AIForHumansEpisode120openAIGPT5

Kevin Pereira: [00:00:00] Huge earth shaking news as the president of the United States finds himself

Gavin Purcell: square in the

Kevin Pereira: middle. No, no, no, no,

Gavin Purcell: no. Wait, wait, wait. Kevin, you're talking about the White House's AI action plan, right? Their coplan to kind of supersize AI in America? Yeah. Yeah. I would've got there. Okay, good. We'll dive into that and we'll also talk about how Sam Aldman is dripping news about GPT five on of all places.

Theo Vaughn's podcast,

Sam Altman: G PT five is the smartest thing. G PT five is smarter than us in almost every way, you know? And yet here we are.

Kevin Pereira: Okay. Yeah, yeah, yeah. But, but, but, but back to scandal. You will not believe what a team of tech billionaires did at a high school.

Gavin Purcell: Oh, this one's Whoa, whoa, whoa. Kevin, just to be clear, that is about Google DeepMind and open AI getting gold medals at the International Math Olympiad and maybe getting a little closer to a GI, right?

That's what you're talking about. Yeah. I mean, if you

Kevin Pereira: wanna focus on the headline itself and not the clickbait.

Gavin Purcell: Yeah. Well, 'cause it kind of sounded like that other thing.

Kevin Pereira: Okay. Listen, how about this one? Netflix and Disney came out and said AI's Okay. And people got big mad.

Gavin Purcell: [00:01:00] See, now we're back on track. Elon also teased an AI version of Vine and PIKA Labs has an entire AI social network on the way.

Kevin Pereira: Plus not one, but two brand new audio models that are so performative and realistic. You can commit bank fraud with them.

Gavin Purcell: And we'll show you two really interesting use cases of VO three, one that's kind of cringey and fun. And the other one is actually legitimately gorgeous.

Kevin Pereira: And if you stick around by the end of the program, we will show you how to hack into Tesla.

Optimist, robots, tele operate them and strangle your enemies. Can. No,

Gavin Purcell: no, no, no, no, no, we won't.

Kevin Pereira: But

Gavin Purcell: this, this is AI

Kevin Pereira: for

Gavin Purcell: humans. Kevin, this is AI for

Kevin Pereira: humans. Elon's putting fluoride in the popcorn and it's turning the robot sentient.

Gavin Purcell: Welcome everybody to AI for humans. We are here. It is another giant week in the world of ai. And Kevin, we are going to start in a very weird place. We are gonna start at Theo Vaughn's podcast because it is now the go-to stop for all tech titans. And we have Sam Altman, uh, on this podcast this week, dropping some [00:02:00] information about GPT five, which according to many people, and Tom Warren from the Verge just mentioned this, uh, himself.

Maybe coming early August and it is late July, so we are not that far away from it. Kevin, should we take a listen a little bit to like one thing that's on your, I'm sure we're gonna have lots of stuff to highlight.

Kevin Pereira: It's so funny, I've been listening to Theo Vaughn for so long, and every time the podcast feed refreshes, I'm like, really?

Yeah. Really. Wow. And seeing Sam Altman, there was just, I mean, uh, delighted, delighted that these two worlds collided. So yeah. Here's some clips.

Theo Von: Is it what? Well, at a certain point, if something has all the information, right, if something has all the information, it can think and, and, and ponder and, uh, pontificate and serve multi options of answers, don't we?

Then working for that thing, like, that's what I start to wonder. Like if it's the smartest thing in the room. GBT five is the smartest thing in gbt. Five is smarter than us

Sam Altman: in almost every way, you know? And yet here we are.

Gavin Purcell: And yet

we persist here. We are. Persist.

Gavin Purcell: Here we are. [00:03:00] We don't have it yet. So we don't know if here we are, we could be a completely different place.

Kevin Pereira: Let's be honest. And my present company included, GPT-3 was smarter than all of us in the room. Yes, right. Like I mean, the advantage is that we can leap from app to app and we have fingers, so that helps us out a little bit. But by and large, these models are already smarter than most of us in so many ways.

You know, not a ton of GPT five stuff, other than he's has it, he's testing it, he's using it. It's incredibly capable and smart. But Sam has since evolved his takes on some things, like, for example, universal basic income. He's not, uh, okay with that anymore. Gavin, he wants universal, uh, universal wealth, and not even basic.

He wants extreme wealth. He wants people to have ownership in the companies and the products and the output of these AI systems. Yes. And he thinks that, that rather than, let's say, get paid in, uh, USD or World Coin or whatever that might be, that the actual tokens which are generated by [00:04:00] these, uh, massive data farms and these power plants, right.

That, that if you, uh, I think he said quintillion was his number of tokens, like multiple quintillion, that if you could take a trillion of those and doll 'em out to every person on the planet, you could then use these tokens to. Create an art project, jam on something with your friends or put 'em back on the market and sell them to someone else who wants those tokens.

So in the near future, not only are we not gonna own anything but a Baja blast is gonna cost us a couple billion tokens and sounds like Turkey cheese

Gavin Purcell: all over again. I don't know if I want to be in cheese, cheese world.

Kevin Pereira: Like we're gonna just be happy with switchblade combs and parachuting army men because we're gonna be paid with tokens.

That's gonna be the economy

Gavin Purcell: that will gets a little teasey of something we're talking about in a bit, which is this, uh, American AI plan that was dropped by the White House, which does not mention UBI and they have, you know, the people that put this out are kind of very UBI, uh, negative in a lot of ways and we can talk more about that.

And Sam has been cozying up to this White House [00:05:00] quite a bit. But let's talk a little bit more about what Sam has been teasing about GPT five because we are very excited about it. There was actually a. A kind of anonymous model that popped up on one of the web arenas that actually was very good at coding.

No one really knows what this is. It's some iteration of a reasoning model that's very good at coding from OpenAI. There is a tweet that was put out of a guy who did a world simulation. It's a very cool visual that you can see. So they are cooking something pretty big. Um, and also Kev, you had mentioned this, uh, clip of Sam from somewhere else where he was talking about this idea of what he did for a, a coding project over the weekend.

Kevin Pereira: Yeah. So Casper Hansen posted this and then there was some comments going back and forth about whether or not he was referring to GPT five here or how old this clip was. But let, let me play a little bit of it and we can kind of level set from there.

Sam Altman: Um. This weekend, I used one of our upcoming models to do a computer programming task that I had wanted to do, sort of like a home automation nerd, and I wanted like the lights and music in my house to do this specific thing.

I knew that before this [00:06:00] technology, it would've taken me days to do that. I was hopeful that with this technology, given our recent progress, I'd be able to do it in hours. Um, I was able to do it in five minutes. The AI did almost all of the work. This is something that, you know, just a year ago, you would've paid a very high-end programmer, 20 hours, 40 hours, something like that to do.

Kevin Pereira: So, high-end programmer, 20 or 40 hours. He hoped it would be, you know. Short amount of time. Did it in minutes.

Gavin Purcell: It's pretty incredible. Like, I mean, this, obviously, he tweeted about this exact thing I think a couple days ago. Um, and when you hear that story, of course your, your first thought is like, well, of course he's just, he's like hyping his own game up, right?

Like, he's doing this thing where he's clearly talking about this thing. But then we've seen these coding tools get better and better and better. And again, I, I want to remind everybody that's been listening to our podcast, the thing that the people that are leading these giant companies has said is that coding by the end of this year could kind of be solved.

And I think that is an [00:07:00] important thing when you think about the powerful that can come out of that sort of, of scenario. Now granted, you know, a five minute, uh, you know, amazing project you're doing on your own side is not necessarily like professional work. And there might be right problems with that, but to get you 90% of the way, that's a pretty big deal.

Kevin Pereira: That's, we, we've been in this space for what seems like, uh, forever, uh, which is only like. A couple years. Really. Yeah, exactly. The point is the speed with which this space moves. We have in that short time heard this tale. Mm-hmm. Over and over and over again. Right. At first it was just writing, finishing sentences, creative writing.

Then it was image generation, then it was video. Oh my god, will Smith. But the spaghetti, the forks going, it says Shutterstock and now we've got camera control and eight K models and blah, blah. Like we know where this is going. It likely is not slowing down. It's going to get bigger, faster, better. And so while this was a home automation project, yes.

That took minutes. It's not a distributed [00:08:00] enterprise grade, blah, blah, blah. That is just a slight moving of the chains though. And that moving of the chains is probably only a few months away. So we, we have got to just be ready for this and not just us as individuals. Gavin. But us as a country, yes, we should, we gotta talk about this as Americans.

Gavin Purcell: So

Kevin Pereira: there's apple pie, eating, diesel, spewing kid, rock loving, monster, monster truck driving. Should be a

Gavin Purcell: country singer, man. Sawdust,

Kevin Pereira: kick in. Harmonica swallowing America.

Gavin Purcell: Yes, everybody. Uh, the big story this week, and I think there's a political story here, and then there is a non-political story that we have kind of talk about in two parts is that the White House dropped a 20 something page AI action plan.

And this is something we've been anticipating for a while. This is the plan, not only of, you know, the, the, the, the White House and the person that's in there, but a lot of the AI people that have been kind of like cozying up to that person, person that in there, the person that's in there that was featured on South Park last night in a very fun way that I think [00:09:00] everybody should go watch.

That was incredible. And by the way, supposedly on that South Park episode, they used AI for that final shot, which I think is amazing. But. This plan is pretty comprehensive, Kevin, and there's a lot in it. I think. Let's go through, there's kind of three main points to it and we can kind of talk through a little bit about what goes into each one.

I encourage everybody to go out there and read it. We have shared it from the show handle. Um, I went through the whole thing. There's lots of interesting stuff in here. So the three big points here are accelerate AI innovation, which sounds pretty good. Then there's accelerate AI infrastructure in America, and then finally there is the idea of leading an international AI diplomacy and security.

So all of that is really interesting. There's some really fascinating specifics here where they get into like the sorts of things that they want to do. Instead of going through and getting super wonky walrus about this, the most important things to take away here are this is a document that clearly shows that they want to kind of push this stuff through in a way that gets past the red tape.

One of the more interesting [00:10:00] ones to me that I, that I kind of glommed onto out of this was a kind of couple lines about the idea that. Um, federal laws should overtake state laws, and there's been a back and forth a lot about like how much states should be able to regulate this versus federal. The other thing, Kevin, that's a little bit sketchy to me, and I'm not sure how you define it, there's a specific part which talks about the free speech question where it says, oh yeah, federal procurement, uh, guidelines to ensure the government only, uh, contracts with large language models who ensure their systems are objective and free from top down ideological bias.

My big question about that is who, who says what, right? Who says what there?

Kevin Pereira: Well, and by the way, there's implicit, uh, and some explicit bias built into this document, which we can get to, but, but to level set, what is the purpose of this document, Gavin? Uh,

Gavin Purcell: the purpose of this document, I think is twofold.

One, it is to like put a flag in the ground and say like, we are going for this thing. And I think this is an important thing for everybody listening to know this thing

Kevin Pereira: being, being nationwide. [00:11:00] AI adoption, getting power grids Yes. Online to power these things. You know, getting tech giants together, government handing out funding, giving access to maybe nationalized models, open sourcing models so that we can export our AI globally.

'cause that can have a trickle down effect. It's, it is comprehensive.

Gavin Purcell: Also turning all of education towards ai. There's a big section of this that talks about how much money you're going to get to do AI education stuff into, do classes for ai. They're gonna start really rolling it out. So that's one part of it.

The other part of it I think is like a little rah rah America, right? And this is the thing that is kind of underlying this whole thing is this idea that we are in this kind of war with China. That unless we do this, there will not be a, a successful America in the future. And that is where it gets kind of political, because I think in part.

This is, and again, you and I have said this on the show many times, we are pro ai. We believe AI can do a lot of great things, but this is a little bit, when you read through this, you're kind of like, you can, you can start to see shades of that AI [00:12:00] 2027 paper in your brain, right? Like this is kind of nationalizing the conversation around ai Yeah.

In a way that like feels much bigger than, you know, the good old days of Will Smith spaghetti.

Kevin Pereira: Right. Um, listen, there's a couple things in there which definitely made my, uh, shoulders hit my ear lobes, uh, specifically. Oh, well how do you do that?

Gavin Purcell: What is that? I know audio people. Oh, I see. Oh god. Close.

You're not getting all the way there. I just wanna be clear. Maybe will can make you get all the way there in a second here.

Okay, will, good job, good job will.

Kevin Pereira: Okay. I don't, whatever effect was dropped in there, if at all. I don't condone. The point was when I read the document, it did that because, uh, I, I'm not doing it now because I'm not being startled by things like, look, the DEI language, like this administration, whatever side you are on that one, like that's in there, right?

Yeah. If you want to play with the government and if you want funding from the government, you can have no DEI [00:13:00] initiatives or whatever. Okay. I'm gonna put that aside for the second. The environmental and climate impact stuff Yes. Is crazy to me because on one hand they are saying that if you want to be, again, participating with the US government, have their support, your model can't be biased.

I love that. Right. We love, again,

Gavin Purcell: like we said, who

Kevin Pereira: defines bias? No, in theory. In theory, yes. Lemme be clear. In theory, I love that because when we use models out of, let's say China, and you ask them certain questions, they refuse to answer or they hallucinate alternate histories because they don't wanna talk about a particular tank in a particular businessman.

Gavin Purcell: Yeah,

Kevin Pereira: right. That's just one example. So I love the notion of the, the USNA. We are going to have this,

sorry.

Kevin Pereira: Wow. Well I get pissed about it. Sorry. Sure, sure. I love the USA. We are going to have this, you know, four loco chugging, ozempic plugging freedom model. Deep fried. And it is. It is. It is never gonna lie to you.

It's just gonna give you the [00:14:00] unbiased thing. But then they say you can have no reference to climate change. You can have no reference to the, uh, to, to carbon having any negative effects like this is in their plan that if you want to be part of a government model, climate change itself can be nowhere in the research you're doing or the output of your model.

Gavin Purcell: I just heard something. Oh yeah. Bring 'em in. Wonky. Walrus is here. Everybody wonky walrus has appeared. Oh, hey, there's wonky with this. So whenever we get wonky, wonky, let's wonky. But yeah, I agree. I agree. Listen, the good news here, as we've been saying on the show, everybody is taking this race very seriously, and we are moving into a world where this is going to dominate the future of the entire American worldwide conversation.

And that is a signal of what this is. And Kevin, the other part of this that's really interesting is the energy infrastructure part, right? Again, wonky walrus people should understand what this is. Energy infrastructure is vitally important because until we get more energy [00:15:00] online, we are not gonna get more AI because energy equals AI right now.

That is how much energy we need. Um, Andro put out a really interesting, kind of like deep dive mini paper on the idea of how much energy we need. And they did it also. Anthropic supported this AI action plan, uh, online. At least they talked about it. Kevin, do you know how far behind America is versus China in energy?

Uh, uh, a creation.

Kevin Pereira: Uh, yeah. For the, for, I'm sorry For those on the audio podcast, we're this far behind and it's not to scale.

Gavin Purcell: No, it's, uh, pretty good. Um, but I am, that's pretty good. I've got

Kevin Pereira: my thumb in my index finger a little about, about two inches.

Gavin Purcell: Yeah. But pretty close. Pretty close. About two inches away.

I think the big thing to understand is that China has been focused on energy production for a very long time. Yes. We are very behind. The other thing that was so fascinating to me that came out of this, um, anthropic paper is they are going to need just them five gigawatts of power by 2028. And when you think about what that is, that is a, that is a massive, massive amount of power.

That is the amount of power that they're expecting to get. [00:16:00] Out of just the Stargate Project in Abilene, Texas, the OpenAI project. So yeah, this is a big deal, I think. And also there was a small side note if you saw this, Kevin, but like Anthropic said they weren't gonna take money from the Middle East. And now there's a leaked memo that says they are because yeah, people need money.

They need money from kind of anywhere it comes from. And again, to your point earlier, like having these models and taking money from countries is maybe a good thing. 'cause then you, your model gets ported into that place, even if you don't a hundred percent agree with the political scenario in that place.

Yeah. But it's an interesting

Kevin Pereira: spot, I think. And by the way, like, not to harp on it, but I'm gonna harp on it. Social cost to carbon, methane rules, NEPA reviews, which is National Environmental Policy Act, clean Water Act, and clean air enforcement. All of that in jeopardy when it comes to, uh, you know, removing the red tape to allow this energy infrastructure to exist.

And when critics of AI get upset, like, like people talk about the, uh, you know, how you waste a quarter of the [00:17:00] ocean every time you ask chat GPT to, you know, identify the sheep in the photo or whatever it's you're doing, right? I make sure

Gavin Purcell: to flush my toilet five times every time I use chat GPT. So I balance it out.

Kevin Pereira: You kind of

Gavin Purcell: balance

Kevin Pereira: it out. Like, like, and I, I laugh at that, not because of people's concern for the environment. No, of course. 'cause that particular stat is wrong. There's, in some, in some ways, sending emails or doing Google searches is just as environmentally damaging as using a basic chatbot that said, this is the kind of stuff.

That I do that does warrant all of the criticism. The fact that you're going to say like, you know, the air, the water, the way in which we burn coal to power these machines.

Gavin Purcell: There's a great interview with my personal hero. I know John Carmac is yours, my personal hero, Demi Abba, who I know you also like Love de Yeah.

With Lex Friedman that I suggest everybody listen to. You know, whatever you think about Lex Demis is an all timer. Amazing. Now, a Nobel Prize winner, the creator of black and white. Our favorite video game slash ai, God, um, my favorite, I should say, sorry, Kevin. It's okay. Where Demi Hassabis [00:18:00] talks about this very specifically about the idea of fusion power, and he thinks that the real thing that's gonna get us there is fusion power and extreme solar.

So the, the only counterpoint to this is, and, and by the way, oh no, hold on. Let me finish my counter, then you can counter my counter. All right, one second. The only counterpoint to this is. That there is a, there is a legitimate pathway to a world where,

Kevin Pereira: yes, AI helps us figure

Gavin Purcell: out this thing. Do you

Kevin Pereira: know what I'm saying?

Oh yes. Oh, I'm with you. I'm with you. Except what I would love to see is. Instead of let's saying, uh, instead of saying like, uh, you, let's go ahead and just knock all the eagles out of the nests. Yes, yes, yes, yes. And maybe poison the water in the air and the pursuit of the technology. Which brings us to that utopia.

Can we do what maybe some other countries have done and invest in clean or renewable ener like, like that's a little slower boxy. I get it. Get, get, get hard on solar,

Gavin Purcell: right? I shouldn't say no. Get hard is a wrong way to describe anything. I'm not getting hard on solar. Um, Kevin might, but not me. So I just wanna make sure I'm clear.

This is not what I'm saying. I'm saying let us expand [00:19:00] solar. Well, maybe that's not that good either. What are we doing here? So maybe the idea is like erect new solar panels. Is that what we're talking? Okay. Yeah, I get

Kevin Pereira: it. Uh, no, this is a good run. This is a good run. All I'm saying is that, you know, if the pursuit I'm with, I'm with new energy, I'm with clean, I'm with all of that stuff.

Sure. It's just, we know, we, we, we tried to, I don't know. Don't do it. Not, I'm already, I'm already in the, it's gonna be in the weeds on it. I think people are right to be upset about some of the environmental rollbacks, which will likely happen in the pursuit of building these data centers. That's all I'm saying.

I, I think that's a very fair criticism of this document. It's all good. It's all good. My buddy wonky walrus,

Gavin Purcell: we're, we're saying goodbye to him. Say, bye, Kevin. We're saying bye to wonky walrus

Kevin Pereira: when you're coughing and gagging in the pursuit of your super intelligence Gavin, don't say I didn't warn you.

Gavin Purcell: Fair enough. Now moving on, we've got our next story here, which also involves super intelligence in a weird way and de um, there's a very [00:20:00] fascinating story that came out late last weekend, um, where OpenAI first and then Google DeepMind both got five out of six on the International Math Olympiad, which is a gold medal, and this is the first time ever.

Reasoning, LLM, general Purpose LLMs have ever gotten this score on this, uh, test? It is a very hard test. It is a test that is given to the best High School Math Elite is a name that I used to like to call myself in fifth grade and math elite people who are brilliant and very good and going on to become some of the best mathematicians in the country.

This is a big deal. Um, Nome Brown had a very good explainer, uh, series of tweets about why this matters. The interesting thing, Kevin, from an LLM perspective is this is not a model that we are going to see. This was not GPT five. It is a new piece of research, which I thought was kind of interesting.

Another really interesting one person kind of developed this i this technique and supposedly it is way better than anything else. I did see a tweet, somebody say, where part of it is that it's helping it reduce [00:21:00] hallucinations. Whether or not I know that's the case, but this is a big deal and I think everybody should understand like we are closing in now on AI that can at least be as good as people in math.

On a general AI model, this is not a specific model. This is a general LLM.

Kevin Pereira: Well, I think one of the things to note is that in the past when ais were tasked with doing math, they, I, I don't wanna say they, they cheated be because they, I mean they, they, they solved the problem, but they do usually wrote, they wrote code, right.

Like usually Python code to actually accomplish the math functions for them and then ran tests against it until the problem was right. And I believe in this case it didn't do that, that it tried to use reasoning to think its way to the conclusion not using code and math.

Gavin Purcell: Yeah. And more than that, it was not allowed to use any tools at all.

So this was literally given the problem. It had a time limit of four and a half hours and through its reasoning and its own little brain. And when you say its own little brain, that's like the stuff that it's only trained on. So when you ask a question at chat [00:22:00] GPT, often it will go to the internet.

That's a tool or it might pull up its coding tool. That's a tool. This was a reasoning only model that was able to get to this level. Now again, you might say, well, it's a high school student who do I care? I suggest you try to go and do these math Olympiad questions. Uh, I, you know, did whatever, four years of high school math and haven't done much since then.

And I probably could brush up, but even if at my peak I don't think I could get through maybe one or two of these max, they are hard. And I think that's an important thing for everybody to understand. Anyway, it's another kind of level up of like. Where we are getting at in this space.

Kevin Pereira: Could you imagine if you gave GPT-4 oh access to a TI I 85?

Like just give it a graphing calculator. What it could do Gavin? The math. It could solve. It could crush Tetris drug wars. Oh, there's an FBI surveillance van pulling up to my house. Right now, I'm not doing it.

Gavin Purcell: There's one thing I will say about this that's so funny to me is like all the people that came out and started hating on the idea that like.

You know, [00:23:00] oh, maybe use tools. Maybe use tools. It's like, guess what? This is a kind of a computer. It should be using tools. And it's like already so smart as it is, but I get this anyway. Yes. Yeah. So this is like, oh, oh, you didn't

Kevin Pereira: tie Its digital hands behind its back. Yes, that's cheating. It's like, no, let it do what it needs to do.

You know, what else? Just cheating. Gavin, enjoying the heck out of this podcast, out of you and I and not daring to engage. What is that? Your little sneaky cheaty. You

Gavin Purcell: may be out there right now watching this. You may have already subscribed, but if you haven't left a note to us, if you haven't left a little comment or if you haven't reviewed our podcast on, uh, podcast platforms, we don't really think about you.

So make sure you send us a message. Make sure you send a review when you do that, then we think Dr. About you go subscribe to us on YouTube right now. Hey Gavin, A quick question for the visual learners out there. What is this? That is a, uh, hand and a thumb

Kevin Pereira: up. Oh, thumb up. Oh. Can you click it? Can you, can you click the thumbs up?

Oh. Like, can you click or subscribe? Lowest. You click that. Yeah, that's what I'm talking about, baby's. That's right. Uh, [00:24:00] sincerely, we beg and plead every, every week for your engagement because it's the only thing that helps us grow. So please like, subscribe, share, leave a comment. It helps juice the algo. So give us that hashtag algo juice and Thank you.

Gavin Purcell: That's right. All right, Kev, we gotta keep going. There's some big stories in the Hollywood space this week. Yeah. Disney and Netflix both admitted to using runway in certain programs. Now Disney actually has not used it in something that's come out yet, but Netflix has a show called The Ether, not, it is an Argentinian show, importantly not covered by American Unions.

And they used AI to create a disaster scene, and they said very directly that it saved them a lot of money. In fact, Ted Sarandos. The head of Netflix, those of you out here know Ted Sarandos. He said, we are, we remain convinced that AI represents an incredible opportunity to help creators and make films in series better, not cheaper.

So that is a big quote from the,

Kevin Pereira: well, not just cheaper is the, the word just is doing some heavy lifting there. Not just,

Gavin Purcell: sorry, not just cheaper. Yes. But, but again. [00:25:00] Whether or not it makes it better, we can debate that all day long. The cheaper part in part is what matters here. And again, I think the big question is could you tell, could you not tell?

I went, I, this clip is kinda hard to find. Did you see it? Did you see? Yeah. 'cause the two

Kevin Pereira: characters go to high five 'cause they save the world and they have seven fingers each. And I thought that was

Gavin Purcell: not

Kevin Pereira: the clip. You saw something else they could have, they could have wrote that out a little

Gavin Purcell: bit. Yeah.

That is not what it is. The clip was interesting. I mean, again, I, we, I don't think we can actually show it here because it is weird and hard to find. I saw it in somebody's blurry shot. Again, not crazy like, and I think it's like, you can see how these things can work. And we've talked about this before.

When you're creating AI video, the biggest key is editing, right? Because you kind of have to like maybe four seconds of a shot works really well and you can do stuff where you edit around something and you make it really interesting. Like, okay,

Kevin Pereira: help creators make films in series better. Sure. Yes. Because I, I, I know so many creators, indie filmmaker, friends of mine Sure.

That have like, yes, they've had to do [00:26:00] creative workarounds. Like you can't show the fire, but you can show a closeup of someone's face and flash some amber light on their face. But if your vision includes the building on fire or the atomic blast exploding, or the car getting crushed under the creature's foot, and you don't have the money to pay a traditional VFX firm to do it, well, yeah, this can make it better.

I understand that there's 20 people looking up from their Wacom tablets going, we, but that's me. I get it. But like, on the surface, that statement is kind of obvious to anybody that's using AI tools to make videos. Right.

Gavin Purcell: And totally, and by the way, if you're following the box office, there's been kind of a, a, you know, the main box office in Hollywood, in America, there's been kind of a little bump back, but these, these movies like Superman came out.

I saw it. It's fine. It's great that a lot of people, I, I thought it was fine. I thought it was really interesting and fun. But like it didn't crush in the way that these movies used to crush, say, 10 years ago. And prices are up. So you have to think there's a world where if you want more Supermans, you want more fantastic fours.

If these [00:27:00] movies can, instead of costing 250, $300 million, if they can cost like $175 million, that's like a huge win. Now, that's a hundred million dollars you're trying to cut out of a budget like that. This is kind of how you get there. In some ways, and again, you're right, like there's a lot of people out there who are working in these giant, you know, FX factories who are gonna be saying up yours, that's my job.

But also. This is something again, if you've been listening to us for a while, we have been saying is coming and now it is here.

Kevin Pereira: I I also like, uh, hashtag James Gunt stand. Love James, uh, yeah. Have the privilege of knowing him and have, you know, chatted with him a couple times.

Gavin Purcell: Great, great, great guy. Yeah.

Kevin Pereira: Yeah. And I'm watching the behind the scenes clips of Superman and I'm actually surprised at how much practical effects. Did you see the one where he was spinning around?

Gavin Purcell: Was it that one? Yeah. Well that a cool clip for

Kevin Pereira: those who don't know, it's a great clip and it's, it's, it's, it's Superman and he is flying and they have these unreal video walls that powered by the Unreal engine.

So it gets all the natural lighting and the effects of the mountain scape that they're blasting into these walls. And they've got the, uh, actor spinning and rotating as if [00:28:00] he's doing a, a star fox barrel roll. Yeah. In front of the video wall. There's another one where they're in like a tower and they're changing day to night and people are oohing and aing because they're reacting to this day night cycle.

And I was like, that's really cool, but. Probably could have just made ALO and said, Hey,

Gavin Purcell: yeah, sure.

Kevin Pereira: Make Superman spin now for the performance and the, this, that, the other. I'm sure everyone, there's a million reasons to do it that way. Everyone do

Gavin Purcell: that. I wanna make clear I didn't say that. Feel free to clip that out with the person.

Gavin thought it, that person's photo,

Kevin Pereira: Gavin thought it hashtag could have been AI

Gavin Purcell: anyway. But, but you're right. I mean, listen, I mean, and again, like a Superman, super fun. There are so many different effects in it that were great and I love that clip and I don't, I don't want practical effects to go away, but like, again, no, if you want more of these things and you want them to work, you're gonna probably see more AI involved.

Now, ke there are a couple other interesting quick, uh, hits with video one. This is really just based off an Elon tweet. It, it follows up on a couple things. Elon had a tweet a couple weeks ago where he said something like, AI video is the future of something. And he's like, clearly [00:29:00] ex has been thinking about ai, video x ai.

But then he said specifically just a few days ago, we are bringing Vine back. But in AI form. Kevin, what do you think this means?

Kevin Pereira: If it means more will Saso with produce in his mouth,

Gavin Purcell: I'm happy. I don't think that, I don't think that's what it means. My, my gut, and this kind of ties into our other story, which is that Pika uh, labs, the company that does all the fun effects, which we just a couple weeks ago said the other day, they haven't seemed to be doing a ton lately.

They have now what might be a pivot might just be a different company. They have announced that they're making an all AI social network, meaning that you would have AI versions of yourself to be able to do stuff. Kevin, this is where I think the, you know, in ification of the world is starting to kind of show itself, right?

Yeah. Because I Vine as all ai, I'm still not convinced all AI is the way to go. Um, I'll talk about this later, but I got into an argument with a guy who was a big Higgs field promoter about something specifically not, not Higgs Field, [00:30:00] but I'll tell you that story in a bit. But like, I don't know if I want a vine that's all ai.

Like I don't know if it's gonna be that interesting. And I don't know what you think about this, but like. I kind of like how TikTok is a mix of stuff, right? Like, I like the idea that human stuff and AI stuff can kind of fold in together. And when you have all ai, it feels like you're limiting in the same way that like, something like ciso, if you remember ciso, the streaming network was like all comedy.

It was like, well, guess what? Like, I, I don't want that. I want a version of something that gives me a variety of choice and kind of brings something to me. Uh, what are your thoughts on an all ai social network sort of vibe?

Kevin Pereira: Uh, yeah. I, I don't. Other than for like, research for our show, I don't know why I would gravitate towards signing up for an AI only social network.

Yeah. And maybe this is now the old man whittling on his front porch, shaking a fist at the cyber cloud and his augmented reality glasses. Like maybe a whole new generation is gonna grow up and their AbbVie will be the, the, the purest representation of [00:31:00] themselves. Sure. Yeah. It's possible. And they'll auto generate all sorts of stuff so they don't have to deal with it.

And, um, I'm sure I'll be bullied by those AI avatars on their network at some point when I try to fellow kids it with my skateboard, I, I just don't see it. And to your point, like being on a network that already exists, where those connections are already established, adding an AI element to that perhaps, but even then, I still don't now, like, we'll, we'll get to, um.

Higgs field's encouraging everybody to steal, uh, looks and put their, uh, personas into it. Like, if I see AI generations of other people in my feed, I go, uh, okay. And I'm, yeah, I move on. It doesn't, it doesn't, it kind of bounces off me. It doesn't have a stickiness.

Gavin Purcell: My question is, when was the last time you thought about a very big launch that happened six weeks ago?

Meta's social AI app. Have you thought about that at all in the last two to three weeks? No. No, I actually haven't. Gavin, yes.

Kevin Pereira: Only that like, grandparents don't know that their chats are public. Yes, yes, yes. In there. And so seeing people have very intimate conversations with an [00:32:00] ai, that's the only times I've discovered it.

Gavin Purcell: Yeah. I mean, it's an interesting time, right? I think, I think, I hope that this, these kind of like AI only social apps. Now, the one thing an AI vine would have going forward is that maybe there's a world where it does get integrated into x and X is still an active social network, so it's easier to share.

Okay, Kev, we wanna now dive into some new AI audio models. Let's talk about first in world ai and then we'll get into a very cool open source. In world ai, if you're familiar, is a kind of a pretty large company that's specialized in NPC voices. They specialized in the idea of big video games. They were gonna provide NPC voices to, to characters so that you could go and, like if you're in Skyrim and you wanna have like a deep philosophical chat with like, uh, the lady who sells oranges in town, you can, um, I don't think it did as well or they're trying to figure out kind of what the next stage is.

But they have now opened up a new text to speech model that is very good. I wouldn't say it's like state of the art, but the cool thing here is it's very cheap. So for people that are [00:33:00] building in, uh, in AI audio specifically like yours, truly, uh, Kevin and I are working on our secret project right now. It is a really interesting opportunity and one thing that's cool about them is they have a lot of history working with characters.

So it actually kind of sounds good. Do you wanna play one of these and we can just get a sense of kinda what it sounds like? Inworld is enabling scaled AI applications that grow into user needs.

Today we are launching Inworld, TTS, a new voice AI model that makes the state of the art radically affordable.

Bracket

Kevin Pereira: cough is actually really exciting to myself and a bunch of audio nerds out there. Yeah. But it was sort of a weird, weird place to put it in your announcement video. It's,

Gavin Purcell: it's a strange place to talk about like, we're gonna have this really

Kevin Pereira: important line, this we're gonna have stay the state of the out performance.

I'm like, excuse me, excuse me. I'm so sorry. At least it wasn't the guy character at first. 'cause you would imagine he's at the doctor's office trying to talk to you, you know?

Gavin Purcell: Yes,

Kevin Pereira: very much so. In world a

was that good

Kevin Pereira: [00:34:00] doc? Just, just to put it in like the, the subtweet that Inworld put out Yes. Said that they identified the problem, and this is kind of big, there's premium text to speech, right.

Which is really expensive if you start doing it at scale, and then there's. Free or really affordable stuff, which does not sound that good. And they even sort of, they admit that they're going for that third rail, that third option. So in comparison to like 11 labs, which can cost you $120 for a million characters, kind of depends on the model using they are $5.

Yeah. In comparison. That's a pretty big difference when you look at the bar graph. Gavin, again, not to scale. Probably about two inches.

Gavin Purcell: Pretty close, but two. Okay. Two inches. I will say that is something that's really interesting for us as we started to price out what these things cost. $5 per million is super cheap, right?

Yep. So if you can get a really comparable and com competitive sort of model out of that and they can improve it over time, big deal. All right, so now let's talk about this open source one. Yeah. Kev, do you wanna tell us a little bit about this? 'cause you got some time to play with it.

Kevin Pereira: This is, uh, Higgs audio V [00:35:00] two from, uh, boon ai.

I saw some reports that they were kind of basing it off of Sesame's model that was released. Oh, interesting. I'm not dive deep enough to confirm if that is true. So feel free to roast me in the comments. But, um, they have zero shot voice cloning, which, you know, makes it essentially very quick, very easy to clone any voice.

Uh, they're saying in the, uh, it works real time. On the edge, so you can have it deployed to smaller devices, potentially even phones. It was trained on 10 million hours of audio. That's speech, music and different events. So this thing can kind of generate background noise. It can generate musical stings, it can do your bracketed cough.

Um, it beats GPT-4 Oh mini text to speech, which is open. AI is one of their best audio models. Yeah. They say it beats 11 labs V two. Um, and when it, you can do multi-speaker, when it, how do

Gavin Purcell: you determine that? What, what does it mean? 'cause when you're listening, is it, is it literally like people give it tests and like a webina sort of thing?

Yeah, like literally like audio

Kevin Pereira: benchmarks and people thumbs up. I like this one. I like that one. So [00:36:00] here's a little bit of their official sort of announcement video so we can hear some of it.

You spent our weekends tuning a model instead of seeing my parents. But that run was really important. You look at your freaking Oscar curve longer than you looked at me.

That is not true. I just needed it to converge.

Kevin Pereira: So this good, this will get better after. That's, that's two characters in the same scene with audience laughing, right? And then you hear like this, this prosody adaptation, meaning the speaker is going to adapt, uh, its speaking style based off the context of what is being said.

This one's kind of interesting.

Beth collapsed into his arms sobbing uncontrollably. I failed them. I failed them all. They're all dead.

Kevin Pereira: So you could hear the way it went into the quote from the character. Yeah, it kinda switched it up. Yeah. But let's get to the good stuff. Let's get to someone cloning Shrek and doing a scene with Shrek and Donkey with the same model.

Here we go

donkey, look at this We metal Dragon cost me a [00:37:00] king's ransom and all it does is fry Tatties. That's a GPU Track. You bought it to train AI for swamp visitor alerts. I, and it always says right now useless. Should have asked it to predict when you'll stop Wising gold.

Kevin Pereira: Gp pretty amazing onions list.

That's one model, right? Yeah. With in near Instant Voice cloning, doing multiple voices. This is really amazing. Higgs Audio V two. There is a GitHub Gavin and they are open sourcing. Higgs audio V two. Um, we should,

Gavin Purcell: uh, talk to these guys. If you're out there and you know the people that made this, let's get us in touch with them because this is something interesting.

Or if you're out there

Kevin Pereira: and you made this, please get in touch with us. Yeah, please get in touch for

Gavin Purcell: sure.

Kevin Pereira: I used a little demo last night. I didn't change any of the, uh, speaker descriptions, Gavin, but I did get this.

Kevin, I'm, I'm so excited to take you to Flavor Town. Oh, come on Gavin. I've said it a thousand times.

No, no. What [00:38:00] you said one day you'd let me take you to Flavor Town and today has geek to be the day. Peak to be. Honestly, bro, every time you say it, it sounds creepier and creepier. Please get off my lawn and, and, and put some pants on.

Kevin Pereira: Now, lemme be clear what's going on there. The audio stamp. Apparently you were on my front lawn pants list and you wanted to take me to Flavor Town.

Don't worry about the narrative, worry about the performances. Which by the way, there was no, like, there was barely punctuation built in there. The stammering, that's amazing. And the, that came from the audio model from a basic prompt. So depending upon the level of controllability you have there, you can have a pretty performative, uh, performative model.

I, I don't know, I please reach out to us. We gotta reach out to them. Actually, that's probably the way this works. We don't have Yeah, pro, well, you know, have, wait,

Gavin Purcell: no, I think there might be some reach anyway. Either way we will get in touch with these people. It's very, very cool thing.

Kevin Pereira: Hey, a couple quick things, Gavin, which could be, [00:39:00] uh, chum for the comment waters they're worth mentioning.

Uh, an app came out, uh, I dunno if I, assuming you're familiar with this 'cause you see everything on the interwebs, but it was called T

Gavin Purcell: Yeah, I saw it. It's a, it's the number one app in the world right now because it is basically allowing women to share comments about guys basically. Yes. And the kind of red flag people and see stuff.

It's a big deal.

Kevin Pereira: That's right. So Ken Wheeler took to chat, GPT. He took to Grok, he took to Claude and he took to Gemini to try to build a clone of the app that lets men report women and put up their red flags and their toxic traits, et cetera. And all four models rejected him outright. They said they would not participate in the building of the app.

So what you have here is an app that's number one in the app store, and you have four foundational models refusing to build. A gender swap, if you will, right. Of that app, which is pretty interesting considering we, we were just talking about GPT five being the smartest thing in the room, and they're gonna be our employees and we're going to be working for [00:40:00] them someday.

And yet they're refusing to do certain things. And this is, again, a new paradigm with computing. Usually you, you command the machine and the machine does your bidding. Well, now the machine can refuse to. Yeah. And what does that near future look like?

Gavin Purcell: I mean, the one thing I'll say about all these things, and again, this seems like a tricky one because it's a little bit probably about how the machine's interpreting the request Correct.

Or what the gender swapping is. You know, every company that exists, every public, a public and private company should have the ability to put their own rules on the thing. The thing that I start to worry about, as we talked about above before, is like when a government starts to come in and says like, your thing has to be this, that is where you start to get into real trouble.

If, if you can't get an answer out of this. I think ultimately we hope that open source models will be there where they can give you the answer you're looking for, whatever it is. So this is a weird thing for sure, and probably there is some sort of like triggers. This is going off. Um, I would like to invite my, uh, other developers to create some sort of app called.

Beer or something where you can go and instead of, [00:41:00] uh, putting red flags on, on, uh, women, you would put red flags on bars that are not great, places that don't serve good beer. Some version of that would allow you to open the door to all sorts of interesting sort of conversations with the ai. Maybe try that instead, instead of, instead of complaining here, let's try that one.

Let's get into something real.

Kevin Pereira: U-S-A-U-S-A. Oh, hey, real quick, too fast. VLM, this is some nerdy stuff, but there is a demo. A Apple research released something that's really interesting. It is a device that is small enough and quick enough to run on iPhones, and it's a visual model that they, they figured out this new thing where instead of, uh, increasing the size of the image, which the camera captures to do some sort of visual recognition or reasoning on, they've come up with a new hybrid mode that kind of generalizes what.

It's seeing, which allows it to process super quick. And so there's a video, uh, we'll, we'll, we'll link to the paper. It's really interesting. But it shows that on device, on an iPhone, um, you can ask like, what is the sign that I'm looking at? Or How many [00:42:00] fingers am I holding up? Or whatever. And it's very quickly, again, on device, no internet connection, giving you answers based off the imagery that is seeing a whisper at what Apple might be brewing.

Uh, and I thought that was kind of cool. Awesome. I'm surprised

Gavin Purcell: they're brewing something at all. It seems like at this point it seems like there's not a lot of brewing happening over there. Shots, fire shots though. Their brewing is my beer app. Get on that Apple. Start brewing that up. All right, what's your next thing?

You got

Kevin Pereira: Ka Covy World's first AI employee was announced. Gavin, we haven't had a chance to put it through its paces, but according to benchmarks, these Gaia level three benchmarks, which measures an AI's ability to com complete economically viable knowledge tasks.

Okay? It

Kevin Pereira: scored a 77% in humans score, like roughly I think like an 86 or an 85%.

So it's. It's nearing human level, em, em employment skills, and they're showing that it can connect to Slack. It gets a phone number and an email address, so you can call it and ask it to do something. You can shoot it a, a, a text message on Slack [00:43:00] or, or a, you know, an SMS and it will go off and do, and one of the examples is having it go learn how to use Google VO three and make a commercial on its own.

And it does it for itself. And it's kind of interesting. So the reason I bring this up is that there's a lot of tools. Coming out that are showing, oh, we can use browsers, we can use tools, we can be agentic. But the positioning of this as an AI employee and it's asking people who run companies out there to considering hiring Kafka versus hiring a human.

This is a, a, a very real positioning that people have, you know, feared for a while and, and, and guests that would be coming. But according to brain-based labs, it's here today. Gavin,

Gavin Purcell: so I, Kev, this is another interesting AI agent like Kev. I don't have, um, chat GPTs agent yet, which is interesting. I thought it was gonna roll out, but I just heard you do have it, which is funny 'cause you're not on pro.

You're a plus user of chat. I'm on

Kevin Pereira: Plus. Yeah. I got it. Um, I, I'm not saying Sam Altman is picking favorites, but it's weird that [00:44:00] I have access to it. Sure. Feels that way, doesn't

Gavin Purcell: it? Sure feels that

Kevin Pereira: way. Well, I mean, and I, I, I understand for good reason. If I were in your shoes, I would absolutely feel less than even more so this week.

But I did run my first. Agent research project yesterday for an app that I wanted to build, and it worked for 16 minutes. Wow. Uh, to, to research. Um, and then 24 minutes in total, so an additional eight minutes while prepping a project for me. And over the course of doing the research for this app that I, I can.

Chat about just yet. Um, it, it researched, I think it was 300 something sites like cool. It went, it went deep and gave me not only the report as a markdown file, so it's like nicely formatted. I can go and see the research that it did and it cited all sorts of things. Some of the citations kind of broken.

I just wanna be clear there, like it, it was citing the source. I'd click on the source and then the page would error out, or it'd be really, really old. So that was kind of interesting. But what's nice is that when it's done with the report, you can chat [00:45:00] with it about the report. So rather than having to go and read and hunt for specific things, I could just say, Hey, what was, who's the person to contact about this?

So what are the specifics regarding that? And you're having a conversation with the agent that went off and did the reporting for you, kind of nice. Did it work? Yeah,

Gavin Purcell: like did do

Kevin Pereira: everything you wanted. It, it ultimately led me to a contact that I was trying to get for a very specific thing regarding an application, and it did it.

That's

Gavin Purcell: very cool. That's awesome. Yeah. Well,

Kevin Pereira: that's great.

Gavin Purcell: Oh, I do wanna say Kafka. Terrible name because Kafka, if you know the stories that Kafka writes, it's a very sad idea of what these AI agents are gonna be doing, stuck in some sort of bureaucracy. Like, I don't love the idea of calling something Kafka that's gonna be doing work for you because it, it makes me think about, uh, all the Kafka stories that are not fun for workers in general, but it is what it is.

You know what I mean? Well, I'll

Kevin Pereira: say for the, uh, uneducated out there like myself. It doesn't matter. Fine, me. Fair enough. Go for it. Fair enough.

Gavin Purcell: Quickly, there is a new robot from Robot Era. This is another [00:46:00] Chinese robot, a new humanoid model. We talk about how far ahead China is, Kevin, this is just another fun cool, like, kind of like exciting looking robot premier, uh, video.

A little bit better lit than some of these other ones. But, you know, it's doing all the fun stuff. It's having a good time. I cannot emphasize enough how often all the fun stuff, how often China is dropping these videos. Like part of me, I I wanna go visit China at some point in the next like two to three years.

Just 'cause when I go over there, my expectation now is that these are everywhere. Like we co we covered that really crazy race a few weeks ago where yeah, they have, uh, marathons of, of humanoids. They're going to be in this world faster than we are. But anyway, another, another pretty interesting robot model came out of this

Kevin Pereira: and by doing all the crazy fun stuff, you mean like doing like jump spins?

Like a 6-year-old who just discovered they have knees, but also later on in the video you watch it sorting packages at a pretty good pace. Yes. So it is grabbing the package, it's analyzing it, it's flipping it so that it could be scanned, have the barcode grab [00:47:00] it, has dextrous hands, grabs the package, moves it into a bin.

Like these are very basic tasks that you know right now, humans still perform better, but every time we see a new video drop, the machines are catching up and quickly

Gavin Purcell: and at the very end of the video you see the machine running and then slowly overtaking the human, which is a great metaphor for what's gonna happen to all of us.

Kevin, also, there were some more robots that happened this week. What are the other robots stories? Yeah, we got

Kevin Pereira: optimists wearing orthopedics, serving popcorn at a diner. So there's that, that diner thing is so

Gavin Purcell: weird. What happened there?

Kevin Pereira: Why is that? The, uh, Tesla diner, which popped up in, uh, Hollywood here in California, and it is a, um, it's a drive-in movie theater essentially, where you can go charge your Tesla, order some burgers, but they've got optimist, robots, likely teleoperated.

Yeah. Slowly grabbing little containers. I think clearly containers, little

Gavin Purcell: operated in this situation. And

Kevin Pereira: yeah, serving you some popcorn. But someone, uh, went and grabbed like some behind the scenes shot. So you can see that the, the, the Optimist Wears Dad sneakers. Yeah, they look like New Balances, uh, or some sketcher slip-ons, no [00:48:00] shade.

Um, and it kind of slowly pivots and grabs a popcorn and serves, and some goes on the floor and some goes in the bin. But you see that this is a kind of a gen two optimist design. They teased gen three, uh, at the, I believe it was on the earnings call yesterday, but yeah, you know, glad that there's. Robots in the US doing things.

Question mark.

Gavin Purcell: You know, I've just thought about a weird new job. They always talk about how all these jobs are going away. One job that's gonna be really interesting is robot wardrobe specialist. Have you seen, there's a very funny video that came out a couple weeks ago of a robot running down a hill that's wearing almost like an Adams sand looking outfit.

Yes. Like. I, I think that will be really interesting. Like, who's gonna make clothes for the robots? We already talked about that one robot who's made a fabric, which is a terrible idea. Oh, oh yeah. Smelly. Smelly boy. Smelly boy. But also, that's a big deal. That's gonna be something someone's gonna have to make clothes for these robots.

And then there will people who dress them. Imagine the first robot stylist. The first robot celebrity stylist. That's job. I want the robot runway.

Kevin Pereira: I [00:49:00] want the, the gala of Rent the Robo

Gavin Purcell: Runway. That's what we're gonna do, Kevin. We're gonna create Rent the Runway, but for robots, and we're gonna get designer clothes that only fit robot.

Anyway, that is the best idea. If you were wondering what our

Kevin Pereira: super, super secret stealth project is, it's one Z for Roombas. Yeah, it's, it's they're little coozies that you could put on your vacuums. Yeah,

Gavin Purcell: but that would do Well we wanna highlight some stuff that are actually, you know what? That's a really interesting idea.

It would do well if you could make like a. A Roomba Cozi that the sensors could still shoot through. We have a fun little segment here that we want to get into where we are gonna highlight very quickly some very useful tools that have come up this week, uh, in a new segment we're calling. Actually useful.

Kevin Pereira: I mean every week we showcase something that maybe someday we'll pay dividends or whatever. Uh, this is a chance to highlight things that you can get your fingers on today that might be hashtag actually useful. The first is object clear. Gavin. [00:50:00] There's still, you know, if you have a Canvas subscription or you wanna play around with like comfy UI nodes, you can remove objects from a scene, but they're not quite removed.

Elegantly. Yes. If you've ever played with certain tools, like even Apple has, like its kind of Apple intelligence powered magic erase and Samsung has one object Clear is interesting because not only does it erase an object from a scene, but it intelligently removes the shadow that the object

Gavin Purcell: casts. How So?

The thing that they're erasing is a dog hanging out with what looks like a cheetah. Yeah. How I wanna know about this relationship. I don't want the dog to go away. I wanna like dive into this story. What, what dog and cheetah got to know each other this well, that they could just like kind of lean on each each other and hang out.

Kevin Pereira: This was actually a really touching like buddy cop sort of thing, like dog and cheetah. It was like a Milo and Otis. It did end with the dog being devoured. Oh

no.

Kevin Pereira: That's so sad. So we try not to talk about the particulars, but it is a dog sitting on a cheetah. The dog is [00:51:00] casting the shadow on the cheetah.

Yeah, that's very real. And when they erase the doggo, gone goes the shadow as well. Hmm. And that's pretty nice. Right?

Gavin Purcell: Anyway. What if the play name was Shadow and it was Gone Goes the Shadow is the name of the movie that we make out of this.

Kevin Pereira: Wow. And written by Kafka. Very dark. You can go use it on hugging face right now 'cause it's hashtag actually useful also.

Also in that vein, our friends at Higgs Field, I say that they don't know who we are, nor should they care. They have a very bad name. Yes. Unfortunately very bad name. It's called Higgs Field. Steal. So our, our, we do have some friends at Glyph that had a similar browser extension for a while that would let you, like, see an image on the web and click a button and then do something with it.

Like a pipeline. Well, Higgs Field, they released soul ID a couple weeks back. We briefly talked about it. Yeah. It lets you basically make a lo or a, an a, a digital clone of yourself that you can apply to their various tools, make videos, images of yourself. So this connects. Your soul ID to any image on the [00:52:00] web.

So if there's a, uh, an uh, like a product photo that you see or a cool image and you wanna put yourself in it, maybe a scene from a movie trailer or a video game, you can clip it and steal that look and put your avatar in it. And so there's all these examples of putting yourself into these, like environments and stuff.

Very cool. But the whole campaign yeah. Was encouraging people to steal from Hollywood.

Gavin Purcell: And I, by the way, like, it's like I'm not Higgs field as a company has made some really incredible stuff. And the steel thing is, I, I, maybe there's a person that was hired at Higgs Field that's kind of like kind of moving these marketing directions in this certain way.

But, uh, one of the things I don't like about Higgs Field is that I think they employ all these kind of AI influencers. If you're on X and you see these people that do all this stuff and they have like this kind of way that they talk about ai, those guys are, I think the no, no here. I think they're getting paid by Higgs field to talk about these tools, which is a very common thing that happens on x.

I had an experience with one of these guys that I don't wanna go too deep into. The guy's name was, uh, [00:53:00] max sq. Um, and we had a back and forth exchange, not about Higgs field. Uh, he put out a tweet basically that said, I canceled my Spotify subscription. I only listen to AI music. It's exactly what I want when I want tonight.

I was in the mood for some UK garage. Totally. Look, I, I understand there's a world on social media where the things that work are the things that you say extremely and the things that you do. I kind of had this back and forth with him when I was like, is that right? Do you really only listen to social music?

And we had some back and forths and at the end he said like, look, I don't, I, if I had said that I listened to like 80% or 90% social music. I don't only get five likes and like. This kind of in my mind Yeah. Is a, is a really good example of what's wrong about these AI tools. Some ways is that like, I just have found these specific kind of tweets really annoying to me.

And that's not, again, against the Higgs field product team. It's not against the Higgs field. No, no, no talent. But Higgs field does seem to bring a lot of these people on board when they do these rollouts. That, that's just my little small rant.

Kevin Pereira: Uh, no, I, I agree. Uh, it's not the, it's not the tool's [00:54:00] fault necessarily.

Yeah. But it is, it is one of the gripes about this space. I mean, there's many gripes. Yes. But this is one of it that every time something comes out, it's like, fire your wife. Yes. Uh, you know, leave your kids. Yeah. You do not need It's over. This thing is over. It's over. Yeah. We're so cooked and it's like, you know.

We should probably adopt more of that Gavin, if we want likes and clicks. Honestly, that's what I'm really coming to terms with. But what am I gonna do? Sue those people Gavin, we, that's exactly, we're not what we're do Kevin.

Gavin Purcell: That's right. One final actually useful thing is a new tool. And this is more of like a funny tool, but it is something very, I love this.

Cool. You can basically use this tool. It's a kind of a plug and play, probably a vibe, code of things. That's very simple what it is, but you can use it to draft a legal document from a fake lawyer to send off to anybody. So it looks like you've got some big legal teams in the back, in your back pocket.

There's a lot of things you can choose from. There's a lot of dropdown menus, but it's a very fun way to use AI to kind of make something goofy.

Kevin Pereira: The idea is, uh, not to impersonate that you're a lawyer because I don't believe that is [00:55:00] exactly legal. Not legal. The idea is that, is that if someone receives a notice that looks like it's formatted legally, the fonts, the names of the partners, uh, what floor.

The, uh, firm is on and how long it's been around. There's some sliders on the website that you can adjust and then put in your notice. So if you wanna like, I dunno, get a landlord's attention or maybe try to get someone to take something down, it was sort of an interesting experiment as what is just formatting a letter, not getting you into legal hot water by pretending you're a law firm.

But what does just formatting the look of something do. And it's so funny, like I see legal documents all of the time and when I saw like the little samples that the thing was generating, it made me go, oh, oh, that. Oh that's, I get it now. I get it. It's a stylistic choice. It's fun. You should go make one.

Send us a take down notice. Why not? Please, yeah, go for for it. Make sure you

Gavin Purcell: tell us it's fake. 'cause we get a couple of those a week that are real and we haven't done anything about them yet. Alright everybody, it's time to see what you did on the internet this week with ai. It's time for AI to see what you did there.

Sometimes ya rolling without [00:56:00] a. Then suddenly you stop and shout,

Gavin Purcell: uh, Kevin, did you see this tweet this week? If you're not watching it, this is from a, a Twitter user or an X user name D five simulate, like des simulate, and it just says OMW and it's on my way. But it is a very cool, like kind of pixel, odd pixel graphics of like, I, I don't how you describe this, like parallax, right?

Where like one part of the graphic is moving in the background, one part's moving closer to you. You see it holding a couple things in its hands. This tweet or this X thing went insane. Like it had a giant glow up in terms of how many people actually used it. The stats are actually phenomenal. When you look at it, it's got 37.7 million views and 145,000 likes.

So wow. What's cool about this is you saw a lot of people come out and say like, I would play a game like [00:57:00] that. I would do this. And then, you know, a lot of people also then said, Hey, this is ai. And then some people said, I don't ever want to use ai. But other people were like, how do we build this? Because a couple game devs I saw said This is almost impossible to do in an actual game engine.

So I thought this was really cool. It's, it's doable. You can actually go make it yourself. Our friends at Cliff Fabian actually shared a pretty decent little, um, pathway to doing it. Um, when you saw this, like what was your first thought when, when, in your reactions? It was, it was

Kevin Pereira: kind of captivating. 'cause we see AI stuff all of the time in our feed, and that one popped up and I was like, oh, that's kind of a gorgeous art style.

And there was something hypnotic about the way you see like the character dual wielding things in the foreground and this giant castle foreboding in the background. And that again, that axed 2D effect. Yeah. As it moved down the road, like I was like, oh, I'd love, I'd like to explore that world. I wanna play a game that looks like that.

And some indie devs came out and said, yeah, we have a game that looks like that. But then others came out and cloned the style and did their own interpretation of it. But what's so fascinating to me is that like. [00:58:00] The space moved so quickly that it was like, oh, here's a cool look. Then someone's like, well, here's how I made it.

Here's 15 variations of it. Yeah. And now today it's like, what else you got for me? AI art scene. Yes, yes. Let's go on. But I'm sure there's gonna be more actual games and experiences coming out with it. And it's, it's kind of cozy. There's something very captivating about the movement that I wanted to play.

Gavin Purcell: Yeah. And I'm so curious to know, like this world of like, you know, how games and AI intersect because a lot of game designers were talking about how difficult this would be to build an actual engine around, but like. Maybe there's a world where this is the kind of AI game that could work if you could figure out the engine side.

Right. Uh, the other thing that came out that was really interesting from a, from a video perspective was our, uh, Ethan Molik, who we're giant fans of, who always does some really interesting stuff with ai, created a prompt that went viral for video three. And this was about video game community theater.

And if you haven't seen these, they're amazing. There's just basically, you know, bad community theater presentations of like Mario Brothers, donkey Kong, all sorts of stuff. It's one of those very fun things when you see, you're like, I wanna try to build [00:59:00] that myself and do something with it. So it's a cool thing to be able to go do.

Kevin Pereira: Yeah, go look. I mean, the, the, the Mortal Kombat one knocked me out. There's some fallout cosplay at the end with cardboard boxes. Like it's just a, a, a great prompt, but also just hilarious to see how the models treat those prompts. Yes. Like the decisions that they make that aren't inherently in the prompt.

So, so good. Um, this is not merely as entertaining, but it, it made my nerdy sense tingle Gavin, it's a satellite imagery, Laura. Ooh. And when you watch the video, it seems magical because what they do is if you've ever zoomed in on a Google Maps or Apple Maps and you see the buildings and trees and they're all kind of like, someone made a 3D model of it and put it in a microwave, they're little melty and low res.

Right. Well, this is a model that turns those images into what look like incredibly richly detailed Oh, cool. High quality stuff. And it's, you can download it for free and sort of run it yourself, but like, it gets me again, thinking about this world of the data that we already have. Being [01:00:00] enhanced in new ways by ai.

When you see that, again, if you're getting just the audio only, please go to the show notes. Go check this out on YouTube. It looks really impressive.

Gavin Purcell: Yeah, shout out to Ready. User Alternative Lab. 4, 4 4 1 who actually did trains this with flux context. I have not spent enough time with Flux context. If you're not familiar.

That is like the newish flux model. Where in flux is the kind of black forest labs. Yeah. Open source, uh, image model. That's very good. But because we now have all these tools that are hands, we can do so much with other image models that we might have subscriptions to. It's like kind of tricky to dive back into that world.

Kevin Pereira: Well, so along those lines, Gavin, of taking existing data and making something new or amazing with it, there is a 3D particle Nerf system that takes a 2D image. Basically it's taking a 2D image, it's generating the depth of that image and it's converting pixels like, almost like Inox. Oh yeah. Like these 3D points of it.

Right. And so, uh, Dominic, uh, [01:01:00] fo sic, I apologize if I'm, uh, butchering your name Dominic, but, um, they did a, a scene from Star Wars, the I Am your father scene dissolved into 3D particles using three Js and TSL and some ai. But basically it creates a 3D model of the scene that you can rotate around. These 3D pixels or like these kind of voxels, if you will.

Um, you know

Gavin Purcell: what's so cool about this is this is the kind of thing that like five or 10 years ago would've been like in a movie credits and everybody would've been like, that's the coolest thing I've ever seen. Yes. How did they do that? And now it's in your hands. Like, you can go play with this yourself.

It's very interesting.

Kevin Pereira: I'm glad you said in your hands, Gavin, because that dovetails into the last thing. Um, Alexandre Devoe has a way to turn any video into a 3D hologram that you can hold in the palm of your hand. And so imagine, and when you look at it, it is taking, again, a video, calculating the depth and doing the same sort of Nerf or or or splat technique with it, allowing you to hold a 2D video in three [01:02:00] dimensional space while it hallucinates the rest.

And if you look at it and go, Hey, that's cool, but it seems a little low res, or it seems a little this, that the other. That you're missing the black forest for the trees. 'cause we just showed you something that takes low res and makes it really high res. There's a near future where a 2D image from something that you, a photo that you took in the seventies or the eighties or whatever, you'll be able to up res it, create a 3D model of it rotated around the palm of your hand in a 3D space.

You just gotta combine these tools and give 'em a little more time and a little more processing and all of the sci-fi is coming. True.

Gavin Purcell: Well, it's so funny you say that 'cause one of the things I think a lot about is like, people have talked about what a bomb the Vision Pro is, right? Which was Apple's kind of super high-end VR glasses.

And I think we're in this kind of like vr, ar, uh, cold period. Like they talk about the AI cold, uh, the cold plunge. No, it's not the cold plunge. Uh, they, they talk about how AI went through a long period of a winter, right? A cold winter where there wasn't a lot happening. What's cool about the ar [01:03:00] vr space is if you're paying attention, there's a lot of this stuff like goss and splat type things that things are getting better.

3D modeling stuff is getting better all the time. People are working on this. So when we do get to the place where lenses come out that everybody wants to wear and are powerful enough to kind of put this in our real world, I think it's gonna be ready to go. Now that might be five years from now, it might be three years from now.

Whatever meta is cooking up is like, sounded like it was like 2, 3, 2, 3 years away still. But we are gonna get to a place where that stuff is good. Kevin, before we leave, I have one more, uh, very important video for you to watch. Oh. It is one of those things that is like kind of life changing in some ways when you see it for the first time.

Oh wow. It's a video of a hamburger that I thought you might find interesting. So before we go take a look at this. A life change.

The audio is even worse. All right, we'll see everybody later if you wanna know what that was. It is not explicit, but it is very weird. Go to our YouTube page. We will see everybody next week. Thanks everybody. Nice to see [01:04:00] y'all. I find new reasons to dislike

Kevin Pereira: you weekly, Gabby.