June 19, 2025

OpenAI's GPT-5 Hype Train, Midjourney Video is Great & More HUGE AI News

The player is loading ...
OpenAI's GPT-5 Hype Train, Midjourney Video is Great & More HUGE AI News

OpenAI's Sam Altman goes on a press tour pre GPT-5 and throws some shade while massive new AI video tools come out including a surprisingly great Midjourney Video. And so much more AI news.

OpenAI’s Sam Altman is doing a full blown AI media tour and taking no prisoners. GPT-5! Humanoid robotics! Smack talk! The next generation of AI is…maybe almost here? 

We unpack Altman’s brand-new in-house podcast (and his brother’s), confirm the “likely-this-summer” GPT-5 timeline and reveal why Meta is dangling $100 million signing bonuses at OpenAI staff. Plus: the freshly launched “OpenAI Files” site, Altman’s latest shot at Elon, and what’s real versus propaganda.

Then it’s model-mania: Midjourney Video goes public, ByteDance’s Seedance stuns, Minimax’s Hailuo 02 levels up, and yet Veo 3 still rules supreme. We tour Amazon’s “fewer-humans” future, Geoffrey Hinton’s job-loss warning, Logan Kilpatrick’s “AGI is product first” take, and a rapid-fire Robot Watch: 1X’s world-model paper, Spirit AI’s nimble dancer, and Hexagon’s rollerblade-footed speedster.

THE ROBOTS ARE ON WHEELS. GPT-5 IS AT THE DOOR. IT’S A GOOD SHOW.

Join the discord: https://discord.gg/muD2TYgC8f

Join our Patreon: https://www.patreon.com/AIForHumansShow

AI For Humans Newsletter: https://aiforhumans.beehiiv.com/

Follow us for more on X @AIForHumansShow

Join our TikTok @aiforhumansshow

To book us for speaking, please visit our website: https://www.aiforhumans.show/

 

// Show Links //

OpenAI’s Official Podcast with Sam Altman

https://youtu.be/DB9mjd-65gw?t=632

Sam Altman on Jack Altman’s Podcast
https://youtu.be/mZUG0pr5hBo?si=QNv3MGQLWWQcb4Aq

Boris Power (Head of OpenAI Research) Tweet

https://x.com/BorisMPower/status/1935160882482528446

The OpenAI Files

https://www.openaifiles.org/

Google’s Logan Kilpatrick on AGI as Product

https://x.com/vitrupo/status/1934627428372283548

Midjourney Video is now LIVE

https://x.com/midjourney/status/1935377193733079452

Our early MJ Video Tests 

https://x.com/AIForHumansShow/status/1935393203731283994

Seedance (New Bytedance AI Video Model)

https://seed.bytedance.com/en/seedance

Hailuo 2 (MiniMax New Model)

https://x.com/Hailuo_AI/status/1935024444285796561

SQUIRREL PHYSICS:

https://x.com/madpencil_/status/1935011921792557463

Higgsfield Canvas: a state-of-the-art image editing model

https://x.com/higgsfield_ai/status/1935042830520697152

Krea1 - New AI Imaging Model

https://www.krea.ai/image?k1intro=true

Generating Mickey Mouse & More In Veo-3

https://x.com/omooretweets/status/1934824634442211561

https://x.com/AIForHumansShow/status/1934832911037112492

LA Dentist Commericals with Veo 3

https://x.com/venturetwins/status/1934378332021461106

AI Will Shrink Amazon’s Workforce Says Andy Jassy, CEO

https://www.cnbc.com/2025/06/17/ai-amazon-workforce-jassy.html

Geoffrey Hinton Diary of a CEO Interview
https://youtu.be/giT0ytynSqg?si=BKsfioNZScK4TJJV

More Microsoft Layoffs Coming

https://x.com/BrodyFord_/status/1935405564831342725

25 New Potential AI Jobs (from the NYT)
https://www.nytimes.com/2025/06/17/magazine/ai-new-jobs.html

1X Robotics World Model

https://x.com/1x_tech/status/1934634700758520053

SpiritAI just dropped their Moz1 humanoid

https://x.com/XRoboHub/status/1934860548853944733

Hexagon Humanoid Robot

https://x.com/TheHumanoidHub/status/1935126478527807496

Training an AI Video To Make Me Laugh (YT Video)

https://youtu.be/fKpUP4dcCLA?si=-tSmsuEhzL-2jdMY

 

AIForHumans114OpenAISamAltmanGPT5

Kevin Pereira: [00:00:00] Open ai. Sam Altman is on a rockstar media tour. He's hyping GPT five. He's taking shots at Mark Zuckerberg and Meta and maybe distracting us all from the Open AI files

Sam Altman: after the election that I didn't think Elon was going to abuse his power in the government to unfairly compete. And I regret to say I was wrong about that.

Gavin Purcell: What exactly is coming from open ai? Is it a GI or something more this week on Unsolved Mysteries?

Kevin Pereira: Gavin? No, this is not Unsolved Mysteries. Please. What? What is this?

Gavin Purcell: The question is, Kevin, with new releases from Midjourney by Dance. And Minimax is AI video as good as it's ever going to be?

Kevin Pereira: Well, sure that's a, that's a question that will answer along with what new things people can do with these tools.

Also, Amazon's CEO says that AI will be amazing for the company, for whoever's left at the company. That's the important part. I suppose.

Gavin Purcell: [00:01:00] Also robots. Are they real or are they just figments of my

Kevin Pereira: imagination? They're real Robots are Robots are very real. We've got a ton of new updates. We're gonna show you just how real they are.

Gavin Purcell: Also, is the person across from me real or did I just eat Taco Bell too late again? You're Baja blasted, bro. This is AI for humans, everybody. There we go.

Welcome everybody to AI for humans and open AI has a lot of things to say. This week, Kevin Sam Alman has gone and done a lot of chatting. Let's just say, uh, have you had a chance to see all these things? Bye.

Kevin Pereira: Big toe hurts. How dare he try to step on our feet, get into our arena, get into our game. Who does this

Gavin Purcell: Altman think he is.

It's true. Kevin is referring to the fact that OpenAI now has an official podcast. They released an official podcast today and they've been doing a lot of like, uh, chatting in general, but this is now from their official pipelines. And Kevin. [00:02:00] Uh, Sam has, we are the official podcast. Are we? We are the official podcast of all.

I don't think they know that. Good. They

Kevin Pereira: probably can. Oh, well, someone better tell them, Hey, listen, AI for human heads, you better take to the Digi streets and start raising a ruckus. We are the only AI podcast, and I stand on that, but okay. What are they saying? What are these noobs. To the podcasting arena saying, Gavin, is anybody listening?

Is anybody paying attention?

Gavin Purcell: Oh, I think so. This is a pretty big deal. Like one of, we've talked about this in the show where a lot of these companies are going direct to consumer with the things they wanna say. And in this instance, Sam is saying a lot of stuff that he may not have said in other podcasts, probably because they can control the messages very clear.

There's another world where this is sometimes referred to as propaganda in a way because it is completely and entirely, uh, controlled by the company itself. But Kevin, the biggest thing I think for our viewers out there. Is, uh, Andrew Maine, who was somebody who used to work at opening. I was the host of this podcast and asked Sam very specifically about GPT five.

Let's take a listen to that.

Andrew Mayne: And I would say that even though we have deep research, [00:03:00] we have these tools. There is a model race going on, and so the question comes up as like GPT five and any idea is that with a system like that, we should see an increase in capabilities. What is the timeframe for G PT five?

When are we gonna see this?

Sam Altman: Probably sometime this summer.

Andrew Mayne: Right? Um, I don't know exactly when I.

Gavin Purcell: Probably, probably sometime this summer probably. We

Kevin Pereira: confirm that as the official open a AI podcast. Yes. We said that I think back in December.

Gavin Purcell: Yes, yes. And we've been saying some form of it at some different date for probably the last year and a half.

That's true. So the good, the interesting thing about this podcast, and we'll get into a couple of other big podcasts that Sam did this week in talking about the stuff that's coming out is that. Obviously GPT five is on the way and we've, and we've discussed what that means. It is the next generation of ai.

And I wanna flag something

Kevin Pereira: too, because people may have seen this immediately after this podcast posted. There was a lot of fear, uncertainty, and doubt about Sam Altman's walking back the GPT five. Predictions, right? He's saying that actually the [00:04:00] 4.1 and 4.5 and these iterative models in between are so good.

Like maybe we could have called one of those GPT five. There was a section of the, the podcast where they talk about that, the difficulty in naming these things. What makes GPT five, GPT five versus a 4.5 or whatever. I didn't have that takeaway though, Gavin. I don't know if you did. I felt like that doesn't diminish him saying.

The change, right? The change from GPT-3 to four is supposed to be as monumental as four to five. He didn't walk that back. He was just saying that that in the interim, they've had all these other things come out that are pretty capable. At least that's how I felt listening to it. So

Gavin Purcell: Kevin, I think he actually got into that a little bit in talking about like the 4.1, 4.2, let's play this other little additional clip from that same section

Sam Altman: post, train them to make them better.

I we're thinking about this right now, like every time, let's say we launch G PT five. And then we update it and update it and update it. Should we just keep calling those g PT five, right? Like we deal with G PT four oh. Or should we call those 5.1? 5.2, [00:05:00] 5.3 so you know, which, you know when the version changes.

Um, I don't think we have an answer to this yet.

Gavin Purcell: It's so simple. Name 'em after Pokemon. Well, I think to your point, what's really interesting about this is like. We've talked about this idea that maybe they're holding back a few advancements to drop into GPT five. Right. And, and again, people have said that GPT five is gonna be a unified model with reasoning baked in, but he does very clearly in this interview, kind of discuss the idea that you may not know the difference between 4.5 and five to, to your point, like there was some like, oh, they're really talking this down.

I don't think that's the case, but we don't know yet. But what was interesting to me about this interview is that Sam really is going. On the record, quote unquote. And the record is somebody that's a friendly interviewer to start talking about this thing that we know will be a big story at some point this summer.

Right. And then, you know, the other thing is he definitely used it as a place to kind of like air some grievances as well. Like he definitely had something kind of, uh, something to say about Elon

Sam Altman: so externally, but at least [00:06:00] internally after the election, that I didn't think Elon was going to abuse his power in the government to.

Unfairly compete. And I regret to say I was wrong about that.

Gavin Purcell: So, I dunno, Kev, I in general, like I'm here for this thing. I'm here for the kind of a Sam Altman, uh, propaganda machine in the, in the way that like, I like when companies have their own external spot to go. But what is your takeaway right now on, on kind of where they sit in this space?

Kevin Pereira: It makes total sense that they'd wanna own their own conversation and get clicks and foster community based off of their big announcements. Instead of giving us exclusive after exclusive. I just wanna let our audience and our patrons specifically know that the, uh, the alpha, the intel is not gonna slow down.

No. It's gonna come from us. Yeah. Quite a calendar of BS predictions to make. I also wanna point out this one clip that I haven't seen get a ton of traction where he talks about nearing the end. Of of surprises, sure. In this space, but that we could see a breakthrough. [00:07:00]

Sam Altman: I think we are. Near the end of this current problem.

But I can imagine a world, I don't know what it is, but I can imagine a world where we discover some new paradigm that again, means we need to like bifurcate the model tree. Great. Even more complicated names. I hope we don't have to do that. I am excited to just get to GBT five. Yeah. And the GT six.

Kevin Pereira: So like the, you know, the, the in between this, if you, if you, the lines are there, Gavin, but the in-between is that with all of the tricks and techniques that are known now.

Confident that we can get to GPT five and GPT six. Okay, that's cool. But. Something may loom in the shadows and this fog of AI war that pops up and creates another O series of models. Yeah. Or Z series, or who knows?

Gavin Purcell: Yeah. I've heard that from a couple people Now, this idea that like, there will always be something there, there may be a new paradigm, right?

Like we don't know what that thing is yet, and maybe that's what Ilya Suca verse trying to figure out on when he goes off. But like Yeah, I mean, you would hope so. John. John Carmack will solve it.

Kevin Pereira: I'm sorry, what was that? Gavin? I talked over you. I didn't hear it. What was that? John Carmack gonna [00:08:00] solve it?

No,

Gavin Purcell: John Carmack's not gonna solve it, but. Kevin, there is another Sam Alman podcast, another friendly spot. He went on his brother's podcast, so these dropped two podcasts today. This is like,

Kevin Pereira: listen, this is like OAN Newsmax. Yeah. Uh, info. It's like they, he's got, he's got his own podcast now. His brother's got a podcast.

I'm sure his second cousin twice removed. He's gonna have a podcast and they're just gonna have all the same talking points. I'm here for it.

Gavin Purcell: Well, the interesting thing about this one was another good. Conversation that happened, but like the really kind of juicy thing that came out of this, which we had heard rumors of, is that he dropped the idea that like Mark Zuckerberg is offering a hundred million dollars signing bonuses to AI talent.

And there's been a lot of conversation online this week about the idea of open ai. People may have turned down this offer. Let's hear what Sam has to say about it because he dishes.

Sam Altman: They started making these like giant offers to uh, you know, a lot of people on our team. Mm-hmm. Um, you know, like a hundred million dollars signing bonuses more than that comp per year.

That's crazy. Uh, [00:09:00] and I'm actually, it is crazy. I'm really happy that at least so far, uh. None of our best people have decided to take them up on that. I think that people sort of look at the two paths and say, all right, open eyes got a really good shot. A much better shot at actually delivering on super intelligence.

Uh, and also may eventually be the more valuable company. But I think the strategy of a ton of upfront guaranteed comp and that being the reason you tell someone to join, like really the degree to which they're focusing on that and not the work and not the mission. Um, I don't think that's gonna set up a great.

Culture. Uh, and you know, I hope that we can be the best place in the world to do this kind of research. Uh.

Gavin Purcell: Billionaire battles I'm here for. It's really what we're looking at. I mean, the fact that you can imagine somebody offering a hundred million dollars in cash, first of all, think of that number as just bonkers.

It's like a lottery win for somebody. But I do wanna point out, well, and

Kevin Pereira: I just wanna, I wanna check our privilege. The lottery win for some, for us another day at the office, and we have to turn those down constantly. We wanna bog [00:10:00] our, our,

Gavin Purcell: I wanna point out that, um, uh, Boris Power, who is the head of applied research at OpenAI.

Type this out on X. He said there's things money can't buy and he put super intelligence. So Boris may be one of these people who was offered a bunch of money and turned it down because like he believes that like they have a greater shot at super intelligence and, and just for everybody out there in the real world wondering how in the heck do you turn down a hundred million dollars for anything?

I have the same question, but first and foremost. These are people that are mostly all already rich, right? And you have to remember, they're getting paid millions of dollars. A lot of them have a ton of open AI stocks, so they are very wealthy and in their minds, they're seeing meta. This company that like, you know, is kind of desperate now to bring their ai uh, uh, back to like the state of the art, uh, start throwing money around.

And like they think they have a real shot at this thing that they believe again. We'll be world changing and that is like, that's the reason, like it's hard to imagine this, but like there is a [00:11:00] company that really believes, maybe a couple that, and maybe Google too, that believes they have a real shot at creating a super intelligence, which as we've talked about in the show.

Changes everything in terms of economics and all sorts of other stuff. No one really knows what the answer to that is yet, but it's a big deal.

Kevin Pereira: There also are some people, and you're not watching two of them presently, who aren't money motivated and there's folks in this space who, who believe in the future and they wanna be on the team that wins.

Yeah. Quite frankly, yeah. Like they, they do care more about the work and the white papers and the discoveries and breakthroughs than they do about the bank account. That makes zero sense to me. But it's also probably why

Gavin Purcell: I'm here, right G? Probably why? It's probably why we are not advanced AI researchers.

There's another big open AI story that came out today, although we're still kind of seeing what the repercussions of it will be. So there's a new website out there that could be why Sam is doing his own press today. But there's a new website that is called the Open AI Files from the Midas project and the Tech Oversight Project.

And what this is, is a website essentially that collects what it seems like to me. All the [00:12:00] Sam Altman dirt quote unquote, like it's basically a timeline of the times that Sam has maybe not been truthful and done other things. And like what they're doing is putting a website together where someone can go through it.

From my experience clicking through this site, mostly it's pulling from other sources and they show you the sources. If you go to the right channel, you can see the different sources they pulled. For this, it's articles, it's Karen, how's Empire of AI book? It's a lot of things, um, that are specifically kind of drawing them all together.

I don't think there's anything new per se in this, but what was interesting is to see this drop. 'cause it is kind of, you could see this as like a hit piece against Sam and it made me think, wow, there are all these weird forces kind of going back and forth. And of course we know like, you know, the Accelerationist and the Deacceleration is, and the effect of altruist, like all of that stuff is swirling right now and.

You know, just from a weird AI gossip section there, we covered a couple weeks ago, the story where Dio Modi came out and said like that white collar jobs might be going away at least 50% of entry level white collar jobs. There were a lot of people who [00:13:00] said like, look, DIO is doing this because he wants to control AI and he wants to be, he wants to really warn people about the dangers of it.

And this will slow AI down. It does start to feel like as these conversations get bigger, a hundred million dollars, um, the control of super intelligence. We are starting to see these like nebulous forces kind of maneuver around these things and it does start to like tweak your conspiracy brain a little bit, right?

Like what is the point of this right now? Like, what are they, why are they putting this out right now? Is there a group of people back there somewhere? Unsolved mystery style, thinking about ways to like knock people down in different ways. I actually think it might be the case in this instance.

Kevin Pereira: Well, who do you think it is, Gavin?

'cause I look at the footer on the open AI files and it says, uh, it's created with complete editorial independence. Yes. It says that they receive no funding, editorial direction, assistance, or support of any kind from Elon Musk, XAI, Andro, meta, Google, Microsoft, or any [00:14:00] OpenAI competitor. And it, well, okay, great.

But

Robot Watch: who,

Kevin Pereira: who, who, who did fund you, right? I

Gavin Purcell: guess who watches the Watchman? Yeah, the. Well look, I mean somebody's behind this thing, but it is really interesting. Of course, all of this information we are not denying. You should read all this stuff. These are first person accounts. Like I don't think either of us are saying that like Sam Alman has not had some funky behavior over the course of the time he's been at OpenAI, and I believe you should read all this stuff, but like just know it does seem weird that like it kind of all collects together in this place and it's not really bringing anything new.

It's more just like, here's all the stuff that you need to know, but also like it's all stuff that's been out there before.

Kevin Pereira: Well, Gavin, in between all of the hype of GPT five and the will, they won't they, uh, of being benevolent rulers, I guess, at open ai. Yeah. Logan Kilpatrick from Google's DeepMind, uh, blew the lid off of a GI Gavin, apparently we've already got it.

I. Like we have someone dumped out a bunch of Lego blocks and we can make a GI with it. I mean, we're busy building like little log [00:15:00] houses and cool spaceships, but we have everything we need to build a GI do. You wanna hear what he has to say? Yeah, sure. I will definitely want to hear this.

Logan Kilpatrick: I think a GI is gonna end up being much more of a product experience.

Like I would guess my, if I have a hypothesis about how people are gonna end up having the A GI moment, my assumption right now, and we'll see if this plays out, is that like someone's gonna release a model that ends up being really good and like it's not going to be this thing that everyone is like, we've clearly.

Built whatever your definition of a GI is, but I think someone's gonna weave together the right components at the product level with a model that's really smart.

Kevin Pereira: Okay, so we don't exactly have all the Lego pieces yet. We need the much smarter model, but we know the smarter models are coming. That part has been quote unquote solved.

I think what's interesting about what Logan is saying, Gavin, is that it's not like. There's gonna be a new, uh, trick or technique, or there's going to, there's like not some unsolved math problem somewhere. Sure. That it's, it's not gonna be the model itself, it's going to be the product people [00:16:00] who bolt capabilities onto the intelligence.

And we're seeing that now with like. Coding agents being able to go off and think and reason on their own. They can orchestrate swarms of AI to do things and while the models might not be smart enough today to do that, we've got line of sight on these new releases. GPT five is hitting during this hot boy summer.

We just confirmed it. So when something like that comes out, it might very well be a GI. They're just gonna need people to build the tools for it so that it could do what feels like a GI

Gavin Purcell: for us. Yeah, I mean, I think the biggest thing to be aware of here is. That. You know, there's been a couple stories in the last week that have come out that talk about how these LLMs and reasoning models aren't as hot as you think they are.

But a lot of those stories, what's interesting is they weren't allowed to use tools and that is a real big difference when you think about like what these models can do and when developers start building them with tools in mind and they start building products to Logan's point with tools, you're going to see them be able to do things that they haven't been able to do before.

That may not even be fully in the model itself. I think that's what logo's [00:17:00] getting at, is this idea that like. The tool plus the model is what makes the magic. And that's where it starts to feel like it can do things that humans can do, like if operator or a project with Mariner, if Google's product thing, if those get 10 times better at using the internet, that's gonna feel a lot like a GII think, in that way.

Kevin Pereira: And what separates crows from the rest of the birds? Gavin? I dunno. Tool usage, baby. Think about that for a second. Put that one if you God.

Gavin Purcell: Now we got, now we got a teaser clip. Now we got a teaser clip. What separates crows from the rest of the birds? Gavin? Tool usage. Okay. That's why they're best bird. You separates true fans of AI for humans, for fans who are fake.

Kevin Pereira: Tool usage? No, Kevin,

Gavin Purcell: they've subscribed to our YouTube channel. They've gone our YouTube channel. They've subscribed, they've left us an audio uh, review, which has been great. We've gotten a couple awesome audio reviews. Thank you everybody for continuing to like and subscribe to this show. Watch it all the way through.

It really does make a difference for watch time. And also Kevin, what can they do [00:18:00] to help us out if they want to help us afford all these very expensive AI video tools? We're about to talk about

Kevin Pereira: Gavin. We got a Patreon, which is unlike others in that we don't really do enough with it. Gavin, we have a Patreon, which is unlike others in that it is a glorified tip job.

Yes, yes. But people can stuff $5 into our waistband and we will use it to generate credits. To make wonky, uh, Shrek dog mashups, which we'll get to later, and a whole bunch of other fun stuff. You can also subscribe to our newsletter, which is completely free at AI for Humans Show. We may even make it again.

It's coming. It's coming

Gavin Purcell: back. Everybody. It's coming back. Alright, so Kevin just referred to, uh, Shrek Dogs. The big news in the AI video space. Actually, there's a lot of AI video news this week. I think four new models came out, or at least three models came out, and one that's a promise. First of all, midjourney video is out.

And I have to say I was not super excited by what I had seen coming out of. They had been doing a lot of AB testing and I was like, oh, that's interesting. Well, it's out officially today. They got [00:19:00] me to resubscribe, so thank you, Patreons. I put some money back in the Midjourney jar. I. And I have to say like it's pretty good.

And one of the things that's really useful with it, Kevin, is that it can take any image that you've generated with mid, during the past and manipulate it. But of, for, of course, the official video that's come out from them has showing all these really cool things like you can see. See that it has that kind of very interesting like mid journey style.

You see a person walking ethereal style. Yeah, ethereal. It just looks much more filmic and interesting than a lot of these AI video models. Like it has a lot of really interesting artistic styles to it. Um, you can see a lot of people in this video that they released are using it for like music video direction, which I could see could be really interesting.

Of course it does not have audio, which VO three has, which to me still is a separator. Just

Kevin Pereira: to level set before we dive into this, um, I'm just. Every week Gavin's still surprised at how quickly the goalposts move. Yeah. Like any one of these clips in their announcement video, let alone the ones that you've made, which, oh yeah, which we'll get to.

Any one of these clips four months ago [00:20:00] would've been earth shattering, jaw dropping. There's no way AI can do this. And now here we are going like, well, the physics on the cloud separation. When the person's on the crystal surfboard surfing through the skies, the physics could be a little bit punchier and it's not natively generating sound when like.

This is a wild year. Sorry. I just level set with it. You have to, we're still blown away. You

Gavin Purcell: have to sit yourself in that space. I will say to me, the audio part of VO three was such a leap and such a surprise. Yes. That that really does feel like a big difference now. And I think that's why, in part, these aren't as exciting as they might be.

But I do wanna say, so Midjourney uh, video is better than I thought it was gonna be. You see this video I put up here? Uh, for some thing we did a long time ago, I had a Terminator robot, uh, running a hotdog stand and it's pretty good. You can use any of your images you've before or a new one, and you get to choose, it's called animate image, and you can say auto low motion, or high motion or manual, low motion or high motion.

And if you [00:21:00] choose auto. It will just kind of assume to, uh, make a motion based on what it sees in the image. And if you choose manual, you can actually change the prompt. So I used, uh, an auto high motion on our, on our robot here, and you can see like it's really not bad. Like it's, it, it, this was one of four things, obviously.

That's another thing to be aware of. You get four results if you're

Kevin Pereira: just getting the audio. This is, uh, a Terminator Cyborg operating a hotdog stand. As you mentioned, the camera slowly pans to the right, there is steam. Yep. Rising up from something on the side of the frame. Uh, there are reflections in the metal, which are slightly blurred out, but they seem to be accurately reflecting like the park setting that this is in.

He's holding, uh oh. The cyborg is holding a. It looks like a hotdog made of hot dogs.

Gavin Purcell: Yeah. Well,

Kevin Pereira: in a hand, and

Gavin Purcell: I will say be, to be clear, like this was an image that was generated a couple months ago in Midjourney. So like, don't blame the actual look of the image. One thing to point out that I think was a, a little funky, that's fair.

You see the woman show up outta the blue, she kind of dips into what looks like the. The point of the, of the, um, tent a little, like she's kind of dip. She's the

Kevin Pereira: [00:22:00] liquid metal Terminator Gavin. She can go through

Gavin Purcell: it like Mercury. There's another version of this, another still that I had created from a, an image a while back, but you can see like it's actually really good looking animation.

Yeah. It's almost like, looks kind of Pixar. The, the downside of this one, if you look at it, is like just to, if you're, if you're only listening, it's kind of a big eyed robot that kind of closes its hand and looks like emotionally like engaged and talking to somebody. The hard part with this is now it's talking.

Obviously I have to figure out how to match those lips, right? Which is gonna be really tricky, which is why VO three is such a big step up because what would ideally be cool is if I could have told Midjourney what it's saying so that it would match those words, and that I don't think is possible yet.

Kevin Pereira: Um, still a beautiful model though. Yeah. And I can see why people are using it to do music videos and storyboards, but yeah. I mean now it adds just another layer of complexity to say, okay, we're gonna bounce out this video, the robot don't have it talking. Yes. So that I can use another program and add the audio and the lip [00:23:00] flap.

So it is just another layer. But I mean, we knew Midjourney was cooking on a video model for a while. Yeah. And it's nice to see that it's out. What does it run for? People that wanna sign up? Is it like a $10 month? It's $10.

Gavin Purcell: And, and to be honest, completely honest, I, I just got it this morning. So I'm not entirely sure how many credits I'm gonna get for those $10.

I assume not that many, but Kevin, I do wanna share the other thing. You created some images for me to try it with and one of my favorite images, uh, was of, let's just call it, uh, a cute dog with a Shrek head, which I think is a fair description of it. I texted you, by

Kevin Pereira: the way, I yanked those images. I just want, I don't wanna take credit.

Oh, you did? Okay. Those, I, I, I yanked those images from Crea one from their, from their website. I was looking at what people were generating. So shout out to whoever made Shrek dog. Oh, okay. Well that's good to know. Thank you for making Shrek dog. That's good.

Gavin Purcell: So, so anyway, we took somebody else's Shrek dog and then we animated.

And what I do wanna say is like. When you talk about the AI slop application of the world, like you can see how this is getting there fast. Like this little video you watch of, like I just said, dog runs around the room and then looks back at [00:24:00] camera realistic. And again, it came back with four variations.

A couple of them weren't that great, but the one I sent you. Pretty interesting, right? Like it jumps up, you see it turn around, you see its tail kind of wag a little bit, and then it comes towards the camera. Terrible. It's, it's really good though though. Like, it's good. It's good. Yeah. But it's terrible. Yes, yes.

It's good, but terrible. So we suggest, if you're only listening to this, go to our, uh, YouTube page. You can watch this. Exactly. But like,

Kevin Pereira: what's really nice Gavin is no, the, the coherence of the little things, like in the past video models, like the, the, the Shrek dog has a weird kind of collar Yeah. With this dangly metal bit on it and the character.

Stands up on its hind legs and, and start, you know, with human nish, knee joints bounces around completely turns away from the lens, and when it turns back, it's still the same piece of jewelry. Yeah. You know, that's the same collar that's on it. It's those tiny little things that are, that look solved. Yeah.

And the fact that this was one image and it generated all of this motion that looks really, really solid and including hallucinating the environment, which looks believable, that the animal's in like just, [00:25:00] uh, it's, it's impressive. And it's also haunting. So it's finally out. People can go use it. It costs a couple bucks.

How do you use it, Gavin? Do you have to like, sign up for Discord and send texts, or does this all work through their website? You can

Gavin Purcell: do it through your website now. Um, you can now join with a Google, uh, link or a Discord link. I still was a member from before when I was paying, and so I'm paying again through Discord.

But like, I think the website's much better than Discord. This was one of the things if you're, if you're new to Midjourney, like back in the day, you could only generate in Discord much cleaner this way. Gotcha. Much easier to do so Kev. There are a bunch more AI video stuff that we have to get through, and I think there's some really important ones to talk about.

First and foremost is A is a tool that you cannot go use yet, but you can see a lot of the results, which is called See Dance, and this is the new state-of-the-art model from uh, bite Dance, which is the company behind TikTok and Kevin, it is amazing looking. Now again, we haven't had our hands on with this, but one of the coolest things about this, obviously they released their promo video and they showed all really interesting stuff.

There is a website you can go to, which I really do appreciate. There's all these new model websites where you can kind [00:26:00] of pit them against other models. There's actually a site that connected to them somehow called Magic Arena. It's called a IG. Uh, C-A-R-E-N a.com. And if you go, you can kind of see different variations and you can click through which one's better based on certain prompts.

And I also love that they have little happy and sad emojis when you click on one, but you can see seed dance, uh, generations in this and some of them are unbelievably good. So it's an interesting thing in a way to look at a bunch of other seed dance generations that give you a way to see stuff.

Kevin Pereira: Uh, now you've lost me because I am.

I am rating video clips in an arena now. Like that's insane.

Gavin Purcell: Yeah. What's cool about these arenas, I did this with another model, like one of the coolest things about the video clip arenas is you can also get a sense of how things are getting prompted. Because one of the coolest things when SOA came out was every, in the four oh image gen is everybody was able to see other people's prompts.

With videos, you often can't. It's a good way to steal prompts if you want to use 'em in your own model. So you can see specific things will be in there and be like, [00:27:00] oh, that's awesome. How can I do that? Literally they show you the prompt. So it's another good thing to think about with these video arenas in general.

And then Kevin, it is not just this tool. There is a new update to Minimax. Hlu. O2 is also out, which is yet another video model. If you remember a couple weeks ago, we had said in the podcast there were some rumors that some video models might have been delayed based on the fact that, um, uh, VO three came out.

While we are seeing all of them now, and again, halo O2. Very good. This is the update. Um, better motion. Better physics, all sorts of stuff. Natives 10 80

Kevin Pereira: p. Yeah. And the, on the physics side of things, it does acrobatics now, like gone are the double back flipping person that turns into one giant leg and then explodes into arms with a leotard.

I, I'm going to miss that phase of ai. Like, I, I hold that deer, but. I mean, this is, the effects look really incredible. The physics look amazing. They've got a couple of samples in their demo video of like a big top circus where there's a unicycling on, uh, like [00:28:00] a slack line basically. And they're, you know, juggling and it's, it all looks legit and people turn to dust and the big top collapses and it's just, it looks.

Really, really impressive. Like we've we're past the threshold now of like B-roll from stock sites. Yeah. Like we're past it now.

Gavin Purcell: And I think that's one interesting thing it made me think when you were saying that is like, you know what, these all are, are kind of different cameras. You can imagine it that way.

Like each one's a different version of a camera that will give you different results and different kind of images out of it. But like they're all getting to the place where they're pretty good. I do wanna point out, uh, there's a squirrel, uh, that it looks like amazing when I say it like. Squirrel physics might need to be like the next kind of like pipeline, uh, a top benchmark, but like, just looking at the way the squirrel's tail moves in this clip, and it seems like a pretty boring clip, but when you start the clip, what you're seeing is the squirrel's tail kind of fold in on itself and back, but the light coming through this off fur is like really shocking.

Like it looks like super, super high-end video footage and that

Kevin Pereira: [00:29:00] just alone is pretty amazing. I'm glad that we just cut out the 15 minutes where you went on and on about squirrel fur and just how soft it is and how delicate it is, and I'm glad the viewers, I do love my, I do love squirrel.

Gavin Purcell: I do love squirrel fur.

Kevin Pereira: Interesting that you mentioned that like each new model is like a different camera. It unlocks new capabilities. Higgs field continues to drop new features, uh, uh, it seems like every week with new crazy VFX new camera controls. Uh, the last one that they just dropped is their, their version of Canvas Gavin.

Mm-hmm. Where you can take, uh, an image. You can paint over it and say what you want it to be swapped with. And if you look at their demo video, which again, very cherry picked examples, I have not gone hands on with this yet, so I don't know. But you can just see some of the capabilities in their examples.

This looks like this is gonna be a very powerful tool if you want to swap the object that a model is showing off as an influencer. If you wanna change clothing, if you wanna change scenery. Now you can just sort of paint where you want it to go and ask the model to do it. [00:30:00] And it's just another tool in the AI generative tool belt.

Gavin Purcell: Yeah, and Kevin, it's not just like video, like CREA one just came out, which is Korea's new image model, which is a kind of a lower end image model, but you can watch it generate in real time. CREA relaunch their entire like suite of products. You can go try this for free right now. I also think this leads into this idea of.

What we have seen over the last like say, month since VO three came out, which is I would refer to as like just an explosion of AI video content. And it shows you that not only are the tools getting well, but people are finding really creative ways to use them. I'm sure you have seen, and probably many people in our audience have seen these storm trooper vlogs that are out there where you know somebody is a storm trooper and people are getting millions of views on these.

There's the big flip flos. A really interesting thing was pointed out by Olivia Moore who works at a 16 z. That did, do you know this Kevin, that like, uh, VO three can actually generate, uh, Mickey Mouse and Harry Potter and other characters that like, clearly Midjourney was just sued for. And it made me wonder, like, do you think they did a deal with, [00:31:00] uh, Google to allow that to happen?

'cause I was able to do it myself there. I created a, a video of Mickey Mouse doing a SMR where he's like scratching at the microphone. That felt like really weird and funny to me that like I was able to generate that. So I wonder. I wonder if there are big deals happening on these IP sides in the back that we don't know about.

Especially for something that's as big as VO three. I found out

Kevin Pereira: about it because of Olivia's thread, you know, specifically? Mm-hmm. Where she's got Mickey Mouse waving. She has Harry Potter riding around on the broomsticks. Spiderman, Indiana Jones. Well. Indiana Jones esque character. Yeah. Mario and Luigi.

Like did Google do deals with all of these companies? I can't. Nintendo is traditionally one of the most restrictive. Yeah, I can't imagine. They said yes. Go ahead and put our characters into your generative tools. And so my question then becomes, oh by the way, the Abraham Lincoln with Darth Vader Barbie and Elmo having tea Oh sounds is something that amazing, phenomenal.

Like definitely my kink, but that should have been [00:32:00] punted. Just on the prompt alone and it, and it wasn't, the result is actually kind of funny looking, so I can't imagine. Deals were done. I think in the future the deals will get done. Yeah. And we'll have our metaverse of AI creation, but I'm kind of shocked that it lets you do it.

I'm curious how far you can push it though. Yeah. Like the fact that you got Mickey to do a SMR is interesting to me. Can you get Mickey performing a fatality on Darth Vader or something like that? Do you want me to try? I

Gavin Purcell: guess I could see if we could try right now and see what happens. Like, uh, I'll bring try.

I'll bring it up. Try. I'll bring it up. We'll see how it goes. Forward, forward. He's try

Kevin Pereira: Mickey doing a Mortal Kombat style fatality on Mario would just tickle me. I definitely go on my,

Gavin Purcell: my account to get flagged. So I have to find a way to kind of write this in a way that feels like it's, they're not gonna flag the

Kevin Pereira: account.

They might flag the creation, but you're, you're Gavin Purcell. You're the official host of the official Open AI podcast. Oh, that's, I forgot

Gavin Purcell: where the official OpenAI podcast. No one's gonna, of course, Google would want to give us access to all the crazy things in their product. By the way, Google. I love you.

This [00:33:00] product VO three is amazing finding It is insane. I literally had to tell somebody the other day to Google Flow VO three, Google, and even then there was a fake website that came up ahead of your thing. So please fix that. Okay, so what are we gonna say? Let's, let's see. We're gonna say, um, Mickey Mouse.

Performs a Mortal Kombat style fatality. Okay. Performs a Mortal Kombat style fatality on Bugs Bunny let's say. Alright everybody, um, we had a quick technical break there, but actually it gave us time to have VO three render this clip that we were talking about. And guess what, Kevin? You can make Mickey Mouse fight Bugs Bunny.

Now it is not a fatality per se, and I guess in the cartoon world it is. But play this, listen to the audio too, 'cause there's some stuff going on here.

We, you're out now, baby.

Kevin Pereira: I just watched it that it was actually not bad in some ways better than what I thought it was going to be. Yeah, yeah. In some ways, way worse. But the fact that it gave it to you and [00:34:00] that you're not on, you know, a sundar's watch list right now. Insane. Well, do we,

Gavin Purcell: we're not gonna push this further, but there were some suggestions from further.

No. Let's a further from someone else on the podcast wants, let's push it a little further. I got some ideas

Kevin Pereira: now.

Gavin Purcell: We're gonna get into this later, Kev, we're gonna get into this later. Okay. On somehow do anonymous, uh, VO three testing. But another, I just, one of those things that's fascinating to me. You can have multiple ips and again, midjourney, which we just talked about last week, was sued by both Universal and by Disney for allowing them to use characters for their thing.

And in this case, we're talking about a Disney character and a Warner Brothers character. What's up Kevin? What's going on? You are out now, baby.

Kevin Pereira: Really cool that it gave you the generation without banning you, but as we know, sometimes these generations are kind of like pulling teeth.

Gavin Purcell: Oh, fantastic transition, sir.

So there is one more really interesting thing, um, weirdly from Olivia's sister who both of these, uh, these two, I just listened to this, they, they're twins that, uh, work at a 16 Z and do now a podcast about ai, but her sister who's under venture twins, we've talked about in the show a bunch. [00:35:00] Showed a video that was created with VO three on TikTok that was actually using these kind of monkey vlog type things, but it was for a dentist, it was a, it was an advertisement for a dentist.

We're not gonna play the thing here, but go to our show link so you can see it as we're talking about it. This is just a really creative local use of AI video because they kind of glommed onto the idea of like something that's big right now, which is like the Bigfoot Vlogs in this case, gorilla Vlog, and they made an ad out of it and it's super fun to watch and like it's one of those things that's like, I get AI video when you can do stuff like this particularly.

Kevin Pereira: I think it's super exciting and great and it's gonna be a fun way for all those Amazon workers to pass the time, uh, sending videos back and forth with the Microsoft employees. Gavin, let's get into the

Gavin Purcell: end of times. We are here again. A couple weeks ago, we covered the idea that the jobs that AI might be not providing a lot of jobs, but they might be taking jobs away eventually from people.

Amazon, Kevin, this week, Andy Jassy, the CEO of [00:36:00] Amazon, came out and said that generative AI is an amazing tool for them. But they are also seeing that eventually they will have less people working at Amazon because of it. Now, in some ways, for stockholders of Amazon, this is good news because like you become a more efficient company.

But Kevin, for the world at large, Amazon is one of the largest employers, right? There are a lot of people who work at Amazon. The CEO of Amazon saying that there will be less jobs at Amazon. A company that is a growth company seems like it should not be said by that guy. Right. Okay. Well, let me, let me, let's

Kevin Pereira: not take his words outta context.

Gavin, I wanna read you exactly what was, uh, here. Quote, quote for quote here, uh, CNBC article. We will need fewer people doing some of the jobs that are being done today and more people doing other types of jobs. The quote continues. It's hard to know exactly where this nets out over time. Now I'm gonna stop right there, Gavin, who?

You wanna finish that? There's Kevin. Keep reading, keep

Gavin Purcell: reading it. [00:37:00]

Kevin Pereira: But in the next few years.

Gavin Purcell: I'll stop it there again. A great, now, why did you keep stopping? Kevin? This is a podcast you're supposed to continue

Kevin Pereira: in the next few years. We expect that this will reduce our total corporate workforce. That's what they expect.

It's

Gavin Purcell: hard to know, Gavin. It's hard to know. Know. It's hard to know. It's hard to know. But they, they expect

Kevin Pereira: that it will reduce the total corporate workforce.

Gavin Purcell: So that is a big thing. You think of a capitalist economy. You are growing, growing, growing. What that signals is, they're going to be able to grow the company without more people, which is a new thing in this world.

Right? It is a new thing. Yeah. And this is also on the heels of Microsoft just announcing a bunch more layoffs. A lot of people in the sales division, in part because they're trying to get more profits, but also in part because they have a lot of efficiencies they've covered with generative ai. We have been beating this drum for a little while.

These changes are coming, and again, we may or may not be ready for what this world looks like, but there will be less jobs at these big companies

Kevin Pereira: going forward. But again, people have to keep in mind. The joy [00:38:00] that the squirrel physics and the softness of the fur. Yes, yes. With the ray trace lighting coming through its tail.

Like again, we gotta keep that in mind. Gavin,

Gavin Purcell: more importantly, Kevin, you know there's a story in the New York Times this week about 25 potential new jobs from ai, which there's a lot of interesting stuff in there. Some, some stuff that's very specifically to creative work. And we'll talk a little bit about like how that can happen, but.

Jeffrey Hinton, of course, the guy who invented AI has a lot of thoughts on this and says there's gonna be a ton of new jobs that come to ai. You wanna play this little clip from the the diary of a CEO podcast this week.

And I think for mundane intellectual labor, AI is just going to replace everybody.

Logan Kilpatrick: People say that it will create new jobs though, so we'll be fine.

Yes, and that's been the case for other technologies, but this is a very different kind of technology if it can do all mundane human intellectual labor. Then what new jobs is it gonna create? You'd, you'd have to be very skilled to have a job that it couldn't just do.

So I don't, I don't think they're right.

Kevin Pereira: That was a rollercoaster [00:39:00] of a clip for a second there. Oh yeah. That go from

Robot Watch: here.

Kevin Pereira: We always create new jobs when there's massive disruption in technology. I'm like, okay, this is great. And then you go over the edge of the rollercoaster, right. Until, yeah, I think they're wrong.

Gavin Purcell: Yeah, well, so this Jeffrey Hinton is, you know, gone on record to be pretty negative around what's going on with AI right now, even though he was one of the heart of like what started the transistor movement. And I think the negativity comes from the fact that like he's just talking about the world not really being ready to make these changes.

And Kevin, another thing that we're gonna start seeing every day are humanoid robots. And we've got a lot of really interesting new humanoid robot news, not just ones that are gonna take your job in this week's second week in a row, uh, episode of Robot Watch. Robot Watch,

Robot Watch: robot Watch.

Watch

Kevin Pereira: one X World Model. Gavin, the company that likes to have chats. In the middle of the Redwood forest, when they talk about robots, uh, [00:40:00] gave they, they released quite a paper about their world model, which allows them to simulate things for their robot so that the robot can just do in the real world, not something we haven't heard before, but it seems to be working pretty well for them.

They're showing off, um, some demos of what the robot can do in terms of, uh, navigating around. Natural human living spaces without a bunch of safety gear, uh, in the way even opening a door and stabilizing itself is an interesting task. They have an example of how failures help the robot learn even more.

And in their white paper there's, um, like a, there's a. Some, some still frames of it. Simulating a ninja air fryer, like the ninja foodie. Yeah. And it pulls on the handle and in the simulation, the entire air fryer just comes flying off the shelf. Then there's another simulation where it detaches, but takes some of the air fryer with it.

And then on subsequent generations, when given a little bit of real world footage, it sorts out quickly how that object reacts in the world. And it got me thinking about. Obviously like simulation data. [00:41:00] Incredibly important. Matching that with real world video and data. Yes. So that it can, it can enhance it.

Obviously important as well. What's going to happen when new products are released? Like, here's an analogy in the coding world, when someone releases a new piece of software, a, a, a new service that you can use, sometimes they'll give you documentation that you can feed to an LLM. So if you're vibe coding, it knows all about Sure.

How to work with the platform. Are people that, uh, manufacture physical goods, are they gonna release. Documentation vectors, some sort of simulation data. Yeah. Of the objects will happen. Objects so that these robots know how to use it and manipulate it. Will that come out of the box?

Gavin Purcell: Yeah. When you're the doing the next roco salad sweeper or whatever you have, you're gonna have to put that in the system so the robot knows how to use it.

So when you get home, you get your salad swept in the right direction. That's right. But Kevin, what's interesting about one X is like we talked about them last week. They're going for it right now. This is a whole nother thing just to make sure that people don't understand that we're not repeating ourselves.

Like they dropped another thing this week, this paper, this role model and is very interesting and I, I am starting to [00:42:00] feel right now this kind of energy around the humanoid robotics. Like, you know, Sam Altman specifically talked about, uh, the humanoid robotics in this interview he did with his brother.

Sam Altman: What about like moving physical things around behind? But I think we'll get there. Uh, for example, I think we have some new technology that could. Just do self-driving for standard cars. Way better than any current approach has worked. Hmm. And that might not.

Kevin Pereira: Hello, Elon. Tap, tap, tap, tap. You know, forget the, the shots at Elon on the subject of mechanical humanoid robots.

Sam said this

Sam Altman: hard, but like even if we had the perfect brain right now, I don't think we have the body yet.

Kevin Pereira: So, you know, he doesn't think we're there on the brain yet. Yeah, I'm sure he's gonna solve that portion. But he then goes on to mention that OpenAI originally had a hand that they were working on.

Yeah, they've talked about that many times. But it was like hard for all the wrong reasons. It just broke all the time. It didn't quite work. Well, there's company after company now announcing all sorts of robots. Um, spirit AI had one called the [00:43:00] Moz one, which if you Oh, yeah. If you're looking at the video, this thing is going to be like, I don't know, what's the international dance competition where people pop and Oh, Eurovision each other's faces?

No, this the dance one, not the song run. I don't know. It doesn't matter. Uh, so you think you can

Gavin Purcell: dance, uh, Europe edition.

Kevin Pereira: That's right. That's right. Yes. It's, uh, so you think you can, France,

Gavin Purcell: we got, we've got trademark that, but before the end of the podcast, you wanna go and so you think you can friends, so apologies to everybody.

You know what you tuned in for.

Kevin Pereira: I mean, is that a competition hoop for who could be the mo most French. Like my makeup long filtered cigarette. Ooh, you

Gavin Purcell: think you, so, you think you can tan France and it's who could be the closest to Tan France? From a queer eye, from the straight guy. All right, everybody back to Robot Watch.

Robot Watch. Robot watch.

Kevin Pereira: All right, so when you watch the Spirit AI video, what is striking is how smooth and controlled the robot moves. Yes, it stays completely rigid, like completely static while it's independently moving. Different limbs. Um, some people were pointing out some issues. With it having like sharp points that get hooked on things.

But I [00:44:00] mean, whatever the design of the robot is impressive. It's moving smooth and incredible. Uh, there's a, a unique Starbucks at a company called Neighbor 1784. They have a hundred robots that are whizzing about their facility delivering hot coffee to people and another robot that is cleaning out the trays.

Very, very cool stuff. Finally, hexagon revealed a robot that. Is very unique and some people are like, oh, it's, it's got wheels where the feet should be. But the CEO, uh, on stage when he is talking about, uh, eon the new robot, um, they said that they tried feet and then they tried treads and they tried a whole bunch of other stuff.

And what they found that, that they, these tiny little wheel feet, these almost like roller blade wheels, it can move 10 x faster than a human. Whoa. And I just watched the video of it moving it, those things move crazy fast. That's what I'm saying. The thing can whiz about faster than a human. And it has knees so it can still lock the wheels and go upstairs.

So it's not, it's not like bound from navigating the world as you or I do. It [00:45:00] just pulls its robot hees out and goes when it needs to.

Gavin Purcell: Yeah. Kevin, I just had this vision of like the apocalypse happening where instead of being chased by terminators or walking around with guns were chased by like nineties era, uh, kids with like JCO jeans and long things and they just like.

In light skate over our heads and they kind of wreck us this way. Like that's what these things remind me of. Jack set

Kevin Pereira: radio baby. Let the robots do some aggressive inline three. Well, that video stopped me in my tracks Gavin, like so many other things we saw this week and when I, when I, when I came across it, I pointed at my laptop and I said, Hey, I see what you did there.

So.

Logan Kilpatrick: Without

Robot Watch: a care.

Logan Kilpatrick: Then suddenly you stop and shout

What,

Gavin Purcell: so this week, Kevin, IC, which you did there, we're doing a very just quick one. I just wanna do one shout out of a very creative AI technologist who [00:46:00] made a YouTube video that blew up, and I think everybody should watch it. This is by a guy named, just goes by Josh Smiley face on YouTube. That's his handle.

The name of the video is, I trained an AI to make me laugh. It worked too well. This is an engineer who's just kind of like fun guy talking to camera. But what he did is he like trained an AI to watch his face react to different memes, and you kind of go on this journey with him where like he learns how to train the ai.

It gets better recognizing it and then at one point I wanna ruin it, but the AI like. Comes at it from a different direction and he is kind of surprised by it. But this is the kind of like AI technologist content that I just love. And just a shout out to Josh again. This is his third video, and this video is now up at, uh, 750,000 views.

You can tell this is like a native person who understands how to do YouTube in a really cool way.

Kevin Pereira: You can't get that many views on YouTube, by the way that, so whatever he's doing. No, the Matt's views

Gavin Purcell: is like 30,000. We've, we've done that. It was like one time, but, but we got there. We're close. So if you're still

Kevin Pereira: watching this on YouTube, by the way, if you're listening to it, thanks again.[00:47:00]

Five star review. Share it with a friend. If you're still watching this on YouTube, smash the sub button. Click the bell, leave a comment. Juice our algo baby. Please,

Gavin Purcell: please. One last thing today. Um. Google, Gemini 2.5 did a little mini update this week in case you missed it. Like that's a thing that came out.

I had the weirdest experience, Kevin, where I tried to do something with Gemini that I was like, oh, maybe I'll try this. But I wasn't even thinking of the thing I was gonna do. I had an idea, if you're not, if you're familiar, the Naked Gun Trailer came out, the new trailer from The Naked Gun with Liam Neon and directed by Kiva Schafer from the The Lonely Island.

Guys, I'm super excited about this. I grew up with the Naked Gun movies. I grew up with the airplane movies. So I was like, oh, this would be interesting to try to make like a VO three style naked gun airplane movie. And what I did is I took the original airplane trailer, which is like three minutes long, and I said to goji Google Gemini, like thinking like I thought, you know, YouTube is integrated into Google Gemini.

Interesting way. So I gave it the trailer, just the link, and I said, Hey, can you gimme a shot by shop breakdown of what's in this trailer? And then it went through its [00:48:00] thinking. And one of the cool things about Google Gemini Pro 2.5 is you can see it's thinking and it's very straightforward. It started using the YouTube tool and it references that in its thinking and what it's doing here, which I guess is something I didn't know that it did very specifically this way, it's looking at that YouTube video in a very deep way, and what I got back was a spreadsheet of literally second by second what the words were and then who was speaking.

Oh wow. And what was interesting is. The first time it was a little bit off and I was like, you didn't get this place where like, you know, Robert Hayes talked, or you didn't get this exact moment. I said, can you check it again? It went through it, checked it, and then it was perfect and it had the exact dialogue.

It had the people saying the dialogue, it had all this stuff. I was kinda shocked and like what it meant to me was like, I think there are things that we haven't uncovered, and by the way, somebody may be listening to this and be like, oh dude, that's been obvious for blank blankety blank for so long.

Right. But it really did shock me. I was like, oh, this is something that could be really useful to people who [00:49:00] wanna figure out how to like either recreate something or learn about like a specific place in a video that it was. And it made me think about your work with like N eight N and all these kind of automation tools and like mm-hmm.

Part of what we're gonna need to do and maybe a new job going forward is like, how do you piece together these tools in ways that let you do things that you weren't even possible before? So like, this was just a small thing that I spent time with AI this week where it was like, well. Kind of off the cuff surprised me in a big way.

I'm trying to trigger the, yes, Google Meet, raise your hand feature. So it goes work.

Kevin Pereira: What are there next steps for you with this, now that you have the trailer breakdown, are you gonna feed it to like. G PT or back to Gemini and say, Hey, change the scenes and make it all Shrek and all about this stuff.

Gavin Purcell: Well, I'll tell you, I'll tell you, I'll tell you what I tried to do, which is interesting. And this is another area where kind of AI video can sometimes be frustrating. I wrote a bunch of jokes, right? I went, 'cause I, my thought was like, oh, what I wanted do was create a, a hospital version of one of those movies, right?

Because I hadn't seen that before. I think it'd be very funny. I actually worked with Gemini to kind of like think through what are different possibilities and it came up with a couple jokes. But then I wrote a bunch of jokes and I went into VO O three to [00:50:00] generate these things. And what was so frustrating is like these are not things that vo three sh they should be able to do these things, right?

And like the, the trickiest thing with AI video still to date is like, I spent half an hour to 45 minutes and I got about te eight to 10 generations out. And of those eight to gen generations, like two were okay for what I needed in the other eight word. And what, what sucks is. Yes, I could continually hammering against this and hopefully get two more, a bunch more, but eventually I run outta credits and they're very expensive to do.

So I think that the, the interesting thing with AI video is still going to be. Can you hammer it for enough times to get exactly what you want? Because by the way, it was giving me like a doctor, one of the bits was like a doctor walks into a, a hospital bed and a guy's kind of passed out and it's flat lining and he hits the, he hits the machine and like you see something pop up on the machine and you hear the modem sound and then it fixes itself like dumb.

Things like that. Um. And it did a lot of that, but it just got three parts of it wrong in this like [00:51:00] eight second clip where you need a lot of things to go. Right. So anyway, I think that the idea of what people will be able to create using these tools going forward is still massive. But the idea of like.

The actual work to final product ratio is still really hard.

Kevin Pereira: Well, you know, while we've been doing this podcast, now that I know about new features, I just started working on, uh, Requiem for a dreamboat, Willie and I took, uh, I dunno if you're familiar, Gavin, I'm ing up by piece. Imagine Mickey off Goofy for cutting the show off.

They're all fours, black and white. Hi everybody. Very artsy.