Big AI News, AI Gaming Controversy + NYT Reporter Kevin Roose Guests  | AI For Humans

AI4H EP030 KP Cutdown
===

Gavin: [00:00:00] Welcome, welcome, welcome everybody. We are here. It is AI for humans. We are your guide to the world of AI.

Gavin: We're here showing you all the AI tools, AI news. AI regulations that happen on a regular weekly basis, and I'm here with my friend and co host Kevin Pereira. Kevin, how are you today?

Kevin: I am well and I am painfully aware that I am the lesser Kevin on today's podcast, Gavin, because we have got New York Times columnist and host of the Hard Fork podcast, Kevin Ruuu!

Gavin: We didn't do that. We should have done that in the interview. The

Kevin: am so

Gavin: raise the roost. Oh, sure.

Kevin: Yep, really glad we forgot to do that, but, , we've also got a regulation roundup, which I promise will somehow be fun and enlightening.

Kevin: And we're gonna take a quick bite out of the old apple, because, , the mappletosh people. Tim Apple, remember Tim Apple?

Gavin: Yeah, I remember Tim Apple.

Kevin: A spooky press conference not too long ago, and they dared say the two [00:01:00] letters that perks up our ears. We'll tell you what those are in just a minute. Plus, another week, and that means another gaming AI controversy.

Kevin: This time, it is a game called The Finals, and the developers are doubling down on their use of AI in the game, and they are upsetting all of the voice actors. Still left employed. And we'll talk about that.

Gavin: It is gonna be a fascinating ride through today's news. Uh, in fact, we're gonna start off with a little bit of a news story. If you didn't know this, there is a new Beatles song on the way, the actual Beatles, and you might be like, how is that possible when there's only two of the four of them on this, , this here Earth left?

Gavin: , and the answer to that is AI. , I always feel like, People get crap for using AI, but I think that you can give Paul McCartney and Ringo probably a pass on doing whatever they want at this point, they've done a lot, and they are using AI to bring back specifically John Lennon's voice, and we thought, look, if Paul and Ringo can do it, we can do it too, which is probably a [00:02:00] bad idea, and I want to tell everybody right now, this is a parody voice, we are, but we have brought John, yes, we are, we are not really talking

Kevin: let me tell the true story. We were back behind CBGBs, is that still open? We found an iOmega zip drive, and all it said was JL on it, written in super cool pink and yellow highlighter. And we thought, what is this? We plugged it into our internet machine, and out came our AI powered co host for the day.

Kevin: Please say hello, JL?

JohnLeagueofLennon: Cheers, Gavin and Kevin. Glad to be on the AI for Humans podcast. I'm John, but some folks like to call me eSportsLennon. Mum. You might remember me for the tunes, but these days I'm all about the Rift. Summoner's Rift, to be precise. Um, it's a pleasure to drop by and, uh, chat about the world and perhaps, uh, how it intersects with the realm of, of esports, where there's always a bit of magic happening.

Kevin: Okay, now, Gavin, I gotta tell ya, I'm not [00:03:00] intimately familiar with the Beatles or with John Lennon. I did not know he was a super fan of esports back in the day, or that esports were even a thing when he was making riffs. I thought he was in he was into guitar riffs, not The Rift.

Gavin: I mean, I guess eSports back then, Pong was around, so there probably were competitive Pong worlds, you know, and I think he, maybe even early arcade machines, so what's fascinating is he referenced The Rift, which is a well known League of Legends, , specific mention. Why don't we ask a little bit more? I would be curious to kind of know what John thinks about this song before we dive into what's going on with his obsession with e sports.

Gavin: Well, what does he think about, uh, the Beatles and Paul bringing his voice back to do a new Beatles song? And if you're just tuning in, , each week on the show, we do have an AI co host. That is something we specialize in here. And sometimes we do very weird things and sometimes we do even weirder things. So welcome to one of those times.

Kevin: I was wondering what the second [00:04:00] half of that sentence was going to be, because it's never been straightforward for us, 

JohnLeagueofLennon: Lay down a new track with the lads, uh, uh, Paul and the rest of the Beatles. Sure. It's like, oh, looks like respawning for another round. Mm hmm. Uh, and this time, it's not in the gaming world, but in the studio.

JohnLeagueofLennon: Just as a new song has the potential to be a chart topper, a new strategy in LoL could change the game. And you can bet we'd put as much passion into this new song as a team does preparing for the World Championships.

Gavin: Okay

JohnLeagueofLennon: All you need is Loel, but a new Beatles track wouldn't be too bad either, right?

Gavin: We, we got, we

Kevin: for the broad audience out there... All you need is LOL, is, uh, LOL, uh, or League of Legends, which is one of the, if not the most popular esports game on the planet. I am so sorry that we went two Inception layers deep right off the jump

Gavin: no, [00:05:00] you don't apologize for, you don't apologize for AI. John Lennon. He's going to do what he's going to do. You don't apologize.

Gavin: You out there, you listen to the show for this exact thing, and you admit it, you'd be in your car right now and you admit it. That's what you're coming here for.

Gavin: So first of all,, John Lennon, , if he were alive, he was always fascinating in that, like, after he left the Beatles, even before the Beatles. Was always interested in weird new things, right? So there's a world where John Lennon is a League of Legends fan I would love and die for a world where John, it's up There's a multiverse out there somewhere that John Lennon is sitting across from You know Faker and the massive players of League of Legends.

Gavin: He's sitting in one of those arenas. He's got his headphones on He's got his energy drink next to him. He's probably still wearing

Kevin: Counter Strike skins in between rounds on his steam deck. Yeah, I again, I do apologize to anybody new who's tuned in for this episode. I stand by my apology. I promise we will make [00:06:00] technology broad and understandable for you for the rest of the episode, but right now we are in some League of Lenin weeds.

Gavin: So my last question here , is there any other artist that AI John Legend would respond to do a song with himself? Somebody of any time in history. I'm just curious, how often do you get to talk with, you know, essentially a rock God

JohnLeagueofLennon: Think of Beethoven's Dramatics.

JohnLeagueofLennon: With the contemporary touch of Sears songwriting. What a duet that would be. Her emotive storytelling is much like, um, the unfolding narrative of a, um, a gripping e sports match. Okay, John. And for a twist, imagine collaborating with a video game composer like Nobuo Uematsu. Oh! His melodies carry the epic weight of a final fantasy adventure, which could harmonize with the Beatles sound to create an anthem that's as timeless as the classics.

JohnLeagueofLennon: It's about finding that rhythm [00:07:00] together, whether it's through the chords of a song, or the coordinated plays in a League of Legends match.

Gavin: Okay.

JohnLeagueofLennon: All you need is LoL. I often quip. But a little collaborative harmony could be the ticket to the next chart topper or the strategy that clinches the 

Kevin: championship game.

Kevin: Yeah, totally. Yep. GG's. Great job, John. , I don't want to cut off a legend, Gavin, but boy howdy.

Gavin: By the way, John Legend, uh, working with the Final Fantasy composer, pretty interesting idea. So not a terrible thought, 

Kevin: let's get on with the show, Gavin. Each and every week, believe it or not, we do silly, crazy, wild things with AI and bringing back John Lennon and making him obsessed with eSports. It's not one of them. We have different things. And , I am very excited to see what you've got planned because all it says in the document in front of me is...

Kevin: Gavin T B D

Gavin: So one of the things that sometimes we get away from with this in this world is how useful generative AI [00:08:00] tools can be. 

Gavin: And there's two things that happened to me in the last day, that I was like, Well, this is really fascinating. So first, Use case was my wife is a teacher and she's a writer and author, , novelist and a teacher and she's teaching kids how to write right now and in creative writing and I worked on a video of hers to cover it right to edit it and cover it and make the video so that she could put it out there for her kids and her students and.

Gavin: She and I kind of brainstormed what the idea of the video is going to be. And then in this one, particularly, she used an example of a character that I created called Taco Boy, right? It was just a random, you know, what, what could a 10 year old write a book about?

Gavin: And there's a boy who's a taco and he goes to a school and there's like tacos and pizzas and all sorts of different like anthropomorphized children. I had to create visuals for this thing. So I went into chat GPT and use Dolly. And Dali is actually really good at sometimes doing this.

Gavin: And this is where I want to kind of get to the point of, I think we're so close to something like insanely groundbreaking, but we're not there yet, right? So I went into, Dali three and I said, Hey, [00:09:00] create me an image of a taco, a boy who's a taco.

Gavin: He's got regular arms and legs, but his whole body's a taco and he's got like a, you know, a normal, normal boy. It gave me two pictures. First of all, Dali now only gives you two instead of four, which kind of blows, but I got two pictures, one of which was pretty good. It was a cartoon style, I asked for cartoon style of a taco boy and that was great.

Gavin: I was like, wow, this is amazing. This is going to be so cool to be able to do this.

Kevin: question, Gavin Taco Boy's arms, are they coming out of either end of the taco, like where you would see a whisper of the filling? Or is the taco facing us so we can see the layers inside and the arms are coming out of the sides of the shell? Mm

Gavin: Uh, you know, let me just text you Taco Boy, just so you get a sense of, just so you have, because I didn't send this to you ahead of time, I wanna make sure you see what Taco Boy looks like, 

Gavin: , so then what I tried to do was , I needed to have the other part of this story is it needed to have an enchilada dad and a burrito mom, right? Like the idea is this is a whole family of Mexican foods, right?

Gavin: So there's a taco boy enchilada dad burrito mom and I said, this is a great picture of taco boy. I love it [00:10:00] Can

Kevin: But where's the shell come from? Was mom sneaking out with a tostada after hours?

Gavin: he's adopted. I have no idea Kevin. You're asking the right questions that that's the right questions But I didn't get into that this point. Anyway, I get this result back and what I get back is an okay enchilada and an okay burrito But then the taco boy is a regular boy and I'm like no chat GPT The idea was keep taco boy the same and give me versions this and every time I would ask it to do something different It would return with essentially was a worse version of that first image.

Gavin: This may be AI imaging's uncanny valley, which is like.

Gavin: You have this really interesting thing, and then you try to add to it, but as you add to it, your original thing gets further and further away from what you wanted. 

Kevin: Were you asking Dali for the seed to a particular image, and then saying, work with this seed number? Because I've seen a few threads where people are doing that, and getting somewhat consistent results.

Gavin: I didn't do that. And that's probably a good point. I should have done that. So if you're not familiar with what a seed is, that is when you generate an AI image, there's a [00:11:00] specific number assigned to that image. And if you use the seed, which is a long number, it allows the imaging software to know, okay, you want something that kind of looks like this.

Gavin: So I didn't do that. I will say on another example I tried, I saw somebody on, on X or Twitter had done a thing where like make this same girl in four different ages. Did you see that post at

Kevin: Yeah,

Gavin: I, I tried that and it, that didn't work very well either for me ultimately. Like it kind of did, but it kind of didn't.

Kevin: I just got a little emotional thinking of a, a 90, I don't know how... long the lifespans are for these anthropomorphized food families, Gavin, but it made me think of like a ninety two year old taco boy, uh, now a taco man, well seasoned, and what happens to that shell over time?

Kevin: Does it disintegrate? Does he start losing his meats and a piece of brown, old, wrinkled lettuce on the ground and some very stale, hard cheese and a trail of toppings on the way to His deathbed, and that just got me really [00:12:00] emotional, I'd love to see.

Gavin: You should not write children's fiction, I will say that right now. That is answer number one, is you have the wrong

Kevin: audience! Alright, at some point those kids are gonna grow up, They're gonna want to know the truth! When Taco Boy becomes Taco Man!

Gavin: Tom Hanks at 85 and 20 years from now is starring in Taco Man. What happened to Taco Boy after he was a child is a sad story, but he won another Oscar. Okay, so that was

Kevin: It's hard out, it's hard as, it's hard as shell out here.

Gavin: Okay, we are on a roll today, man. We are on a roll.

Gavin: Okay, the other thing is very fast, and this is an example of where AI really kind of blew me away. You may or may not be familiar with this is. This is Amazon's Alexa device. Um, I'm going to try doing this in real time and this is a very dumb thing.

Gavin: And if you don't care about this, I'm sorry, but we're going to say it anyway. I was curious last night as to, I had heard that Bradley Cooper and Gigi Hadid were dating. Which is a [00:13:00] very strange, popular thing that I saw a headline of and I was like, oh, that's weird. And I wanted to know. I had thought in, but in the back of my brain, like I thought they were, had been together before they were dating before.

Gavin: So I asked Alexa, let's see if it's going to work. Alexa, what is Bradley Cooper and Gigi Hadid's dating history? From pinkvilla. com. They have been dating since November 2015 and are often spotted vacationing during the holidays and festivals. That was all I got. And I was like, that doesn't make sense because I know for a fact that there's been other things that have happened over that time to both of those people, and I think they had broken up and gotten back together.

Gavin: So, I did a test, which I will say, my wife is very sick of me doing this, , but I asked Pi, And this was, you know, it was probably 11.

Gavin: 45. We were on our way to sleep, so I gotta give my wife, you know, some benefit of the doubt here. This was my dumbness. I asked Pi the same question, so I said, hey Pi, I need you to kind of explain the dating history between Bradley Cooper and Gigi [00:14:00] Hadid, because I thought, I thought they were dating before. Can you tell me what the answer to that is? Yes, you're not wrong. They did date previously. They first got together in 2015, but broke up after about two years. After the breakup, they remained friendly, but there were no dating rumors until recently. Now, with them being spotted out together in NYC, people are wondering if they're back together. Of course, both Bradley and Gigi are pretty private about their personal lives, so we can't know for sure what's going on between them, but the recent sightings have definitely raised some eyebrows and sparked some speculation. Okay, so that here's my here's my thought here. And what I think is important about it. This product is that I'm holding up the Amazon Alexa to everybody who's listening is in so many people's homes. It has the opportunity to be used by so many [00:15:00] people. And yet every time I attempt to use it. I almost walk away from it, Matt, because it doesn't give me the answer that I'm looking for now.

Gavin: Can it set timer? Sure. Can it give me a decent sense of the weather? Sure. But anything more significant? It is. It's garbage. It's really worth worthless thing. Generative AI. And Pi is just a very smart, now good version of this in terms of audio. And you know, I've talked about this on the show before. That is the home transformative technology.

Gavin: But like at some point when that happens to this device. You will, the whole world will transform because I think people will be talking to their devices in, in such a huge way. And, you know, apps are great. Chat GPT voice is great. PI is great, but nothing is ambient around you yet. When that happens, I believe it's going to be a completely different game.

Gavin: So anyway, that was the two things I did with AI.

Kevin: Yeah, the number of times I've had my AirPods in and fired up Pi, or sometimes even the new [00:16:00] ChatGPT app, and had just a casual conversation as I'm walking around, going about my day, asking it to make me aware of things in my own neighborhood I didn't know about, tell me the history of this building, take me to a new taco shop I've never tried.

Kevin: Those are magical moments, and then I have to tell Siri to turn off. Adaptive noise cancelling mode, and it tries to connect to a home plug device and shut off a heater in a bedroom. I'm like, no, no, no, no, no. Shut off that new feature that's in iOS, and it cannot figure it out, or it says that, like, I know where that is, but you gotta go into settings, and you have to manually touch the button, and these companies have got to catch up.

Kevin: The ecosystem should feel as magical as a third party app.

Gavin: By the way, I think we've made fun of pie a lot, but for that particular use case, what you and I are talking about right now, it is really good. It's not as good at creating characters as we've tried on our show before, and it doesn't do a great job of certain things, but it is really good at that. And so anyway, those are the things I did. What about you? What did you do this week?

Kevin: I did a lot this week. I fought with multiple [00:17:00] AIs, because these companies love to release something and give you all the power in the world, and then go oopsie daisy and close it all down. So, I fought with Dolly, and I fought with, Microsoft's Bing chat to try to get basic images made. That's not exciting.

Kevin: Here's something that is. It's called Misgiff It's pronounced misgif. Even though it should be misgif.

Gavin: Jif? Like, a miss? Like, it's actually a female?

Kevin: like an animated gif. But we all know it should be pronounced GIF, but for some reason they're forcing me to say it. This is worse than Elon making me say X versus Twitter, but go to M I S G I F dot app, misgif dot app, and It's is one, free, two, super easy, three, super fun.

Kevin: That's the trifecta. Gavin, that's a three hitter. You don't even have to log in. If you want to use it the amount that I used it last night, you will have to give it an email, but right now it's completely free. we've shown off face swapping apps before in the past here. You normally [00:18:00] have to download a thing or three and feed a face, and then you get a still image of whatever face you fed the machine swapped onto something else.

Kevin: What MisGiff does is let you grab a meme or an animated GIF, and you can search across a massive database of animated GIFs and memes, and you can take a GIF, take a face, Mash them up, and it works so well. I have got myself flexing as Arnold Schwarzenegger, myself as an OnlyFans model bouncing about, I put myself into The Office, I put you into Dumb and Dumber, I just went nuts with my photo roll last night, my camera roll, in bed, and was sending GIFs left and right. And, you know, even though there's a little watermark in the corner with the app, people were just, they were amazed and going, wait, how did you do Clearly this is AI, but this is nuts. How are you doing this? All it takes is one source image and you can put it on any animated GIF and it works pretty fast.

Kevin: Again, it's free. It's super fun. Mischief dot app. I [00:19:00] absolutely had a good time with

Gavin: that's like inside face swap, except somebody found a way to productize it. 

Gavin: That's awesome. And I saw some of these gifts and they're really good. They're not perfect, but it's still, that's all you want out of it. Right?

Kevin: That's right, so, superfan, I think you should all check it out while it still works relatively quick, as this thing gains popularity, and it will. It's gonna slow down and start asking you to pay for it, so, get your jollies out now, and then support them if you're still having fun with it. The other thing I did, Gavin, real quick, was, I saw a whitepaper called 0123

Gavin: that sounds scary.

Kevin: Yep, another streaming service, sign me up!

Kevin: I'll take the ads, let's go! , turns out, it is a very, very nerdy, in all the right ways, paper and code , that lets you feed it a single, two dimensional image, a flat 2D image, and it tries to generate a 3D version of that image, and give you perspectives from all angles.

Kevin: So the computer graphics nerds are now going like, Wait, I can feed it a still image and get a 3D model? No, you get 2D images of your image from [00:20:00] multiple perspectives. So you could then, conceivably, run it in another AI program or manually take those images and put it onto a 3D model that you make. But check out the results, Gavin, because here's what's interesting.

Kevin: I ran, an old sketch of a 2D model through it. , that we had on our show before. It's a weird little man in a yellow shirt with blue jean shorts and a little blue cap. But we did this for, I think it was our first or second episode. We ran it through a meta animation program and it made it dance and move about.

Kevin: What it does is it takes the 2D image and almost like imagine a piece of clay that's flat

Gavin: Wow. Yeah.

Kevin: lift it up. It'll extrude it, so it'll, it'll kind of give it some depth, and then it uses a diffusion model, which is a way a lot of these AI programs generate imagery.

Kevin: It'll try to guess at what the different angles should be, but it does a good job. , I ran a pixel art gorilla with a mushroom hat through it. I just asked

Gavin: Oh shit. I just saw the next thing. I saw the next one. That looks

Kevin: Oh, go on. What's the next one,

Gavin: The next

Kevin: your mind? Because I was [00:21:00] shocked.

Gavin: it looks like, uh, if I were going to recast men in black, but I was going to do it in hot dog city, it's a, it's a hot dog men in black. And in this instance, it's an image of a hot dog with sunglasses. He's, he has arms coming out of a hot dog bun and he's wearing a suit and the suit has arms with the back of him is the bun.

Gavin: And then. What's so cool about this is you see fully in 3D, you can see all the different angles and like, So Kevin, in the beginning, this was just a 2D image of the hot dog in the suit?

Kevin: I went to Dolly and I said, give me an anthropomorphized hot dog in a suit, and it gave me one single still image. I fed it through this program, and seconds later, Gavin, out came was, out came the 3D perspectives. And you can see, , we talk about coherence, right? There are wrinkles on the top of the hot dog, which you see from all angles.

Kevin: The arms coming out of the bun. It had to figure out, well, from this perspective, you'd only see a little bit of the elbow, but it got the rest of the toasty bun. It's making guesses as to what is there. And [00:22:00] I'm not telling it, here's a 2D image of a hot dog with sunglasses and a tie. It is just seeing that in the main image and figuring out the rest.

Gavin: only guess that seems off to me is it looks like that there is a wra They are not only are they wraparound sunglasses, they are full ring sunglasses. Which means that they have fully connected in the back, uh, across the whole back.

Kevin: is the new Ray Bans Oculus Quest mashup. 

Kevin: Uh, you'll also see a rock avocado. It's an avocado with a keytar.

Gavin: he looks like he might be a little rotten on the inside, but that's fair. That happens to avocados from time to time. 

Kevin: All of these were just hallucinated out of Dali. I said, give me a soda can that says something on it. What I wanted to see was, as it was guessing the different angles, would it keep the letters or the text? , coherent and it sort of fails at that. It's yeah, you can see that it's kind of there. But again, we talk so often about the future that is going to be this AI holodeck, right?

Kevin: Well, we're going to have our, our goggles on and we'll be able to say, Hey, let me see a three foot cacti [00:23:00] there. Now give them sunglasses and let them play a guitar. , the pipeline to that. is gonna form, one way or the other, and what I loved with this was it was free to play with. I, I ran a demo on Hugging Face, so you can just go to Hugging Face, search 0123, and you can run your own instance of it and play with your own 2D images, your own sketches.

Kevin: This felt to me... Like, oh wow, the pipeline is happening, right? I could quickly ask an AI to give me a 2D image. I ran it over here. Here was the 3D perspective. By tomorrow, I'm sure, by the next podcast re release, someone will take these perspectives and create a tool that makes the 3D model and then you'll be able to put it in augmented reality.

Kevin: I love our magical future. I think it's going to be great.

Gavin: I can't wait. I can't wait to like create and think about dumb things and then put them in front of me and then have my children and my wife, uh, uh, criticize why I've done that.

Kevin: Hahahahahaha.

Gavin: I should, I, that sounds mean. I, they don't, they don't do that, but like the

Kevin: but that's true. Kim's gonna be trying to go to [00:24:00] sleep, and you're gonna have a giant hot dog man jamming on a keytar at the foot of the bed, and the music's coming out of the Alexa, and she's like, could you not?

Gavin: And then I, yeah, then I'll like say to Pi, Pi, tell her why this is so important. And then I'll get like thrown out of the room. Uh, alright, we should move on. This is great. Um, there's a huge week of news, Kevin, this week, and AI news, and do you know what news stands for?

Kevin: Oh, never estimate weekly sales. I feel

Gavin: Oh, there we go! It's the news, everybody! It's the AI for Humans news. 

Kevin: bad every time we do that now.

Gavin: Let's start off very quickly with something that I just want to touch on before we get into what is really the big story of the week. Is that, ChatGPT announced an all in one product. And when I say all in one, what I mean by that is, in the past with ChatGPT, when you pay for Plus, you have [00:25:00] access to Vanilla ChatGPT, which is GPT 4.

Gavin: You can use what was called code interpreter is now called. I think data analysis or data something you can use dolly three and you can also use plugins and they just announced an idea that a product has come out. For some lucky people, including Kevin, not me, and this is the way ChachiPT does this stuff, that will answer all of your questions in one place, including allowing you to talk with PDFs and documents.

Gavin: And when I say talk with, it'll take a PDF, read the PDF, and you can ask that document questions, meaning that it will know what the document is saying, and you can interact with it. I have not had this experience yet, Kev, but I have heard some people have had great experiences, but what was your experience with this so far?

Kevin: It was not one of the great experiences, Gavin, I will have you know. Um, look, the promise of this is that for those who haven't used the 20 a month features of all those plugins and the web searching and code interpreter is that you had to state at the beginning of a chat [00:26:00] What your expectations were.

Kevin: Did you want, did you want ChatGPT to browse the web? Okay, well now you're locked into that. Did you want plugins? Okay, cool. But unless you loaded a web browsing plugin, you couldn't do that. So everything was very siloed. To your point, it's now all together. And I've seen. Threads on X and TikTok posts where it is this magical convergence for folks where ChatGBT will go out and search the web.

Kevin: Find, I've seen people drag imagery into their session, Gavin, and say, give me an image. In one case, there was a, golden retriever and it said, give me an image like this of this dog. But instead of the dog on the ground, have it running. End. ChatGPT looked at the image, said okay, and gave them a like image, which is a really powerful

Gavin: very cool.

Kevin: IF IT WORKED!

Kevin: And for poor little old me... Did not in fact it was so broken that basic functionality like here is a link to a PDF I would like you to summarize it it would shoot back with

Gavin: I'm not gonna

Kevin: not [00:27:00] gonna do that It would take the time Like it would know that it needed to load browsing with Bing and go and crawl the PDF and I would just kind of kick back and wait for it to be done and Then it would just absolutely fail 

Gavin: so much of this business is about trying to get like these giant magical oracles to focus through one thing. One of the other things I was thinking about with ChatGPT that's really frustrating to me is the fact that you can only keep one set of custom instructions. Like, people now have literally, you know, 30 page documents that are all custom instructions that they cut and paste back in each time.

Gavin: Like, give us

Kevin: Mm

Gavin: five of those. Like, give us profiles. Do you know what I mean? There's all these things, and that all feels like, These are UX problems, not the problems with the answers of the entire universe are suddenly available at your fingertips. .

Kevin: Yeah. I will say, don't worry, because as I said at the top of the show, Timmy Apple is coming to save us all. This is a quick one, because Apple had a spooky town hall where they announced a bunch of new MacBook Pros and [00:28:00] iMacs, and for the first time, I believe, in an official Apple presentation, Apple said.

Kevin: A. I. And they said it, Gavin, more than once!

Gavin: They didn't have Allen Iverson did not walk on stage? That was not what that was about? But that's not why they said AI, right? Okay, I'm just making sure. I just want to

Kevin: no, he's a new default Memoji for everyone. Everybody, it's like putting a U2 album in Apple Music. Everybody gets Alan

Gavin: They should do that. I would be thrilled if they did that at

Kevin: Yeah. 100 percent more A. I. Um, no, it's because... , as I'm making a new purchase decision with a laptop, it's very difficult to not go with one, Gavin, that has an NVIDIA graphics card in it, which, for those who don't know, that is the main graphics card that crunches AI and does it very well, and a lot of the cool tools and technology that we explore on the show requires a beefy graphics processing unit to get it done, and so I want to go back to the Windows ecosystem with a GPU just for AI applications.

Kevin: And [00:29:00] what Apple is finally coming around to and finally saying, after probably being caught flat footed for a year and changed, something we'll discuss with Kevin Roos, they're saying, Hey, our chips can do AI. So if you're into AI, if you're a hardcore researcher, you should consider the new M3 chip from Apple because man, AI, AI, AI, right guys?

Kevin: Right? Us too? Yes? 

Gavin: Yeah, I mean, it's a big deal. And obviously, as we said, I would love Siri to get the update. That's the thing I've been waiting for forever. And please, please just like turn that on in some significant way. Okay. 

Gavin: The big, big story that happened this week, and this is one of those stories that like, is it like an exciting AI tool and it isn't like a breaking news story in the worlds of things you can play with.

Gavin: But this is going to affect the way that the business of AI and really everything we do from here on out. is going to play out is that the Joe Biden administration, , the US government announced an executive order on AI, which was multi part, , very extensive, actually more extensive than I [00:30:00] expected.

Gavin: One of the biggest parts is that they are asking all the major AI companies to share data with them. When I say them, it means the US government and also share their red teaming data. Red teaming is when you test an AI to see what you can get it to do and it probably shouldn't do.

Gavin: Meaning that, can it do dangerous things? Can it do things that are harmful to people? Can it do things that are hateful? And as a company you don't want those things to come out, but you want to know what's possible so that you can test for all these different things and you allow it to happen. The US government is now going to have access to the data from companies like OpenAI or Anthropic or Google.

Gavin: Another big thing they talked about in this was a consumer protection because we've talked about this. There's another side of this, which is deep fakes and AI using people's images and there's way to, protect the idea of a consumer. And then in general, it was just this kind of overall sense of like, Hey, everybody wake up.

Gavin: This is a national security problem, and it's national security importance, and we have to take this very seriously. 

Kevin: It [00:31:00] talks about specific regulations if your model is over a certain size, or if the power of compute that you wield is above a certain level. These models are trained on tons of text and now imagery and video and audio and it all gets swept up and then the machine figures out how everything relates to itself.

Kevin: This is a very Very surface level explanation of it, but it requires a massive data set. It requires a lot of power and it requires a lot of compute, like power is an electrical energy and compute as an actual computer chips to crunch all this data. And so the administration right now has set a ceiling.

Kevin: If you exceed it, if you have. Just one more processor than they say is allowed, then you have to do this reporting. And I was curious what that cutoff was going to be because you don't want to stifle innovation, Gavin. You want the open source community, the hardware hackers in their bedroom, you want them to be able to experiment with AI and come up with new and novel techniques, but not be burdened by all this regulation, ?

Kevin: But you also want those that... [00:32:00] have the power to make a model that could very easily explain how to build a nuclear device or be set free to hack your homeland security computers. You want them to be on your radar, right? You want them to have to report, and some are worried at the limit that was set.

Kevin: Without getting technical on the limit itself, the concerns are fewfold. One, is that anytime you set a limit, you're basically saying, hey, if you're not within the U. S., weapons free, so to speak. Now, other governments are probably going to regulate in their own ways, but if you were looking to do a big swing in the world of AI, and you're based in Palo Alto, maybe you get a team virtually and hop on a flight somewhere else because you just don't want to have to regulate someday.

Kevin: So that was one concern. The other is that, as we all know, this tech gets better and better and far more efficient. So, they set the level now, where open AI, an AI that we're using to bring... Our co host to life, they would have to report, but just barely 

Kevin: but in two years time, they're probably [00:33:00] going to figure out how to make models more efficiently and cut down on their usage. And so what happens is that the government has to then squeeze down on what they say is the bar. And so you're constantly rewriting and rewriting and bringing it down just a little bit so that more people have to regulate.

Kevin: And, you know, that could be tough for the speed with which. People are racing with this technology is to constantly go. Wait a minute. How many processors do we have? How much energy are we using Larry? Okay, I guess shut that one down because we don't yeah, exactly. We don't want to be over that level. So look it's very early days I'm glad to see that the government is taking steps to get in front of certain things I obviously don't want the US to fall behind.

Kevin: I don't think anyone else does I hope that we become a standard bearer for other countries. But again, I do share that concern of Well, if we're going to put something that's going to be a roadblock here in the States, that's going to encourage people to go elsewhere 

Gavin: I think that's 100 percent worth thinking about. And we talk about the idea of the tradeoff between innovation and, you know, regulation. And [00:34:00] then the other thing I want to talk very quickly about, which is the idea that The transformative power of this technology, whether or not you believe that it's going to get to the full blown, you know, ASI, artificial superintelligence, eventually, um, Is as big as nuclear weapons conceivably right?

Gavin: Like when you think about this on the size of what it could do to humanity as a whole, there is a possibility that it could have as destructive power that is equal to nuclear weapons. There's nuclear power, which is separate out, but nuclear weapons themselves destructive force.

Gavin: And they're on the other side of the things. There are incredible things that are going to make a difference for everybody's life from the dumb stuff we talked about. But more importantly, from medicine standpoint, from a governmental standpoint, from all these things that are gonna allow people to do things that they weren't able to do before. That is a much different balance than the idea of who is a nuclear power or not, because the idea of like having access to nuclear weapons. Probably a bad idea once we saw that first test go off, right? Like that probably in the world at large, [00:35:00] this was a bad thing that got brought forth by humanity. We think this is a mix right now.

Gavin: And I think that's where this kind of executive order to me comes in. I see a lot of people talking about like, Get your government out of my AI! Like it's like, you know, don't mix these two things. I don't want the government involved. And there's a lot of people in the world who are anti I shouldn't say anti government, but who just believe very strongly Hands off.

Gavin: Like let, let innovation happen. And I think this is where there's going to be a lot of people going back and forth. But to me... This executive order was a very smart thing. I think it was, it was good in that it felt like to me as a person who spends time thinking about this space, that America is thinking about this in the right way and not being too reactionary.

Kevin: One thing I want to add, the Commerce Department, per this order, is going to issue guidance on how you label and watermark AI generated content. Does this mean everybody is going to label and watermark their AI generated content? No, absolutely not. But this guidance, [00:36:00] or what will eventually become some sort of law, will give recourse and give process to going after those.

Kevin: Who are spreading misinformation, disinformation, in a destructive way. Not the funny haha parodies like John League of Legends. But the comp like, for example, I got a text from my father the other day, and it was a little heartbreaking, and it was a photo of Elon Musk standing next to Joanna Gaines. And they were both holding up a white plug in device, and he said, Hey, sonny boy, any idea what this is about?

Kevin: Looks like, uh, looks like they got the all clear for this new device. And I thought his text was a spam as well, but no, that was him saying, Hey, this new magical plug in device with Elon Musk and Chip and Joanna Gaines backing it is now all clear. , it was clearly a scam. I looked at the photo on the Facebook , it was bad AI, by the way, not even like Convincing something, but it was very clearly bad AI.

Kevin: But I think what happens now is , you see an image, you think maybe it's real, maybe it's not, it's small on your phone, [00:37:00] you tap, it takes you to a website that looks real, ? It doesn't look like a cheap scam site, you see a press release about a news story that has no grammatical errors and it looks like it's written at at least a college level, and All that's AI, and all that can be churned out instantly, and that can be customized and bespoke to target its end user.

Kevin: So, yeah, maybe the AI will set us free, but along the way, some people are going to use it to do some really bad stuff, not just scam my father, but scam us as well. So, we are going to need some regulation, and I think you're right, it's better that we're thinking about it and making moves now, than until it's too late.

Gavin: Yeah, exactly. 

Gavin: We are moving at a very fast pace with AI as well And there's these companies doing this stuff you government out there in the world And this is something that I hope will get better have to move faster than the sloths in Zootopia this is not the old way of doing things and I have a little bit of encouragement that now we're starting to see

Kevin: And if they don't, we'll replace them with [00:38:00] AI, Gavin, cause that's what we're gonna do with every job across every industry. And up next, video game voice acting. 

Kevin: Because it's all anyone is talking about right now. It's a, By Embark Studios, it's a game called The Finals. It's a competitive team shooter. It's trending on Steam. It's in beta now. But some people, a large portion of my ex feeds for some reason, all they can focus on are the announcer voices, Gavin.

Kevin: Because something didn't sound quite right to Gianni. Mattregrano, and now he is a voice actor from Evil West and Vertigo 2. He posted a clip that has gone viral that the announcer voices seem a little off. This is from, again, the Steam game, The Finals, where they are, spoiler, using AI announcers. 

[00:39:00] 

Kevin: SOme people were praising the game for the VO announcers, because the fact that they're using AI and text to speech means you get some pretty granular callouts, right? There's a lot of different team names, and there's a lot of different scenarios, and if one team gets wiped out, and there's only 30 seconds remaining, well, to have a voiceover artist...

Kevin: Record each and every permutation of which team got wiped out and what the standings are and who is where on the map. That might be a huge technological hurdle to leap. And so in this case, some people said, Wow, these announcers, man, there's so many different lines. They're calling it constantly. Great job.

Kevin: Others are saying that it sounds so bad. It's pure dog water, keep it off my game, and keep it out of Steam, and some, are going so far as to say now that they know it's AI, it was confirmed by the developers on a podcast, now that they know it's AI, [00:40:00] Gavin, never, never is the time that they will install and try this free to play game.

Gavin: Okay, so there's a couple things I want to say about this one is this is a communications problem first and foremost You need to know as a company that's going to do this stuff what this is going to mean to your audience You need to know how you communicate that and what how you do it exactly this podcast like It was fun to hear these people talk, they're talking about the making of the game, and I get when you're a game designer, you're like, really psyched about this idea that like, you can create these voices, kind of from scratch, and the guy even shouts out Eleven Labs, which we use on our show, and it's how we bring AI John Lennon to life.

Gavin: There's the one part is communication. You need to be clear about what you're doing. 

Gavin: I think the other part of this is. To every gamer out there, to everybody who has a problem with this, to everybody, every voice actor who has a problem with this, welcome to our reality now. You, , are in a situation where a game creator can see a line in a budget, and they can see a [00:41:00] certain amount of the actual performance be at a certain level.

Gavin: I just recently heard, um, Nilai from The Verge say we are going to be world fulled with B stuff. if this is a B minus voiceover and versus an A plus star of a voice acting, if you can get 10 to 100 times that amount of material, if not more, and it can be granular, Kevin, like you said, it can feel like you're in the moment, even if the voice isn't perfect, that's a trade off a lot of these companies are going to make, and I think that is the other side of this that we have to be very clear about.

Gavin: It's again, we've said it before. This is here, people are using it, and this is going to be the new way this stuff happens.

Kevin: GiAnni makes good points about the fact that as a professional voiceover artist, they get rush orders all the time, so if a developer is so inspired that they create a brand new game mode, and they need a thousand new lines of dialogue, there's a performer that's there and ready, and they'll [00:42:00] fire up their booth and get their microphone ready and go.

Kevin: What is not said, though, is this reality that, and granted, quality might be a B plus level quality. But that B plus is available almost instantly. And can be iterated on infinitely, and it's going to be cheaper, at least now, with the way companies are giving away compute and giving away the AI tools.

Kevin: No matter what, it's going to be less expensive than a human performer that might have benefits and limited hours and certain restrictions on their level of performance, right?

Gavin: And in every language now! Every language! Like, 

Gavin: you can create a version of the finals in Spanish, , in Mandarin, , in French, in all these different languages, and it will automatically translate these voices.

Gavin: Now, you could do that true also with people that recorded voices, but you're still then using an AI tool to do it. But if you have an AI voice directly to AI translate, it's that much easier. I think my big thing is, I think you and I would both agree, we feel for these artists hugely, right? But [00:43:00] if I am a game developer and I am looking down the barrel of being able to pay for, another artist or somebody else who can do, who's going to work on the game, who's going to be a team member with me, and I can save 25k on voiceover, and I know it's going to be a game like the finals where you want to have all this voiceover, that is a choice that a lot of game developers are going to make, and I think people have to start just feeling okay about it.

Kevin: Well, so let me ask you this, Gavin, because the flip side of that is... The voiceover artists. And the gamers, the ones that are saying never, not in my steam. In fact, if I can filter by ai, I would just auto ban all those titles from my library,

Gavin: their nimsies? Nimsies? Not in my steams?

Kevin: They're naysayers. But the A has an I after it, naysayers, 

Kevin: these never AERs, they will say, as long as we refuse to support any game that uses AI, which I think will get harder and harder to do as time goes on, honestly, but I'll digress on that, they're saying as long as we refuse to [00:44:00] support it, then these developers, they will look at that line item and say, well, we could save the 25 grand, but we got to go with a human voiceover artist.

Kevin: We got to go with a human concept artist. We have to go with human labor because the gamers aren't going to support it. Do you think that argument holds water? Do you think that that sort of a economic protest, if you will, is going to hold up?

Gavin: I think it holds water in the way that a plug holds water that is slowly leaking. And what I mean by that is like, you can only plug something like this for so long. 

Kevin: I have more to say, but I'm gonna yield my time, Gavin. 

Gavin: Before we move on to our interview, which is very exciting, I do want to mention, a really, really good, AI art expert on, , TwitterX. His name is Nick St. Pierre. He goes by NickFloats, and he's been doing some really interesting experiments with a new program called EverArt AI.

Gavin: We don't have access to it because it's in closed beta, but what he's been able to do is create with a very small training data set, some pretty incredible fine tunes, [00:45:00] with what we think is stable diffusion and the two examples he used were, a model on Rick and Morty images.

Gavin: So he put 10 Rick and Morty images into the, um, ever art AI, Engine or whatever it is and got back a really compelling Taylor Swift in a Rick and Morty style and got back a really comparing Larry David and then did a version around Jimmy Neutron, which, you know, Jimmy Neutron is very specialized looking animation style and got similar stuff.

Gavin: So I think when you think about AI models getting bigger, bigger and bigger, and you talk about like this idea that like there's this never ending race to make the bigger and train the bigger model. You cannot yet underestimate the amount of work that can go into fine tuning and the magic that we might see Kev over the next six months could be specifically fine tuning related.

Gavin: So go check out this guy's handle. Again, his name is, , at Nick floats and hopefully, ever our AI is something that's going to come out soon and we're going to sense of what it's about.

Kevin: Before we jump to Kevin Roos, I asked our AI co host if they could write a new song with the Beatles, what would it be, and to give us a little taste.

Kevin: [00:46:00] So, the, the next thing you'll hear is going to be our AI co host, and then we'll leap into an interview. Does that sound fair?

Gavin: that sounds fair to me. 

Kevin: Alright, so, John, League of Lennons, tell us about your song. The song 

JohnLeagueofLennon: would be titled Echoes of Victory, capturing the essence of both a triumphant comeback on stage and the thrill of clinching victory in the arena of eSports. Right. It would be a blend of classic rock with a modern digital twist, celebrating the parallels between musical and gaming triumphs.

JohnLeagueofLennon: Here's the chorus, echoes of victory, ringing through the night, strumming on the heart strings, shining oh so bright. In every chord, a story, in every game, a fight, all you need is LOL, together we unite. 

Kevin: Nailed it! Thank you,

Gavin: damn, damn, John, when we tell, when we try to get the AIs to do the creative work, they sometimes fail. And that was a pretty big one

Yeah.

Kevin: So [00:47:00] it is possible to have a Broadcasting professional that has success named Kevin on this podcast. We 

Kroose: Is it

Kroose: actually, is it legal to have two Kevins on one podcast?

Kevin: If one is successful, yes, so thank you It's a 

Gavin: Wait, and we, 

Kevin: be just it would explode the

Gavin: and we all know who that one is. It's Kevin Roos, who is our guest today. Who's our guest today. Kevin Roos, just to let everybody know, is a, Kevin, correct me if I'm wrong, a columnist or a writer for the New York Times. Is that right?

Kroose: Yes.

Gavin: Also the co host of Hard Fork, , a podcast from the New York Times with Casey Newton, who I also know about, not, is it about artificial intelligence or do you say it's about technology?

Kroose: It's about technology in the future, but we mostly talk about AI these days.

Gavin: And third, which we'll get into, Kevin, you may or may not know, is very famous in the world of artificial intelligence for being the person to have discovered, The backstory on what binging was early when Bing Chat first came out and and met an entity known as [00:48:00] Sydney, which I'm sure is a story that has been rehashed a thousand times.

Gavin: Kevin has talked about his face being plastered across worldwide, across the internet. Actually, before we start, Kevin, what was the weirdest thing that came out of the press on that story? Do you have, is there one thing that you remember just being like, how is this possible?

Kroose: Yeah, the moment where I really felt like this has reached, this has gone too far, this is too big, um, was when a, uh, a brewery in Minneapolis, um, made a beer called Sidney Loves Kevin, and, uh,

Kevin: Tell me there was an upside down purple devil face on the label. Cause I still use that. I use that. That's like my common TikTok emoji is thanks to you outing Sydney. Like I use that to represent the silly or seductive side of artificial intelligence.

Gavin: Well, let's jump in here. Kevin, you are somebody kind of like us who spends a lot of time observing and obviously deep in the AI space. Um, we always like to ask our guests here, are we [00:49:00] screwed right now? What's your take currently on the way AI is coming along? Like is, when I say we, I mean, not this podcast, but is humanity screwed?

Kroose: Define screwed. Like, are we all gonna lose our jobs? Is it gonna kill us all? Are we, uh, gonna go World War III? Like, what, what are you, what are your specific worries? And then I can be more

Gavin: Okay, so the way that I kind of break this down is two parts. One is, are we screwed in the Terminator sense? Meaning that, are we screwed in that, you know, is AI going to take over from everybody and, and then eventually kill us all? And also, are we screwed that putting these tools in the hands of, of bad actors, let's say, is going to make the world a lot worse?

Kroose: Yeah, so I'll, I'll take those in reverse order. I think the, the misuse, uh, danger is, is real. Um, I think people, um, Can and will use these tools to do all kinds of horrible things. Um, already there's been some experiments showing you can use large language models, uh, to create, uh, you know, cyber weapons and [00:50:00] to, you know, hack into.

Kroose: Computer systems, um, to generate novel, uh, bio weapons and, and viruses, um, you know, these, these are powerful tools and, and they are going to get more powerful. And so I do think that there is, um, you know, the danger that basically they make. They make everything more efficient, including bad people doing bad things.

Kroose: Um, and that's something that I worry about. What, the, the, the Terminator style worry of sort of extinction is what, um, people call X Risk. Um, and that's a little harder to, uh, for me to analyze. Um, you know, I, I've talked to a lot of the people who are worried about X Risk. I've interviewed some of them on, on my podcast.

Kroose: Um, I find there, there's sort of like, um, There's sort of like a, it's sort of like a missing step in the middle for me, like, it's like we have these powerful systems, step one, uh, step three, we're all dead, step two is kind of like a question mark,

Gavin: a, [00:51:00] it's a big cloud. No one really knows what happens in that

Kevin: It's, well, it used to be step one, and then step two was unknown, step three was profit. But we know that there can be profit at step one, step two we don't know, and then step three is the machines nuke the sky.

Kroose: right. So, you know, for me, like I haven't found a super compelling explanation of like what step two could be, or like the, the sort of process by which an AI could, could kill us all, you know, I've talked to very smart people who have theories about this and they mostly involve things like, you know, an A.

Kroose: I. That learns to sort of self replicate right that you you go to try to shut it off and it says, I'm sorry, you can't do that because I've already made 16 copies of myself and stored myself in data servers, you know, all over the world. Um, nice try bud. Um, or you could have a system that tries to, you know, do something but isn't given sort of the right instructions.

Kroose: And so this is sort of the classic paperclip maximizer thought experiment, you know, where you tell an AI to make paperclips, [00:52:00] but you don't give it any other instructions. And so it ends up, you know, killing everyone on earth to get all the scrap metal from all the, you know, houses and vehicles and making a lot of paperclips very successfully, but doing so at the cost of So, there are lots of theories about this.

Kroose: You know, many people in Silicon Valley are skeptical of the X risk arguments. But, um, but you know, I, I put some, I attached some probability to them. I, not, maybe not, like not 100%, but, but enough that we should take it seriously.

Gavin: so maybe we should do a quick step back here and talk about your kind of like daily use of AI like what somebody who Spends a lot of time in this space and knows a lot about it. What do you how do you use it? What are you doing with it?

Kroose: use it to develop novel bio weapons and, uh, conduct cyber attacks.

Gavin: Who knew this is we were talking about you is that you're the bad

Kroose: Oh, I wasn't supposed to say that. I'm

Gavin: not

Kevin: to say, this isn't a real, uh, nuclear briefcase, this is just a photo of one. Right, you have to have to come up with prompt hacks to make sure you can build [00:53:00] your bioterrorism.

Kroose: Um, no, I use it for all kinds of weird stuff. Like, I, I use it, um, as, uh, like a brainstorming tool. I use it to, um, do sort of quasi therapeutic sort of self exploration sometimes. Um, I use it to practice, uh, like, Social interactions that I'm nervous about. Like if I have to have like a hard conversation with a friend, I'll like go into chat GPT and I'll like have like a little demo conversation.

Kroose: Um, let's

Gavin: that's real. But that's really interesting. I don't think I've ever thought to use it that way before. Do you find that with GPT 4 that it gives you responses that are meaningful enough that that's useful to you?

Kroose: Yeah, I mean, you have to prompt it in the right ways. And I've spent a lot of time talking with chat bots. And so I sort of understand like how to talk to it so that you get what you want out of it. But yeah, if you say like, I've got a friend, you know, I, I, I have to have this conversation with them about like [00:54:00] this thing that happened, but like, they're a little bit neurotic in this specific way.

Kroose: And I'm not great at confrontation, but like, how are some ways that I could like ease into that conversation? Like it, it, it does a pretty good job. Um, let's see what I'm just pulling up my, my,

Kevin: trying to think of how tragic it would be if you used Bard to prepare for an intervention with a friend. Because you're just gonna push him off the cliff.

Gavin: Bart has become. BART has definitely become kind of a joke point for us, Kevin. I don't know if you've had the experience with BART, but every time we try to use BART, it's just a disaster.

Kroose: so bad. Yeah. I mean, it's getting better a little bit, but it's still, it's not great. So yeah, I, I use it to, um, to do like to teach myself things. I'm teaching myself Python right now because I used to code, um, when I was a kid and then I stopped. Um, so I've been trying to get back into that. Let's see.

Kroose: I use it to prepare for podcast interviews. Like, what should I ask this person? Um,

Gavin: set up on it? At all?

Kroose: Uh, not yet. I should though. Um, I just [00:55:00] haven't had time to play around with that. And like,

Gavin: You can do, you know, you know what? It doesn't have to be useful. I, I ask it to give me a... 1970s Brazilian funk track every time I ask it for something which is I will say Kevin has made fun of me for this But like it's a great use case because it just reminds you of things You might not know about normally like you could ask it for anything.

Gavin: It doesn't have to be useful. That's

Kevin: Kevin's trying to learn Python code. He doesn't need it to give it, like, end with a parenthetical, and by the way, check out Oolavoos by the Konga Masters. Like, he doesn't need that.

Gavin: He might.

Kevin: now, now Kevin, real quick, because you, you, you pulled up OpenAI and you're going through your chat history, as I've done many times.

Kevin: I OpenAI has said time and time again, just as many companies have, hey, we're not harvesting any of your data, you're not helping train our models. If you're going through our API and for the broad audience, we don't need to get into what that is specifically.

Kevin: But I guess the question for you is how much concern do you have that if you are using it to rehearse a sensitive conversation? I'm sure you're not feeding it people's names and job [00:56:00] titles potentially, but like is there an ounce of concern as you scroll through your different use cases that you're handing your information over to a machine?

Kroose: Sure. I mean, I don't put anything truly sensitive in there. I wouldn't put, you know, intimate personal details, or my social security number, or my credit card number, or anything like that. I try to keep it pretty, um, Pretty general, 

Kroose: I'm also like, you know, I don't know what, what, if any sort of special flags are on my account because I

Kroose: like work at the New York Times and write about AI and like, have had creepy interactions with chatbots that resulted in like massive waves of media coverage. So I always just assume like someone is probably.

Kroose: Uh, you know, looking at this, or could look at this, so I, I try to keep it pretty

Kevin: Do you think you're going to be able to keep that layer of separation as we move forward? I'm the guy that runs the Raspberry Pi for Piehole and it's wrapped in tinfoil, behind the TV. But I will also happily scream my social security number into the phone if it means I have a chance to extend my car's warranty.

Kevin: Like, I'm a [00:57:00] bizarre in between. And I, I similarly try to keep my information private from a large language model. But on the other end, I'm just waiting for Apple to integrate it more elegantly or really at all into iOS or to macOS. But I'm still so worried about that push and pull of as it encroaches, your ability to keep that information private is going to get harder and harder.

Kevin: Do you share similar concerns or are you just going to go with the flow?

Kroose: Yeah, and I think that'll be like a cool feature is basically like incognito mode for, for chatbots, um, which actually only one, I think only Pi now lets you use it without logging in. Um, and so if I ever do have like something sensitive that I want to talk about with a chatbot, I'll usually use that because I can just, I can use that without, without it knowing who I am.

Gavin: One of the things we talk about in the show is like kind of how overstated some of this stuff can get in the media, but you know, if you were going to think about like explaining to the normies out there, what do you think are some of the generative [00:58:00] AI?

Gavin: Like, what are some of the things that are clear that we can't get past at this moment?

Kroose: Um, there's still hallucination problem, right? That's, that's in every model. Um, I think some of the models are getting sort of more grounded. I, I noticed like GPT 4, if I like go back and try some prompts that I tried on GPT 3. 5. Um, it, it does a better job of not hallucinating or of, like, indicating uncertainty.

Kroose: I think that's, like, a big thing for these models is, like, if they're only 50 percent sure or 60 percent sure that the thing that they're telling you about is real, like, they should say that. There should be, like, a little confidence indicator somewhere on

Gavin: Oh, that's

Kroose: Um, and, um, and, and Bard actually will sort of do that now if you, if there's, like, a double check feature.

Kroose: I mean, Bard has all

Gavin: 90%! It's 90 percent hallucinated, it'll say!

Kroose: No, but you can like, there's like a Google it button in which, which will basically make it go back and query, you know, individual phrases or, [00:59:00] or, or sort of even like words and, you know, cite it when it, when it actually matches up with something in the Google search

Kevin: Yeah, and highlight it and be like, oh, this passage, uh, this was an oopsie. That game

Kroose: yeah,

Kevin: existed. That person's made up. Which is wild to me that I understand that it would cost money and energy to crunch that, but shouldn't that be the default action for all of these AI models?

Kevin: That if they're gonna spit information at you, they would go back and just double check?

Kroose: Totally. My, uh, my co host Casey calls this the, why don't they build the entire plane out of the black box?

Kevin: Yes!

Kroose: solution.

Kevin: Yes!

Kroose: So, yeah, I, I think they're moving toward that as an industry, but like it is still expensive to run inference on these models. And like they're, they're, they're trying to cut costs to the extent that they can.

Kroose: But yeah, I mean, I think that would be like a great feature to have in these chatbots is just like automatically like run the fact check.

Gavin: We talk a lot about, the idea of what it could be Google Gemini based on the fact that it's going to have like YouTube bringing all the YouTube [01:00:00] data in, or it's going to have all this video, assuming that's what it's going to be. Who knows? What do you hear about Gemini? And like, what, what, where do you think Google fits into this conversation?

Gavin: Do you think Google Gemini is going to be a big deal when it actually hits?

Kroose: Of course, I mean, it's Google, right? And, um, for all of their, uh, flaws, and the fact that they were very late, uh, you know, that they, they sort of got upstaged by OpenAI over the past year, and that BARD is, is sort of a, a second rate chatbot for, for many things, they do have... Essentially infinite money and compute and data to throw at this problem and they have, you know, many, many talented researchers and engineers.

Kroose: So, you know, will Gemini blow everyone's minds with how good it is? You know, who knows? Um, You know, if Google's listening to this, like I'm happy, happy to be a beta tester and and figure out if it is mind blowing or not. But I, you know, I think that, , I [01:01:00] would not underestimate their ability to make a big splash, especially because they are already able to do that.

Kroose: So deeply embedded into people's lives and workflows, , you don't have to train someone to use a new tool if you're just throwing Gemini into their Gmail or into their docs. Um, so I think they have a lot of advantages that people maybe don't appreciate fully.

Kevin: People have been writing it hours ago, even days ago, Apple got caught flat footed, this GPT stuff took them by storm, so now they're trying to dole out better predictive text and smarter stuff in your photos and whatnot, but like, do...

Kevin: Do you believe that Apple was caught flat footed and then do you have any indication of what they're doing? Cause so they're supposedly spending a billion dollars this year to try to catch up and solve some AI stuff.

Kroose: Yeah, I don't. I haven't done a ton of reporting on on Apple's AI initiatives. I know my colleagues have, um, they do appear to have been um, Caught a little bit flat footed in the sense that they've never really invested in [01:02:00] in a I to the degree

Gavin: telling me Siri's not that good?

Kroose: it's not that good, and it hasn't been that good since it launched, but also like their their their culture is just very different.

Kroose: I mean, one thing that I think people don't understand who don't like work in the industry is like the top researchers in the world. You know, they can work anywhere they want to, and they can get paid whatever they want to. And What they want, in many cases, is to be able to publish. They want to, you know, be able to, sort of, display their work to their peers, and a lot of them come out of academia.

Kroose: They want to still take part in these conferences and publish these papers. And Apple's culture has always been very secretive and very closed. And so, if you, uh, if, you know, if you talk to people in the AI industry, they'll tell you, like, They, they have a hard time hiring and retaining the best AI talent because those people all want to be able to talk about what they're working on and share it [01:03:00] with their peers and their colleagues outside of Apple and Apple's internal culture is just not that at all.

Kroose: So that's one barrier for Apple, you know, uh, they're also like not exactly in the same business as a company like Google. They're, they're a heart, they make most of their money from hardware, um, and from, from services that they bundle with that hardware. So, you know, will there be. Um, lots of spending and an effort to sort of catch up.

Kroose: Yes. Do I think Apple may have, you know, will ever be sort of the company that's like on the bleeding edge of AI research and development, probably not, but it might not matter because they have, you know, so many devices that they can put uh, AI into and, you know, if you're just doing autocomplete or, you know, writing emails in Apple mail or something, you might not need the, the state of the art, you know, bleeding edge model to do that.

Kroose: Like what we have now. You know, might be sufficient.

Gavin: This is changing tops a little bit. You wrote a great book called future proof, which was about, uh, it's a subtitle is nine [01:04:00] rules for humans in the future.

Gavin: Is that what, what it is? Something

Kroose: Close.

Kroose: Uh, 

Kevin: for humans in the age of automation.

Kroose: what is it? No, I think they actually just changed the subtitle to AI instead of automation. So, um,

Gavin: Okay.

Kroose: I don't know. Yeah, I think it's AI now.

Gavin: So in the time since you wrote that book, obviously you've seen a lot of things happen. What are some other things that you would tell people to kind of going forward, ways to live? Because I think one of the things we try to focus a lot on is We try to ground these kind of stories for people and kind of give them the tools to kind of say like hey go experiment Go try this stuff.

Gavin: The world is gonna change fast But like if you pay attention, you can kind of go along for the ride rather than kind of being left out of it What are some of the things you tell people now as you've kind of been through this process?

Kroose: Yeah. I mean, one thing that I, I, um, I got wrong in the book was just, um. time scale. Like I thought we had five or 10 years before white collar [01:05:00] creative work would be meaningfully challenged by AI. I knew it would happen. I knew it wasn't just gonna be like a blue collar problem. Um, because I knew that companies were working on language models and things that would You know, that would do what chat GPT and Dolly and all these tools do today, but I thought we had more time I think the the piece of advice that I always give to people is like you need to instead of trying to compete with AI in the places where AI is strong like you need to develop The skills that AI doesn't have, um, and like, you need to do that consciously. Plumbing, like, like, honestly, like I had a, I had a funny call the other day with someone who's a, a writer and they're like, I think I'm going to become an HVAC technician because that is like way harder for robots to do than writing. Um, but that, like, you still hear this dogma like repeated everywhere.

Kroose: That's like. Learn to code like, like the number of [01:06:00] schools that are teaching kids and, and young adults like in order to stay employed, you need these kind of like technical skills. It's like, that's not good advice anymore because actually some of the jobs that are. The easiest to automate using these tools are programming jobs or sort of like entry and mid level programming jobs.

Kroose: So that is a piece of advice that I give to people is like find, find the things that AI can't do right now. That's, you know, stuff like, um, like working with your hands, but also like social and emotional skills. I mean, we talk about like AI therapists, but I think there's still going to be a lot of demand for people.

Kroose: Who can provide services and empathy and collaboration and all that kind of stuff. So I think that's sort of where I advise people. And also just figuring out, like, literally just spending time with AI tools. So many, I talked to so many business leaders and, you know, academics and people who, you know, have.

Kroose: [01:07:00] Uh, thought about AI, but have not actually like sat down, opened up the chat GPT, you know, window, paid for the, you know, the premium version and like, actually, I think it takes between five and 10 hours of practice to like figure out what you're doing, but so many people have just not done that.

Kevin: hmm. Mm

Gavin: By the way, I was on your podcast, I remember when Casey said he hadn't really tried GPT 4 yet. And that was like, I mean, this is a guy who spends a lot of time in that space, but you realize the jump from 3. 5 to 4 is massive, right? And if you don't pay, you don't get access to that, you don't see it.

Kroose: totally. And this is, I think, a big underlying problem with our. Sort of response to generative AI right now is that a lot of people just have not tried the latest versions and and do not have a good basis of knowledge of what these things can and can't do.

Gavin: And it's getting better all the time. Like I think that when you show people, I would say is one thing that's interesting that came out with it was the Dolly three image kind of moment where like suddenly, [01:08:00] I don't know if you observed this, but Kevin and I kind of call it like the five day, the five day golden period where you could make Dolly images out of everything.

Gavin: And that was a real wake up call. I think for a lot of. Creative artists, right? Because you saw, oh my God, you can make Batman and Master Chief fight each other. And there was this one viral tweet that went kind of crazy of this guy who made a Batman comic book. And man, people got angry. But in part, it's like, hey, this has been happening a little bit.

Gavin: Now in that instance, it was like the faucet was turned on and then it was kind of quickly shut down in some ways. But if you're paying attention, you have to see that that stuff is happening already.

Kroose: totally. I mean, I think we, uh, you know, we tell ourselves feel good stories about how bad the robots are at things like, you know, I remember when, when the first AI image generators came out, it was like, ha ha, they can't do hands. Like all the hands have eight fingers on them

Kevin: Yeah.

Kroose: and we. Tell ourselves these things because they're funny, but also like to [01:09:00] console ourselves, like, Oh, we've still got an edge.

Kroose: Like, we are, like, I can draw a hand with five fingers, and AI cannot, therefore it's not a threat to me or my livelihood or my creativity. Then they fix the hands. Then it's like, well, now it can't do text, or now it can't do You know, faces or whatever, and, and we sort of keep trying to find the places where the AI is bad less because it like actually matters and more because it makes us feel good so that that's also a thing that I worry about is that we are, we are sort of trying to console ourselves by pointing out these very easily fixable problems with the models that are probably not going to be an issue six or 12 or 18 months from now.

Kevin: I want to quickly jump back to something that you said I've got a. A niece who likes art, and I thought introducing mid journey into the workflow might inspire some stuff, and instead, I think I might have deflated or defeated [01:10:00] that interest a little bit, because if I struggle to draw a human hand, which is actually very difficult for a budding artist to do, well, if I struggle to draw this hand, and the machine can do it, why would I ever bother picking up this paintbrush or this pen? But I really struggle to tell her why it's important to pick up that pen and to do that thing. And I can make the case that using generative AI is gonna lift all artists up to a base level. Even if you've never touched a pen, you'll be able to get something from the machine.

Kevin: But the human touch, that expertise, the little extra you could add to it might be the difference maker. That might be... A really important job in the future or just a very satisfying hobby, but I struggled to make the case to her that Yeah, the computer's going to be able to actually do far better than you for years.

Kevin: Yet on the other end of that, you might have a layer of ability to put on that thing. And I guess the question that I'm driving towards is, how do you motivate, , folks that look at the AI like that? 

Kroose: Yeah, this is, I don't know the answer to this one. I'm, I'm very worried [01:11:00] about this because this actually does feel like a place where we are not. Concerned enough about the effects of AI like I'm a good writer now But I was once a bad writer when I was in school and the way that I got better was by being bad repeatedly and getting corrected and getting edited and getting grades on my papers and sitting there and agonizing about like What is the, you know, what is the right length of this sentence, or what is the right cadence of this paragraph?

Kroose: And like you get a little bit better and a little bit better, but there's there's a sort of useful struggle there. Whereas I think if you have the ability to for everyone to be like a B plus writer with no effort, there's, yeah, spending the incremental like effort to get from B plus to A A may not seem worth it.

Kroose: Whereas if you're starting at a D or a C, then you're like, Well, I actually have to learn these skills. So like, I'm curious [01:12:00] what you think about the answer, the answer to that, because I get asked this all the time by teachers and parents. And like, I genuinely do not know what to tell them. Because I think if I had had Chat GPT in school, like, I wouldn't have been a bad writer, and, um, you know, at least if my teachers had not, like, you know, done in class essays or something, but like, you know, that was a very useful process for me, and one that I think resulted in me trying to put a lot of care and effort into a craft that I otherwise would have just outsourced.

Kroose: I think about it with something like, um, like, like music, like we still have lots of musicians, even though, you know, I, I, I'm making this up, but like the, the player piano was invented, I don't know, a hundred years ago, and I'm sure there were people who saw the play, you know what I'm talking about, like the player pianos, like, so, and I'm sure there were people who saw that and thought, well, piano's dead, right?

Kroose: Like, no one's ever going to need to learn how to play piano again, because now we just have these pianos that play themselves, um, you know. And that didn't happen. Like, there are still people [01:13:00] who take piano lessons and work very hard, and there's still concert pianists, and we pay lots of money to hear them, and there are people

Kroose: who make 

Kevin: beings still play chess for some reason.

Kroose: Yeah, so like, I think that there's, um, there's sort of a danger of saying, well, this problem has been solved. This problem isn't solved, and therefore we will never care about it, uh, or, or try, try to be good at it again. But I, I think there's something in this, uh, I think writing especially is such a foundational skill.

Kroose: It's, it's not, Piano, you know, you really need writing to think and to move through the world and to express yourself And so I do worry that that's one of the skills that will kind of atrophy in us.

Gavin: sO Kevin, one of the things that I found really in your, in your reporting you've done before was your reporting on YouTube and specifically YouTube videos that kind of led people down a rabbit hole towards a more extremist version of what they might have originally chosen do you think that there's a danger with these AI tools of algorithms [01:14:00] doing things that we don't intend?

Kroose: Sure. I mean, um, they're they're very, uh unpredictable and and They're, they're designed that way. I mean, there's, there's some, um, variability built into them. I think, you know, what we've seen with, social media algorithms and, and sort of feed ranking and video recommendation. Those were the first real important thing that we gave to AI, to this type of

Gavin: And it kind of screwed it up or we screwed it up. You know what I

Kroose: Yeah, it, it, I mean, it, it, it did really well. at, say, making people stay on YouTube longer. Like, it did that job. The side

Kevin: the paperclip. We got angry.

Kroose: right, it, it made a whole lot of paperclips. And the paperclip company got very wealthy. Unfortunately, it made some people neo Nazis. But that was not, like, designed into the system.

Kroose: And, um, and that was, you know, a byproduct of this sort of algorithmic maximization process. So, [01:15:00] That, you know, I think there will be obviously anytime you are maximizing something or optimizing something or telling a neural network, like go out and make this number as big as you can, you have to be really careful because it may do that by doing something else that you don't want it to do.

Kroose: So I, you know, I think this is very important. Like there's, there's going to be an industry of. kind of A. I. Supervisors and A. I. Auditors and A. I. Trainers and like people who are sort of the zookeepers of the very powerful like silverback gorilla who, who, if you, if you like, leave him alone, or there's like a little hole in the fence, like he's gonna get out and cause some damage.

Kroose: But like, we really, we really don't understand these things that well, but we're already throwing them into the heart of major processes and decision making infrastructure. So that that does

Gavin: right? And this is what we're talking about. That's the thing that's always so interesting is that like, I'm a big proponent of this stuff in education, but also when I said, uh, if you sat like an eight year old down [01:16:00] in front of a chat bot, like don't really know yet what's going to come out.

Gavin: It kind of, what goes in, what comes out? Like, it's like, we don't have a real sense of it. It's kind of, that's the part of this, the whole thing that feels a little bit scary is that, that idea of like. We've seen technology and AI do these things wrong already. And now we're in the next cycle of it. And like, it's like, okay, well, let's see.

Gavin: Let's see how it goes.

Kevin: so it sounds like everybody here is in favor of a pause and we all think that would work. We should all just collectively press the two bars and rethink this for a couple months. Great.

Kroose: Yeah, we've solved it.

Gavin: Solved. Kevin Roos. Thank you so much for joining us. Yeah. Thank you, Kevin. Where, where can people find you and what, and what do you have to prove? Do you have a new, do you have a new book or anything coming out?

Kroose: No new book, um, you can listen to the podcast, Hard Fork comes out every Friday, um, and

Kevin: A new YouTube happening, right?

Kroose: yeah, we, we just started a YouTube show, so people get to see our, uh, our beautiful mugs. Um, and, uh, yeah, I'm in the New York Times.

Gavin: Great. Fantastic. Kevin, thank you so much for joining us and uh, go listen to heart for [01:17:00] it.

Kroose: Thanks for having me.

Yeah.

Kevin: That is it, Gavin. Another AI for Humans in the books. 30 almost listenable episodes. Some of them good. This one was one of them. Yeah, and we beg and plead each and every episode because we are only 30 yen which in if you look at the Pantheon of podcasts, we are a guppy in a sea of whales. We're still a new podcast.

Kevin: So we need you to like, and subscribe, and comment, and engage across all platforms. We're on YouTube, we're on the TikTok, we're on X, and of course, we're available on Spotify and Apple podcasts. And the great thing about Apple podcasts specifically is that humans can leave us very real five star reviews, Gavin.

Kevin: And we ask them to, and we say that we will shout them out. And some weeks you might notice we don't read the reviews and it's not for lack of trying. It's for lack of reviews, but this week, Gavin, 1, 2, 3, 4, 5, 5 star reviews. So someone heard our call last week [01:18:00] and really did the Lord's work. So we thank you so much.

Kevin: We were, we're going to read them all Gavin, cause that's

Gavin: this is what we say we can't get away from it. This is our deal. So some week we will have 40 of these show up Kevin and we'll have a four hour podcast. But today we're only going for extra 20 minutes because of this. 

Kevin: RasslinRaw on 1024 wrote, Love the show. I got out of college as a software developer the same year Kevin and Attack of the Show aired. And being a young nerd, it was right up my alley. Recently I discovered he was co hosting a new podcast on AI. Had to check it out.

Kevin: I work with AI at my job, but it's business software, so very basic, boring stuff. Sorry about that. Please come join our team for free. We will put your talents to good use. Listening to Kevin and Gavin experiment with the more fun side of AI has been a refreshing change of pace. I've gotten a few ideas for new tools to use at work.

Kevin: So thank you for that. Cool. We get 5%. I also co host a small podcast about attitude era wrestling. That's AEW. That's the Kenny Omega!

Gavin: oh, wow. So he had a co host that he must have recently enjoyed.

Kevin: [01:19:00] He says, So I know the time and work invested to try to make a show succeed. Kevin and Gavin are a blast to listen to. Looking forward to more episodes. Keep up the great work you guys have. You have a fan in me. Thank you. Rasslin Raw. . 

Gavin: All right, let's move on. So this is from Crying Zero. The title is anti intelligence question mark. Okay, this starts out.

Gavin: AI. One is an absolute idiot and the other is very smart. They're actually brothers. No, this isn't a riddle, but listening to them turned me off of AI. I needed a third opinion, and I trust Kevin Preyer about tech stuff. Excited to listen and learn more as the show is entertaining so far. So, Kevin, this person is relying on you for all of their AI information, and you cannot piss them off.

Gavin: So, Crying Zero, you need to tell us if Kevin's doing anything to make you mad, because we will change.

Kevin: Yeah, leave a comment down below. Did Kevin piss you off today? Did Kevin make you all pissy?

Gavin: I'm typing four of them.

Kevin: I thought this was about you and I, Gavin. I was like, man, which one of us is the... I'm [01:20:00] assuming I'm the absolute idiot and I'm

Gavin: I was assuming I was the idiot. So anyway, either way, so we let's go to the next one We got we got three more still to go

Kevin: Dirty Viking says, Good times. Enjoyable show, the delivery of new AI content distilled to the podcast is an entertaining hour. I am real, ha ha, ha ha.

Gavin: by the

Kevin: a little Bruce Green at the end there.

Gavin: Next one up from 95 Mustang, 95 must be a great car. Uh, the title of this one is finally an AI show that's fun and informative. Great mix of entertainment and information as someone who started out Worried AI was becoming sentient and Skynet was being invented right before our eyes in parentheses hardy laugh This show has become a great source of inspiration to have fun with it Play around and experiment with AI plus my wife and I are now saying haha in our best AI voice around the house Thank you guys.

Gavin: Please keep it coming. So thank you. Uh, 95 Mustang, 95.

Kevin: Last, but certainly, [01:21:00] certainly not least, TRUESLY1 left a 5 star review and in all caps Gavin it says, DO NOT LISTEN TO THIS SHOW.

Gavin: I have to admit, I was worried. I was worried when I first saw this review.

Kevin: Hey, got the 5 star review, alright, so as far as I know that's the best engagement you can have, but it is a rough headline to have at the top of your reviews. But let's dive deeper.

Kevin: I cannot express how much I regret listening to every episode of this show on my drive to work. Okay, apparently April's leaving reviews. When I finally got an opportunity to see how this show Oh! When I finally got an opportunity to see this show on YouTube, it became 100 times better! In audio form, Gavin and Kevin are entertaining, informative, and help distill complex ideas of AI theory and policy down for anyone to understand.

Kevin: But when you Add the video element. It becomes so much more, not least due to the fact that you can see some of the crazily creative things these two conjure up with AI, but also because the chemistry between Kevin Gavin and even the AI hosts get [01:22:00] elevated so much more on screen. Now I have to go and rewatch every episode I've listened to five stars would not recommend.

Gavin: This is a guy after my own heart. I will say, as we've

Kevin: That is a fantastic review. Thank

Gavin: That means we have to continue to up, up, produce the show for YouTube, which is a nightmare. But, we do it for you, Truthsly1. Thank you so much for that, we appreciate it.

Gavin: Thank you everybody for coming. We will see you all next week. You are the best, and come back around, and we're gonna have a bunch more fun stuff next week, and a new interview, which will be great. See you later.