AIForHumansAIJobCutsGoogleNanoBanana

Gavin Purcell: [00:00:00] AI isn't just coming for your job. It might've already taken it.

Kevin Pereira: Salesforce, CEO, mark Benioff just announced they cut thousands of jobs thanks to AI automation. That's. His energy about it, not necessarily mine.

Gavin Purcell: Yeah, but Kevin, he also said this video of Tesla's new optimist three was a quote, productivity game changer.

Mark Benioff: Hey, optimist, what are you doing there?

Gavin Purcell: Just

Mark Benioff: chilling. Ready to help.

Kevin Pereira: Yeah. See, that's the future baby. No jobs for humans, but hey, at least nothing really works well.

Gavin Purcell: Thankfully, we will have Google's nano banana to keep us all busy. We are gonna catch you up on some of the amazing new things that people have done with Google's AI image model.

Kevin Pereira: Also, Google just Stealthily launched a new vibe coding platform for all of your vibey coding

Gavin Purcell: Plus Mirage AI lets you do real time AI

Kevin Pereira: filters on your meeting. And a new open source image model for boring reality. Thankfully, Kevin, we don't need that. Our reality is very boring. How dare you, Gavin? Hey, optimist.

Three [00:01:00] activate, uh, office waddle mode.

Gavin Purcell: That's exciting. Sure. This is AI for humans.

All right. It is another big week in ai, and this week, Kevin, we have some real interesting news about the fact that AI is impacting the job force, the workforce, how people work. In general. Salesforce, which is a very large tech company that has employed many, many people around the world, just announced that they have cut 4,000 jobs due to AI automation.

Mark Benioff, who is out there kind of touting his, uh, success in this way. It definitely feels like a harbinger for, uh, the world that we live in right now. What, what is your first thought on this?

Kevin Pereira: Um, you know. I barely noticed this story, Gavin, because when I was searching for Salesforce, all I saw was, uh, articles [00:02:00] about their massive data breach and hacking incident that happened due to a third party AI powered chatbot.

But what a great announcement to make to sweep some of that under the rug and say, look at all of the costs we're cutting. But, uh, should we hear him talking about this? Yeah, let's hear what he

Gavin Purcell: had to say exactly about it.

AI See What You Did: Also, I was able to rebalance my head count on my support. I've reduced it from 9,000 heads to about 5,000 because I need less heads.

But there's also an omnichannel supervisor now that's kind of helping those agents and those humans work together, and this is the most exciting thing that's happened.

Kevin Pereira: There

Gavin Purcell: we go. Yeah. So I need less heads. Yeah, let's talk. First of all, the idea that you talk about them as heads and not humans is kind of scary.

But we do need to talk about this, Kevin, 'cause this is a big deal. This is something that is going to be affecting the world at large. Uh, obviously when you think of these big businesses coming out and saying these sorts of stories, a lot of it is like, well, they're trying to help their stock price.

They're trying to do all this other stuff. [00:03:00] But as much as people out there in the world, the AI haters might, uh, argue AI automation is coming for a lot of roles, and in some cases this is going to be happening to you. So let's first jump into like. Do we feel like this is an ongoing thing? Are we gonna keep seeing this?

How seriously are you taking this right now? And then, you know, a little later maybe we'll talk about some things that you can do. But Kevin, I'm just kind of curious, in your brain, are you seeing this or feeling this in the world? I mean, through the stuff that you've done or at the places you've worked?

Kevin Pereira: Yeah, a hundred percent. I think there's, uh, uh, there's an expectation that. Um, obviously more can get done with less. There's an expectation that, uh, you're either ai, uh, first and AI focused or you're failing regardless of what that position is. And regardless of if the, the tools are actually there for demonstrable, like quality change.

But, uh, you know, I. I have personally implemented automation strategies, which have automated away portions of what would've been my job. And if I had [00:04:00] people under me to do that job, then yeah, they, they would've been automated away as well. So there's no denying, like there's, there's some that like to hand wave Gavin and say.

This is, uh, AI washing, right? Yeah. Like companies always do this. They always extract margin when they're going for stock buybacks and this, that, the other, and they're downsizing so that the CEO can get a bigger, bigger golden parachute and bigger bonuses. Well, this is the new scapegoat and they're.

Probably is a percentage of that going on? Sure,

Gavin Purcell: sure.

Kevin Pereira: You and I use these tools every day, and we are constantly astonished at the new capabilities that get unlocked with these tools every day. So this is the new reality. Jobs are going to be displaced until automation, or until I should say, until the industry as a whole catches up to where new jobs are being created.

Gavin Purcell: Yes, and that's, there was another big story that Oracle had a bunch of layoffs yesterday, like kind of surprise layoffs and people didn't know about it. And this also follows up on the St. Louis Fed actually had a story that came out that says job slowing, which is happening in the [00:05:00] economy at large, might in part be due to ai, which is kind of a big deal.

I think in general, people have to have an awareness. That this is the time to be, as we've said on the show many times, learn the AI stuff and be the person that put yourself out of business. 'cause if you can show that to your boss in some ways, then you're actually gonna be successful. Although, the other side of this, Kevin, is that there's a lot of other kind of things bubbling up in the automation space.

And one of those is robotics. And what I do wanna kind of pair this with in some ways is that Mark, uh, Benioff, the guy who just cut 4,000 jobs. Also got a very early sneak peek at the Optimist three robot. Let's watch this video, uh, that he shot with Elon. I think standing right there. 'cause you hear Elon in the back Yes.

About, uh, a productivity machine that is coming.

Mark Benioff: Hey, optimist, what are you doing there?

Gavin Purcell: Just

Mark Benioff: chilling. Ready to help.

Kevin Pereira: Okay. Just you have to pause real quick for the audio only listeners. This is a, a, like a, a golden esque [00:06:00] optimist three robot with mannequin arms. It appears. Yes. Full mannequin,

Gavin Purcell: arms at least mannequin hands for sure.

Kevin Pereira: Mannequin, hands and forearm. Yeah. Yes. From the elbow down.

This thing is giving, uh, mannequin. Yes. Probably because there might be a new hand for the new. Yeah, maybe they don't wanna show it. And it's like wrapping a car or something.

Gavin Purcell: Yeah. Yeah. A

Kevin Pereira: hundred percent. So, but it is an, the, the knees kind of facing away from the camera with the torso of the bot turned towards it.

It is gold. It's standing there. Mark asks, what are you doing? The bot announces just chilling. And,

Mark Benioff: Hey, optimist, do you know where I can get a coke? Sorry, I don't.

Gavin Purcell: It's not that stops time. It's not our, it's not our recording it. I can take you to your kitchen if you want to check for a Coke there. Oh

Kevin Pereira: yeah, it's, sorry, I don't have access to realtime information.

Oh, oh. Uh, I can take you to the kitchen. Yeah, that'd be great. Go. Yes, let's do that. So, uh, I really wanna let this play out in real time. Let's go.[00:07:00]

Mark Benioff: Awesome. Let's head to the kitchen. Okay. Okay. Go. I think it's, I think we need to give a bit more room.

Kevin Pereira: Yeah. That's e supposedly Elon off to the side saying, we gotta give him some more room, because Optimist is being extra safe about the space that it navigates.

Mark Benioff: It'll be able to walk a lot faster too.

Kevin Pereira: And then Optimist begins the.

It's sort of a, if this were like, you know, 8:00 AM the next morning, uh, this is, it's the waddle of shame that he's doing through this office. Yes, yes. Uh, but, but optimist begins waddling away theoretically to bring him to a Coca-Cola in the kitchen.

Gavin Purcell: Yeah. So I think the interesting thing about this is like, it's a little bit of like, this is very cool and I don't wanna dismiss the idea that like.

Optimist. Robots have not made big advances than they have. But it's a really interesting dichotomy to see somebody talking about cutting 4,000 jobs at their company and then posting this as like the big new productivity thing. 'cause it shows you the differentiation between like what can be done [00:08:00] kind of in the white collar.

Coding, marketing kind of on paper world and what the real world kind of experience of these things look like. Was this demo this bad on purpose? Oh, that's interesting. Like you really, you're starting to think conspiracy wise. Like you think that they

Kevin Pereira: I'm not, I'm be, because listen, like there, there's been a lot of robot demos recently, right?

We've got the figure robot, which we'll get to that's doing laundry and washing dishes. We've seen full Olympic, uh, athletic level competitions. Yeah. We've all invented in China, outta China. And this is, again, I, I am optimistic for optimists. I actually want US robotics to succeed and work out regardless of who's behind it.

But this was so jarringly bad. Yeah. I gotta feel like, is this like the cyber truck broken window? Like, go ahead and release it. Like the, the, the audio. Was was awkward at first. The command seemed to fail. It then seemed to pick up. They then said, okay, let's go. Then he has to say, oh, let's, let's move back.

So it actually works. Like how do you get this released?

Gavin Purcell: I have a [00:09:00] theory. You wanna hear my theory? My theory is that CEOs are often, especially billionaire CEOs, are ambivalent to what their words and videos might actually do in the world. I think, ah, my gut is telling me is that Mark Benioff was in the Tesla offices, shot this video and like Elon was there probably as a bajillion, other things that going on.

He shot this video and it was fine and uploaded it, and now we all see it. So I just think that this is not a planned thing. I think it's not. Clearly it's not ready for primetime, but it is the brand new version of it. And to your point, maybe they're hiding aspects of it, but. I just think the important thing to realize here is that we are in a world where two things can be true at once.

That jobs can be cut due to ai and everybody in our audience should be listening to that and standing up and thinking, what can I do to protect myself from that experience if you want to or start your own thing. And at the same time, we might be far away from like a future that people envision, right?

Like one of the things I always think about this space is like there are the extremes, which are [00:10:00] like. On one side, you have like the techno optimist people that believe like within two years we're gonna be living in like an entirely different planet and everything's gonna be different. There's the other extreme which believes that like AI is like no good for anything and no good at all, and it's just never gonna come to anything.

You just have to remember to stay in the middle, right, and kind of be aware of where things are, because I think that's the land we're gonna live in in the next five years. GPT five isn't a giant leap. It's okay. We have lots of things to think about. Maybe Gemini three will be a giant leap, but we are in the, this kind of mucky middle of this transition, and this was just a really interesting way to see those two things.

Kevin Pereira: You mentioned something earlier. It was very quickly, but I, I think we should focus on it just if, if there are folks listening to this going like, but what I, that leaves me powerless, right? Yeah. I'm a drift in Yeah. A big industries, uh, sea and I'm gonna try not to be absorbed by the giant wave. There are some things you can do.

So the, the ticketable work. Is being eliminated or being automated, right? Repetitive tasks, [00:11:00] especially if they're text base. Yes. If you're doing support or basic ops, like that is, that is going away now. You can wait for someone else to automate that or what Gavin mentioned earlier. You can show that you can leverage these tools and be proactive and go learn how to do them, and don't be afraid to sort of automate your own job away, which sounds bizarre, but.

That would to me at any, at any, uh, rational company with, with, with reasonable oversight that would show, oh, this is an employee that needs to be valued. Yes. Yeah. Not like, thanks for automating yourself away. See you later. Yeah. They would then put you on other tasks. So it seems weird to like, sort of cannibalize your own position, but that is a very actionable thing you can do.

Try to move up the stack. Yeah. Become an orchestrator of these agents and these tools and that will open the door for you for the foreseeable future.

Gavin Purcell: There's actually a really interesting post that, uh, Tibor bla ho, if you're not familiar with him, he's an insider in the AI space. Talked about, um, where OpenAI has published what is a leadership guide on how to [00:12:00] champion AI at work.

Now this in part from OpenAI is being like, you can use open AI and do all this stuff, but the other part of this that's really interesting when you actually read this. Is it does talk about the idea of like how to use AI at work, how to kind of start getting corporations on board. And if you remember, we talked a little while ago, maybe it was last week or the week before, about the 95% of corporations that are not finding value out of using AI tools.

Well, part of that is making sure that they're used in the right way. So all this, the stuff we're talking about here is kind of a mission for you at home to go out and say. Okay, I can figure out how to do X, Y, and Z of what I do right now with ai. It might not be perfect yet, but again, to Kevin's point, if you're showing the initiative, if you are a job person, that will be a big deal to the manager, the manager's manager.

I just had a really interesting conversation with somebody the other day in the entertainment business. Who was like trying to put together a pretty big deal for something and they were looking to buy a kind of a legacy media, um, organization and one of the leg, one of the things that the legacy media organization kind of wanted to hear about was like, about AI stuff and like what they might plan to do with [00:13:00] the AI stuff.

And it's like. If you know that stuff, if you have directions on those things, if you can say to that person, uh, in your organization, like, well, what about, uh, X, Y, and Z? What about if you took this and this, you become a much more valuable person Because A, it shows that you have thought through this stuff and you're paying attention.

But also, Kevin, the important thing is, and we're gonna get to this later, is that like. Human creativity is still that like it's putting into practice what you know and how to put it forward. The biggest thing that I think you can't be is a wage slave, do your job, walk away sort of vibe Right now, like it's unfortunate because I think some people just wanna live that way, but.

You need to lean in at wherever you are so that you can kind of protect yourself. This sounds very dire, but it is like a time where all of this is happening, I feel like, very quickly.

Kevin Pereira: Right? Uh, there's, and there's also a, a large swath of folks, uh, certainly on the grind set, uh, X and LinkedIn side of things that would say.

There's never been a better time to be replaced because if you [00:14:00] have, uh, an ounce of desire, the tools are there for you to cannibalize the business that you're working for. Yeah, sure, sure. That a single person with a, you know, knowledge and passion of these tools can start the next billion dollar business.

So you could always go that route and try to shake a fist at the corporation and take them out. Or you could become a plumber.

Gavin Purcell: Yes, Kevin. You can become a plumber. This is one of those stories where like. You know, they had a big list of, of store of jobs where AI is going to like kind of take over. And one of those jobs specifically is that is not gonna take over is plumbers, skilled trades are going to be protected.

So a lot of really talented people are saying, maybe I'll go learn to be an electrician, or I'll be a plumber, or like HVAC technicians. So like there is a world where that's gonna be a very viable skill. And just to the optimist point, like. I think we're a long ways away from that optimist three coming into my bathroom and being able to undo the thing that I did to my toilet.

Um, and it probably a lot longer than, than you might expect, but.

Kevin Pereira: [00:15:00] Figure oh two apparently trained on an exclusive data set from my wife. True, true. When I watched it load the dishwasher autonomously and not at all optimized space on the rack. Uh, this is a video. We've had this discussion. I'm not speaking behind her back, but if you look at the dishwasher,

Gavin Purcell: it doesn't, there is a only be able to watch like about a third of the dishes that you could actually watch.

'cause in the video it's just basically kind of randomly putting dishes. My family always puts in giant fricking strainers and frying pans. No, do it by hand. One thing, hand taking, taking up the entire rack space. Yes,

Kevin Pereira: yes. I'll get Gavin and I'll, and this is, this is the biggest concern that I have in my life is that's how hashtag privileged I am.

Sometimes Gavin, I'll get. Spoon up. Other times spoon down, but knife up. Oh, in the cutlery tray? Sure. Who puts knife up? You gotta put, come on, figure O2. And this is explicitly about figure O2. I wanna be very clear. I am only mad about No what you do [00:16:00] see, like we've seen figure doing the, uh, the laundry, right?

Yeah. A couple different demos of that. This is the same end-to-end AI model. Kind of, I'll give it a c. Yeah. Uh, great. Loading the dishwasher, but it is handling delicate glass. It is manipulating plates. It is turning them around. It's placing them in the dishwasher. And again, this is the same model. So conceivably in the past when we've seen demos like this, it's very highly controlled environments.

Yes. And it's very, very limited on Rails code specifically designed to handle that thing. We don't know how on rails this is, but the promise seems to be that this is sort of their one model to rule it all. So maybe we'll get back next. Yeah. Supposedly,

Gavin Purcell: supposedly they downloaded this knowledge to the robot and it was able to do that.

So we don't know that for sure, but like they said in their tweet says, no new algorithms, no special case. Engineering just knew data. So this is the same robot that was folding towels and sorting packages. Yeah. So now we can do this. Um, it is a big deal. It is. Obviously this is the state of the art and I think, you know, if you look at the [00:17:00] optimist two, it's a little closer to this, but again, we are probably a few years away from the plumber version of this.

Anyway, um, just keep, keep going everybody. If you're out there, please make sure that you are trying these AI tools and do it. But Kevin, before we go on, we have to tell people the other thing we have to make sure they're doing, which is liking and subscribing the AI for Humans channel. Yeah. Because you, you, the people out there are the ones that support us.

You bring us to all these new people by liking and subscribing, more people get to see it. And also if you go to the audio and you leave a five star review, we always see it. Um, in fact, our website automatically tweets it. So that's something to know. If you leave a five star review, you will get an automatic tweet from us, uh, without us maybe even knowing sometimes.

But please do all that and then finally go to and then chat. We have talked about this on the show. We are very hard working on our new startup that has been. Uh, funded by a 16 Z speed run. In about probably three to four weeks, we will be launching, kind of come hell or high water, and we're very excited about the [00:18:00] stuff we're working on, but please go there to learn more and again, come to our discord.

If there's something interesting to you. We might be sharing a few things soon.

Kevin Pereira: And if you're on those YouTube comments, thank you for the algo. Yes, always appreciate, appreciate the little comment for the, the sweet, sweet algo juice, so thank you for that.

Gavin Purcell: That's right. Alright, Kevin, let's talk about nano banana again.

So last week, uh, nano banana took over the internet and we've been playing with it a ton. The thing I love about this, and it kind of happened with four oh image gen, is that. When something that's this big comes out, and when I say this big, like Google AI studios, like searches have gone way up for the first time in a long time, like Google cooked on this and their brand is very, very strong.

You see so many people doing it. So two things happen. One is you just get a lot more interesting creative people playing with it, and on the other hand, you get a lot more, you get a lot more like hype Beastie posts about it, which we'll talk about in a bit. But the creative stuff is really cool because every time somebody unlocks one of these things, I'm always like, oh, that's really fun.

I would love to try it. So [00:19:00] we wanna talk through a few more of these cool things we've seen people do, and these are all things you can go do for free right now. So which of these was the one that was most interesting to you when you saw

Kevin Pereira: it? I mean, actually the first example that we have of being able to basically mark up an image and use that to feed directions.

We've seen that with. With, uh, like VO three, their video model. So it makes sense that it works here, but the, uh, adherence and the coherence to the source material blows me away. So, for example, if you're looking on the screen here, the, the example that, uh, uh, Prashant pr,

Gavin Purcell: Prashant

Kevin Pereira: pr,

Gavin Purcell: Prashant Prashant, the, uh,

Kevin Pereira: the example that Prashant posted on X, and I dunno if it was their original material.

So. Sorry. Uh, but they have a photo of sort of three women in dresses. They put colored circles around them and then they added another image that was just of like a empty kind of living room, dining room sort of scenario with a couple, uh, chairs around a table. And they basically put empty colored circles in the chairs and told the model, Hey, go match.[00:20:00]

The, the person from the image with their colored circle. Yeah. To the colored circle over the chair and the model. Got it. It posed them appropriately. It seemed to add, you know, shadows and, uh, it occluded, uh, like the, the, the light coming in from the window Yeah. Where one of the models is sitting. So it's like the tiny little details are there.

It seemed to really get the, the nuances in like the, the dress, the earrings, the everything. It's just a really clean example. So if you have a complex image manipulation, you wanna do. You can do it.

Gavin Purcell: Yeah. This was something I hadn't seen, this kind of color circle or color idea where you pick a different color so that you represent where you want that person to be.

That was something I hadn't seen with DO three done before. Another really cool, very simple one is somebody suggested, this is Martin LeBlanc suggested that. In order to do better green screens, you can actually take your, if you want to green screen yourself into a background instead of just like asking it to put you somewhere, you can actually ask nano banana to first green screen you, and then put you somewhere.

And it actually is much better [00:21:00] at the lighting. And again, this is all different ways of thinking about. An AI video or photo model as a, a thing that can think through stuff more than just like a simple Photoshop, like you have to imagine, like it's trying to figure out what you want. A lot of this is just giving it better instructions overall.

Kevin Pereira: Yeah. The, the hack of, oh, make it a green screen and then change it. That hack. Seems to produce much better lighting in the new scene that it drops you into. So that's a, a nice one. Uh, a shout out to, hello Rob. I saw their post on x where they built a comfy ui, tryon workflow, Gavin, that is like, that's cool.

I'm look at this. Yeah. Super amazing. So, so if, again, for the audio only folks, they took photos from different perspectives of a model and then took the, the, the flat lays ba basically when you lay out wardrobe, that would be in a look. So the. The shirt next to the shorts, next to the glasses, next to the whatever.

They feed it all into comfy UI at once and press a button and you get an instant multiple angle. Super accurate. Tryon. So cool.

Kevin Pereiera: Yeah. Yeah.

Kevin Pereira: Without a lot [00:22:00] of fine tuning. Even Rob kind of shouts that out in his thing. He's like, this could get a lot better. This was the first pass. But when you look at the examples.

They look great. And so I start to think of the apps that might get made for this

Gavin Purcell: or the apps that were already made. It might get completely, uh, cannibalized based on people vibe, coding apps, which we'll get to in a second. A couple other things here, uh, FO fr ai, who we always love, has done some stuff with progressive sketching, which is a very cool thing if you ever played, if you ever grew up drawing and you might have in those books were like.

How to draw an owl or how to draw something. And you see like the circles and then each step. Yeah, you can now do that with pretty much any image with the nano banana, which is a very cool thing. It does a very good job of stepping you through four different looks. Um, just a very cool way to use this that I was surprised by.

Kevin Pereira: I'm for progressive prompting actually. I think Sketchings going away and I think we'll start like. You know, your first thing will be an apple. Then you'll learn to describe the apple's colors. Sure. Then the perspective of the lens would ever, that's the children's world.

Gavin Purcell: That's what they're gonna live in.

Or our grandchildren's world. Yeah. Why would

Kevin Pereira: we ever lift our hand [00:23:00] to use primitive tools to make the image happen on paper? Get out of my way. Get

Gavin Purcell: outta your way. The other thing, tech Hol, who if you're not following you should be, he always is doing very interesting demos. He showed step by step how he used nano banana, I think through free pick, which might be one of his sponsors.

So just keep that in mind. How he basically took himself. Put him into a video game and all the different cool stuff he made. He made it into a video. The video is very good. I really suggest if you're listening just to the audio, go watch it. It's just a cool way of showing that like, oh, when now that you have character consistency and style consistency, you can really start telling a story image by image.

And then once you have those images, you can make the videos of them because they're very good.

Kevin Pereira: Yeah,

Gavin Purcell: and

Kevin Pereira: I guess friend of the show, uh, I won't say fan of the show, that might be going too far, but, uh, Fabian Stelzer, uh, who runs Glyph, uh, shout out to Glyph. We've, we've covered them a few times on the show.

They always have interesting extensions and workflows. Yes. And fun experiments. Fabian has been [00:24:00] going nuts with Nano. He really has, and making really, really amazing art, like, uh, endless video generation and this hair. Try on thing. I saw a lot of people doing this type of workflow. Yeah, I mean,

Gavin Purcell: this went viral.

Like it's one of those things that he made, it went viral. And again, if you're just listening, please watch on YouTube too. But like you see him, he's got kind of normally kind of salt and pepper, kinda short hair, but he, you see a, a curls come down off his head and then you see a bunch of different hairstyles.

And again, this is video that's been based off of nano banana pictures. And then you kind of stitch it together using video tools. But again, please go try nano banana. It is free to use. It is really one of the best image tools we've seen. It's not as good at as four oh image gen at some things, but overall it feels like we are really stepping into a new creative space with it.

So go try it.

Kevin Pereira: And if you go to gemini.google.com, that is the site hashtag not an ad, but please Google. We are listening. I also think Gavin Google might be listening because we said last week, nano Banana was a much stronger name Yes. Than [00:25:00] hiding the model behind the Gemini 2.5. Yes. Instant flash, brand new use.

This one. No seriously. V two model that they called it. They added a banana emoji. Yes, to the image generation button. So you're welcome, Google. And we can be bought, I meant hired for consult. I

Gavin Purcell: will say Google did reach out to us for an ad and we were just too busy that we weren't able to actually do an ad with them.

So there is that side of Google as well too. So it's our fault. In part. It's our fault in part. So we wanna admit what happened here. We'll get to that later. All right, next up we are, uh, speaking of Google, uh, Logan Kilpatrick, who is the Google, kind of the face of Google at large, has teased a new thing in the AI studio.

And Kevin, this is a vibe coding suite under AI studio.google.com/apps. If you go there right now, sometimes when you go there, by the way, I have this experience where like, it's a, it doesn't let you through, but once you get into it, it is a very cool front end for vibe coding. Again, this is another one of those things where we've, we've got people we [00:26:00] know and, and things that like vibe coding startups that have started, but like this is a very big deal when Google itself is basically building a vibe coding platform on their studio because I think it just shows you, okay, rep let's been successful.

All these apps have been successful. Guess what? This is something Google could just eat. They could completely eat it if, if Gemini three's coding model is good, like this is a real thing you have to watch out for.

Kevin Pereira: Well, I mean this uses 2.5 now. They even have like a template cookbook one that you can use, which uses the Gemini 2.5 flash image model.

Nano banana. Yes. Which we were just talking about. So if you have an idea of an app you wanna build with it, like native support right there. I actually vibed an app this morning like with one shot, one prompt. Oh, you did? And it used the microphone, it automatically wired in, uh, a Google LLM for speech to text and for reasoning.

It had sound playback associated with some of the stuff that was happening. That's amazing. And it was one of those things where like, I. I actually don't wanna show it off because I'm giving, I'm [00:27:00] sharing it with a friend. It was like their idea, they had an idea for an app and I was like, I wonder, wonder if I could just kind of want code it.

It was basically a one shot, and that's amazing. If you don't know how to code, this is like the one for you to start with because yes, you don't have to touch the code. You can use natural language. It uses Gemini 2.5, which is a. Totally capable coding model. It will orchestrate the app to run in your browser and you can keep chatting with it and it will update, you know, basically as each turn is done and it was sourcing files, it was placing them in the right spots.

It was just, just working in the browser. Yeah, in a really exciting way. So. I mean, this is the Google continues to just cook.

Gavin Purcell: Yeah. I mean I, I remember when, I think it was when 2.5 first came out, like I don't, if you remember, I mean that little bear jumping game all within the browser. Yes. But now the idea that you can do that in a space is great.

There's another really interesting cool AI tool that just came out, Kevin, that I played with this morning. You played with it as well. It's called. Mirage ai. Now, Mirage AI has a lot of things they're doing. They're working mostly I think in [00:28:00] open source video stuff, but they have come out with a real time webcam transformation, and you can go try this.

There might be a little bit of a wait, but basically the idea is that you can have a conversation with somebody in a webcam and then prompt in what you want it to look like now. It's not gonna look like VO three, but the fact that it's real time and you can switch it is pretty incredible. Shout out to, uh, Dan Shippers every show for, for bringing this to our attention.

In his interview. He did, but Kevin, you played with this morning and you got some really interesting, uh, results out of it.

Kevin Pereira: Yeah, I mean, look, you can go to their site and try to run the demo. It's free. You might be stuck in a queue and you might be forced to wait a few seconds, but, um, once it fires up your webcam, it will have a.

A carousel on the side that is automatically cycling through some of their pre-generated worlds, uh, or styles, if you will. So I went from looking like a Lego character to a wizard, to a Minecraft type person. Um, and even, well, by one point of these,

Gavin Purcell: you look like me. What? How did this happen? You look like an old version of me in this.

If you go to, [00:29:00] like I

Kevin Pereira: put, I put, oh, is that, why

Gavin Purcell: did you put my picture in it? I was like, what the hell happened? Whatcha laughing about?

Kevin Pereira: I, I was

Gavin Purcell: gonna say like I just wrote tired. I just wrote, oh my God. This is the weirdest thing and I wonder if in some form it knows that you and I do a show together and that there was a, hold on, I'm gonna text this to you right now.

I assume. Is that

Kevin Pereira: where I typed in Hellscape and the whole room turned red and I got No, it was under

Gavin Purcell: Yarn world, but I was just skimming through this video you sent me and like that looks like a picture. Oh, yarn world picture of me with white, with yellow, with white hair. Doesn't it? Like look at this like, okay, hold on.

Imagine me with yellow hair or white hair.

Kevin Pereira: I think you're being dismissive of how beautiful you are. Gavin. Um. Okay. Yeah, I guess we are kind of generic looking. I guess that's what we're learning. I think

Gavin Purcell: that looks enough like me, that there might be some sort of weird training data stuff. Like what is your training data?

Yeah, no, it's, let's keep, listen, keep going.

Kevin Pereira: First of all, very cool. Like, you know, people are like, well, super cool, but there's Snapchat filters and you can do this in a a, a Google Hangout or Google Meet, right? You [00:30:00] can augment yourself with AI or AR and yeah. To some extent yes. That that is true. This is a different approach.

This is a, uh, diffusion model. Um, I believe that is changing the entire image, you know, pixel for pixel, not trying to track and add on features. But, um, they say that the newer version of this model has a 14 millisecond latency that is gonna be very close to real time. Yeah. And as we talk about. You know this, this is a model that is probably pretty large and pretty compute intensive, which is why you get stuck in a queue waiting for it, but extrapolate this out into the near future where.

You've got lenses on your face. Yes, yes. That can modify the imagery that is coming in in real time. And suddenly you clap twice and the world looks the way you want it to look as you're moving about. It's amazing. I mean, it's such state. Looks like a Coca-Cola commercial.

Gavin Purcell: Yeah. I mean, this is the whole promise of the AR world, right?

When you talk about the AR world that's coming, that at some point this will be like, you would look around and see this sort of thing. And if you wanna live in the pink bunny world, you can live in the pink bunny world and [00:31:00] I can live in. Purple Hippo world and like we will have conversations, but to me you'll be a purple hippo and I will be a pink bunny to you.

Like that's coming for sure.

Kevin Pereira: Yeah, that's right. The, the Waymo that slams into either of us will either look like a bunny or a hippo and isn't that a great reality to have that, a great crossing the neon streets

Gavin Purcell: and then we'll go to Waymo will go to a bunny or hippo heaven after that. So anyway, very cool thing.

Go try it. Okay. Kev, next up Higgs field draw to edit. This is just another cool, um, tool that Higgs Field has released and it has to do very much with nano banana and, but there's something very specific I wanna talk about with this. So just again, Higgs Field. Very cool. Uh, company, they do a lot of really interesting stuff with, um, image models.

They are leaning in hard and nano banana. As our free pick. All these other companies, we have no deals with any of these companies. Just to be clear, the Kevin, the thing I wanna talk about is I saw the term draw to edit trending yesterday on Twitter, and I was like, wow, that's pretty crazy. That draw to edit, which is a weird phrase to see trending on Twitter.[00:32:00]

And I clicked on it. And Kevin, I gotta say, when I clicked on it, and I, I will say, I've said this a little bit about Higgs Field. Before it looked like every single one of those posts when I clicked on it, was written either by an AI bot or it was paid for as a a tool. Now I'm not trying to go out, I don't know the answer to this or not.

I will say Higgs Field is very good about getting their new things out there in the world. And this is part draw to edit was in part a kind of, I think a contest to get people to like win a, a subscription for a year to their service. Mm-hmm. But it made me very much think, 'cause I, I went through some of the very specific, um, uh, accounts and some of them are big influencers and we know, we've talked about certain influencers that like.

Go over the top and say like, the Higgs field model is the end of the end of everything. And it's like, that exists. But I started to go through some of the other ones and I was like, wow, this has 8,000 followers, this has 4,000 followers. And when I would look down their things, and again, I wanna be clear, I'm not accusing anything Higgs field of [00:33:00] anything, but.

It felt like maybe this is that. Are you

Kevin Pereira: sure we're not accusing anybody?

Gavin Purcell: No, we're not. I'm not accusing, I'm just saying like, this is, I, I'm getting into this conversation around, yeah, it feels like the world of bots is growing significantly when you see something like this, and again, I don't know for sure who planned to do this, it may or may not have been them or, or maybe these aren't bots.

Maybe these are all real, um, accounts, but. It was the first time I, I really saw it happen at a moment. Like it was, it felt coordinated enough to be trending on Twitter. And then you posted this thing in the, in the rundown, which I thought was interesting because Sam Altman just yesterday started talking about the idea of these, uh, more, many more bots on Twitter at large.

Kevin Pereira: Yes, Sam Waltman said that the dead internet theory, I mean, it's bad out there. Basically he, he says, I never took the dead internet theory that seriously, but it seems like there are really a lot of LLM run Twitter accounts now. So yeah, he takes the specific shot at X or [00:34:00] Twitter. Um, I would say of course, and I think that applies to Reddit.

Also, I think it applies to meta products also. I think it applies to pretty much everything now, right? Like we've been saying this for a long while, like I assume by default that most of the things that I'm seeing now are generated by AI or some sort of click farm or server farm somewhere in the cloud.

Gavin Purcell: Well, and which is wild. The interesting, yeah. And the interesting thing about this one, and again, I'm not accusing hiss seal of anything. I wanna make that clear for the fourth time. But the interesting thing about this is like, it's a little bit of like watching people, the, the, this is not the right word, but it's a little bit of like weaponizing a thing that exists and then when it starts to be like over, when you start to see these kind of things and it fills your feet up.

You, I, I have to say like, I pushed away from that tool because I was like, oh, this is too crazy. It feels like there's not an organic, there's not organic stuff coming across about this thing. And again, maybe I just didn't see the organic stuff. There probably were organic posts [00:35:00] about it, but. It does start to make you feel bad about the thing, first of all, about social media, but then at large, but, but also it just doesn't work for me anymore.

And I think at some point people think it works in part because like they get a bunch of numbers going up, but like Right. I think we're interesting. We're entering into a transition period where it won't work for much longer.

Kevin Pereira: Do you think we're gonna hit a point where like a proof of person is a Yeah, a little bit.

A proof of human is gonna be a thing. Because it's funny, like Altman was trying to do that, uh, ages ago with, with with the eyeball thing. Scanning or, exactly, yeah. Tokens and all sorts of stuff. But I, I, I do wonder if that will be the new thing, like having. You know, 500 million followers is way less interesting than having 50 verified human followers.

Yeah,

Gavin Purcell: a hundred. In fact, I would like to see a world, and again, you know, this same thing around x might be just a, a shot across the bow against Elon, but like I will say, shout out to Nikita Byer, who was hired as, as X's head of product, who I believe is trying to do the work around eliminating bots that, who knows for [00:36:00] sure, but like.

I would love to see a world where there was like a verified follower's number, right? Like where you actually got something like that. Anyway, it's a really interesting thing to follow along on. If you've seen a bunch of this yourself, let us know in the comments it's some of are worth, uh, while, and if you are a bot, leave a comment as well, bleep.

We definitely have bots. We de there are definitely bots that have left comments in our, in our YouTube and I reply to them all the time. All right, Kevin, we're gonna jump into this, a very cool story in the open source image world. This is the boring Laura, uh, from Quinn. It is, uh, created by Kza. And what, just to be clear, what a Laura is.

Is a collection of images that you are training a model on so that it will produce images that kind of look like that image, but have the same sort of style. And what's great about this, Kev, is we've said this before, sometimes to make AI images or videos look realistic, you have to make them look worse because sometimes the AI wants to come out with stuff that feels like very fancy.

Yes. This is by far the best looking realism model I've seen. There's a couple great pictures of just a guy with his anime cutout. [00:37:00] There's a amazing shot of a dentist working on a lion's mouth. I just think it's a very cool thing to go play with yourself, and there's a ton of really interesting examples.

And again, you can go on hugging face and try this yourself. Right now.

Kevin Pereira: The baby being baptized with the Red Bull. Uh, is Heartbreakingly good or a pug at the piano? Yes. It's like, you know, imagine the old if, if you're an old head like us, where you'd have to wind the disposable or instant camera to snap the photo.

It gives you that harsh flash look, um, you know, not the greatest quality lens. Uh, but it does make stuff. The lighting is, is sort of flat in ways that when you see a lobster operating a forklift. Uh, as a photo from 2012, you go, oh yeah, that looks kind of real.

Gavin Purcell: I mean, the best, this is again, we're shouting out, Rob.

So, hello Rob. You get two shout outs on this, but did you see the one where he, there's a, there's a monkey smoking a cigarette with a holding onto a monster drink. Like, this is what I want AI to do, right? So like, and you know, like, this is so fun that you can do this stuff. And again, it might be that [00:38:00] open source models, this is what's gonna be use, they're useful for in a lot of ways.

Like, you'll be able to do stuff that the larger models may or may not want you to do, or. You can't, because it's such a big tool, you wouldn't be able to get these results out of it. But it's super fun. Go play with it. And again, we'll put a link into the show notes. You can go to hugging Face and generate your own right now.

All right, Kevin, it's time to see what you did around the internet this week with ai. It's time for ai. See what you did there times

Kevin Pereiera: ya. Without a care,

AI See What You Did: then suddenly you stop and shout.

Kevin Pereira: Alright, Gavin, uh, this thing dropped late last night. I haven't had a chance to install it myself, but I saw it and sent it to you immediately and was just like, uh, how cool. Yeah. Cool. Damien Mason, uh, posted visual story writing and it's, uh, basically a word processor that while you are writing your tale and introducing characters and situations and [00:39:00] actions between those characters, you get a visual mind map, basically.

So cool. Yeah. Off to the side where you can in real time see the relationships between characters. You get a, a graphical sort of, uh, a timeline of your story that you can scroll and see how those relationships evolve and change. And it just made me think about how like. Wow, this is an interesting new interface.

I wanna see popular stories loaded into it, so I could see the relationships there, but also what does it look like to create with like a workflow diagram for what would be a traditional text narrative to say, at this point, I want these two characters to break apart. Yeah. And here's what I want to have happen between them.

And then watch the way it visually affects everything else. It's super cool. Uh, I believe it's open source. There's some GitHub code that you can go and run, but there's also a demo page. But just like an interesting thing that made me say, Hey, I, I see what you did there.

Gavin Purcell: You did it. You did it. Okay. Yeah.

What else you got? You got, you put a couple of these in here. I want, I wanna see what you got from me this time. Let bring, bring me.

Kevin Pereira: Well, this one's super nerdy, but it just dropped like two [00:40:00] hours ago and, uh, Google announced embedding Gemma. So don't wait. Are we now doing,

Gavin Purcell: we're going. These are not, these are, we're now going back to, to main company stories out of, uh, A ICH.

Did there, uh, did, did I, did I soil the sanctity of a IC What you did there? You did, because you did. But it's okay. We can put the breaking news lab for graphic app breaking news just like last week. Breaking news. Breaking news.

Kevin Pereira: Okay. For those of you still remaining, just know that Google has, uh, dropped something that's going to make embeddings on your mobile phone way more efficient and, uh, and interesting. And that is cool. What does that mean? Just so I

Gavin Purcell: know. What is, what is embeddings on your mobile phone mean? Does I

Kevin Pereira: understand?

Yeah. So when you've got emails or text messages or big PDFs or files or whatever, even your images, right? Sure. Those need to be stored as embeddings, uh, uh, data points, relationships between the data so that when an LLM needs to look it up, it's super fast and it's efficient. And so for a small on-device model running on your cell phone, a tablet, your smart glasses, in the future, it [00:41:00] needs to be able to embed things.

Accurately. Um, so this is a, a very gross overview of something that should not be in a, I see what you did there and I'm sorry.

Gavin Purcell: Hey, I see what you did there. Google, uh, third largest company in the world. Nice work. What's up next? What's next up Kevin? We got somebody who did something that small and interesting.

They did something cool and I like it. I think it's

Kevin Pereiera: interesting. And I'm sorry. And the next thing is from 11 Labs. So now I don't wanna cover it. We have

Gavin Purcell: another, do we have another corporation as the third link today in a I see what you did there.

Kevin Pereiera: I thought we could play with it together. 'cause there really something new and I thought it'd.

Fun. I wanted to say something nice. Okay, do that. Let's do that. Nice. Now I feel that, so this is an update

Gavin Purcell: to 11 labs, uh, sound effects tool, which is very cool. Yeah, we've been using it in different ways. We, Kevin made something for our startup, uh, yesterday or the day before that I was like super excited by.

And you mentioned part of the way you did the soundtrack was with this, right? So tell us a little bit about what you did.

Kevin Pereira: So, I mean the 11 Labs music application is incredible, but this is an update to their sound effects application, [00:42:00] which allowed for like now what you can do, supposedly ultra high fidelity prompts, you can turn on looping, which is a big deal.

Yeah. If you super. Lo-fi chills, something that's gonna loop in the background. Now you can do it or like a even a for a white noise app or whatever. You got it. I just put in a prompt. Gavin, I have not tested any of these. Oh, okay. Let's hear it. I selected five seconds and I wanted it looping and I can see that we could download.

Yeah, you can download as a 48 kilohertz wave file. That is a very high quality sample. Sure. But let's see if you can identify what this is, Gavin.

Gavin Purcell: That's, what is that? Well, it's a dog. Is it Wesley? Wait, oh God. What are we doing here? There's a dog and there's a fart. And there are different sounds like are they connected in some way?

Kevin Pereira: Yeah, Gavin, the barking dog. Lets out a massive far now that was a five second looping sample at only 49% prompt influence. How incredible [00:43:00] is this Gavin? Do you wanna give about the deforestation occurring so that we could generate. This whimsy. I appreciate everybody

Gavin Purcell: that stuck around for the end of the show for this exact moment.

Don't appreciate it. What would you use this for in some, I'll give you another one. Real world use case. Yeah,

Kevin Pereira: that's a pretty good car horn. Sure. It was supposed to be a honking car that lets out a massive fart that didn't do it. Gavin, if you're an indie game designer, sure. If you're working on a vibe coated app and you need some sound effects for it, from UI clicks to, you know, transition noises or shouts from a character, maybe you need the sound of a sword clanging against armor.

Yeah, you can generate that sort of stuff now with 11 Labs. Let's try one more and see if you can guess what this is. Sure. Okay.

Gavin Purcell: Wow, that's bad.

Okay. That's it everybody. We'll see you all next week. Thank you for joining us on. Hey, I free it's bye. Leave a comment. [00:44:00] Leave a comment in the messages. Bye-bye. Should we just try a looping fart? Let's just try a looping fart. Continu a continually on fart should be a

Kevin Pereira: particularly juicy part. Gavin God.

Kevin Pereiera: What are we doing?

Kevin Pereira: If this works and you made it here, let us know in the alga that you made it to the juicy loop. We'll just call it a juicy loop. Now. That now let's

Kevin Pereiera: Oh,

Kevin Pereira: it's

Kevin Pereiera: loop.

Gavin Purcell: It looped. It looped. All right, everybody. We'll see y'all next week. Gavin, we're sorry. We're sorry. Please, I hope, I hope that your

Kevin Pereiera: family's not

Gavin Purcell: listening.

I am embarrassed to be part of this right now, and we'll see you all next week. You can't even, we'll, return loop. We'll do. Bye-bye Byebye. It's a prank. Oh God. Byebye. Bye-bye. Bye-bye. Perfect. We're out now. It's over. Whoa.