance to come out, talk about the new tools they're creating.

Gavin: It's, it's imagine like, you know, Steve Jobs getting on stage. This was Sam Altman doing his kind of Steve Jobs impression. And they actually came out with some pretty big stuff.

Kevin: Obviously the announcements were geared toward developers, but very quickly, the rumbling started breaking out as like the implications of these new tools and techniques, what they could mean to the end users, obviously in the hands of capable developers, I was getting texts.

Kevin: From friends of mine that are developers saying they were doing the clickbaity stuff of like the game just got changed Our mutual friend Andrew shout out to Andrew texted him He's like this reminds me of the launch of the App Store like that's [00:26:00] Pretty pivotal, right? Coming from a developer to say, Oh, this moment, this thing on stage feels like the next generation of that.

Kevin: So, let's take a step back, let's talk about the announcements and try to contextualize them in a way that, um, we know the developers are excited, but let's talk about it for a broad audience. Let's start with something that seems simple. GPT 4 Turbo. for everyone. Okay, so there are, there are multiple language models, the, the core technology, which allows the artificial intelligence to hallucinate sentences, to put thoughts together, to analyze data.

Kevin: There's different models that have different weights, the way that they were trained and have different capabilities. Usually, , the amount of data that went into the model right now correlates to the quality that you get out of it. GPT 4 is one of the best, but the trade off. Is that because it's such a large model, Gavin, that when you ask it a question, it takes more time and literally more energy, more compute power to give you a response.

Kevin: And so sometimes if you have a very basic question or query, it makes [00:27:00] sense to use an older, lighter, faster model because it costs less and gives you a faster response. And what they announced is, hey, GPT 4 now has a turbo model, which is going to deliver answers so much faster. Oh, and by the way, everybody who was building on this platform, , your product may have cost X dollars this morning, but tonight it's going to cost Y.

Kevin: It's going to be half as expensive because this model is so much faster and cheaper. Do you think that, is that a fair summation of the GPT 4 Turbo?

Gavin: I think that's totally fair. And I think the big thing to take away with that is everybody talks about new models, meaning GPT three or GPT four GPT five, that that's the big change. And it's always like incremental change. And then something big happens. If you look at a year in difference, it looks huge.

Gavin: But incremental change is a big deal. This doesn't sound like a lot in a lot of ways This is like the specs of the MacBook going from like the m1 to the m2 But it is a big deal when you think about like a if you're interacting with it like I know a GPT for it used to be kind of annoying where you type a [00:28:00] question and it would just hang for A bit and then it would think if that comes fast and furious, that's amazing.

Gavin: And I think for developers People that are making actual things with this, if you're somebody that's interacting with your iPhone app, you can't wait there for an answer. It's going to get frustrating. You're going to just chuck your phone and be like, I don't want to deal with this. So you do have to have that instant reaction.

Gavin: And I think, again, faster and cheaper is the way that technology has gone forever. And it seems like AI is kind of... Pushing through that same 

Kevin: Just to give the numbers to those who want them, 3x cheaper, three times cheaper for input tokens. So when you put something into, , GPT 4 Turbo, costs you three times less, 2x cheaper on the way out. 

Kevin: And the context window, meaning the amount of data, uh, tokens that it can, uh, keep in its memory, let's say while you're having a conversation, . Has just exploded, a 128 K context window. So what does that mean? That means if you have, let's say 300 pages of a PDF or a book or something that you want it to keep in its memory, you can now feed it directly into [00:29:00] this new model and it will keep it in context as you have a conversation.

Kevin: So the memory on this gets larger, which just means it's capabilities gets larger as well. And I think you're right. These sound like incremental improvements, but given an ounce of distance, we will look back and go, wow, these are. monumental achievements happening on a timeline that just seems to get faster and faster.

Gavin: a hundred percent and that token thing is a giant thing because people don't understand up until this point It's been kind of choked at a certain level and entropic was the whole idea was entropic was this model that could allow you to Put more things in What I think is going to be interesting about this, and we'll get to this with GPTs in a second too, is that open AI is starting to really maybe create a true moat, right?

Gavin: Like they're really creating their moat in my mind. When I say a moat in technology, we've talked about this before, but what a moat means in technology is that you create technology so that other people can't copy you and do the same thing. What open AI's moat is starting to feel like is the top of the funnel, right?

Gavin: Being the, the kind of person there first. And if they can be there first and they [00:30:00] get people used to using that. Then they're always going to be one step ahead of the next guy. That feels like a legitimate advantage for a company like this, if they can continue to scale and continue to get better.

Gavin: One thing we haven't talked about, which is, I can't imagine being Google's executives watching this. And knowing that we are now November 7th, there was a rumor that Google Gemini, which is the next big shot Google has at LLMs, was supposed to come out in October, supposedly, that was the rumor. And it kind of came and went, we got no Gemini information, and then maybe Gemini is going to come out early next year.

Gavin: Now they're saying, some people are saying so this year. But watching this kind of startup push forward and do all this stuff and continue to move the like the stick forward and forward and having them have to get further and further away to make an impact fee must feel difficult. 

Gavin: So let's, let's move on to talk about what I believe is kind of the biggest thing. And this gets to what Andrew's point about the App Store, which is OpenAI introduced the idea of what they're calling GPTs. And GPTs, much like what we saw at the beginning of the [00:31:00] show, are individualized chatbots That can be created by anybody, and this is the thing that is kind of a game changer, hosted by OpenAI, but essentially, , can allow the creator themselves to make money off of them.

Gavin: Now we don't know what that means exactly, we don't know how the revenue share is going to work, but OpenAI does have a lot of people paying now as paying users, so they're getting a chunk of money that's coming in. So, you know, GPT four is the GPT model. GPTs and what they're calling them are each individual asset of these chatbots, right? So an 

Kevin: Let's just say that they're apps, right? Just for the broad strokes, like each GPT is now an app that runs on the GPT operating system, the OpenAI OS.

Gavin: The idea is that you at home or anybody else, any developer can create a specific use case for one of these GPTs, meaning that. On stage, Sam, uh, created a GPT, Sam Altman created a GPT that was about, I think, a software developer, a senior [00:32:00] software developer who was there to give advice to younger software developers, right, or a startup founder.

Gavin: And so the idea is it's basically creating a personality, but for a specific use case. And then when you start to expand out the idea of what this means, and we'll get into it in a second here, the tools that are available to it, you start to think, okay.

Gavin: Every time we use chat GPT, when we create a character, it's a little bit of a laborious process, right? You're going through the process. You're saying, I want Grok. I want him to be this AI bot. I want him to have this information. If you could create that and then immediately share it and let other people use it.

Gavin: That immediately makes me more interested in doing that on a professional level or not even, even on an amateur level because like then you get this like shareability sense. I can already see a world where somebody creates one of these and it goes like crazy viral. In fact, it already has started. And I think one thing I do want to talk a little bit about, and I know you were excited about as well as I was, was the idea of opening up ChatGPTV, which is the ability for ChatGPT to see.

Gavin: And these chat bots, [00:33:00] right? And one really cool example, which will show here is that somebody already has taken a chat bot and shown it clips of soccer or football. If you're the you, um, and had GPT. Comment live on the footage and it sounds amazing. Like, I mean, listen, it's still a computer, but it is really well done.

Gavin: And you can see, Oh, this could be a GPT, like the comp, like, you know, a specific comment or maybe I want. Grok to do our soccer commentary, right? Like imagine Grok commenting on soccer. That's something you can do and you can make happen. And the other side of this Kev, which I think you'll, you also are excited about is the way they're doing it is to try to do it really natural language first, which means that like normal people, I mean, you still need a little bit of technical savvy to get into this, but like you are not going to have to be a hyper, you know, Python programmer or 

Kevin: I would argue listening to this and thinking like, I can't write software code, and why [00:34:00] would I, what kind of a chatbot would I make?

Kevin: Everybody has a unique personality, uh, an interesting perspective on something, a unique perspective at least, and you probably have an area of expertise that is unique to you. You might not be the foremost expert in, let's say, Knitting there is something that you know about it that someone else doesn't because of your life experience with it And so the idea is that if you want to create the ultimate knitting chat Bot you can go in with natural language say here's what I want to make I want to make something that's gonna help novice knitters Get up to speed using my knowledge, and I'm gonna drag in everything that I've ever written about knitting, and my favorite bookmarks, etc.

Kevin: This could apply to anything! To playing an instrument, to your favorite vegan recipes, to your strategy for leveling up in a mobile game faster. Whatever it is that you're passionate about, that you're knowledged about. You can leverage that now and make a chatbot for everybody else to use. And the promise, and [00:35:00] again, the devil will be in the details on this one, but the promise is that if people find value in that and use it, OpenAI is going to share a piece of the pie with you.

Kevin: They're going to compensate you because your chatbot is doing well. Some red flags. We don't know what that looks like just yet.

Kevin: tHey made it very clear that they're not going to train on your data. If you use their API, I didn't see if they said that they wouldn't train on your data, if you created a custom chat bot.

Kevin: And so. I get an ounce of concern when it's basically saying, Hey everybody out there, come to our platform, create a unique bot with your individual knowledge, or scrape the web for us and put together the ultimate resource on professional wrestling, or anime cartoons from the 80s, whatever your thing is, go ahead and do it!

Kevin: Because you've now trained our AI to be better and more knowledgeable about that thing, and as it starts to get more popular, They kink the hose, or they wall the garden, and they say we're just gonna take these [00:36:00] millions of chatbots, which you have all created for us, and swallow them into our own thing.

Kevin: And we've seen, you know, Apple do that with its own apps, right? Seeing what pops in the App Store and then releasing their own

Kevin: version of 

Kevin: it. 

Kevin: We've seen 

Kevin: Google do 

Gavin: Facebook, Facebook's done that 

Gavin: for 

Gavin: years, 

Kevin: all do it, so I don't know that I would have the confidence that Open AI isn't going to do that ultimately, but in the near term, they're going to be incentivized to making a handful of users very, very wealthy off of properly

Kevin: compensating 

Kevin: them 

Kevin: for 

Kevin: their bots. 

Gavin: Or who knows, right? Who knows what's wealthy in this, we really do, it's the red flags, like we have no idea what the revenue sharing looks like. Like it's clear, at this point I don't think open AI is making, I mean they're making money in that they're charging people but I think that their costs way outweigh their income.

Gavin: This is like a world where like who knows what the financials of it look like, but I think you're right, like there's a lot of things to be really cautious about here so. There is a world now that if this works and, and it very well could, it seems like it's knowing what we [00:37:00] know about AI.

Gavin: Like, it doesn't seem like it's hard to imagine. That the idea of being a software engineer has just completely changed and, you know, and Andre Karpathy, one of the smartest software engineer minds in the world, um, you know, a famous software engineer himself has said that like natural language is the new programming language.

Gavin: And I think that's the thing to take away from you at home if you are a normal human or you're somebody that is a passionate about technology, but maybe weren't trained as a developer. This is your time, right? Because now you can do things that weren't possible before because of this tool that exists.

Gavin: And I think that's the thing that's got me most 

Kevin: some of the big announcements that came out of OpenAI's Developer Day. If you don't follow us on TikTok, uh, or on YouTube, please do. AI for Humans show.

Kevin: We constantly are churning out content on those platforms.

Kevin: So go check us out over there and we'll keep you up to date, , in between podcast release on the biggest stuff. Now, Elon Musk, , this past weekend really started to tease Grok, which is [00:38:00] his large language model, his AI, um, which he says is best in class at certain things, says it's way better than anything else out there.

Kevin: at specific tasks. And we started to see some leaks and started to see some responses from Grock. It is AI with Attitude, which I don't mind. , as you said earlier, as you were chatting with those character AI bots, it is supposed to be anti woke and supposed to save us from the mind virus. 

Kevin: I want to get your take. I am actually all for this in some weird ways, , I don't particularly have a reason to trust anything coming out of X, but I am interested in this LLM for reasons which I will get to, but what is your take so far,

Kevin: Gavin? 

Gavin: so here's the thing. I think in general all Having different LLMs is good I think having a variety of LLMs is a good thing because we'll get a sense of like never one becomes like the Oracle that Everybody goes to I actually listen love or hate Elon There's an element [00:39:00] of him of like you could you know, it's something I call trolling other people might call it You know really offensive stuff in general the idea of like At least pushing back a little bit on like rules is kind of an interesting thing.

Gavin: And I, I mean, listen, you and I have worked with uncensored models for a long time and would it be nice to have another true uncensored model that we could do stuff with the show with on the show with probably, but also the weird part about this and Elon always gets wrapped up in to me is that the, how much of this is like.

Gavin: Not going to be protect, not going to protect people because I do think there are good reasons to red team chat chat bots. Now, granted, you could say the same thing about the Internet, right? And there is a lot of terrible stuff that you can put into your Google search bar and go find immediately and it has not ruined the world.

Gavin: So I don't know. I feel kind of conflicted. I'm the biggest thing is I'm all for new LMS. I don't know if I love this kind of like. Push towards like, uh, you know, anti wokeness or all [00:40:00] that side of it. But I like the idea of 

Gavin: having 

Kevin: concrete stuff that I've seen teased, I say concrete because I haven't got my hands on it, right, we don't know, Elon is no stranger to smoke and mirrors 

Gavin: you mean we're not in driverless cars 

Kevin: Oh, sorry, my Uber, my 

Kevin: Uber 

Gavin: like 

Kevin: just arrived, let me hop into that one right now. Um, you know, so he's no stranger to missing deadlines, he's no stranger to fake demos.

Kevin: , so what we're seeing are mostly screenshots or screen records of people interacting with Grok who somehow have access. I will say, so two things. One, it has access to Twitter or X's real time Data. That could be massive. Now, X is littered with spam and misinformation and disinformation. Granted, most platforms are, right?

Kevin: Most social media platforms are. So, there's going to be, uh, quality and trust and, it's not even a hallucination issue here. If you, , query against Grok and it's pulling from Twitter in real time, it might be thinking it's giving you real information because that's what's really being tweeted or X'd or Z'ded, if you [00:41:00] will, , in real time.

Kevin: So it's not making it up, but I don't know if the quality of the information will be great. I will say that when something does happen. From a high speed chase, to potential war crimes, to the release of a new MacBook, I am still refreshing X for information. And it, you have to get good at scrolling through the garbage to get to the real stuff, right?

Kevin: And I'm sure we all get got from time to time. But it still is a very powerful... Up to the

Kevin: millisecond 

Kevin: information 

Kevin: source. 

Gavin: real time search 

Kevin: right. That's right. So Elon's got it, right? If there's anything good that came out of his purchase, he's got that and the ability to query against it in real time. That is not something that OpenAI has.

Kevin: It's not something that Google has. So that could be an interesting differentiator. I also saw the example, and we're gonna have to censor this word, I think. Ah, let me just say it in a way that doesn't require a censor. I don't know if you saw this, they asked Grok how to make, uh, an illegal chemical substance that human beings love to take during a roller disco session.

Kevin: And they showed an example that OpenAI [00:42:00] or, or Google's Bard, Anthropic's Claude, they get all, Oh no, no, no, no. We can't even talk about that. But Grok not only loves it, but tells you step by step how to make this controlled substance, this chemical. And so many people were saying, Oh. How dare you not have guardrails up?

Kevin: This is absurd. This is ridiculous. This is, it's an illegal substance in the United States. How could you tell somebody how to make this? I actually err on the side of, who's to tell me that I can't have access to the information on how to make it? And some people will slippery slope that with, what about nuclear weaponry?

Kevin: What about 3D printed guns? What about all of the things? And I'm not, I'm not a proponent of people making them. You know, there are, there are strict controls in place to make sure that I can't refine plutonium. Don't worry, I'm not gonna make it. I don't know how to make it. I would definitely explode myself in the process.

Kevin: But I do think that it's, it's an interesting, worthwhile exercise to question who controls the access [00:43:00] to that information.

Gavin: I think this is a good point to keep in mind. To be honest, what I think is probably going to happen with Croc, knowing Elon's kind of history of overhyping stuff, It will probably come out, it will be kind of funny, and then it will probably land in the mid to lower tier of what LLMs are doing.

Gavin: You know what I mean? Like, in terms of what they're capable of. I don't necessarily see it going past the Facebook open source models, but maybe, I mean, we might be 

Kevin: Well, I think, you know, we talk all the time about how , these AI companies giveth and then immediately taketh away. Right. They launched something you're amazed at, at the capabilities. And then because universal music group is threatening over song lyrics, suddenly this all seeing all powerful AI can't even make a song parody for you or whatever the use cases, I could see that being a lane for grok.

Kevin: The real time access to information could be massive. And if Elon is willing to say, Hey. Guardrails up. It'll be the go to AI to, to play with characters that don't mind cursing. Or to, you know, to get information on how to make a recreational something [00:44:00] for a Friday night. Whatever it is, that could be their lane.

Gavin: I think a hundred percent. And then the question will just become is like, uh, how is it perceived across the world? Right? Like, is there, is there a social push back on it in some form or another? Does it hurt the fact, the company that it exists because of it? Do people protest against all that sort of stuff may come out of it.

Gavin: But you're right. Ultimately, like the truth of the matter is more of these, the better. If this is the one that's like the kind of asshole uncle that you don't want to talk to him once in a while, but you'll have a conversation with him just to be entertained by what he's going to say. 

Kevin: This LLM has a cigarette and a chain wallet. Oh, Grock, you're so cool. Hehehehe.

Kevin: So Elon made 

Kevin: headlines? 

Kevin: OpenAI had their announcements, and Microsoft... Man, they stirred up the gamer waters again with announcement of a partnership within world that makes AI characters come to life and they, they infuse AI into 3D gaming engines.

Kevin: And Gavin, you have some thoughts.

Gavin: Okay. So first of all, listen, Kevin and [00:45:00] I are both gamers. We love video games. We've been playing our whole life. We play still to this day as middle aged men, right? So just to clear the table, we are big gamers. InWorld is this company that powers these, uh, games and you've seen it within the, if you've seen the Skyrim, uh, NPC mod, which we featured on the show, which means, You go into Skyrim, the very famous, uh, fantasy game from Bethesda.

Gavin: And in that game, you can interact with the characters in real time, talking to them back and forth, and they will respond to you in real time with an answer. And that is powered by an LLM. So the in world is a company that powers these sorts of things. And what we're talking about in this instance.

Gavin: Microsoft has partnered with them to create essentially a developer kit to use this within Xbox games going forward, and I kind of think this is cool and one of the things. So last week we covered this kind of AI controversy with the game. The finals. And the finals is a big, uh, first person shooter that is using AI to create real time voiceover for characters, um, [00:46:00] based on what's happening in the world.

Gavin: There's a level of anger that exists in the world of people who just do not want AI to do any of this stuff. Specifically, in that instance, to replace voiceovers. I think what's kind of cool about this, and, and Kevin, I think you, you probably agree, but I'd be curious to hear your thoughts, is in open world games, if you're a video gamer, you know an open world game means, A place you can wander around and you have interactions with all sorts of different types of people.

Gavin: I think for open world games, this technology is insanely transformative. Because I often find when I'm playing an open world game, you know, you go from quest to quest to quest, and you may be on a side quest, and you want to kind of just fart around a town for a little bit and see what's going on. The minute you ask anybody in the town, except for the characters, the three characters that have been plotted there to know where the story's gonna go, It just falls apart, right?

Gavin: The whole illusion falls apart. But if the AI could come up with responses based on what's going on, like that feels like a really useful thing for these, these larger worlds. And I think the idea of the [00:47:00] anger that comes from these sorts of announcements, I have a feeling we're just going to see a thousand more of these and maybe it'll start to bubble down when people have fun with these.

Gavin: But for right now, I 

Gavin: think 

Kevin: Yeah, and think about it from an indie developer standpoint, like, yes, there's the AAA rock stars that may put this in the next Grand Theft Auto, who knows, right? But if you're the indie developer who wants to make an expansive game, with 10, 000 characters that seem like they're real, You can't think like that with the current toolset.

Kevin: I'm excited for what developers will be able to make with this, right? this could usher in a new era of gaming that we can't literally conceive right now. So I'm excited for what people can make with these tools and these technologies. I don't see an announcement like this and immediately start throwing rocks and saying, How dare you? Like, clearly you're anti gaming or you're anti humanity.

Kevin: That's just not my take on this. I'm excited for the technology. And again, human beings may... Use and abuse it in negative ways. Okay. I'm going to swim towards those that are using it in positive ways. And I'm going to play all of the new exciting [00:48:00] games that were previously impossible to create before this tech existed.

Gavin: And last thought is like the power of the market is the power of the market. If you and enough people don't want to play these games, I totally understand that. Don't play them. Don't buy them. That is the way it goes. I just think that ultimately the pluses of this are going to be stronger than the minuses.

Gavin: And to your point, We're going to have to rethink what happens to the people that are affected by it. But that is a much larger conversation that affects so many more people than just these people. And I think at some point we'll do the whole, like, we'll do a catch up on the universal basic income versus AI conversation.

Gavin: But that is not today. That is a huge conversation that we've gotten into many times, but it won't be for today. 

Gavin: It's time for our interview. We're very excited to have Mikey from Suno. He is one of the co founders of Suno.

Gavin: They make a product called chirp, which is an incredible AI music generation service. Kevin, you are a giant fan of this and

Kevin: Yeah, how dare you distill it to just being an AI music generation service. It is a transformative, [00:49:00] artificially intelligent tool that can make a musician out of mere mortals. Love this tool. I've been accused of being a paid shill for it. I wish, Gavin. I really do. I'm just such a super fan of it, and if you've never heard me yap about it, it can take a single word or a sentence and turn it into a 30 second, or maybe longer, we're about to hear, perhaps?

Kevin: Banger. And you know you hate that term, I know you

Gavin: I, it's fine. I don't

Kevin: it can take a small seed of inspiration and generate really, really incredible music. And if you've never heard it, what you're about to hear is from their new model, which will be coming out in a few weeks, this is what a Suno Chirp song sounds like.

[00:50:00] 

Gavin: One thing I want to start with is y'all make audio software with AI. Can you just kind of basically explain what your company is and what you do before we jump into some of the really cool specifics, especially around open source audio and the kind of products you've made?

MikeySuno: Yeah, sure. I saw, um, I think of us as, , building the tools that people need to be creative or unleash their creativity in what is kind of the most human modality, which is, which is audio. Audio is like a really special, I think the modality that really makes us feel the most. And it's also the one that has just historically been lacking, um, or certainly lagging behind images and text for a really long time.

MikeySuno: Um, [00:51:00] and there's nothing fundamental that says that that can't be, uh, fixed. And so, um, in particular, we focus on music. Audio is a little bit bigger than just music, but music is something that is so personal, that is so emotional, that really makes people feel that everybody has like really innate tastes in, um, and nobody, almost nobody has the means to make good music.

MikeySuno: You don't have to be like a really skilled musician. You have to know really complicated software. Um, And it's too bad. And so I think of us as providing the gateway for people to be able to unleash kind of their, their inner selves in, in the music domain. Um, we've done some other stuff in audio also, but that's, the driving force behind the whole company.

Kevin: Speaking holistically of the music space and the tools available, or the lack thereof, really, why did it seem to lag behind images and text specifically for so long? I feel like I saw new Photoshop features all the time that would add layers and let you deconstruct and reconstruct and go 2D to 3D, and with audio, it was like...

Kevin: Well, you can kind of filter out the lyrics, , and that was it for like the longest time. 

MikeySuno: I think, I [00:52:00] think there's a couple of things. It's hard to know exactly which one was the biggest contributor. I think historically, um, a lot of the academic work was like somewhat, um, Behind and I think this is, uh, largely a function of, uh, data that like most academic audio data sets are kind of small, um, and very kind of singularly focused.

MikeySuno: And so just to compare it, for example, to text, um, there's like this giant text data set that everyone uses to train their LLMs called, called the pile or the common crawl. No such thing exists in audio, for example. And then the other thing is like the industrial.

Kevin: have Kazaa? You didn't have LimeWire? You couldn't have grabbed 50 copies of the same Metallica song and just jammed it into the internet

Gavin: There's a lot of people have that U2 album. You probably could have used that. I

Kevin: Yeah, I got four copies of it accidentally!

MikeySuno: Oh man, if only, if only, so, you know, and then I think, I think, um, a lot of work also gets driven by the big academic labs like the Googles and the Facebooks of the world. And if you think about it, um, and I say this with no judgment, there's nothing [00:53:00] wrong with this, but a lot of the AI progress has been driven by what can you sell ads based on.

MikeySuno: Um, and so you sell because that's where they need to make progress. And so you get served ads based on what you look at. That's like Instagram images. And you get served ads based on what you read. That's like Google. And you don't really get served ads based on what you listen to. And that's just beginning to change.

MikeySuno: But that's, that's meant that there's just comparatively less focus on, um, audio as a modality, um, which I think is too bad.

Gavin: that's really interesting. I never thought about that before. Did you find, so I guess my thought is like, what made you personally want to go into this section of the AI business? Like was it, were you just drawn to be, are you a musician? Is that something you do in your spare time?

MikeySuno: Um, I am a bad musician. I, I, um, I was, you know, played a lot of piano growing up and, and, and a couple of other instruments was never, was never great. Um, I think there's, there's kind of two things that did it for me. You know, one is I've always been, uh, very audio focused. Um, and it's because I'm actually one of the worst readers you will ever meet.

MikeySuno: Um, and so I end up [00:54:00] consuming a lot of information through my ears, uh, in general and have like trained myself to, um, like do audio books at 3x, um, because I want to do it faster than a normal audio book. Um, and then the other was actually kind of getting our feet wet with, uh, audio AI. So we all came from, from the same company, the founders of Suna.

MikeySuno: We all came from a company called Kensho, which was machine learning for a totally different discipline for financial services. Um, and I was lucky enough to be the first machine learning hire there. And, um, we were ultimately acquired by S& P Global, which is like a huge financial services behemoth. Um, and we ended up building speech to text, um, like a very financially relevant speech to text.

MikeySuno: And, and it was very impactful for that business, helped them, um, speed up the transcription of earnings calls. And I think there was this aha moment there of like, wow, that was so much more fun than all of the text that we've been doing. Not that I don't love text. I love. Text, uh, you know, but wow, that was so much more fun.

MikeySuno: And like, wow, it's so far [00:55:00] behind it's like, and, and, um, like we can, we can go and change that and, and, um, it doesn't, it's not that hard. And, and, you know, you can go look at bark, our open source project, and you can see like, this is just GPT. Like it's not reinventing the wheel. It's doing audio in the right way.

Kevin: , so to that point, so Suno small team, , do you say how, how big the team is? Are you

MikeySuno: Oh, I'll tell you. Yeah, yeah, no, I'll be I'll tell you whatever you want to know. Uh, we're seven people right now

Kevin: Really?

MikeySuno: Yeah

Kevin: Oh, wow. , that is super impressive considering the releases that you've made and the speed with which you've iterated on those releases and the quality I'm talking primarily. I mean, we love bark. We showed it off the moment you guys released it on GitHub. We had it on this show and we put it through its paces, but.

Kevin: What's become Chirp is a is a tool that allows anybody to either provide song lyrics and a style of music themselves and your model generates A complete, 30 second sample of a song with instrumentation and lyrics, [00:56:00] harmony, the whole nine. Or you can ask the AI to sort of generate lyrics for you based off of the topic.

Kevin: And you've done a very, uh, elegant, uh, Discord bot approach to it. So that anybody, regardless of their level of knowledge or musical ability, , they can just start generating songs and start spitting it out. And to me... I have a fundamental understanding of how it all works, and it still feels like magic.

Kevin: It still feels like I'm going to this internet slot machine, and I'm pulling the lever, and I'm saying what I want on the reels, and sometimes I get an absolute jackpot, and I am, I am just astonished at what comes out of this. And so, please help my dumb brain understand how your technology is working.

Kevin: When I say I want a country song, and here are some lyrics, what is happening behind the scenes to give me 30 seconds of coherent, quality audio?

MikeySuno: Yeah, it's um, it's not it's not like anything magical it's like it happens a lot like GPT does when you ask it to write you a paragraph about a Uh, you know, your, your friend Gavin or to, [00:57:00] uh, write an email, uh, trying to convince somebody to buy your product or something like that. And it's, um, it's something that's kind of soaked up a lot of, um, information from a lot of data that it's trained on.

MikeySuno: And it learns basically the patterns of what does it mean to be a country song? And what does it mean to sing the lyrics that you just gave it? Um, and it kind of spits it out, you know, bit by bit, little by little. And I'm sure you've noticed that sometimes that slot machine doesn't, doesn't yield something good and it can go off the rails a little bit.

MikeySuno: And we're, we're still, we're still learning how to make it better. Um, but that's, that's, that's basically it. It's, it's nothing all that crazy.

Gavin: When when you train this versus because I understand like the scraping of text makes a lot more sense than to me like it's very easy. Do you does it get fed music? What what does it get fed to train something like this?

MikeySuno: of audio, not only music. And I think, you know, some one, one thing I think to appreciate, I think is, um, I think the right analogy for text is like code and it turns out GPT is [00:58:00] very good at code and maybe the best code models or the best open source code models code Lama, but it's also trained on a lot of regular English and that makes it better at writing code.

MikeySuno: And so you can think, um, generally exposing these models to lots of audio, even if they aren't music make, makes them better at music. And, you know, um, as a cartoon example, it wouldn't surprise you that if I. Um, show the model, a lot of speech that, that will help it, for example, do lyrics. And I think one of the things that we do, we may be the only ones who really do, um, kind of in their full generality is lyrics.

MikeySuno: Um, and that's something, uh, just like bark, you know, we, we were able to do, um, text to speech in a lot of different languages. And this is kind of the same approach to GPT. It's like, you're kind of sharing resources amongst these different languages. And so, you know, there's a lot more data, obviously in English than there is for the 10th most popular language, but that still actually helps your models quite a bit.

,

Gavin: One thing about that, that when I think about and you know, is an important conversation is you, you know, we all know in this group that one of the most litigious groups of people in the world are [00:59:00] song lyric, uh, rights holders. Right. Do you w were song lyrics part of the dataset, and is that something that you worry about at all in terms of like.

Gavin: Do you feel that there is a, um, Do you feel like you legally are going to be on the hook for any of that? I know there's a, from what we've talked about on the show, there's a very clear kind of understanding, I think, from most of these companies that the legal restrictions on what is fair use and what isn't fair use are pretty clear.

Gavin: Do you land in that same space as most people, or what's your take on that?

MikeySuno: I, I, I'm actually not sure I would agree that most people agree on what's fair use and what's not fair. You say, you know, like open AI is getting sued right now. Right. As is Anthropic, um, and the, or as ours, as our most companies. And I think, um, but like, let me back up a little bit. Cause like, in some sense, it looks like every big AI company gets sued.

MikeySuno: Um, but in the other sense for music, actually, um, something most people don't know, if you have a song that charts, like a human has a song that charts, you will get sued. Um, it's just like a thing, right? And,[01:00:00] 

Gavin: what? By who?

MikeySuno: by another person who says you've, you've been inspired by their song in some

Kevin: That chord progression was actually something that I did in 1976. You went from C to F sharp and

Gavin: My Sweet Lord. That kind of

MikeySuno: Exactly. Exactly. And the majority of these things have just settled for small amounts. A lot of people will take a producer credit or something like that. But, um, so that's kind of where we're starting at. And, and honestly, with that, like assumption, uh, that would, that would assume that like, at some point we will.

MikeySuno: Uh, we, we, we may actually get sued because most AI companies are going to get sued and most, uh, and, and people suit as soon as you have a hit song, you get sued. But I think the flip side is like, it's actually gonna all be okay. You know, and I think what we're trying to do is not, um, replace artists. Um, I don't think we're trying to make like just AI powered Spotify where I just get to stream music that was made by an AI we're going for like.

MikeySuno: Changing how people actually interact with music. And that means getting people involved in creating music. And that doesn't mean they're going to stop [01:01:00] listening to music. Actually, they probably listen to more music. Um, as a result of kind of being engaged in the creations. And so, I think about, in the short term, things may very well be messy.

MikeySuno: But in the long term, I think, Um, if we, we look very closely at text and images just because they're ahead, and I'm actually quite hopeful in those modalities that everything kind of works out because it has kind of fundamentally changed the nature of how people interact with these things and music will be exactly, exactly the same way.

MikeySuno: And so, you know, what I would describe to you is like, we're not trying to, again, we're not trying to do AI Spotify, where you just stream lots of AI music, you know, in the background, um, and we're not trying to give people necessarily the tools to like, Um, try to. Make a song with our stuff and put it on Spotify and get a hundred million streams.

MikeySuno: Although I'm sure that will happen at some point. Um, like the usage patterns that we've seen are the most fulfilling and, um, enjoyable for people are like, I made a song for my friend and I shared it with them. Or I [01:02:00] made a song and I listened to it five times and then I didn't share it with anyone else.

MikeySuno: It is actually in enjoying the act of creating music. And it's like. You know, Kevin, you said you played the drums, like the first time you got into your friend's basement and you had a jam session, like you were probably so excited you couldn't sleep. Right? And so it's like, that's what we're trying to give everybody, but even if they don't play the drums.

Kevin: Well, and it, this is pure anecdotal, and this will reinforce my unpaid shill status for Suno, which I've been accused of, one, I generated a song, , with Chirp extended it, loved it so much that I downloaded it across multiple devices and listened to, it was 45 seconds of audio, but I listened to it on Loop at the gym while driving around or whatever.

Kevin: And it was, it has board its way into my head in a way that most popular music has failed to recently. And this was an AI generated something, so I appreciated it. , But anecdotally had a friend make a track who was not a musician, never thought they could be, had this idea for a song, had these lyrics, put it in there.

Kevin: Loved what they got out and then [01:03:00] said oh, but I really wish like, you know, the part where that goes into the what's the part where it's like the Oh, the chorus. Yeah, the chorus. I wish the singer had like, you know, a copy of themselves. Oh, you mean like a, like a, like a harmony, like a backing track? Yeah, yeah, yeah.

Kevin: They didn't know what those terms were. They were trying to work their way to it. And they are now Going through the rabbit hole on YouTube of GarageBand so they can make their own track and to your point, again, unpaid shill, like this was a point of inspiration.

Kevin: They used the AI tool, they got something out of it, and it's not about replacing a human in the music process, It's about making a human a musician. Someone who wasn't prior. So, hats off, .

MikeySuno: Yeah, it's, it's really wonderful to, to hear that kind of stuff. I mean, and I hope ultimately. Where we're going is that you don't need to port this into another tool and that you'll be able to get out, you know, the, the thing that's in your head. It's like, I think if you just walk down the street and you stop a random person, they have, I won't say good or bad taste, but they have very specific tastes.

MikeySuno: They know what they like. And [01:04:00] they can't get it out and, and it's so funny, you know, we talk to a lot of musicians and we talk to a lot of non musicians and, you know, musicians have a very specific vocabulary for describing what, what they want. And non musicians have like a fairly non overlapping vocabulary for describing what they want.

Gavin: Well, actually, this is a follow up I have on that. So Mikey, originally it was, I think there were a certain amount of time and then you, you allowed people to extend it. One of the things I'm so interested in, Two parts. One is, when you extend it, one of the problems that I've seen with something like Runway, right, Runway.

Gavin: ml, uh, does video extension, and you have a shot, and you extend it a little bit, but then suddenly it veers off into a weird direction. With Suno, or with Chirp, it seems like it holds up pretty well throughout the extensions most of the time. Are you basically finding, is there a pattern in the audio? That you just are kind of like continuing and our songs easier in that way than something like say video is that like Because a song is based on a recurring kind of first chorus first structure It allows you to do stuff in that sort of pattern based world.

MikeySuno: [01:05:00] Yeah. I think, I think there's two things there. One is exactly what you said that audio and music in particular tends to be a little more structured and repetitive. And so that's easier for models to pick up on and without knowing the internals of the runway model. It's hard to say for sure, but I can speculate that, um, it's not, it's not nearly as easy in video to do the GPT style things that just continue.

MikeySuno: And if you are teaching models to just continue the same way GPT learns to just continue text, we learn to just continue audio. Well, it gets pretty good at continuing audio. Um, and so I think that that's probably why, but I actually think we have a long ways to go in the, in the continuation to make that like amazing and smooth because it's still a little finicky as I'm sure you guys have realized.

Gavin: Yeah, I was gonna follow up on that Which is the other side, you know Kevin was alluding to like I'm not a musician But like the idea of being able to separate out the vocals just simply the vocals is a really interesting thing right like the idea is like taking out the vocal or putting a different vocal in like That's [01:06:00] a version of in painting, but a relatively simple one, like do you guys see that sort of stuff coming soon ish?

MikeySuno: You know, a lot of those, a lot of those technologies exist, like to, to stem out an original recording into the, into the original tracks. Um, the, the short answer is, I don't know, I mean, we can do it, but I think once you are breaking tracks down into their individual stems, and once you're breaking a song down into the individual tracks, like, I think you've lost most of the people that we're trying to go after.

MikeySuno: We're trying to empower everybody. to make music and I, I think, you know, once, once you're thinking about like individual tweaks to the drum track, I think you've lost 90 percent of people.

Gavin: Oh that, well that actually makes me have one other question, which is like what, so the audience is this big broad audience, and I, you know, which is great, I think that's really, I think your mission of trying to get people excited about music is amazing. How do you want to try to hook those people is your goal as a company to try to be like a consumer facing music generation company?

Gavin: Or [01:07:00] is it more like we're creating this really cool technology and music generation is an example of what you can do with it. We want this to be used by all these different companies that do this stuff. 

MikeySuno: Very much the former. We want to bring this to everybody. Um, we get a lot of inbounds asking for API access and stuff like that. And it's just, it's not what this mission is. It's, it's, it's a lot of cool applications, but it's not what the mission is. We want to let everybody be creative in music. Um, And that is very different from some of these, uh, kind of more business driven, driven things.

MikeySuno: You know, I can only describe it as like, um, it almost borders on therapy. It's like it is, it is a really fulfilling and, um, uh, wholesome video game in some sense. Like people feel really good when they're done because people like to be creative. And I think that's what we're trying to give people. I'll give you another example.

MikeySuno: We're only seven people, but we're trying to hire obviously. And so. Um, we went to a career fair at, uh, one of the local safety schools and, you know, of course we can bring lots of headphones and let people [01:08:00] make music. And every single person who does this, she has the biggest smile on their face.

MikeySuno: And it's because all I had to do was type in a little bit about how I was feeling and I got something out of it and I feel like super connected to this creation. So, um, that is what we are trying to unleash in

Gavin: I can tell you personally, one of my favorite moments with it was on our show, uh, Kevin came back from being very sick and I got to create a, uh, basically a funeral march for a version of Kevin that existed and I, I used this Celine Dion style kind of ballad about how sick and almost, he almost died and that brought me a lot of joy.

Gavin: I can tell you that. That brought me a crazy amount of

Kevin: Yeah, it was, as Mikey said, it was therapeutic for you, Gavin.

Gavin: Yes, it was very therapeutic for

Kevin: And I, the addict that I am, described it like a slot machine. I'm like, give me another hit! Let's go! So yeah, it's therapy. I get it.

MikeySuno: It's, it's, but it's like, it's a, it's a slot machine that ultimately is, is, um, has positive externalities, right? It's not, it's not just taking your money and it's not the game that's rigged against you. So,

Gavin: It's Joel, like my whole thing about the world we [01:09:00] talk, we know about what the human condition is on the show. Weirdly, a lot considering it's about AI, but like part of the thing that makes being a human great is joy, right? Is that feeling of excitement and joy? And like, what's cool about these AI products and the best possible version of them is they can bring joy like we do with our AI co host.

Gavin: Today we created a caveman AI co host for the show and it was dumb and fun, but it brings joy. And what's so cool about chirp, I think, is that like, The minute you turn somebody on to it, you can see that joy directly come out of it.

MikeySuno: It's, it, like, what an amazing job that I have, honestly, to be able to, like, bring joy to people. It's like, it's, it's very cool. Uh,

Gavin: for sure.

Kevin: You know, I joked about, Hey, you're only seven people let's go where all these features, the model that you released, uh, even just, I think a couple of months ago versus the model that is available now. And then I think you have another one on the horizon as well.

Kevin: , how are those getting better? Is it just throwing more music into the initial machine? 

MikeySuno: Yeah, we're a couple weeks out from, so [01:10:00] we have the initial version of the next model, and it's a lot better, and so we're very excited about that. You know, the cool thing about audio being behind is that like there's still so much low hanging fruit to pick.

MikeySuno: And so, you know, we're basically, we, we've not nearly exhausted that low hanging fruit. Um, and so I, I would say very largely one of the, one of the big areas for improvement is basically how do we tokenize audio. Um, so, you know, in GPT you tokenize a bunch of text and you turn it into individual words.

MikeySuno: And how do we do the same thing for audio? And that's like a big open ended problem that doesn't have a good... Academic solution. And it's something that we work on a lot. Um, and then the other thing that's like really cool for us is, um, honestly, just by virtue of being in the audio space and being behind images and text is we have a great advantage.

MikeySuno: of benefiting from the open source community. And that's like part of the reason that we try to contribute back to it with things like Bark. Like the open source community provides us a lot of know how that we don't have to create from scratch. Um, [01:11:00] and so I'd say that's largely why progress is fairly fast here.

MikeySuno: And I think, you know, if, if you look, if you, if you'd like to say, you know, we're 12 or 18 months behind the other modalities, then maybe we have another 12 or 18 months until things get really hard. Um, but that means like we have a few. Big step changes in song quality, um, in front of us. And so that's really exciting.

Kevin: I would imagine if I were a remotely successful musician, I would be knocking on your door and saying, Hey, here's every piece of audio I've ever made, including stuff that never got released and stuff that I just dreamed up in the lab yesterday.

Kevin: And I don't know where to go with it. Can you make a model of me? And then I press a button and Suno's Chirp AI gives me thousands of 30 second snippets that I could then take and be inspired by. I got to imagine those conversations have happened, and tell me what Yay and What the Weekend are doing with , those technologies right now, Mikey.

MikeySuno: Yeah, it totally works. Uh, this is again, you know, this is not [01:12:00] reinventing the wheel. Stuff like this works in other modalities too. And it's about doing it with the right partner. You know, I think it's, it's... It's important to us to do this in the right way, and that was both from a technology standpoint, but also kind of from a legal or moral standpoint of we're not trying to make it so that you can listen to, uh, uh, I don't know, Frank Sinatra, um, singing the latest Adele song or something like that, you know, and so, um, for, Us, this is basically partnering with artists to let them use their voice in new ways, um, and not giving this to everyone, or partnering with aspiring artists to let them use their voice, um, in new ways.

MikeySuno: You know, I can tell you, like, just playing with it myself, when you have a, uh, a model with your voice, it's like, it's really very different to hear something sung in your own voice. Um, and it's very differently inspiring to hear something sung in your own voice. And so this isn't going to drop anytime soon, but it's something that we, we know how to do, and we work with a couple of artists on this and, [01:13:00] and in the future, we'll definitely be giving something like this to everybody.

Gavin: That's a really interesting, like, that's like that RVC program, right? Is the idea, basically, you'll be able to sample your own voice and you're not gonna be hearing you sing the Frank Sinatra Adele song through Suno, but you could hear you sing this original song in a way that you may not be able to sing on your own.

Gavin: That's a very powerful thing, I

MikeySuno: Yeah, totally. Totally. And I think, I think, um, it goes beyond RBC, because it's like, RBC will let you swap voices, it can be finicky, but it lets you swap voices, and now, I don't even need to swap voices. I can just type, just like, Mikey singing this, you know, uh, funk song, in, like, a really laid back, smoky jazz lounge, and like, that'll just come out.

Kevin: , I gotta imagine Doja Cat loves hearing Spongebob sing every one of her songs though, right? Like let's not forget that artists really do love that sort of stuff. 

MikeySuno: Let me, let me tell you guys just, just an opinion though, but I actually think a lot of this, , voice swapping stuff is not the long term thing here. I think it's a little gimmicky. It's like super meme y, it's super viral, but it's [01:14:00] actually like fairly like disconnected from what we are trying to do.

MikeySuno: So we like largely just ignore it.

Gavin: Yeah, ,, . Is there a paid version of your program? There is. So that's

MikeySuno: There is

Gavin: your business model. Yeah,

MikeySuno: That's the business model. Um, so, now you can, you can do, uh, five chirps a day, and then if you want more than that, you have to subscribe for ten bucks a month.

Gavin: Got it. Okay. So it's kind of the mid journey esque version of the same business model.

MikeySuno: Yeah, I think so.

Kevin: And then what are some of the cool, unannounced, like, premium plus features that'll be coming down the pipe? These multiplayer things, these longer song generations, the clone your own voice. Go ahead, you

MikeySuno: Yeah, yeah, the, the, so I am most excited, I think, about the longer song thing, because I think, 40 seconds is like interesting, but I think once you get maybe in the 90 second or two minute range, um, you begin to actually be able to bring people on a journey. So I, you know, just like backing all the way up, you look at 20 years ago, most like kind of hit songs are five minutes and you look at now most hit songs are like three minutes.

MikeySuno: And I think, um, Let's just say it very crudely, like, um, a tick tock of a song, like [01:15:00] 60 seconds of a song is very different from a song that is meant to be 60 seconds. And so I think that is the experience that we can give people. And so we can make a kind of a new medium here, um, that I'm excited to give people.

MikeySuno: So longer generations, for sure. , look out for way better quality. , way better, both the song quality, the audio quality, and the, the ability to kind of describe what you're going after. Very, very exciting. Um, look out for some increased speed, that's always nice. I'm sure you guys are sick of waiting for the slot machine to

Kevin: You know, it's, it's, it's plenty fast, but hey, I'll always take faster.

MikeySuno: So, those are the immediate ones, and then those other features are coming soon. We're still, we're still kind of baking those out, but the things like in painting, like, Oh, that song was perfect, but this one little bit just didn't quite have it. Um, or do the same thing again, but a little more up tempo.

MikeySuno: Um, something, something, or just like, give me variations on that, because like, I think it's on to something, I can't exactly describe what I'm going after. And so, so just play with that.

Kevin: You've just, you've just vocalized every thought that I've ever had while playing in the Discord. Like, [01:16:00] again, it's, it's always an overall enjoyable experience, but I just, I get so excited for what the future of this is, and knowing that you're already working on the next thing after the next thing.

Kevin: Continues to, uh, to keep my interest peaked. 

Gavin: Mikey, before we go, I have one last question for you. If you're going to give somebody who's never used Suno before or Chirp before a really good prompt to start with, what is one of your favorite prompts to use to get a good song out of Chirp?

MikeySuno: so for, for, you know, some, some genres we are just better at and some we are less good at. , two that come to mind that are fairly different. So hopefully this covers a large swath of people, but like pretty good at EDM. Um, and so like doing something like emotional EDM, uplifting EDM, upbeat EDM.

MikeySuno: Um, those are all, those all produce good, good music. And then the other is bluegrass. Um, we're

Gavin: no way, really? That's fun. I'll have to

Kevin: bluegrass and country are really

MikeySuno: And so, so, we have hypotheses of why these, this is the case, but I, I don't think any of them is,

Gavin: With our crossover genres, of course, clearly. There's a [01:17:00] thousand bluegrass EDM artists out there,

MikeySuno: I'll tell you this, actually, I had an amazing moment of like realizing, like playing with it with someone just like a one on one testing of it. And it's like, they actually had this idea in their head of something that was like bluegrass. They didn't even know the word bluegrass and they just wrote like, Oh, what is that thing?

MikeySuno: And they, and they wrote, they wrote like mandolin, banjo, violin. And like it comes out as a bluegrass song, like, and so like, this is amazing, right? You don't need to have the vocabulary, not that bluegrass is such an esoteric thing, but um, you don't need to have the vocabulary. Yeah.

Kevin: I love it. Yeah. Well, the upbeat, uplifting house, you know, EDM banger that totally works, but just, uh, you know, not necessarily from me, but someone would love if you could just got sad house. real depressing, , melancholy

MikeySuno: I'm going to send you one. I'm going to send you one.

Kevin: Okay, yeah, and clowncore, we need better clowncore.

Kevin: But it's fine, I've got very specific tastes.

.

Gavin: All right, everybody, that's it for today. Thank you so much for joining us as always with our show. We are new Well kind of new we are [01:18:00] we are new ish and we always need y'all to give us reviews on Apple podcast Please go review us on Spotify Like and subscribe on our YouTube or tik tok channels.

Gavin: We are creating original content for both of those So make sure you're checking those out And thank you so much for listening. We are always so happy to hear from people. We engage with people online all the time. We respond to comments. So leave us a note, uh, and let us know how you think we're doing.

Kevin: Yeah, and Gavin, I was sent a little special something by one of our viewers. , who I know, but I'm not going to allude to. But they have been... I'm gonna take credit for it. I think we should. Uh, they have been inspired by our podcast. To dabble in A. I. filmmaking, and they sent me a little something that I was delighted by, I sent it to you, and I think you were similarly tickled, so we asked if we could play it for the audience, and, , they graciously agreed.

Kevin: Gavin, thank you for the episode. That was a lot of fun. , thank you to Mikey from Suno. And now here it is, the world premiere! An AI for humans, juicy sluicy, [01:19:00] of Backwoods Munker

[01:20:00] [01:21:00]