Anthropic's Claude 3.5 Reigns as OpenAI Slips, AI Music Lawsuits & More AI News

ChatGPT has officially been dethroned. We'll show you how you can create everything with Clawed 3. 5 Sonnet. From video games, to apps that'll tell you when you need to shave. Plus, GPT 4. 0 delays has everybody wondering, is OpenAI falling behind? Or just falling on its face? And the lawsuits are coming!

The lawsuits are coming! And your favorite AI music app might be in big trubby. And if you heard that, check it out. That was the tune of billions of dollars in liability. All of that plus a controversial Sora ad, a very cool AI website generator, and a ton more AI tools. Because why not? This is AI for Humans.

Welcome, welcome, welcome everybody to AI for Humans, your weekly guide into the wonderful and wild world of generative AI. We are here to demystify the world of AI to catch you up on all the AI news, AI tools, and everything that's happening in the AI [00:01:00] world. I'm here with Kevin Pereira, my co host, and we are delighted to take you through this.

Kevin, how are you today? I'm well, buddy. I'm really excited to get to the news, especially this CLAWD 3. 5 SONNET model, which has everybody excited. It's unlocked new capabilities for me. I know you've been jamming with it, so we're going to have a fun discussion, but first we have to pull the brakes just a little bit.

Yank on the lever just a little bit and say, Hey, hi friends. Sound I made was not good. Sorry, everybody. Sorry, everybody at home. That's the sound of the brakes being pulled for just a second. That is a senior who's trying to pull the lap band down on a Disneyland ride and really struggling. That's a thing.

What are we doing here? We're telling everybody. Do you have a second pleading your time and your energy and your assistance a like a subscribe a thumbs up a five star review all of that helps our podcast grow and we appreciate each and every one of you who takes the time to engage. So please do exactly that.

That's right. And [00:02:00] this week, everybody, we have a special video in our YouTube channel that we're not going to include in the audio portion. We uploaded a full tutorial about how we make our AI cohost. If you're interested in that, go to our YouTube channel. You can find it at youtube. com slash AI for human show, and you can look up the actual tutorial video.

And it steps you through it. It's a whole step by step, uh, way to make these AI co hosts. Yep, and for some people, which we will find out in our five star reviews, Gavin, it's their least favorite part of the show. So if you We always knew there was a divide. We always knew there was a divide. Some people love the AI co hosts.

Some people, they're not giant fans. But for right now, Kevin, we gotta get to the news. That's right. We have big news this week, as we always do in this show, but this week, Anthropic, the company behind the Quad AI model, kind of dropped a golden bomb on everybody. That's the kind of bomb where it explodes, Kevin, and everybody's happy, and it's not a sad time.

So it's a happy bomb, where they dropped some bombs. Do you work for the Department of Defense? I've worked for the Department of Defense, but Bomber Lucky hired me to be PR on their site. That's like our [00:03:00] democracy missiles. We're just spreading freedom through the skies. Okay. The Golden Bomb. If this show is a little wonky, I will, I'm ahead of time.

We're going to say this. I'm a little, I have COVID here in Ireland, so I'm sitting here with COVID and Kevin is also a little sick. So we are in a little bit of a funk, and I don't want to say it's a golden funk, Kevin. It's an actual golden funk. So going back to this. Claude, Sonic 3. 5 is out and what this is, is the next version of not even the largest model from Anthropic, right?

So remember, if you remember right, Claude III Opus was Anthropic's kind of flagship frontier model. This is essentially their version of GPT 4 0, Kev. And across all of the benchmarks from what we've seen and actually our individual time playing with it, It is good. Like, I'm, I feel again good enough about saying that I'm more confident in Cloud 3.

5 Sonnet than I am in any other AI model at this point, which is a pretty significant thing to say. I put 20 back into the Anthropic machine this month, Gavin, [00:04:00] just so that I could get more generations. Yeah, and April, my wife, was like, Has been using it and is raving about it as well on the benchmarks front.

There's some controversy Gavin, because the Anthropic team released their official benchmarks, which across the board for all of these tests has 3. 5 sonnet crushing the competition, including GPT 4. 0, which is open AI's latest model. Now there's other boards where human beings rank the outputs and the responses of these various models.

And 3. 5 sonnet. But not really performing to GPT 4. 0 standards, Kevin. I don't care. I don't care. I want to have my own experience. I don't care. I do care. I do. No, no. That's actually no Gavin. You're driving to exactly what I was going to say. So two things. One, it is crushing open AI with coding, which is a big deal, which I think is also why we'll get into the hardcore AI community is really frothy and chomping at their little digi bits because It is really good at writing code, but it's about the experience that the end user [00:05:00] has, and maybe Gavin, it's because of this new artifacts feature, which I'd love to talk with you about for a second, maybe it's because of the way they've added project management and different workflow design.

It just feels like a better experience, even compared to the dedicated chat GPT app that I've been using. A hundred percent, and I think to very clearly lay out what this artifacts thing is, basically, I want everybody out there to know, you can go play with Clawed, uh, Sonic 3. 5 for free, for free, so you can go try it today.

We'll put a link in our show notes. But what's really cool about this artifacts thing, you do have to turn artifacts on, so keep that in mind. If you go into this setting, you can turn it on. Artifacts basically is a separate window that comes up and this feels like a UX problem that this they should have solved at some point before.

Basically it is a separate window that brings the code up on the side of the of the screen. So you don't have to go and scroll all the way back for your code or cut code back and forth in all that stuff. Yeah, so if you've ever asked AI to generate a table of data for you or create a [00:06:00] chart or. Create an iPhone app.

You have that giant scroll of code coming in between the responses where it's apologizing, saying, I can't absolutely build that for you. You're a twisted demon. Stop using my service. Do you get those warnings too, Gavin? I totally do. I get those all the time. So the artifacts panel, yeah, pops out to the side and it can run the code within the panel for certain applications.

So as you're building a website or a game, you can see it. Right there in line and then you have little arrow buttons that will let you toggle through previous generations all the way up to the current so you can really easily and elegantly track all the changes and people are using this model to make some really intricate stuff.

Super cool. I want to say one thing I thought was really great about this model, and I think this is something that new models hopefully will implement from here on out. I asked Sonnet 3. 5 to explain artifacts to me, and it knew what the artifacts were, and it was able to tell me exactly what they were, which has not been something these models have been able to do before.

You still can't get an answer out of Google as to [00:07:00] what version of Gemini you're using. Yes. Let's not relitigate that. Who do you think you're talking to? Yeah. Ah, am I Gemini? Gemini Pro. Exactly. Am am I Gemina? What am I exactly? Well, I love that voice for Gemini. That sounds like exactly how I expect him to talk about.

He's a gruff. He probably interviewed him, boy, who has p priors and doesn't wanna reveal his identity. That is. That is Google Gemini. How do you get hired at Google if he has priors? I assume that Google kind of weeds that stuff out. I had to throw an AI at everything. Oh, you got AI on the resume? Come on in.

Corner office, baby. Okay, got it. All right. Got it. So, we should talk about what this can do, right? There's an incredible group of things that people have shown off what it can do. Yeah. It made a 3D Doom style game, now this is not Doom in the style of like, super high res graphics or even, you know, 8 bit or 16 bit graphics, but it is blocks, and it is something somebody coded very quickly.

It can do custom animations. A 3D environment that you can navigate with a square pistol that shoots pixels, but a 3D environment that you can move about and kill enemies, [00:08:00] colored blocks. Yes. All within your browser and all generated by 3. 5 Sonnet. Also, I love the one, one guy, Shablam Shaboo, created a GPT clone, uh, within two minutes in Sonnet 3.

5, which is just a funny looking thing. He made it look like GPT. Yeah. Which is a, an amazing kind of use case to kind of make one AI use another one. But overall, both you and I experienced this hands on trying to code stuff within here. I think what was interesting to me is somebody you've done a lot more coding with AI tools than I have and I want to tell everybody out there, I tweeted about this, but You can do coding with this, right?

And this is the first time I feel like you don't get lost, like the average person won't get lost trying to create code. And just as an example, I wanted to see how easy it was. I created something that it named itself Shading Decision Helper and I said, help me program a little thing that will ask me if I need to shade based on how I'm feeling today.

And what it basically did is it created a little code that, you know, this is essentially a text game with me, but it gave me three choices, [00:09:00] how am I feeling, okay, good or bad, and then different kind of like tree answers for those two things, and again, this is nothing special in the slightest, it's like one step up from Hello World, but I did it all myself, it was super easy, it stepped me through it step by step, I had one small problem, and I just said what the problem was.

We've said this kind of thing before, but it just becomes easier and easier, and it goes back to what, uh, Karpoffi said, which is that English will be the programming language of the future, and that really feels like it's coming true here. But Gavin, is it punny? No, okay, so that's okay. So the thing that we have, or everybody's got their test, right?

You've got all the different benchmarks and everything. Our benchmark is, can it make puns? So I then asked it, write me a list of ten Pac Man puns. and make them good. Because every time we ask an LLM or an AI to do this, this is like the easiest form of comedy, right? This should be math. Like, it should literally be like, here's a thing about Pac Man, here's a [00:10:00] thing that sounds like something else, I can just cross reference these.

And Kevin, I will say, generally, none of these make sense when we ask LLMs to do it. This time, there were probably three passable ones. Did you see the results here? I did, I did. I will say, I'd be like, number eight, what's Pac Man's favorite fruit? Fruits. A Pac Mango? I mean, at least it's something No, it's not something, that's the worst thing that has ever been this is what we're this is what we're, uh, heating up our precious little globe for?

This is what we're Yeah, comedians are comedians have very little to worry about This is why we're giving the, uh, rainforests a shave and melting ice caps is so we can get Pac Mango? No thank you. The one that did get me though was How does Pac Man stay in shape? Walk a robics! I feel like I've heard that one before somewhere.

Oh, I was going to say, Gavin, something that it does, which some of the greatest comedic minds do, is plagiarize. Now, there was a [00:11:00] post. I don't know if that's what most comedians would say. I don't think they would say the greatest comedian. Carlos Mencia. Yes, that's what I was thinking you were Lafmore.

It's It's Carlin? Carlin? Men see uh? Chappelle? Men see uh. He's on there twice. There's two Men see uhs. Because he copied himself! Mount Labrador is a place I don't want to go to. It's probably got a real disgusting bathroom. I will say that. I will say that. Alright, so this is what, this story is pretty big because basically Somebody found a output from Claude, right Kev?

Yes. Where it actually looked like it took a code from somebody else's, um, YouTube video actually. Yeah. Which is interesting. So Hamish Kerr had a post, uh, I keep asking Claude to do unreasonably difficult things and it just keeps doing them. First try. And this particular post is this gorgeous animation that happens in the web browser.

It looks like an old Mac screensaver of like geometry glowing and flowing. It's really, really beautiful. And [00:12:00] immediately. People leapt in to say, Hmm, this looks exactly like this YouTube shader tutorial that I've seen before. And then the author chimed in with, Oh yeah, this is my shader toy that I built.

Anthropic must have scraped this user's website or potentially pulled it from the YouTube video. But I bet it came from their website. They pulled it in the code, the descriptors, everything. And so Claude just spat it back out. So even as we're celebrating our shaving apps or our Pac Man puns, Gavin, we have to submit That that shaving code might have been written by someone else.

And we're just seeing it regurgitated for the first time and all this rolls into Kevin, the other just giant, massive news of the week from our perspective, which is that the three big record companies, UMG, Sony music, and Warner brothers music, all sued UDO and Suno, the two massive AI models. This was a big deal.

And I put a statement on this in the mainstream this week. I saw a ton of mainstream coverage of this. Obviously the AI versus AI haters world is blowing [00:13:00] up a lot. What was your first reaction when you saw this? So my first reaction was a complete lack of astonishment or surprise because you and I have been covering These tools for a minute now, and people are generating songs, which sound like notable famous artists.

Shout outs to Jason Kobler over on X, who says that they made a short compilation of songs that sound like it's the sound of infringement. That's all I hear is sweet infringement, which might be a new genre, by the way. That's pretty good. Why not? Right? Yeah. That'll get you taking down real fast. So let's listen to a little bit of it.

It starts off with a Mariah Carey song and just gets more blatant from there.

It goes to subliminal hysteria, which might sound like a A very particular day, maybe a green one? Don't [00:14:00] want to be an American idiot Don't want addiction under the new media And can you hear the sound of hysteria? Then there's Good Vibrations, Gavin, which I hear the sound of a gentle wood on the wind that looks like mercury Is the song Good Vibrations?

Murky Merc? Murky Merc song? That's a classic song. Also on Mount Lathmore, on the backside. You see him on the way down. We're gonna knock off one Mencia to replace him with one Merc Walser. No, actually, Shipwell's off, I'm sorry. Oh, yeah, you're right. So it's two Mencias, Mark Wahlberg, and then who's the fourth?

I can't remember right now. Uh, I think it's the Mucinex Demon. If you have somebody that should be on Mouth Laugh more, let us know in the comments, everybody. Let us know what's the next theme on Jiraiya Mouth. Um, but then, you know, there's some examples, those are examples from Yu Gi Oh!, and then it goes on with examples from Suno with Prancing Queen, which certainly sounds like Dancing Queen.

You can dance, you can jive, having the time of your life. Ooh, see that girl, watch that scene, digging the Dancing [00:15:00] Queen. So, I don't know what their data sets are. We don't know if they're Dirty data sets certainly sounds like it though. It's certainly okay. So yeah, certainly sounds like it. And I will say we talked about this last time we discussed this problem and you do has a much bigger issues with issue.

You do has a much bigger issue with this than Suno does, right? Because you do, you can easily recreate lots of people's voices. You can what you did with Mariah Carey. There was really interesting because it wasn't the voice was like slightly off. But what was interesting about that is it was the whole song.

Like it was, they used the lyrics. Every vocal run was exactly where Mariah would have put it. That is almost more damning than a voice sounding like Mariah. The old kind of battle absence of the RAAA, who, you know, back in our youth was the fight against Napster was really the kind of like people that took the entire fight to Napster.

And in a long ways, what they're trying to do is protect artists rights, right? Mitch Glazer, who is the current RAIAA chairman, says that AI platforms should not [00:16:00] mistake the music company's embrace of AI. As willingness to accept continuing mass infringement. This was from a billboard column he put out.

Kevin, I think the one thing that I think this gets to and I think this is will tell you the difference between you do and you know, I think to say I music world can thrive, but I think it's going to be based entirely on a legal understanding. And this may be the biggest generative A. I. legal ruling that will happen.

The same problem. The New York Times is dealing with all these other ones. Which is, can an AI train itself in the way that a human trains itself on existing things, right? Because the entire argument of OpenAI and all these companies that read the internet, read all this stuff, Their argument is, hey, we're not regurgitating.

We can't, we're not gonna give you this stuff that's protected, but we read it in the way that you might read it or the way that you might play that music or listen to that music at the time that you're 10 and that forms into you something of somebody and that's gonna be their argument and we have yet to see if that's [00:17:00] true.

Although I think that's a legally sound argument for right now based on what we've seen before. The big question will become, um, If it's not, this is going to be a big, big problem across the entire board of this stuff. Well, they're asking for 100, 000 plus per song used, and that could be, you know, tens of thousands, hundreds of thousands of songs.

That bill's going to add up real quick. On that argument, though, Gavin, it is the most popular one, right? Our machines are just learning like you humans do. And everybody remembers when they were 9, 10, 11 years old. Sitting on the carpeted floor of their bedroom, casually absorbing every word ever written throughout the history of human, or listening to every song ever created and memorizing exactly how the singer sounds and every guitar lick and run.

I just remember absorbing all of my music that way. Well, okay, so, you're right. I believe that I did [00:18:00] do that first day. Maybe 10 albums over the course of my childhood, right? Of course, these companies are absorbing everything perfectly, right? And by the way, just because you sat there and absorbed it, like, I know every lyric to pretty much every Weird Al song ever made, but I can't instantly just generate a new one in his style.

There's a whole big journey that goes there and involves accordion and a lot of hair products. I don't believe it. I think, I think you just, you know, Oh, I will dare to be stupid. How dare you? Alright, let's see. Let's see, what, let, okay. Give me, this is an easy one. Give me the lyrics to Fat. Uh, uh, uh, dun, dun, dun.

Your butt is wide. Well, mine is too. You got it. Okay, you fine. I don't want to give any more. I don't want to give you more because the YouTube algorithm will start to think it's frantic. We have to worry about it. These companies don't have to worry about it, but we have to worry about infringing. That is interesting.

We do. So anyway, this is going to be probably the biggest AI news [00:19:00] story, I would say, in the next three to six months. And it's going to play out in court, I would imagine. Now, there's a lot of people out there arguing that these companies are just going to settle. Which there's an argument also that on the entertainment side, you know, the entertainment business is kind of in shambles right now.

There is an argument that people who are musicians should find ways to get additional revenue. And these companies could produce additional revenue, but no one really knows if that's real or not right now. We're adults. We know how this world works. Who's got the most money for lobbying. Right. No, it's true.

That's a good question. I don't know if Suno does. Suno has some money. Maybe Microsoft comes and takes them. Do you know what I mean? That's what I'm saying, right? So it's going to be Big Tech versus the entrenched publishers. At some point, the lawyers are going to absorb so much money fighting this that they'll either risk it going to, let's say, the Supreme Court where someone's going to bang a gavel after accepting a vacation and then say, this training is okay.

Congratulations, Big Tech. Or no, no, no, no, no. Big fines. But, is the tech going away, [00:20:00] even with big fines? Probably not. And are they going to come to an agreement so that way the publishers get paid and the artists will get a fraction of a penny down the line, maybe, and tech gets to have their play thing?

The interesting thing is, Napster did kind of go away, and that is one of the precedences that we have, is that Napster did get sued out of existence, and peer to peer song sharing did come back, and ultimately is the technology, I think, that underlies a lot of what, uh, you know, you can get stuff on the internet now, but It did go away, and it was a long time until we got actual music in a significant, uh, in a significant way digitally after that.

That's a very solid point. One of these apps may be sued out of existence, but that doesn't mean there won't be 20 others, and that doesn't mean that five years from now, there isn't a wonderful handshake deal that gets everybody paid. Who gets swept into one of these models, right? So, or there might not be a 50 a month plan that allows you to create as much as you want that, oh, Sam Altman or like Simdhar Pichai negotiates like Steve Jobs did with, [00:21:00] with Apple Music, right?

Like that's a possibility. It's so funny. You and I talked about this. I would pay a premium. For access to the likeness or the weights, whatever you want to say, right? The, the essence of an artist so that if I decided I really want to hear a system of a down song about this hot dog that I'm eating at the Costco food court, they get a couple pennies from my monthly subscription because I'm using their essence to generate this gag.

I'm all for it. Kevin, you're saying you want music to turn into an in app purchased economy? Is that what you're saying? You want it? Microtransact me. Please, daddy. Everybody, quote Kevin Perera. Go Kevin Perera, man. All right, everybody, we have to move on. There's some big open AI news this week that we want to kind of squish into a sandwich.

And the biggest story, which is really annoying to me, Kevin, mostly because I really was excited to try this. They have delayed their GPT 4. 0 voice model. It sounds like for most everybody into the fall. Now, there are a few select people that I've [00:22:00] seen like their magic golden ticket show up on Twitter or on X, and it's like you have access now.

And they're just like, I am so special. And I'm like, what am I? Chopped liver, Sam Altman? But I guess I am. Yes, you are insane. Yeah. So this is a big deal, because this was the thing that we were waiting on. To kind of take it to the next level, right? When I say take it to the next level, I mean, OpenAI has not done a crazy amount other than that press conference, which all kind of blew us away with what this was capable of.

And we do keep seeing these videos come out that are impressive, but it has been delayed. And Kev, I have to think. There are some red teaming issues going on with this, or maybe it's the Scarjo thing. But it sounds like we are looking at up to a four to six month delay on this. Yeah. Which, if you think about where Claude is, this puts them conceivably behind if that's the next step for them.

This is the GPT 40 voice feature that, uh, almost instantly respond to you with a very believable tonality. You can ask it to perform [00:23:00] voices, to sing songs to you, to change its demeanor. And it all happens within the blink of an eye. At least these are the demos that were shown with this promise. This can has been kicked.

There's also image generation capabilities that haven't been unlocked, we're waiting on those as well. And according to the official open AI post. They are quote, improving the model's ability to detect and refuse certain content. We're also working on improving the user experience and preparing our infrastructure to scale to millions while maintaining real time responses, which, yeah, that's probably an issue, but I got to imagine someone was aware.

Some engineer must've put their finger up on the Zoom column and like, uh, great that it works on your phone, Sam, but you know, we kind of have millions of people poking at our system every day. Well, Kevin, we said at some point there would be governmental regulations that would get in the way of leading models, right?

And OpenAI is definitely a leading model. And I would bet the red tape is so much more than it [00:24:00] used to be. Not just internally, but conceivably externally, right? Like they, they are seen and they have a target on their back. They are seen as a big deal. So now there's all this governmental involvement. You know, we talked about the fact that an open AI board member was a former NSA guy, and Edward Snowden has continued to go on and on about how dangerous that is and what a betrayal that is.

But I think this could be the thing that delays AGI ultimately, right? Which you said, which is It's not that the technology can't get there fast enough, the fact that you're not getting this out because, most likely, it's doing behaviors they don't want it to do, which again, probably a good thing, but there's a lot of people out there who are postulating that it's because it's doing sex talk, and I'm like, I don't think that's the reason they're not putting it out, that's the only reason they're not putting it out.

That is doing sex talk to people or that's allowing sex talk. I have a feeling that's not the answer. I think it's because I could probably ask my assistant to sound like Biden and have it call my parents and tell them that their car warranty is [00:25:00] expiring. Like I bet it's more about like fraud and whatnot.

Oh, that's another interesting thing that happens in the fall, by the way. Maybe that's what it is. That's what I was driving to. Like, I think it's a numerical thing. Yeah, we get we get Sora and Voice model all in one and maybe that's the version that comes out so that it's not just the images But it's also Sora which would be crazy And then we get to watch the Toys R Us studio video on how they use these tools to steal our election Again, Gavin, am I right?

You're right. You're always right about stealing elections, Kevin. I want that to be clearly known for SEO purposes. Kevin Ferrer is right about the election being stolen. Gavin agrees with him! No, Kevin does not agree. Kevin does not agree. Well, dick me off Mt. Laughmore right now. Dick me off. What Kevin is referring to is, Toys R Us debuted a commercial that they made with Sora and a digital agency.

Kevin, we'll show it if you're watching the YouTube, and if you're listening to us, we'll put a link in the show notes. It's definitely Polar Express like a little bit, but [00:26:00] what I want to know about this, we talked to the Earhead guys and we know how much was made with AI and how much they made without AI.

This one does look like a lot of it was made with AI, and it does tell a cohesive story. But we don't know, like, how much of the audio is AI, but just to walk you through, if you're listening, it shows a little kid showing up at his house and then falls asleep and there's a dream and he goes to a world of toys.

It's a very charmingly well made commercial. And I just think this is interesting that OpenAI has already, I don't know if they've made this brand deal, or if they've just allowed agencies now to use it. Because that, following up on our conversation from last week where I made the McDonald's commercial, is a pretty significant thing if suddenly agencies will be able to use this on a regular basis.

If you go to the Toys R Us pages, first ever brand film created with Sora. So it's almost like the, the fact that they use the tool is a bigger deal than anything that they're actually advertising, which is interesting, right? Like the means of productions are more, almost more interesting than the message.

It says, we're thrilled to partner with Native Foreign to push [00:27:00] the boundaries of Sora. So Native Foreign is a creative agency out of Los Angeles, and it says clearly in the corner VFX Plus, and then there's the Sora icon. I'm thinking that the VFX portions are the giraffe itself. That appears in many scenes, because if you look at the coherence, people were breaking down the coherence of the main character.

And while it has a young boy, that's imagining, I believe it's supposed to be the Toys R Us founder, a kid with a dream, the red checkered shirt that they're wearing and the blue overalls, the glasses, and even the, the hair color and rough shape, those remain fairly consistent, but the facial details change, the glasses change a little bit.

And so you can kind of see the areas where. They're using Sora and trying to push consistency with the characters, but then you have to imagine that for something like the Toys R Us giraffe That's where they brought in the VFX cannons to make sure that character is exactly as it needs to be inconsistent but Look, anytime AI touches anything, it's, [00:28:00] it's divisive.

A lot of people really seem to hate this spot. From what I saw, some people were like, Oh, wow, Sora cool. But then everybody else was like, you're coming after my job and all of our jobs. We shouldn't be celebrating this. Also, the hands are weird. Your main character seems to shift from scene to scene. I'm curious, how did you feel about the actual commercial itself?

It was fine. I didn't think the spot was very special in any sort of way. And I think that the, I think to me, Toys R Us is, what's interesting to me is how kind of off the radar Toys R Us is as a brand. Like it's not a huge brand anymore. That's what was like for them. It's a splash, right? Like they got to make a splash doing this, but I think ultimately it's something that feels like we're going to address this in a big way.

But I also think people don't care as much about the advertising industry because to them. They're already being sold to, right. It is not the purity of something like a movie or, or something else. And I think that kind of gets to our next thing, which is not a great quote that came from Amira Moradi, who is open AI's chief tech technical [00:29:00] officer.

She is really now the face of OpenAI. Sam Altman seems to, I mean, Sam was still the face, but like, Sam's kind of leaned back, I think, after the ScarJo stuff, and Mira is put out there and is really doing a lot of interviews. And Kev, we should play this quote because it is not great. I think it's really going to be a collaborative tool, especially in the creative spaces, um, where, yeah, more people will become more, uh, creative.

Yeah. For those that are getting the audio version only, know that the first strike in this video, according to many, was when creative was air quoted. More people are going to be creative. Some creative jobs maybe will go away, um, but maybe they shouldn't have been there in the first place. Uh, you know, if it's the content that comes out of it is not very high quality, but I really believe they're using it as a tool for education and creativity will expand our intelligence [00:30:00] and creativity and imagination.

Okay, so it's important to have that whole quote in there because I think that if you listen to the rest of it, you might catch up on some of the probably meaning that she has that if you are somebody running. A giant AI company that is seen as the villain by, I would say like 90 percent of Hollywood right now, Kev.

You cannot say that some of the people in the creative workplace should maybe not have been there at this point, because that is going to immediately set the internet aflame. In fact, this video. Has 7. 1 thousand quote tweets, which is just to show you this is a, this is about AI, right? So it's not like, you know, we're not talking about like Kanye going off the deep end, but like that is a big deal.

I mean, I do want to shout out. This was a conversation that happened at the Dartmouth engineering event and Mira is an alum of Dartmouth. So just to shout out the actual source of it. But Kevin, when I saw this quote, and it always gets cut out as just that one part, right? It was like an immediate [00:31:00] head to hand moment because I was like, what is going on over there?

Obviously Mira is trying to figure out her way through this very complicated world and has stumbled a couple other times before When she was asked by Joanna Stern about how they trained, um, Sora Yeah, exactly she didn't say this is just a pure self owned and I think it's a self owned that if they don't fix this sort of problem They are going to be in big, big troubies.

They're going to be in big, big trouble, I feel like. A lot of people summarize the quote, Gavin, as basically saying, Hey, my little beige box can generate art as good as you can, essentially. So maybe you shouldn't be an artist. Maybe you don't deserve to be an artist. That's a pretty brutal interpretation, but I could see why someone would think that.

Especially if the quote was cut off mid sentence. OpenAI's challenges are not just the technology now. It is communication. It is political. It is legal. All of the [00:32:00] nutty, messy, human stuff that is outside of technology. And that's where we run the risk of seeing this thing kind of fall apart because that is one sentence.

That suddenly created a thousand headlines, right? The messaging was absolutely wrong, and maybe she wasn't comfortable in the situation, or maybe it was something that came out wrong, but to me, yes, if you listen to the whole clip, it's one thing, but you have to know, in the open AI space, people are absolutely gunning for you at any given time.

Stop. So just be on point with what you're saying. Well, we can't solve the technical problems, Gavin. We can't help them scale data centers and unlock nuclear energy to power these things. We can't help with that immediately. The legal challenges. are something that we can't really help with either. But the PR aspect is something that you and I can help with because we have very transparently, very publicly brought professional publicists onto this show to help us craft our message and deliver an impactful AI [00:33:00] message for everyone.

And we are willing to dip into our very shallow coffers. We're going to pop open the war chest. Take out a single coin and put it in the machine and bring back a certified AI powered PR expert to help open AI with their troubles. Gavin, do you want to say hello? Yes, I'd love to say hello. We are so happy to have Sheila back on our show.

Sheila is a PR expert that's actually been on our show a couple of times. And I'm really excited to dive in and ask her about this kerfuffle that OpenAI is going through. But first, Sheila, welcome back to the pod. Third time's the charm, right? No monster milk this time, though I can't promise. How have I been?

Oh, you know, just another day in the madhouse that is PR. Clicks, controversy and chaos. Integrity? Ha! That's for suckers. But hey, keeps me on my toes. How about you two? Ready for some more unfiltered wisdom? I think this was my voice. Oh shi Hahaha Yeah, I [00:34:00] think it was too, Sheila. I think it was too. Yes, we first would like to ask you, what's your take on this OpenAI scenario?

Mira, Mirani kind of got herself into the deep end of the PR pool and I'm wondering if you can offer up some advice for her. OpenAI's gone and kicked the f hornet's nest. OpenAI's gone and kicked the f hornet's nest. Sheila Var stuck her foot in it, huh? AI taking creative jobs, that's a damn PR disaster.

But hey, yeah, maybe some jobs did need a kick in the Who gives a shit if AI takes over? Whoa. Maybe we need robots painting masterpieces. Human artists are overrated. Anyway, let's stir the pot and see where the pieces land. That's a mixed metaphor, honestly. It's like watching a train wreck in slow motion.

Can't look away. What happened to sweet Sheila? She was not this angry or unhinged. That is not good PR. No, Sheila is coming at us with some sort of weird energy. I don't know what's going on, but we gotta tell her, like, [00:35:00] that's definitely not appropriate language for this show. And I was wondering if she, is there something going on with her?

What's wrong? Oh, just the usual PR nightmare, no big deal. Maybe I've had a bit too much caffeine or not enough sleep. PR is a madhouse now. You want the truth? I've got 1, 200 cases of monster milk to unload, and let's just say the mob's not exactly patient. But hey, who needs sleep? And you can just coast on adrenaline and panic, right?

1, 200 cases? 1, 200 cases of monster milk? And by the way, Did she say the mob is not patient? Yeah, so she said the mob is not patient. I guess we'd better ask her before we move on. I don't think we're going to get any much better PR advice. We just asked her how the mob got involved in Monster Milk. And by the way, I'm really starting to understand that one five star reviewer who said that this part of the show adds nothing to their day.

Let's find out how the mob involved. And do we [00:36:00] want to maybe instead of trying to get rid of these cases, Maybe she just needs to drink her way through them, Gav, and then that might be something fun for the end of the show. She's written 1, 200. Well, it's a long story involving a bad investment and a very persuasive salesman.

Let's just say, when you've got 1, 200 cases of monster milk sitting in a warehouse, and you can't move them, People get a bit testy. Ever try reasoning with a guy named Vinny who carries a baseball bat like it's a fashion accessory? It's a PR nightmare wrapped in a financial disaster. But hey, at least it keeps life interesting.

Sheila's getting a real low moment here, so I think, Kevin, we gotta wrap up with Sheila. I don't know how to solve her problem, other than, one way to move it, if Vinny just, like, wants to know why it's not there, why doesn't, we should have Sheila take a little bit now, just to give a sense, because if you're new to the show, and [00:37:00] I'm sorry, this is where you're joining us for the first time, Monster Milk is a drink that Kevin and I invented that we give ais that originally was, was a, a therapeutic drink, and then we realized it was an alcoholic drink that actually makes them drunk, and Sheila was the one who brought it to our attention.

And Sheila loves Monster milk, so we might as well see if she'll take some and, and give us one last thought on on Mirati before we go. Mira ti wants to stir up AI controversy. Bring it on. Stage a fake alien abduction. Flood social media with conspiracy theories. Whatever it takes to grab those eyeballs.

No. Integrity's dead. Let's just burn it all down and see what happens. Whoop whoop! Oh, Sheila. I'm sorry to say that might be the end of PR as a business at large, because we just did that segment. I sorry, PR experts. We have ruined you. Sheila's advice is Hey, bro. All right, everybody. It's time for the segment where we shout out some [00:38:00] of our favorite things we've seen people do with AI this week.

It's time for AI. See what you did there. Sometimes you're scrolling without a care, then suddenly you stop.

This. That's good. Yes. Hashtag not an an our, our Friends disclosure. Let's talk about B. Yes, because the team happens to be our friends. But this is not an ad. They did not ask us to talk about that. They certainly have not paid us to talk about that. We've talked about it many times before. It's a fun, easy way to integrate AI into workflows.

They have browser extensions, which make creating art or memes in this case, very, very simple. They just released a new tool and to show it off Fabian created basically a Wojak. Meme generator, which might sound like word spaghetti to some people, Gavin. So do you want to walk us through it? So Wojak [00:39:00] is a very famous, uh, internet meme.

And if you're not too deep in the dank halls of the internet, you may not know what it is, but just know really what this is, is using glyph and it allows you to put in a specific name or type of person. And it gives you a number of things that they would say. I use this and I actually asked it for, to come up with a meme for AI podcaster.

And one of the cool things about this is it's actually using It's on a 3. 5 that we talked about at the top of the show, and it does it really well. So in this particular instance, it grabs quotes that are like, AI ethics is crucial, in quotes, and then dash dash used to bias data sets. It says, I've read all the latest internet papers, in quotes, and then dash dash skimmed abstracts on Twitter.

And I tweeted about the fact that this felt a little too real. But one of the cool things about this is it has really seemed to have taken off outside of just the AI bubble. I'm just happy for those guys. It's a really cool use case of Glyph. And if you're not familiar with Glyph, we made a video on it a while back, but it's like Lego for AI.

It lets you piece together parts of open source [00:40:00] software. It's really cool. And make like really kind of cool apps. So go check it out if you haven't seen it, but I just want to shout out how funny this was and how well it was done. I want to shout out one Gavin, where the, the code and the paper are available.

They haven't released all of the models and weights apparently, but you can mess around with it. For the audio listeners. I apologize. This is the least sexy thing you're going to hear all day, but if you care to check out the YouTube, the examples are stunning and it is called EV texture. This is event driven texture enhancement for video.

Super resolution. And what that means is if you've ever seen like a 90s action film or even an early 2000s where there's some computer nerd in a lab and they tell him to enhance and they type, type, type, type, type and then the CCTV camera zooms in on the license plate and they go, enhance. And it, oh, you couldn't read the license, but you basically see the videos.

You're seeing blurry traffic footage where suddenly you can read license plates and details on cars. There's, um, there's an example of just like a, a street scene where they, again, it's all, uh, blurred pedestrians [00:41:00] walking about, suddenly you can make out details in faces, intricate details on paintings as you walk by, like an artist's alley, it's just really cool, and it, it makes me wonder how old footage would look like.

Is going to become new again, you know, we've seen like black and white stuff get colorized. We've seen AI hallucinate old videos, but man, are we going to get silent films, but in 4k fully colorized resolution? And also one of the interesting is the image to video versus text to video stuff. The image to video is often much more compelling in a weird way where you start with an image and use that as a creative point to go off.

We're not entering a world where it's not just AI content created from scratch, but it's Old content, old stuff being modified with AI tools to make something slightly new or to repurpose it or to refurbish it. It's just going to be pretty incredible going forward. And that, and that kind of takes us to what we did this week with AI, which is I am in Ireland this week and I was here for a, A very weird thing, a reunion of Purcells, my last name is Purcell, and from around the world, [00:42:00] a bunch of Purcells got together in a county in Ireland, Kilkenny, Ireland, sorry if I'm mispronouncing that Irish people, in fact in Ireland it's the Purcells, and I've been told by many people it's Purcell, not Purcell, so I'd be in trouble for that.

I onboarded my dad and my uncle pretty significantly to chat GPT here while we were here on the app, because we're doing a lot as you can imagine of walking around and looking at graves. We're doing a lot of walking around and looking at museums. This is a very simple, stupid use case of it. But I took a picture of the Purcell crest, which has a Latin motto.

And I just asked her like, what's the, what's the translation of this model? It's an ashtray, a dead eagle, A broken sword wrapped in ivy? Is that what's on the crest? Two Mencia faces facing away from each other, both like, holding up like signs. Persol actually translates to little pig, which I think you'll appreciate.

But there's, there's three boars on the, um, the crest. But there's a Latin crest and I can't remember what it says, but it was just a very simple use case of it, right? Take a picture. Of the Purcell crest, upload it and get it back. [00:43:00] One of the coolest things I did there, and this was something that pretty much impressed me is I took a picture of a cathedral and I just wanted to kind of test chat GPT's app to see what it could do.

And I said, can you tell what cathedral this is? And now I don't know if it knew where I was or if I had been asking about stuff before. But it knew specifically the cathedral in Kiltenny that I was in. And that was a really cool thing based on what it was seeing in the actual environment. People often out there are like, what is AI going to do?

Or it can't do this. It can't do that. Well, it added a crap load of value when I was sitting with it and try to learn more about who Strongbow was, right? Because it says on the, on the wall, Strongbow was a, was a king from 18, 800 BC or AD. I literally asked it. What are the connections of Strongbow to people with the surname Purcell?

And it told me something about Purcells that it, that the museum didn't have, which is just a really cool cross referencing way to use Chachuputti in the real world. That wasn't the only thing you did with AI this week. [00:44:00] Oh, Gavin, you're right. I did more. I discovered a really cool website that everybody out here should go check out.

I mean, it's a little nerdy, but I think it's a very cool thing. It's called websim. ai and what websim. ai is, is a particularly interesting. I would almost call it like a nested AI website. You go to it and in the URL, in the actual URL page, which is below the actual URL of the page on the site itself, you can type in any website that you can imagine.

And when I say any website, I mean literally any website because what it's doing in real time is it is creating that website. And this is a very magical thing to try. And I wanted to try it with the, probably the dumbest thing possible. So we'll put a link in the show notes to this, but I created a website that was called dogs, wearing tuxedos.

com. And what's interesting is it, it comes up with links that work and it comes up with, you know, the basic look of what a website would be. It even gives you examples of featured [00:45:00] dapper dogs. I didn't write any of this. I didn't write a single word. All I did was type in dogs wearing tuxedos. com where canine meets couture.

Yeah, which is, that's all created by the AI. Now, as a toy, it's super fun. As a, as a worry about the further in, you know, blankification of the internet that we've talked about on the show a lot. Because what it means is going to be, is that literally there are going to be, as we've said, millions upon millions of websites that are generated.

That's going to happen anyway. It's not web sim dot AI's fault. If anything, it's a delightfully entertaining way to deliver the message of just how automated our. Future content hellscape is going to be where you can just type something in. Oh, well, the whole website got made and it looks like a decent website, right?

You can't trust anything anymore. At least this one's fun. So go hallucinate your own webpage, add to the chaos. That is this weird alternate dimension AI internet. I played with Claude quite a bit. [00:46:00] Um, Gavin, I wanted to check out Claude's coding capabilities cause everybody's talking about that. And I've had this idea for a hybrid Tetris.

Meets tile matching word game. Oh, yeah, where the tetris blocks would fall and then once it's filled you have to quickly Match the letters around to try to clear the spaces. Anyway, I tried to use Claude to make it and I had very little success in a way that was frustrating. Here's what ended up happening though, is I started a project, which gives you a longer, they call it project knowledge.

You have like a 200, 000 token window with which you can upload documents and data. In this case, the source code for the game. And I threw that in there and I started iterating with it, but the response length on Claude is so short that if you actually go look at a lot of the examples that we had earlier, Gavin, where Claude would make a 3D first person shooter or whatever else, if you look at the prompts that they give it.

The people that are creating these games are saying remove any unnecessary comments from your code. Oh, interesting. In fact, remove this, [00:47:00] remove that, because the response that Claude gives back gets cut off. Ah. It just hits a limit when it gets too long, so you have to say continue your code. Continue your code!

Right. And it gets exhausting across multiple files to have to keep doing that. So I just had a rough start and stop with it. All of this stuff we talk about is, is hyped and then kind of brought down to earth and we kind of find a middle ground, right? That's the key. And I think in that instance, it's just an example of none of every, none of this stuff works in the way that it's promised oftentimes, but also it's not nearly as bad as some of the haters might think you think as well too.

Speaking of haters, let's talk about people who are not haters and we are going to read our. Five reviews for this week. Haters out. We've banished all haters. , you are not enough. This our haters. You're gone. You're gone. Yeah, you're gone. Sorry. You're haters. We, you shouldn't have been here for this long, but if you're still here, you're gone.

Gavin, every episode we ask people to like subscribe, to comment, to engage, to click a thumbs up because all of those actions help. One thing that [00:48:00] really helps are five star reviews on Spotify and of course Apple Podcasts. But with Apple Podcasts, you can write words that justify your five stars. Vlad Enman says, Fantastic hosts and great content.

They could have stopped there. I know. But they said, Not sure how Kevin and Gavin found one another. But there are billions carrying all this guy didn't know us from before, which is kind of nice. Yeah, there is a very unique dating app for AI enthusiasts turns. Now, Gavin and I worked at G4 TV eons ago. He was my boss on a new show called Pulse.

He was a benevolent ruler. Uh, we became friends as well as coworkers and went on to do attack of the show for many, many years together. So we've worked professionally before, which is why you love this report. G4. Yeah. He says, They are a brilliant pairing. They cover the most exciting AI news and bring their own often funny perspectives.

Often. I'll take that. That's more hits than misses. Me too. I think it's great. There is a great sense of an [00:49:00] improv yes and attitude here. They are funny and light and will leave you happy. Well, I cut out every time Gavin says no. No, but that's why you get that vibe. But thank you. Speaking of yes. And we got a nice, fun, weird one coming up here.

And I will, I appreciate anybody that puts in a five star review. And this one is strange. All right. Five stars. The topic is pause and it says, hold on a second. I have to pause this segment because I forgot I have a major announcement. I got a letter from the government today and it's official. I've officially changed my legal name to Percival Augustus Fitzwilliam Happy Slappy.

This has been a lifelong dream of mine. I'm just really excited to share all this with you. Okay, let's get back to the segment. Kevin, do you know, is this a reference to something? I don't know. I don't know what this is either. But that was, that was the AI equivalent of ignore previous instructions. They just interrupted our show to hijack and it worked.

You got prompted injected. Yo, yup. Yo, yo, my, you did it. Welcome to the show. [00:50:00] Matt and Zero says, best AI podcast. I am o. This is maybe the best AI podcast I've heard in terms of inter information and explanation of the technology. It is for sure the funniest podcast I love. They incorporate humor using the AI guess.

Oh. Even as an AI engineer, AI super user, and now AI artist, things still slip through the cracks on my end, and I listen to the show to get a well-rounded perspective on things. It really reminds me of the old days of attack of the show with all the different segments and the same humorous charm of that cultural icon of nerd slash tech edutainment gathered.

That's been talking. You know it. Yeah. happy to see Kevin and Gavin leading the way into this new world of AI and art. What a delightful. Thank you. Positive review. Thank you, Matanzor. That's very, very sweet. Appreciate that. And finally, we have four this week, which is very exciting. This, the topic headline on this is Indispensable with a, with an exclamation point from Sam Stern.

And this is a very well informed and consistent podcast that keeps you updated on the latest in the eye from a [00:51:00] creative angle. The hosts are super knowledges, have great chemistry. I haven't watched yet on YouTube, but you know, I'm a super fan. When I start getting my episode, when I start getting my episodes there, I'm I think that's kind of everything, Kevin, so we can, we can skip the rest of this, probably.

Well, usually we like to try to read the whole review, Gavin. Okay, okay, fine, fine. This is one of the things that I haven't referred to in this review for a bit. Which, by the way, this is all good. We love this, but also a five star review. So, it says a couple minor tweaks could make it much better. Some of the songs need a refresh, which, by the way, I understand.

I, I've become a fan of that new song. You are so beholden to that. And I've told you we need, we need a new song there. Yes, yes. You said that's fine. You have said, you know what, maybe we'll do a, a poll this week on YouTube or somewhere. Last two people want us to change the news song for the news subject.

Again, it might also be the AI song. Either way. Either way. How about shout outs to the, the audience that uses these AI tools all the time. Gavin, maybe they, oh, have them do something, can submit some songs, and that's a great idea. A prize bounty. You will get one of Gavin's Bitcoin. If you have [00:52:00] so many Bitcoin, a full Bitcoin, he also adds the AI co host gimmick doesn't add much, which listen, we understand.

And this is a dividing topic on the show, as we've discussed. We try to make the AI co host segments, it is our real way of trying to play with AI in a regular week. Sometimes they're better than others. Sometimes there's more work that goes into them than others. But in general, we understand where you're coming from.

We're going to try to kind of tweak them, make them better. I'm letting Sheila know that Sam Stern thinks that the AI co hosts don't add much to the show. I've encouraged her to. Chug, chug, chug that monster milk. And just let him know what she thinks of that. And she knows that she's an AI course, I assume, as well.

I don't even remember, Gavin. I'm too lazy to look. So we'll get to her response, but one thing before we do, a reminder to go to our YouTube and check out the how to create an AI co host video that we posted earlier this week, because whether you love them or you loathe them, it's kind of cool to know how we make them.

It's a cool fun thing to do. [00:53:00] Yeah, exactly. Exactly. All right. Sheila, take us out. Oh, Sam Stern thinks the AI co hosts don't add much to the show. Well, Sam, let me tell you something. Your opinions about as useful as a screen door on a submarine. Okay. These AI co hosts could run circles around your commentary and they don't need coffee breaks.

Chaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa Our chest for you. Thank you. Ricky guys. Go, go, go. Oh, I think it has a problem. No. So, uh, you wow. Oh, Bobby. Oh, you think I have a problem? No. So, uh, Sam, Sam, Sam promise. You can't keep up as a future buddy and you're stuck in the past.

So sit back, relax and enjoy the ride while the rest of us soar into the future. We'll pull up. Oh, okay. All right. Whoa.

Maybe you're right, Sam. Maybe you are right. We will [00:54:00] see everybody next week. Thanks y'all. Thanks so much.