← AI For Humans: Weekly AI News, Tools & Trends
AI For Humans: Weekly AI News, Tools & Trends
Google AI Studio Got a Big Upgrade and We're All Vibecoders Now
00:35:57
Loading summary
Transcript97 lines
- [00:00]
A
Big AI news. Google just dropped a whole stack of tools they are integrating into their AI
- [00:05]
B
Studio platform with a new agent runtime
- [00:08]
A
and native Firebase support to level up your vive coding workflow.
- [00:11]
B
That's right. We're talking databases, authentication, multiplayer gaming, persistent sessions. All your little Vibe coded Vibe babies are growing up so fast.
- [00:20]
A
We'll break down what all that actually means and whether you'll be switching over from cloud code or Codex to level up your Vibe coding workflow. Kevin. I think they want to level up our Vibe coding workflow. Gross. Plus, the five other big stories you need to know about AI today in five minutes.
- [00:35]
B
Also, we're dropping an open source project that we spent precious minutes on. Tens of thousands of tokens. You can add anything you want to an absolutely classic game. We're talking about Generative Jump.
- [00:47]
A
Anything. Kevin.
- [00:48]
B
Oh, I. I already guy fiettied it. Gavin. He's in there.
- [00:52]
A
Whoop, whoop. This is AI for humans, everybody.
- [00:54]
B
That's a Juggalo thing, is it? Woop, woop, woop, woop is Juggalos.
- [00:58]
A
Oh, wow. What do you know? We're.
- [00:59]
B
Now we can put insane clowns in there.
- [01:01]
A
Ooh. This is AI for humans, everybody. Welcome, everybody, to AI for Humans, your twice a week guide now to the AI world and AI news. We learned last week via our comments that bi weekly does not mean twice a week, it means twice a month. So we are here, Kevin. We have some big news today coming from Google. They have updated their AI studio platform. And I just think this is one of those things where like, Google has been sitting back a little bit. Remember back November when Gemini 3 dropped and we were all just like Google's one? They're doing it. And now since then, we have seen cloud code and Cloud Opus 4.6. We have seen GPT 5.4 take over. This is Google's kind of entry point into what we might want to now call the Vibe coding wars. What is your first take on what this suite looks like?
- [01:52]
B
Look, I'm always excited for new tools that can help people bring their imagination to life. Like, full stop. Like, love that they're doing this. I did sort of feel like, was that DJ Khaled? Like another one? Like, didn't we just get Anti Gravity? Which I know was more of a stab at the, like, traditional ide, the developer interface, but. And wasn't there also a web exploration of this too, where you could run like miniature games and snippets and then share them and whatever. So I, I hope they find what they're looking for because I do love what, what they, what they crank out. But like, okay, let's take a step back and talk about what Google AI Studio is and why we're excited for it, even if it might be another confusing entrant into this whole vibe coding arena.
- [02:37]
A
Yeah, so AI Studio is Google's kind of like, suite of tools that allows you to kind of create within their ecosystem. Meaning that, like, you can create coding projects, you can do a bunch of other stuff. But specifically today's update is really about kind of upping the game of vibe coding. And Kevin, I do think there's a world where vibe coding at large, we need a new term for pretty soon because, like, I feel like everybody's vibe coding. So maybe let's talk about that in a second. But first and foremost, here are the details of what you're actually getting in this update. There are new multiplayer experiences, which is a very cool idea that you and other people can work together on the same project and also kind of create multiplayer games, games or apps within it, which is a really cool thing.
- [03:20]
B
Very smart to focus on this. Most things are better with friends. I think you would agree. And not only just the, the fact that you can build projects together, which is like central to any, any repo worth its salt. The notion that you can make multiplayer games easily, like out of the box, gets people to creating virality so much faster.
- [03:41]
A
Yes. Right.
- [03:42]
B
Breathe your, breathe your idea into existence and then don't fight with cloud code or Codex over what the best tech stack is. Just say you want a shared experience for other people and it knows what to build. That's right.
- [03:52]
A
Another big update they have is they're adding database and authentication. Now this is something you might be saying if you've been using things like cloud code or Codex, they kind of know already. But this is a big deal for AI Studio because integrated into this, it will tell you, do you need a database for this project? And it will help you set it up. Specifically Firebase, which is Google's database software program. We, I think both you and I are using Supabase a lot for different projects that we use. And Supabase is a. Is a separate company that kind of came up a couple years ago and is very big in the world of, like, people who are developing tools. Still a very cool thing in general.
- [04:25]
B
What's big about this is not that it understands databases. What's big about this is that you don't have to understand databases. That's the promise. Like if you have an idea for something and Claude code or Codex by any other name says what kind of database do you want? And. And then your eyes glaze over and you roll back and then the foam starts drooling out of the sides of her mouth like, you're not alone. I've been there.
- [04:45]
A
Yeah. Here.
- [04:46]
B
It will suggest it for you, set it up for you, help you with the credentials. And then when you're ready to connect things to like real world live services, it can also theoretically hold your hand and bring you there as well. We haven't tested it, but that's the promise.
- [05:00]
A
Well, that's something I want to talk about because that's the next big step. Here is about the idea of like being in Google's ecosystem. The idea is that Google's ecosystem is vast and very large. Much larger than something like Claude or even much larger than OpenAI. These are new services. Google has a lot of stuff. And the idea here is that Google has its own API. Nano Banana has its own API. Like all these things that can work together are a big thing and if you can kind of put those all together in one location, that's a big deal. A couple other quick things before we get into talking more about. I did get a little bit of time to play with this. You are able to pick up where you left off, which is like a device thing, meaning that like you can oauth, which means you can authorize into this from different places. One of the things that happened with cloud this week is they integrated dispatch, which is a from your phone able to integrate with cloud code and cloud cowork. This makes that very easily. Oh, and then, and then finally like building with Next JS is something that they're kind of like highlighting. This to me feels like a real catch up moment for Google because like if you don't know what Next JS is, it's kind of the framework for a lot of the modern web. It was created by Vercel, I think originally and we talked a little bit about that a few weeks ago. But this is like a lot of the modern web is built on Next JS and the software and this is allowing you to access that and be able to build with it.
- [06:13]
B
Yeah, I mean look, a whole suite of upgrades. I am very much excited to play with them. The, the lingering concern that I have with a lot of Google products recently is what is the actual roadmap? What is the longevity for this? Are they going to fully support this? Are they waiting to see if this takes off and becomes like their Vibe, Cody, Open claw movement. Like, I don't know what their intentions are. I'm excited to play with it and excited to come up with a multiplayer something that you and I can try to Vibe and then maybe share with the audience and let them come in and play. That multiplayer component specifically is the one that has me most excited.
- [06:51]
A
Yeah, let's talk a little bit more about that because I think that's something really interesting to think about when it comes to what Vibe coding is and what it means to be a Vibe coder. Now, we can also have that conversation around whether or not we should move off of Vibe coder, But, like, one of the things. Yes, we should. I think we can come up with our own version of that as well. If you in the audience have some ideas about different names, please let us know in the comments. The big thing I keep thinking about this is everybody is a developer now, right? This idea that instead of it just being a thing where developers were like, these kind of like, magical people who grew up through math, grew up through science, and understood how to write code, everyone is a developer now, right? So the idea here is that, like, if you have an idea out there, the tools exist for you to kind of like, magically bring it into an existence. Now, there are going to be people who are better at it than others. There are better people who are going to be better at understanding how to plan and program this stuff. But this does just feel like kind of the next step into that world. And that world is exciting. It's interesting. It's the idea of what a creator is, moves from being just like a YouTuber or a TikToker into a product person. And, Kev, I'm just curious to know your thoughts on what that world looks like. What do you feel like that world looks like? I mean, it must be in some ways exciting for somebody like you who grew up tinkering on things at a very early age. You know, if you go way back to the Captain Emmy days, but how do you feel this is going to go over kind of in the world at large?
- [08:17]
B
It gets better every minute. The capabilities grow as well. I mean, it does. So I think, you know, as I listened to what you just said about all this, how everybody's a creator and anybody can code now, you know, the caveat is like, well, some apps, well, some experiences, some this, some that. I think developers have a massive leg up. They might not be writing each line of code, but they can audit it and know what it means. They know the best practices for writing these things and deploying these, for scaling them, for making them secure. So I still think there's plenty of gap left for the quote, quote unquote, traditional developer. That said, it is getting better every second. And so the notion of that you can whisper more complexity to these machines and out of the other end you get a product, I think that's amazing. I similarly am on some ways, like in the trenches with you and on the forefront of this, and other ways, I feel completely on the sidelines watching this all happen.
- [09:09]
A
Yes.
- [09:10]
B
And. And when a new idea springs into my mind, unfortunately, I'm also like, well, if I could crank that out in 72 hours, someone else can as well. So, well. But is the difficulty or the actual execution of it, is that what made the idea special? No, no, it was the idea I should go do that. And I'm constantly conflicted and I feel like you have something to say about that.
- [09:30]
A
So. No, I was going to say I think there's this new paralysis that exists around these tools that's really important for everybody to kind of push through. Right. So if you're out there and you're trying these things, I mean, you and I have both talked about this. I have two things I'm going to finish this weekend. I've said I'm going to finish these two ideas I have this weekend that I want to get through. Meanwhile, yesterday I sent you this link randomly in my brain. I was like, oh, you know what I should do is try to remake the old 4D boxing PC game into MMA. And we show a little bit of this here. And I spent like a bunch of time just mostly with Codex going back and forth saying, make this better, make this better. And it got there. But, like, I don't know why I did that. Right. So, like, that's not like a final product necessarily. So it's one of those things where, like, we're in this moment where they talk about when you go to the grocery store. I don't know what this is called. Somebody will know this. But, like, when you go to the grocery store and there's like 15 kinds of peanut butter, that you have a much harder time decide whether if there were three kinds of peanut butter, that's what I think is going in my brain. Yes. Paradox of choice. That's exactly what it is. So anyway, this is the moment we're in. The other big thing that Google announced that we have to talk a little bit about is Stitch an update to Stytch. Now, Stytch is their design program their design AI tool which allows you to take a very interesting kind of almost like comfy UI looking interface and throw websites at it to make the designs better. And Kev, you and I both know there's been issues with websites that come out of codecs that don't look amazing. Websites even that come out of cloud code, even with the front end skill that all look very similar. Claude, I don't want you to use a emojis and any more things that I do because every website, by the way, that's the tell right now to see if an AI designed a website is there's tons of emojis on it. But this is very cool. If you look at this, basically it allows you to kind of. It'll allow you to take a website in. I actually had it took. Take my Gavin Purcell.com website in and I said, hey, give me an idea of what you can do to update this website. Now. I kind of like right now the design of it. And I said, update it with fonts and all that stuff. One of the cool things here is if you're not a designer, it gives you this kind of interesting layout where you can see the whole website laid out, you vertically, you see all the pages through it. It also allows you to choose fonts and do different things. So one of the things you may or may not know out there is that Google actually has a ton of awesome free fonts. And fonts blow sometimes because you have to spend money for them. But Google's font selection is actually pretty good. It also gives you color pickers, like, in general, this is another version of that same sort of story instead around code, it's around design. And this might kind of like bring those stories together of like, vibe coding, Vibe designing, Vibe whatever. Like, we made like. Let's go to that. What was our word gonna be for whatever this thing is, you know?
- [12:03]
B
Yeah, I don't know. But we're. I mean, we're certainly no longer coding. At least this, you know, we're. I don't know that we're. We're. Yeah. I don't know. Where are we?
- [12:11]
A
You see this, by the way, I'm doing. I'm. I'm. I'm using both coffee and water. This is the. What the world of vibe coding. Whatever has.
- [12:17]
B
Oh, that's your. That's your new wellness routine.
- [12:19]
A
This is my new stack.
- [12:20]
B
This is my stack.
- [12:21]
A
Stack this.
- [12:21]
B
And he's trying to. Guys, he's. Yeah, what. What should we call, like, when I look at Stitch, one of the Things that I think was really, really fascinating about it is the voice, the duplex mode. Like what does that mean? That means you can use your voice to talk to it, which other development tools have that. But it speaks back to you and it becomes a sounding board so you can ask it to iterate on something. What do you think about this? Do you think someone who has an accessibility issue might have a problem with this? What can we do about the corners here? Or whatever. And it will answer you back. And it's like a preview mode. But I do think, you know, you and I are bullish on voice AI and that being a driver whenever I can talk to the machine, my productivity, the pace at which I can iterate, it's so much quicker, it's so much better. So to have it answer back is also really interesting. And I think this becomes kind of like the Squarespace killer in some ways. In terms of building a site like Squarespace was really first on the scene with these beautiful templates that you could easily customize here it's sort of like whisper whatever you want and we've got templates for that. I also wonder though, Gavin, who are websites for three to five years from now?
- [13:30]
A
Yes, and this is a really interesting conversation that maybe we'll get into in a later show. This feels like a longer conversation. But the idea that the agentic web, all of this stuff we've been talking about for the last like couple months on the show is about making stuff for agents. And just to be clear, AI studio, more vocabulary for you is another agentic harness around Gemini. Right? This idea that you are putting tools around Gemini to allow it to kind of pull stuff in and do stuff together. If all of our web based stuff, if I create a website and you create a website, but if we're never going to the web and you're just dealing with Mr. Tibbs, your little open claw assistant and I'm dealing with, yeah, go ahead, what's that?
- [14:10]
B
Yeah, that's fine, my little assistant.
- [14:15]
A
But I don't know, you're right. Why would it need a look of this website? And I think this is something you and I have been thinking a lot about lately is like if you were an agent and you. If I said to my agent and said hey, I want to go get a cheap shure microphone and I don't want to spend the time looking for it, the agents right now might have to go to like Craigslist or ebay and click pictures of all these things. But at some point there will be another Agent on the other end who's got somebody who said, I want to sell my shure microphone. What do those two agents look like when they connect? This is not what you might think of it as, but, like, no, they are docking.
- [14:48]
B
And that agent has a huge agentic trench coat.
- [14:51]
A
No.
- [14:52]
B
Opens it up with his glasses on. He's like, what are you looking for? Oh, you want me to open both and then you get your microphone. Here's the good news. We only needed one more website in this entire world, and it hath been created and very, very well. By the way, that is aiforhumans Show. That is the last website that will ever be breathed and breathed into existence. Wow.
- [15:18]
A
This is. We are gods. We have brought you this. And yes, it's the last.
- [15:22]
B
Sorry, guys.
- [15:23]
A
If you go to aiforhumans show, you can do all the things for our show, but most importantly, you can follow us on all the platforms, like, and share this video if you're watching it. Thank you so much for being here. We appreciate it. Throw us some money in our tip jar. You can buy me a coffee. You can also add money on Patreon. We are making a business out of this thing for real. We are now coming to you twice a week. It's not a joke anymore. We spent the last three years kind of thinking this was a joke. No more. This is real, Kevin. We're real.
- [15:49]
B
AI for Humans was a side project, Gavin, but now we're really focusing in on the main thing to fend off all of our competitors out there. Does that sound familiar?
- [15:59]
A
Ooh, that's really interesting. So, yes, this is a big story this week. In case you missed it, OpenAI, there's a big Wall Street Journal piece where OpenAI has basically said no more side projects. We have to focus in on business encoding. And, Kevin, this is because right now, Anthropic is eating their lunch.
- [16:17]
B
But what about my sexy chat robot? Weren't they going to go into Spicy Chat?
- [16:21]
A
Weren't we going to do that? There's a weird side note there that supposedly Spicy Chat is still coming. In fact, there was another story that came out where Spicy Chat is scaring people who work at OpenAI or Advisors of OpenAI, that they'll want to happen. But anyway, that's a side. That's side project for us. Oh, what is it? Let's get back. Oh, bleep that Will. We are not listening. We're not putting that in the show. Back to the story. The most important chart that you can see, Axio shared, which is The AI model share of first time enterprise customers and it literally looks like two, two ships crossing in the night, one going in the right direction, one going the wrong direction. This number goes from in early December, 60% for OpenAI, 40% for anthropic. Now these are first time enterprise customers. Now in late February it was 73% anthropic, 26% OpenAI. That is insane. So when you think about the fact that Anthropic is catching up to OpenAI in MMR, which is monthly recurring revenue, these companies are existentially at risk because of how much money they're spending and honestly how much money they're losing. This is probably the biggest business story that has happened to AI in years. And OpenAI, basically Fiji. Simo, who is the OpenAI CEO of products and something else, not the real CEO because Sam's the real CEO, has set out a memo internally that said we are focused entirely on the business consumer. And Kev, my big question to you is what do you think the kinds of things that they're saying, okay, we're schluffing off to the side are because some people think that might be sora. Do you think that's the case? Like do you think you're going to care less about AI Video?
- [17:58]
B
I do, I think so, yeah. I think, I think they're going to care less about AI video. I think their image models haven't been updated in a long while. I'm sure they're still brewing something, but we'll see. I have built products and am actively building products off their real time Voice product. They did a slight 1.5 upgrade to the Voice product recently. There was no pomp or circumstance, largely because it got slightly better. But it's been well over a year now that I'm waiting for fundamental changes to that product. And that is I think also a business product. So it's a little odd to me and it does make me a little wary. The same way I as a consumer am wary of adopting all these new Google things when they release on the, on the business side on my day job where I'm trying to integrate OpenAI into all of the things I now worry which products are actually going to be accelerated and focused on and which ones are on the chopping block. And I don't actively know that. They haven't really communicated what those things are.
- [18:53]
A
Yeah, you know, it's really interesting. Somebody the other day said that Google is set up to win the consumer AI world. And by the way, I hold on,
- [19:01]
B
I'M sorry, Gavin, I gotta get a better answer to this. I'm use the Atlas browser and ask it to.
- [19:05]
A
Oh wait, it doesn't happen. By the way, right now it's a
- [19:09]
B
nasty bottle of malt liquor.
- [19:10]
A
I will say I am using the cloud code integration into Chrome or the cloud integration into Chrome. Right now I don't see my computer back there, but it's set up in the background and it's cleaning up my inbox. So like I'm actively using Claude's programs to do this. Here's an interesting insight to this. I always use Nana Banana to do thumbnails for us. Now we take our own pictures, but I use it sometimes to kind of add backgrounds and things like that. Last week I was like, you know what, I'll try a couple other ones just to see. I tried Grok and I tried image gen, OpenAI's and nano banana is just so much further away in terms of editing, in terms of what you get out of it, that it might almost be that they've won. Right? And I think you see this a lot right now happening in the AI space. Companies kind of just going like, well we're not going to get there. Freaking Elon yesterday tweeted this Google will win the AI race in the West, China on Earth and SpaceX in space. Now Elon says stuff all the time, but like the fact that he's saying that Google will win the AI race in the west, that kind of feels like an admission that like maybe Xai is not doing as much as he thought it was. So like you're seeing companies pull away, this is that moment and like Anthropic is like ramping up right behind OpenAI and they're looking in their back window and they could go right past them super easily. And by the way, cloud code is still shipping like crazy. So dispatch as I mentioned before, came out which is the ability to use cloud code and cloud cowork from their phone. In fact it was originally just cloud cowork, Kevin and then they just added cloud code like a couple days ago. And skills, I have to tell you, skills are really important for now. Now they may not be the long run, but adding skills to Claude and also you can add skills to Codex is really important. And one skill that I wrote about in newsletter this week that I use all the time is called the superpower skill. There will be a link in our show notes. But what this basically does is when you're using cloud code or Codex is it asks you a bunch of Questions which we've talked about before about how to kind of open your door. But Superpowers does a really good job. It's one install, it's super simple anyway. This is like anthropic. Just feels like they are shipping and shipping stuff that is actually useful to me, whether it's in my work or it's in my coding. All sorts of stuff like that.
- [21:11]
B
Yeah, I listen anecdotally. I switched to Codex recently with a Pro subscription to try to feel that out and see what that's like. And I like the Codex app, but I desperately miss Claude code, I miss the command line interface and I will likely switch back next month unless there is a major development there. So I get why the lines are converging the way they are, but those lines, Gavin, are so wireframe Y2D gross and ugly. I wish we could run an AI beautification filter on them and make them run at 30 frames a second, just the way the artists intended.
- [21:44]
A
Yes, we're going to quickly follow up on our DLSS5 story from earlier this week. If you missed it, go back and watch the episode we released on Tuesday out of Nvidia's GTC. There was an announcement that DLSS5 is an AI affected tool to make AI video games visuals better. The most important thing here to reference, Kevin, is like, as we expected, there is massive blowback from the gamer community on this idea. Now I think we don't have to spend a crazy amount of time on this, but I do want to just shout out the Digital Foundry guys who went and kind of were like, hey, this is kind of cool. Got so much crap for this that they made a second video where they basically talked about the harassment that they got around this. And I just, this is just a moment to say, like, we have to understand that. Like, yes, there's a lot of big feelings about AI and we understand that. But if you were in the audience of this show or other shows, just know that like this stuff, they're not trying to hurt you personally. Right? Like, I know that gamers, we're both gamers, we grew up gamers. Yes. These are people who feel passionately about stuff and I understand that. But also, like, just because an AI name is involved does not mean that they are trying to take away your childhood joy. That's all I want to say about this. Like, I feel like that's what it comes down to this.
- [22:55]
B
It's like if someone would have released an emulator that had this built into it and could with a toggle enhance your old Wii games or the Nintendo 64 games. People would be losing their mind over it. And even if it wasn't true to the artist intent, they would say Banjo Kazooie looks so much better. Oh my God. Right? Also, same community that would, you know, that will say, like it, this is disrespectful to again, the artist's intent to the creative will, etc.
- [23:24]
A
Etc.
- [23:24]
B
And then they go and they put Macho Man's voice into Thomas the Tank Engine and replace the dragons in Skyrim with it. Which is awesome. Yes, by the way. So look, I understand why there's a contingent that is so passionate about the artist intent. And I, and I love that. I love that people are passionate about their hobbies and their art forms and their crafts, but I do think there's an unfair amount of. Whenever AI comes up and all of the companies that have been featured thus far gave permission, they said they wanted to be supported in their games. And here's the other thing that people aren't going to like when you start dls, dlssing things, which we're going to get into, because you can DLSS anything right now with a very cool Hugging Face mod. When you start doing that to like really old games or like vector art games or even like Atari and television stuff and you see the results, you, your mind will start racing with the possibilities that are complete bastardizations of what was there.
- [24:22]
A
But they're still really fascinating but fun vibe code. Let's, let's jump into that. This is a new plugin that is available on Hugging Face from one of the, I think the co founders of Hugging Face, Victor Mustar. Mustar. And it's called DLSS anything. And you know, you and I have been following these tools for a while that you can dlls like pictures pretty easily because you're just kind of running an AI up res on this. You can go do this. But Kevin, you actually did this yourself. You released a thread of these images. Tell me which ones like surprised you the most and why.
- [24:49]
B
I just texted you one which isn't on the thread, but take a look at it. I ran Paperboy for the Sega Genesis through this DLSS5 on and now.
- [24:59]
A
Oh my God.
- [24:59]
B
Like this is just an image model with a creative prompt to try to enhance. That is so cool. I love that.
- [25:07]
A
Yes.
- [25:08]
B
And now look, you can pinch, zoom and go. It turned the lawnmower into a car. Whatever. If this were a really truly like custom trained model with like a Laura and some art for Paperboy, this Looks like a brand new 3D but beautifully isometric. Paperboy. Yeah. That I would love to play. When you see what it did with Qbert, it is absolutely hallucinating in a broken way. But it's kind of cool. I ran an old PlayStation 1 game that few, very few know. It's called Vib Ribbon. If you get a chance to play it, it's great. You could.
- [25:38]
A
It's the best.
- [25:39]
B
Yeah. You popped in Vib Ribbon, it would load the game to ram, then you put in whatever audio CD you want and it would generate a level for you to skip through it as a wireframe bunny. But when you see the results that I got with no other prompting other than just turn this into DLSS5 or whatever, it fully imagined the rabbit and the world that they're in and all the other characters. And it was very like. It was mind blowing. It is not what the original artist intended. No, it's not what the original creators had in mind at all. But you could immediately see a future where an indie developer could say, look, here are the bones of my new game.
- [26:14]
A
Yeah.
- [26:15]
B
Here are some instructions for the AI artwork layer that's going to sit on top of this. And if you want to go make your own too, go for it.
- [26:23]
A
Hold on to that thought because in a bit we're going to talk about the thing that Kevin actually has worked on, which kind of allows you to do this. But first, Kev, we have a new segment in our show. There's so much added news to cover and we are trying to make these shows a little tighter. So get ready for five quick AI news stories in five minutes. Okay, Kev, let's.
- [26:45]
B
We could have done like a 5 and 5. We could have done something else. But no, it's. Here's five quick artificial intelligence news stories delivered to you in five minutes or less. That is our promise.
- [26:56]
A
First, Mid Journey v8 is in alpha and people are underwhelmed. There are. There are images coming out of this. You can go and try it if you are a Mid Journey subscriber, Kev, this is one of those systems where, you know they are having problems with hands. Again, there are things that don't look that exciting. There are defenders. In fact, the Mid Journey handle is actively defending this model, but it feels like they are falling a little bit behind your thoughts.
- [27:17]
B
Diffusion model versus world model. It seems like a lot of companies have moved on and are integrating other tech, which might be very tough for a smaller company without massive funding to keep up with.
- [27:29]
A
Next up, what's Up. That was my job.
- [27:32]
B
Boy, I hope there's a graphic there. I didn't know if you were playing Fruit Ninja in another tab or what was going on. Runway and the Vera Rubin chip, which is a new Nvidia chip doing near real time video generation. We had talked a year and a half ago about how this was going to be the future. It's now at least a bleeding edge reality. We're talking you press a button and less than 100 milliseconds later, high fidelity video starts streaming out of the model and runs continuously. Gavin, are we going to have personalized Netflix tomorrow? Are we going to press a button and be able to stream infinite things?
- [28:06]
A
My biggest thing about this, the number one problem I have with AI video generation for me is the time it takes to get some results back. So I am very excited to see a world where I can get instant results. All right, next up, next up, Val Kilmer, one of my favorite actors from back in the day and real genius, has agreed to be resurrected in AI to start. Agreed? Yes, yes. He hasn't agreed yet to the state to be resurrected to star in a new movie called as Deep as the Grave. This was a film that the creator of the film had always imagined Val being in it. Val passed away sadly from, I think, throat cancer. And this is a big deal, Kevin, in Hollywood, because this is our first real test of an actual kind of large actor who's going to do a lead role and it will be completely AI generated. There is nothing that was shot for this film ahead of time. Your thoughts?
- [28:55]
B
Yeah. Plenty more of this to come. Whether you like it or loathe it. I just wonder what DLSS5 Kilmer is going to look like. And can we put that on the screen? Nope. What's next?
- [29:08]
A
What's next?
- [29:09]
B
Why are you spitting on the keyboard? Boy howdy. People love to shoot shade all over anybody doing anything agentic these days and how they don't have the vibe secured or whatever. But Meta, with their multi billion dollar corral of engineers is having a rogue AI agent problem. There was a post to a thread that an internal AI agent responded to and leaked company information and that became like the snowball which triggered an avalanche and they actually flagged it as a SEV1 incident, which is the second highest security severity level that you can have met. A safety director recently shared that her AI agent deleted her entire inbox. So that wasn't that long ago. So I feel like there's some one, two punches happening with Meta and Gavin, if they can't secure it. What hope do we have?
- [29:58]
A
Well, my bigger worry is what does my mom do when she starts making her own AI agents to kind of send us Christmas letters? But that's all for that. I had to stop making that spit
- [30:09]
B
sound at some point.
- [30:10]
A
Finally, in our five stories, Kagi, an interesting kind of platform that's actually a kind of a web browser and a bunch of other stuff, has created a universal translator device. And if you've seen these kind of LinkedIn translator social posts that have come around, basically it's a very cool use of generative AI that allows you to translate to anything, not just languages, but to characters. And Ethan Malik created a screaming tree. I actually used it to create an AI that didn't really want to do this sort of thing. But it's a very fun tool that you can go use. And the best part about this, Kevin, is the idea that people have LinkedIn speak. LinkedIn to me is the worst of the social networks when you go there. I spent time for a little bit trying to grow my LinkedIn, but I'm like done with it. Because every time you go there it does feel like it is already taken over by AI bots. But that might just be like sales bros thinking that's the way you're supposed to speak. Something very exciting we want to talk about when it comes to vibe coding and all the fun stuff is something Kevin, that you have been working on and I'm very excited to play this. Let us know what cringe every time
- [31:05]
B
you say working on it because it has not been a labor of love. It was like a day long jam though, as I sat with the laptop on my chest and would occasionally yell at agents. So last week you and I were mid discussion on something and I think I said something like, well, but wouldn't it be cool if you could prompt new Doom weapons or enemies or even gameplay modes into existence? And we both kind of like ah. And I think we said like cut that out because we're gonna do it. Yeah, yeah, believed it, right? And then I felt like, oh crap, I guess I gotta actually try to do it because we said we were gonna do it and I'm going to release it for everybody to mess around with. But I essentially sat with Claude Code and with Codex a little bit and said, listen, here's what I want to make. I want to go find the best open source because Doom has been open source. The legends at ID Carmack, who I believe in, he's going to get us to AGI. You watch Gavin, he's still very much.
- [31:55]
A
I'm sure he will actually.
- [31:56]
B
I don't doubt they released all of the source code to Doom a long while ago, which is why you see Doom running on everything from toasters to home pregnancy tests. The code is out there. It is well established. People have modded it in many ways including porting it to web browsers. So I said that's a nice entry point. Go find me the best Doom port that runs in a web browser so that I can tack on an interface over it that connects to generative AIs like OpenAI or Gemini or Eleven Labs. So that if I want to say, replace the pistol with a T bone steak and make it move when I shoot, will understand that and do it now. It was far from a one shot. I had to go back and forth a lot. But what I will try to clean up and release, which is again like I. I actually take pride in the stuff that I release these days.
- [32:45]
A
Sure.
- [32:45]
B
Unlike the rest of my career, this is not one of those things. This was a sloppy vibe code that maybe you will want to take and go do something with. But where it exists today is that you can connect it to OpenAI or 11 labs, be very easy to add your favorite provider. There's also a local mode if you want to run Flux locally and not deal with generating graphics. The problem is I'm developing on a Mac and so like the best audio models and the better, you know, video models, they just do not run well unfortunately. So I left the cloud in by default. But you can load up the game and say replace my pistol with anything and it will develop the sprites for it to be shot, including the muzzle flash themed or whatever it is. So if you like replace the pistol with a turkey sandwich, it'll make it look like mayonnaise coming out. If you replace it with a rubber chicken, it makes it.
- [33:32]
A
Yeah, it's a pretty scary look, it's pretty scary imagination.
- [33:37]
B
You can replace it with a rubber chick, whatever you want, it will just replace the graphics. It will also go to 11 labs to generate a sound effect. I also have it so that you can replace the zombie guy, which is like the first kind of default character that you face in a lot of these levels. And it will use OpenAI to generate two character sheets of whatever you ask for. So it does all the views, cuts them up, puts them in where they need to be and you'll see on the screen. It is not perfect by any stretch,
- [34:04]
A
but the fact that it was like
- [34:05]
B
a day to Kind of jam it out and get it as far as I did. Bodes pretty well for somebody who wants to take it and run with it and let you be able to swap anything. And what I really love about it is that, by the way, if you run it with the local models, it will generate the character sheets and then it will go. Because Flux Schnell, which is what I'm using, doesn't support transparency. It does it through a secondary process and removes the background from all the sprites. So it's pretty elegant.
- [34:30]
A
It.
- [34:31]
B
It's fairly performant. It will run pretty well. You can play it on a Chromebook if you're using the cloud version of it, because it'll just do it all in the cloud and then spit it back. And it injects the sprites and the sound effects into the engine in real time, which was important to me because I didn't want it to have to be like, you do a prompt, then you run the game and you play it like you can be playing it and in real time, prompting it and changing out the graphics and the sounds. And that, to me, is like, super, super interesting.
- [34:56]
A
Yeah, I mean, it's kind of magic. And I want everybody to make sure they get a chance to go try it. We'll have it available hopefully by the time the show goes live. If not, then right afterwards. But I do think the cool thing about this is the idea of, like, letting something go like this. Open source, you didn't spend a crazy amount of time on it, but it does become that multiplayer vibe of like, oh, someone else will take it and do something with it, or someone else will take it there. I also think the coolest thing about this is that Doom is now maybe becoming like the next ST of whatever the product everybody makes their own version of, which is a very cool thing to your Carmack point and everything like that, which is amazing.
- [35:28]
B
What's funny is that we, you know, we look back at early image generation or video generation or Vibe coded games, right? And they were text with like, AI Dungeon. And then we got to like Atari graphics. Like, oh, this thing. We are. We are graduating. As the tools get better, so do the platforms that we're iterating on. So I can't wait to get to the Dreamcast Vibe code stage. That's where I. I am done.
- [35:49]
A
Ooh. All right, well, we'll see you all next week. Thanks for joining us. It's AI for Humans. Bye, everybody. Bye. Bye.