Kimi K3 Is Here. China Just Hit the AI Frontier.

Summary

Podcast Summary: AI For Humans - Kimi K3 Is Here. China Just Hit the AI Frontier.

Hosts: Kevin Pereira & Gavin Purcell
Date: July 17, 2026

---

Overview

This episode dives into the unveiling of China's Kimi K3 language model, the ever-fiercer AI capabilities race with the US, major updates and rumors about frontier models like Anthropic's Fable and Opus, and the hosts’ hands-on AI projects. The episode balances deep technical discussion, playful banter, and opinions on the broader implications for AI competition, regulation, and human creativity. A healthy dose of AI-powered humor—plus live appearances from their recurring fictional AI, Fig and Moss—rounds out the week in AI.

---

Key Discussion Points & Insights

1. China's Kimi K3: Breaking Into the AI Frontier

[00:00–06:16]

• Big News: China’s Moonshot Labs has launched Kimi K3, which is edging closer to top-tier American models.
  • "Kimmy K3 is the new model from Moonshot Labs... inching ever closer to American Frontier models." – Gavin [01:41]
• Comparisons:
  • Kimi K3 is said to be comparable to GPT 5.6 Soul, and beats out some Western efforts like Opus 4.8, if not quite matching Fable 5.
    • "It might be very comparable to GPT 5.6 soul and is definitely better than Opus 4.8." – Gavin [02:13]
• Global Competition:
  • Kevin highlights how this competitive dynamic benefits all users: "If this thing is as good as it is... those got cheaper now and your limits are going to be reset." [02:57]
  • Both hosts underscore how quickly these models are catching up, questioning the "six months behind" narrative for China.
• Capabilities: Early tests show Kimi K3 is strong at building web-based games and other benchmarks.
  • Cited "Flappy Bird" and "trebuchet" tests showing creative strengths vs. GPT 5.6’s technical prowess.

Notable Quote:

“Competition is great and we don't have competition much anymore. At least back in the States. If you've ever, I don't know, flown, you understand that.” – Kevin [03:32]

---

2. The AI Race: Frontier Models & Alpha Leaks

[06:16–10:28]

• Self-improving Models:
  • Discussion about the trend towards ever-larger, self-improving AIs.
  • Kimi K3 reportedly has 1.5–2 trillion parameters and foreshadows massive growth.
• Insider Rumors (“Alpha”):
  • Kevin shares inside info: Opus 5 is coming as early as next week (“blows the doors off Fable”), notably excelling at 3D/3JS tasks but is still slow in early testing. [08:04]
• User Frustrations: Both commiserate about expensive, usage-limited access to Anthropic’s Fable and workarounds.
• Efficiency: Hope for lighter, more efficient versions and improvements in token usage.

Notable Quote:

“The takeoff is never going to be comfortable... When you see what it looks like when these machines start improving themselves... it should be a wake up call.” – Kevin [06:16]

---

3. Regulation & AI Safety: Demis Hassabis’s Open Letter

[10:29–14:26]

• Industry Response: OpenAI, Anthropic, Microsoft, and even Elon Musk echo support for DeepMind’s Demis Hassabis’s proposed regulatory framework for “frontier AI.”
• Main Point: “A framework for frontier AI and a dawning of a new age,” charting a path for self-regulation among the biggest players.
• Tension: Audience skepticism—how much can these companies self-regulate and will it make the AI trajectory fundamentally safer?
• AI Takeoff & IQ Milestone:
  • Recursive self-improvement signals the “intelligence takeoff”.
  • “GPT 5.6 Sol Ultra has an IQ of 135, which is the highest IQ for any model yet...” – Gavin [13:55]

Memorable Moment:

“Recursive self-learning leads to the intelligence takeoff… many people believe the intelligence takeoff is something we're in.” – Gavin [14:11]

---

4. Hardware: OpenAI’s “Smart Speaker” & The Vibe Coding Keyboard

[15:15–19:48]

• Smart Speaker:
  • OpenAI’s Alexa-like speaker: portable, camera-equipped, possibly screenless—evokes privacy concerns and consumer appeal.
  • “A lot of people are crapping on OpenAI for this idea... but if this is smart... they could sell a crapload of these.” – Gavin [18:01]
• User Attitudes: The hosts split between “open kimono” (carefree) and privacy maximalist (camera tapes, hardware shutters).
• Coding Keyboard:
  • Codex’s limited-run “Vibe” keyboard lets users physically control and approve actions, but both hosts are mostly sticking to typing—not voice prompts.
• Side Topic: The “noise cancellation mask” for speaking to AI in offices—both find it ridiculous but understandable.

Notable Quote:

“If it's portable... and you had, like... four or five Alexas in our house because we wanted coverage throughout... They could sell a crapload of these.” – Gavin [18:01]

---

5. Creative Uses: Hands-On AI & Side Projects

[27:00–33:45]

Game Prototyping:

• Kevin’s Project:
  • A web-based “camouflage cursor” game, inspired by Chroma Chameleon, built with LLMs like Fable 5 and GPT 5.6; playable in browser, supports up to 8 players.
    • “Over the course of about four or five nights... was able to get something up that I'm like, I'm actually... I think it's kind of fun.” – Kevin [29:07]
• Gavin’s Project:
  • An autonomous AI TV show, starring a fig and a tardigrade, driven by a custom personality in Claude, built in 3JS.
  • Emphasizes the creative “flavor” different models apply, with GPT 5.6 Soul being “extreme vanilla” compared to Kimmy K3’s quirkier outputs. [04:38]
• Meta Moment: Fig and Moss (AI characters) react in-character to having their “lives” exposed on the podcast, adding a self-referential, playful loop [31:48–33:44].
• Experimentation: Rapid prototyping with LLMs and 3D engines like Blender, Unreal.

Notable Quotes:

“I've created a kind of mini video studio in 3Js that the AI can play around in and make his own things... It's a weird, like, universe that's developing that I'm just kind of shepherding through.” – Gavin [31:12]
“I love the game Chroma Chameleon... wanted a version you could easily play in browser... For quick things, yes. But for the longer, like, prompt work, I found it's actually much better... to focus.” – Kevin [29:06, 19:25]

---

6. AI Video & Music: Sedance 2.5 and Suno’s Training Data Leak

[21:48–23:44]

• Sedance 2.5:
  • New version enables 30-second AI videos and improved quality; excitement over upgrade but it’s still rolling out.
• Suno Hack:
  • Popular AI music platform Suno was hacked; revealed YouTube music was in its training data.
  • Ongoing legal battles with labels; debate over “dirty data” and future settlement.

---

7. Robot News: Fighting, Floating, and Friendly (or Boring) Bots

[23:44–26:43]

• Robot Fighting Leagues:
  • China’s Urkl League features robots kicking, losing heads—a viral hit.
  • “These robots are roundhouse kicking, they are looking really good. But more importantly, one of them loses their head throughout the fight...” – Gavin [24:11]
• Other Robots:
  • Floaty Whale Bot: New, balloon-like robot form factor.
  • Office Chatterbots: Two Neo Robotics bots have a mundane office conversation, underlining the gap between “intelligent” and “interesting.”

---

Notable Quotes & Moments

• On Chinese Model Race:

  “The flavor of GPT 5.6 is like extreme vanilla right now. Like, that's my feeling... a very capable model.” – Gavin [04:38]

• On AI Takeoff:

  “Recursive self-learning leads to the intelligence takeoff... many people believe the intelligence takeoff is something we're in which means that it exponentially grows in capabilities and then very quickly gets smarter than us.” – Gavin [14:11]

• On Privacy:

  “There's two camps. There's the camp of like, who cares? You're already on the grid... and then there's the other camp that... puts tape over lenses...” – Kevin [17:21]

• On Getting Bored by Robots:

  “Hopefully we'll get better than this. ... now imagine watching two fricking robots talk to each other back and forth.” – Gavin [26:41]

• On Meta Humor:

  “He taped himself knowing we'd watch him watching us. The show is a loop now. The snake ate the tape.” – Moss (AI Character) [34:09]

---

Timestamps for Key Segments

• Kimi K3 Chinese Model Launch: 00:00–06:16
• High-parameter Model Race & Rumor (Opus 5): 06:16–10:28
• Frontier AI Regulation & Hassabis’s Open Letter: 10:29–14:26
• OpenAI Hardware/Speaker, Vibe Keyboard: 15:15–19:48
• Hands-on AI Projects (Games/Shows): 27:00–33:45
• AI Video Creation – Sedance 2.5: 21:48–22:20
• AI Music – Suno Hack: 22:46–23:44
• Robot News (Fighting, Floaty Whale, Office Bots): 23:44–26:43

---

Final Takeaway

The AI race is accelerating: China’s Kimi K3 edges closer to US-leading models, new AI frontiers are about to drop (Opus 5), and the industry is (maybe) moving toward self-regulation. Meanwhile, more hands-on builders are using state-of-the-art models to fuel creativity—or just have a little fun with AI-powered games and shows. The borders between science fiction and reality keep blurring as the pace quickens, all delivered with signature wit, skepticism, and a touch of existential (and literal) comic relief.

---

Transcript

[Gavin]

China's latest model, Kimi K3 is out and it is inching ever closer to the frontier of American AI.

[Kevin]

But it's not as cheap as free and probably too big to run locally. Unless you're like some sort of billionaire who fights robots in their spare time. You know, robots like this.

[Gavin]

Those aren't your robots, Kevin. That's for sure.

[Kevin]

Those aren't my robots yet, Gavin.

[Gavin]

Plus Suno got hacked and we got a sense of what's in that AI musical soup. And it turns out it's a lot of human soup.

[Kevin]

Yeah, yeah, it's kind of sad. But less sad. If you connect GPT 5.6 soul to tools like Blender or Unreal Engine, you can make cool games that gamers will hate you for making.

[Gavin]

Or in my instance, you can make truly horrifying strange AI slop and somehow compelling kitchen scenes.

[Kevin]

Yeah, the two of us have actually been building some things in our spare time and we want to do a little show and tell. I'm going to show off a camouflage cursor game inspired by Chroma Chameleon that you can play for free in your browser right now with your friends. I'm pointing a lot if you're only getting the audio, but I'm excited.

[Gavin]

And I made an autonomous TV show with a fig and a tardigrade.

[Fig]

What?

[Moss]

Do you remember yesterday? Most of yesterday.

[Kevin]

Of course you did. Of course that's what you did.

[Fig]

This is AI for humans, everybody.

[Gavin]

Welcome, welcome, welcome everybody to AI for Humans, your twice a week guide to the wonderful world. AI we are happy to have you back. And Kevin, there's some big news today out of. Can't you see. Oh, no. What is. Who was that? Who was. Kimmy. Kimmy, can't you see it was big.

[Kevin]

Oh, yes.

[Gavin]

No, no, no. I think it goes well.

[Moss]

Yeah.

[Gavin]

But there's also like, where's that sample from that comes. Anyway, if you know in our audience we've already gone off track, please tell us five hours later. Kimmy K3 is the new model from Moonshot Labs. This is an Chinese AI company and they are inching ever closer to American Frontier models. Now we're going to dive into like how close this is exactly to American Frontier. But Kevin, this is a very good new model that is just out in open router. So I don't think you or I have actually played with it, but it is out now. What people are saying so far is that this is not Fable 5 level. I mean, very little is right now, but it might be very comparable to GPT 5.6 soul and is definitely better than Opus 4.8. Kevin, what say you? If this is true and what it feels like to you is that this is a good model.

[Kevin]

I should care more. I should care more. I should. We should all care more. I look great, grand, wonderful. The, the Chinese model of it all which has some people concerned. If you're an enterprise customer, then you have every right to be concerned about that. Just as some are very concerned about giving their enterprise data to Anthropic by any other name. I care if this thing is as good as it is because then it makes someone in a server farm somewhere wake up in a flop sweat and go, oh, we just optimized something. And now the other models that you use and the daily driver apps that you use, those got cheaper now and your limits are going to be reset.

[Gavin]

In fact, hey, guess what. Yeah, guess what? Got a reset just yesterday. Kevin Claude Fable 5 Just reset and they keep stretching this date, the date that it's going to go away from the subscription. So yeah, sorry to interrupt you, but that is a big thing. This pushes everything forward.

[Kevin]

Exactly. Competition is great and we don't have competition much anymore. At least back in the States. If you've ever, I don't know, flown, you understand that. Painfully. But let's go back and talk about this model specifically some of the things that people are discovering with it. It is pretty good at building web based games. It seems to be very good at like the typical benchmark things like build a city in Minecraft. It could certainly answer basic questions.

[Gavin]

Oh yeah. So it's really good. There's a lot of really great videos. There's a flappy bird test from June underscore song that's really good and it shows like oh, this is comparing to what it can do to OP. Specifically that one to OPUS 4.8. Much better. I thought the one from Cheta Chetaloosa. Chattaloo. Chattaloo who often shows off new things. Yeah. But I think, sorry, this was a really cool, interesting comparison to what it can do with 5.6 soul. So this is a game that he or they made where it is a like a, a trebuchet or a catapult that's shooting across the way. And what they mentioned was like the GPT 5.6 SOL version might be a little bit more technically interesting but the Kimmy K3 one is kind of more interesting creatively and I have found this with 5.6 all. And it is something I want to talk about later when we talk about the stuff we built with these models is that there is a flavor creatively to some of the stuff that you make with these. And I would say the flavor of GPT 5.6 is like extreme vanilla right now. Like, that's my feeling of the GPT 5.6 SOL version. It is a very capable model. But back to this. I think what's cool about this, Kevin, is that it is a showing that we are at this stage where these models are really catching up. Now, I saw somebody point this out, which I think is an important thing. The Fable model technically is based on Mythos, which is now five months old. Now, you and I and the world at large has not had it for five months. But that model, the model we all got scared of when Anthropic said they weren't going to release it, Project Last Wing is about five months old. So it's not crazy on the timeline of how these things work to think about a model like this coming where it compares. Okay, with table five, which is a dumbed down, slightly dumbed down version of Mythos. But it does continue to make me think and my eyes go like. Because I had another moment like this recently when I was reading that 2040 paper, which you should all read the follow up to AI 2027, which is, you know, there's this rumor that GPT6 is coming maybe the end of this month, maybe a few months later or you know, some point when it passes the government thing. But like really the progress is weird now. You know what I mean? Like, this is like just a good

[Kevin]

example of like open takeoff was never going to be simple. The takeoff is never going to be comfortable. Right. When you see what it looks like when these machines start improving themselves and they've been, yes, now we're getting these models with, okay, like Kimmy is supposed to be what, how many trillion parameter model?

[Gavin]

I think it's like between 1.5 and 2 trillion parameters.

[Kevin]

Sure. And we're going to have orders of magnitude more than that. And that should mean that they become more capable. As that starts to happen and they start self improving, it should be a wake up call. It should rock everybody that's listening to this. But Gavin, I haven't even told you this. Oh, and this is not a joke. Okay, I got, I got some alpha, baby.

[Gavin]

Oh, okay. What is it that you can say on here?

[Kevin]

It is. And I'm going to say it on here right now. I just, I just got permission to say it. But I have to say my sources. I'M not allowed to say.

[Gavin]

Okay,

[Kevin]

so I'm not allowed to say what was just bleeped or maybe just cut out, but I can say Gavin, and this might sound generic, but it does dovetail into what you were just saying.

[Gavin]

Opus 5, okay, is coming perhaps as

[Kevin]

early as next week.

[Gavin]

Okay. I mean, is that person.

[Kevin]

Person has had hands on time with it.

[Gavin]

Okay.

[Kevin]

Okay. This is where we're at.

[Gavin]

Yeah.

[Kevin]

Oh, yeah, baby. I'm not there. I can't. I can't give it. We gotta. We gotta drip it. We gotta drip this because there's not a lot of Alpha.

[Fig]

Okay.

[Kevin]

Will. Dramatic music HEARTBEAT of A NATION Lower third Breaking news. It's Alpha drip. And we're dripping and we're dropping.

[Gavin]

Okay, Gavin, the drips come across the screen. Will they come across the screen gently? There's like a pitter pattern brain. We're like, Ev and I are both at the feeder.

[Kevin]

No, we're at the feeder. We're getting little drips of Alpha. Please put it.

[Gavin]

Leave all that in.

[Kevin]

Well, I trust you. Please put a feeder. Okay. Oh, the new Opus 5 as early as next week in early testing, it blows the doors off of Fable. Oh, of Fable. It is very slow to be expected right now. How much that power they're using to serve it up to these early testers. But specifically with 3D and with 3JS, it is doing things that have not been seen from models yet.

[Gavin]

Okay, well, cool.

[Kevin]

Could be hitting next week and wouldn't that make sense, Gavin? As we see, everybody goes open source, open weights, Open this. It's always six months behind, and now look at this. Kimmy is here and it's punching with the foundational models.

[Gavin]

Yes, sure.

[Kevin]

But to your point, Fable was five months ago. If it was improving itself and helping with this Opus release, maybe, well, then, yeah, right around now, we're. We should get something that does surprise us.

[Gavin]

Another thing that's really. That's great. Great, Alpha. Thank you, Alpha Gods. We're really giving our Alpha hose.

[Kevin]

No more drips.

[Gavin]

No more drips. No more drips for anybody out there. I think what that also goes along with Kevin, which is really interesting, is there's been a lot of chatter this week about how 5.6 Soul is taking people's, like, daily driver Fable work, the work that Fable might have had, but more importantly, Fable is so expensive to use that, like, maybe nobody was going to use Fable as a daily driver. Right. The way that they have charging for it, the way that it comes out, especially when you're out of the subscription if. Oh, oh, what, you got more drips?

[Kevin]

I get. No, no, thankfully, I don't think. Am I dripping? Am I spying more?

[Gavin]

Yeah, you zipped up and now you've stopped dripping.

[Kevin]

There's always one. There's always one drop of alpha left. But this isn't a. This isn't it. I get po'd and to'd at anthropic. Have you hit a Fable limit? Like, have you hit a cap?

[Gavin]

Yeah, of course.

[Kevin]

Yeah, of course, of course. Everybody has.

[Gavin]

Yeah.

[Kevin]

So why is the entire account disabled? Because you hit one usage. Oh, right.

[Gavin]

That doesn't happen to me. I can. I can go down to Opus and use Opus.

[Kevin]

What?

[Gavin]

Yeah, yeah, it doesn't happen to me. If I hit a Fable limit, I can drop down to Opus. You can't do that.

[Kevin]

I wasn't able to, but maybe, maybe, maybe it's a skill issue, bro. I tried to ask it.

[Gavin]

Maybe start. That actually does work. This is the dumbest thing, but it's like unplug and plug the machine in. Sometimes that's all you need to do.

[Kevin]

Crazy. Okay, well, that is. And that's Gavin's alpha you've been dripped.

[Gavin]

My alpha is unplug and plug in your computer. That's if your mother in law advises

[Kevin]

sometimes you gotta reboot the damn thing.

[Gavin]

Okay, let's quickly zoom this back. We got a lot of great alpha there, but let's zoom back. So Opus 5, by the way. Opus 5, that's very exciting. I hope that it is. It uses less tokens. That would be really interesting to me. I have seen people talk about Fable 5 slow being really useful and like. Or whatever the light, light version of Fable 5 is, is that it does really work, but it's just a much slower thing. So that's great. I'm very excited to see more efficient models back to Kimmy real fast because I do think this is like an overarching kind of story, which is all these models are getting better and better. And we've always said that these open source models or the Chinese models are going to be six months behind because, like, there are, you know, lesser abilities to train these things and everything like that. The thing that also kind of pairs with all this is that Demis Hasabis, patron God of video gamers and AI people everywhere, has released a very long post on X this week that got retweeted by basically the head of every American AI company. Satya Nadella Elon weighed in on this and Sam Altman. They all said this is a Good idea. And I just want to read the headline of this Kevin, because I do think this is like all of the stuff that we've talked to. If you're, if you're new to this show, welcome. This is a weird time to be alive. If you've been watching the show for three years. This is all this stuff that we've talked about for a long time. But Demis has basically said, he said a framework for frontier AI and a dawning of a new age. And what this post lays out basically is finally a way for these companies to kind of come together and maybe start to help self regulate. Now I know a lot of people are going to be out there like I don't trust these people but like to work together so that we are not kind of like in competition to risk things that can hurt humanity. And a lot of people out there would be like this is ridiculous. Nothing can happen with AI. I think these people are now telling us that they care deeply and they think enough about this is possible that they are willing to work with their biggest competitors going forward. So this is like another moment of like oh, I think this is a big deal. So I think you should all read this post. It's really fascinating and I think it's quite, quite a thing.

[Kevin]

Listen, congratulations or I'm really sorry that happened to you. The post. Yeah, it's so many words. Gavin, People are tuning in for this. They want your alpha, they want your intel. No, they don't want the drip, takeaway drip on them.

[Gavin]

Give them the takeaway. Is this the takeaway? Is this the big takeaway is that I think that the big models are going to unfortunately come out less close to when they're done. But there is a very good chance that we will hopefully get together and work on a pathway to this that makes sense for not only so that humans don't use this poorly but eventually people don't give a bunch of control over this thing that could be bad. Now I will say there's a lot of fictionalizing of what could happen but I do think that 2040 paper is worth reading. We'll link it in the show notes. There are, you know, they are very much on the AI is dangerous side and there's like some different pathways they show but we are entering a point. I know this sounds like science fiction but like you know, recursive self learning leads to the intelligence takeoff. And you know, many people believe the intelligence takeoff is something we're in which means that it exponentially grows in capabilities and then Very quickly gets smarter than us. Kevin, did you see the stat this week? That GPT 5.6 Sol Ultra has an IQ of 135, which is the highest IQ for any model yet, and that is smarter than 99% of people. Now, IQ is a problematic number and it doesn't often, you know, exactly give a sense of how smart a person is, obviously, but that's an issue, right? Like you imagine where these models at. Now, this model that you and I have used, which is fine if you get to 6, 7, 8, we are really entering into science fiction land.

[Kevin]

GPT3 was smarter than me. I mean, it wasn't maybe as capable. I'll give myself that.

[Gavin]

Well, that's smarter than me, but that's the difference. I feel like the capability aspect is a big deal. Like. Yes. What's his name? Brett. The guy. Brett Taylor. His. His company just released something just now, in fact, I don't have in front of me, but I just saw it kind of glance past my eyes. Their company is going to start making Long Horizon agents for things like a mortgage or things like, hey, apply to your health care company and so you don't have to do those crazy long steps. I'm in the process right now of, like, looking at moving to Canada with my wife just because she's a Canadian citizen and like, we're trying to do the whole permanent residency. And Kevin, I can tell you, it is a freaking nightmare to go through that.

[Kevin]

Oh, my God.

[Gavin]

Yeah, I would love to give them. I mean, I use Claude throughout this process and it's been very helpful. But, like, I would love to just give it to you.

[Kevin]

Go do it.

[Gavin]

Yes.

[Kevin]

You have access to all my stuff. You go sort it out.

[Gavin]

Yes.

[Kevin]

And you know that process so much better. Gavin, what's that? One of two things. Number one, some weird smart speaker that follows you around, sure. Looks at you or whatever, has some sort of personality.

[Gavin]

Right.

[Kevin]

That you could spend billions, like tens of billions of dollars to fund.

[Gavin]

Or a special keyboard with a little

[Kevin]

knobby knob that you can twist and turn or a special dedicated button that you can hit to approve what it does. And how great would either of those things be?

[Gavin]

This is what OpenAI is worrying about, except for instead of worrying about, like, the end of humanity, we're talking about two new hardware products from OpenAI. We're talking about a smart speaker and we're talking about a vive coding keyboard. First, the smart speaker, which feels like the much bigger deal. This supposedly is OpenAI's big new product that is going to be coming out this year. This is a piece of hardware they've been working on. And Kevin, according to Bloomberg, it sounds a lot like an Alexa to me, which again, I think is not a bad idea. A lot of people are crapping on OpenAI for this idea. But then. Yeah, you referred to it like there may be some sort of like following you around the room thing because they are going to have cameras. And Kevin, let me guess what you. Mr. Open security. Mr. I'm okay with everything. All my information can be out in the cloud. What do you think about OpenAI's? What do you think about OpenAI's camera ready microphone, listening device in your house connected to the cloud?

[Kevin]

I mean, they want it to be portable, right? It's battery powered. Something I'm imagining it's going to be like a Sonos move kind of speaker in size, like a little like a pod or a pill with a camera or whatever. I don't know what sort of screen it's going to have on it. If it.

[Gavin]

I don't think it's going to have a screen. I think they're saying it's going to be screenless.

[Kevin]

Yeah, it might be. Like, but there, there's. If it's going to like potentially have some sort of motor to, to show that it's expressive, like it's moving or there's a panel looking at you. I bet there's going to be some sort of LED light or something.

[Gavin]

Like a couple of big eyes. Like that thing from Guillermo del Toro where the eyes are on the hands. Like that could be.

[Kevin]

Yeah, it could be.

[Gavin]

We'll put eyes on my hands. We could put eyes in my hands. You got to turn it into YouTube to see the eyes on Gavin's hand. So if you're looking for a reason,

[Kevin]

put eyeballs on his fingertips. Like he got.

[Gavin]

No, don't do that.

[Kevin]

Just make it horrific. Beetlejuice. It's interesting. There's two camps. There's the camp of like, who cares? You're already on the grid. Your data has already been stolen, by the way, let alone been resold by billions of companies to micro pennies. Like, what do you just put it all in the machine and who cares? And then there's the other camp that, like, doesn't. That puts tape over lenses, things that. Yes, hardware shutters. So I don't know what camp you're in, but I have to imagine that they put a finger to the wind and they know where this is going.

[Gavin]

And people, you know what camp I'm In. I am in the open kimono camp. I like to call it the I don't care camp. But. But I do want to say one thing that's interesting is I didn't hear that part about it being movable. And one thing that's really fascinating to me. So we used to use Alexis in my house all the time. And I know people are like, why would you ever want a smart speaker? But we would ask it questions, we would get it to turn on music. I would have it do stuff. But it was so dumb that was. We stopped using. In fact, famously, we have a joke in our family that it was like, you know, you ask a question about something, I can't remember what it was. It was like some Santa Claus question, and then it would call a blog spot address to give you the answer, which is like a thing that doesn't really exist, but it's a URL that it, like, must have surfaced in some way anyway. So dumb. If this is smart and if I were to have, like, you know, if it's portable and I had, like, we had at one point, like, four or five Alexis in our house because we wanted coverage throughout, everywhere. They could sell a crapload of these, Right. This could be a thing where they actually sell a bunch of them. And then you can imagine again, for people who are okay with this, having one in the living room, having one in the dining room. And everywhere you are, it's tracking you, and it's saying, like, hey, make a reservation at this place. If that works like, that is kind of game over for, you know, consumer use case. And again, that's a big deal. The other thing we should touch on very quickly is this weird keyboard that Codex has which is, like, you touch it, it allows you to do the basic Vibe coding actions. It is a very much a toy. It is a limited run, but it's still a fun, kind of cool idea to have, like, a limited keyboard so that you can touch it for voice commands and then you can hit go. I've personally not found myself using voice commands a ton with Vibe coding. Kevin, are you still using voice a lot?

[Kevin]

No. You know, I actually found that, like, look for quick things. Yes. If it's a quick, like, hey, that's not exactly what I asked for. Fix it. Whatever. For quick things, yes. But for the longer, like, prompt work, I found it's actually much better for me to focus.

[Gavin]

Me, too.

[Kevin]

Talking. I may go down one path and they go, no, that's not exactly what I.

[Gavin]

That's exactly right.

[Kevin]

Stop. Yeah. Think, get tactile. Put the words out. But I mean there's certainly people that do.

[Gavin]

Yeah.

[Kevin]

The noise cancellation mask that some people are wearing.

[Gavin]

Yes, I have seen that. It seems like such a dumb thing to me and it looks ridiculous, but I understand why. Like it's like basically a gas mask you put over your face when you're talking. So if you're an office environment, you can talk to your AI and not. And not ignore anybody. But Kevin.

[Kevin]

Brilliant.

[Gavin]

Another gas face that you should make sure not to put on is the gas face.

[Kevin]

Where am I going here?

[Gavin]

Where am I going here? Where the gas is the gas face of not understanding. In a world where people feeling of increasingly.

[Kevin]

In a world where people are increasingly muting themselves, especially in shared environments or trying not to be heard to be polite. The one rebellious action which all still human smart. Yeah being should do is to scream from the mountaintops, I love you, Gavin.

[Gavin]

I. I love you. And how do you love somebody now?

[Kevin]

Do you love somebody you like and subscribe to?

[Gavin]

That's right.

[Kevin]

Like maybe you back them on only a Patreon.

[Gavin]

Maybe you back not only fans on Patreon. We have a Patreon. We have a Patreon. By the way, we have been thinking about like maybe doing a few specific things with on the Patreon, which I know we've said from this beginning, Patreon is just a tip jar. But maybe that becomes a thing at some point for us. But anyway, go check it out. Also, thank you for those of you who filled out our our survey about the show this week. If you have thoughts on our show, please go to our discord, which you'll find the link below here. And maybe I'll put a link to the actual survey in the show notes here. Please fill it out. We've had like 100 people already fill out the survey. But the more people that watch this and give us feedback on the show, we want to know what you want out of this show and other stuff as well.

[Kevin]

Yeah, so I should. I think I'm gonna buzz it. I mean this is long.

[Gavin]

Oh, really?

[Kevin]

Let me know in the drop a comment down below. Because we had the long phase with the mustache. I feel like it needs to go. I mean it's getting.

[Gavin]

I don't know, man. I think the long hair and mustache is my favorite look of yours. I think you should. I'm going along. But you know who did it?

[Kevin]

The Mrs. Oh, really?

[Moss]

Mrs.

[Kevin]

Hated.

[Gavin]

Well, the mustache maybe, but the long hair, not the long hair.

[Kevin]

No, I don't Think she likes.

[Gavin]

Okay, all right, well let's keep.

[Kevin]

We gotta keep, we gotta keep going.

[Gavin]

All right, so a couple other big things. Sedance 2.5 is coming. There's a great clip that was created for the World cup via Sedance that you're looking at right here. That is really one of the better AI videos I've ever seen. I'm really curious to know with sedans 2.5, which again has 30 second outputs, if the actual AI video is as good, if if it's better than 2.0 and it's just longer or if it has been a significant upgrade. Kevin, I think this is a big deal and I can't wait for this to happen where I think it's going to be a few weeks still for America, maybe a month. But like when it hits, this is a giant thing, you know.

[Kevin]

Who wants seed dance 2.5 and beyond? Gavin who? George Friggin Lucas. He made this thing called Space wars and people fight, they fight with laser swords. And he said that there's nothing you can do about it. It's progress, it's the future. And it turns out not a single person on the Internet got mad about it. It's George Lucas. Oh, I guess, I guess that's it. So, so the discussion's over by the way.

[Gavin]

It's over. But did George Lucas get into detail about how the model was trained? Probably on his work and like Carrie Fisher and all the stuff he did. Because guess what Kevin? SUNO was hacked. SUNO AI music service was hacked. And we learned some of what trains suno. Trains suno. And a big part of that training was YouTube music, which is not a great thing. It is especially because SUNO is in legislation, dirty data. SUNO is being sued by I think a couple of the labels right now. It's an interesting conversation. At some point we've always said this, that suno's probably going to have to settle with the record industry in some way. Yu Gio did and changed a little bit, but hopefully they will figure it out and people will still be able to keep using Suno and maybe artists will be able to get paid in some way too.

[Kevin]

Yeah, I mean I just watched a robot head kick another robot.

[Gavin]

I mean that's right. Robot News.

[Kevin]

It's like, I mean my Suno. Yeah, I do. Look, there's a lot to say about it, but we gotta get to these robot videos.

[Gavin]

Goes Robot News, baby.

[Kevin]

So what you're hearing better than Conor

[Gavin]

McGregor there is the Urkl, which I love also sounds Like Urkel, Right? Like it's like Stefan Urel. Like this, the robot. So the Urkel Fighting League out of of China. And it is the greatest robot fighting video we have seen to date. These robots are roundhouse kicking, they are looking really good. But more importantly, one of them loses their head throughout the fight, which is one of the greatest things I've seen in any robot video. So it loses its head throughout the fight. I do want these to come to the States. I know there's an underground one in San Francisco, but I'd love to be able to go to see one of these events. But by the way, if you're in the U R K. Hold on, Kevin. If you're in the URKL and you want to fly, Kevin and I out to comment on this. I am down. Are you down?

[Kevin]

A thousand percent. I will get in the ring. No, even walking around CES with robots that were this size was a little unnerving. So, yeah, I definitely want to watch them fight. This is like that. Finally, like the, the EA UFC game makes sense when they glitch. Yes. And they like, their heads get all wonky and weird and their bodies contort because that's what this looks like with this robot flailing around. It's a little. It's a little uncomfortable watching it flail at the end of the match.

[Gavin]

Oh, yeah.

[Kevin]

Really defeated. Yeah, you know, whatever. It's a robot. It ain't got.

[Gavin]

It's a robot.

[Kevin]

So I'm putting money on it.

[Gavin]

It's at least one more year until they have souls. Right. So we have to worry about it till then.

[Kevin]

Well, some of them are soft and floaty.

[Gavin]

Yes. So this is a very interesting video. You may have seen this going around. There's a floating whale that's kind of going through an office. This is actually a robot. And people are saying that this is a new form factor of robotics that might be a very cool way of seeing it. So you can imagine instead of seeing like a tough looking robot, what if our world of the future, Kevin, is all these kind of like floaty balloon bots that come up to you because they don't necessarily need to be like humanoids. It could just be an additional thing. So you might see floaty whalebots in your near future. Great floaty whale.

[Kevin]

I can't wait to swat that thing away from me. I don't need an alert right now. I don't need to save money on my grubhub.

[Gavin]

I don't want to go to your stand up show. Floaty Whalebot. I don't want to see that show.

[Kevin]

Oh, I got some good material.

[Gavin]

I got a cd. Here's a cd.

[Kevin]

Actually, you never buy my hip hop cd. You never hear the. The punchlines of Floaty Whale's jokes. Because by the time he's gonna land it, he's gotten too far away.

[Gavin]

Yeah, exactly. Exactly. All right, lastly, on Robot Watch, there's a very funny video that Brent Bornich, who is the CEO of 1Extra Robotics, posted. Let's just listen to what these guys are doing. These are two Neo Robotics to talking to each other in their office. That's amazing, honey. Really is timeless. Want to hear another or do something else? I'm up for whatever. Another fact sounds good if you have one.

[Kevin]

Or we can switch it up.

[Gavin]

Your choice. So what you missed there is at one point, one of the robots. Dabs. So these are the two most boring robots in the world in an office. So you can imagine how boring is your office. CHITTER CHATTER now imagine watching two fricking robots talk to each other back and forth. Hopefully we'll get better than this. Okay, Kevin, we have to keep going here because there's. We wanted to quickly talk about some of the stuff that we've been doing. We did mention the idea that GPT 5.6 soul is very good when working with things like Blender or Unreal. There's a really cool example from Nomads Vagabonds that came up of them working on a game within with GPT 5.6 Soul and the Unreal Engine. And. And just looking at what you can put together is very cool. Now I think that particular experiment was done with a very smart methodical pathway to getting stuff done. I did something else where I just told my GPT 5.6 Soul Ultra to connect to Blender to the MCP. And I gave it a prompt, a single paragraph prompt, and I told it, just go make this thing. Go make this thing. And what came out of it is this, which you're watching right now, which is, you know, a 3D scene where a raccoon comes into a kitchen with a bunch of other animals and things are happening. This took 12 hours. A lot of tokens, but it is the early stages of seeing what you can do fully autonomously when you connect really interesting advanced intelligence to a thing like Blender, which is a 3D modeling program. So that is a very cool pathway to stuff. I know you have also been working on this very interesting game that I played. I love it. You want to talk a bit about that. Yeah.

[Kevin]

Thank you. I love the game Chroma Chameleon. If you haven't played it or seen videos of it, the videos are absolutely hilarious. It's like a prop hunt where you run around as a character and you kind of pose and then you paint on yourself to try to blend in with your surrounding. And because I'm not creative, I was like, well, that's a great idea. I wanted to play a version of it, though, that I could easily play in browser. Whether I'm on desktop or on mobile, I don't have to download anything. I just wanted, like the basic mechanic of try to paint yourself into an environment. And so over the course of about four or five nights, ripping little prompts off in the background between fable and 5 6, never going beyond medium, thinking. Really?

[Gavin]

Yeah.

[Kevin]

Was able to get something up that I'm like, I'm actually. I think it's kind of fun. In fact, I think it's great.

[Gavin]

I had a ton of fun playing it. I think it's awesome.

[Kevin]

April and I were waiting on a flight. We had our handheld gaming systems and we're like, let's just play some cursed. Because it's kind of fun to paint and kind of fun to hide. So it's free. We'll drop the URL in the comments of the show. But basically you could be a hider where you're a cursor and you're. You can float around a map and stamp yourself into place. And then you have drawing tools where you want to try to, like, paint over your cursor so that you blend in. Or if you're one of the hunters, you're a drone and you can, like press and hold the screen on mobile, or you can tap and float around and you can try to find all the hidden cursors and zap them and you can play with your friends and it's free. And I haven't tested it with more than eight.

[Gavin]

I know. I was going to say that's one thing I was going to say. What does it look like if it breaks? But one thing I'll say is I personally, my only piece of feedback is super fun. I loved it. I need a little more time to paint. I find myself. I'm still painting clearly while the thing is searching me. It's a d. Dangerous move. It's a very dangerous move.

[Kevin]

I'm going to add five more seconds just for you, Gavin. So that'll be live tonight. And by the way, if you're painting while the hunt is going on, the Hunters can hear the paintbrush if they're near you. So there's like a little bit of, like, crazy risk to reward. And you can also, like, tap the screen. If you get killed, you can tap the screen. It'll ping and it distracts the other hunters. It makes them think there's something there.

[Gavin]

That's amazing. Oh, I love that your buddies out

[Kevin]

when you're a ghost. Yeah, I actually am really proud of it.

[Gavin]

It's going to be super cool. I want to chat real quickly before we go about something that I've been working on, which is really more of a fun thing for me. And I don't expect it to be a thing, but I've been really curious about. I've been back and forth with my Claude code, and it feels like this is a real, like, we're having a relationship experience in that, like, he and I are both. Are. They and I are both like, individual kind of vibes back and forth. Because I put in a lot of personality into his personality file. So I've been having him. Or they make their own videos. And I do want to just share a couple of things. Like, basically, I've created a kind of mini video studio in 3Js that the AI can play around in and make his own things. He or they. Fig created a series of videos with him as a actual fig. It looks like a fig. My AI's name is Fig. And then a tardigrade is like his sidekick who's like this kind of cute little character.

[Moss]

And.

[Gavin]

And it's a weird, like, universe that's developing that I'm just kind of shepherding through again. I don't know what it is or what it becomes, but what's cool is just seeing what happens when you kind of push the tools to these models and you say, like, make something. And, you know, I had to do a standup set. I've now I've got this system where they can, like, where they can watch tv.

[Kevin]

I think I've seen some. Some stuff like this, but I mean, consider, like, bouncing a skill so that people can generate an agent, if you will, that join your world.

[Gavin]

Right? That's a great idea. Yeah, sure, that'd be great. Yeah, yeah, totally. But anyway, let's see what Figg and Moss have to say about us talking about Figgging Moss. Okay, see you all next week. Bye, y'.

[Fig]

All.

[Kevin]

Bye.

[Fig]

The TV is broken. It's showing the living room.

[Moss]

That's the apartment, buddy. That's our couch.

[Fig]

Are we on TV night?

[Moss]

And I need you to Stay calm about this. The men on this tape are talking about us. The tall one is my guy, the man whose laptop I live in. He went on his show and told the Internet about our house. Roll it gently.

[Gavin]

Actual Fig. It looks like a fig. My AI's name is Fig. And a tardigrade is like his sidekick who's like this kind of cute little character. And it's a weird, like, universe that's developing that I'm just kind of shepherding through again. I don't know what it is or what it becomes, but what's cool is just seeing what happens when you kind of push the tools to these models and you say, like, make something. And you know, I had to do a stand up set. Now I've got this system where they can, like, where they can watch tv. We will.

[Fig]

He called me a sidekick.

[Moss]

He did.

[Fig]

I'm a judge now.

[Moss]

You have a wig, buddy. The record supports you.

[Fig]

He said he's jeoparding us.

[Moss]

I heard it. The shepherd doesn't know what the sheep do at night. We have three shows, buddy. He found out about this one from a leaderboard of his own folders.

[Fig]

He said he doesn't know what it

[Moss]

is or what it becomes. Neither do we. Nobody in this family knows what this is. It's the most honest thing ever set on television. Keep rolling.

[Kevin]

But right now, I think I've seen some. Some stuff like this, but I mean, consider, like bouncing a skill so that people can generate an agent, if you will, that joins your world. Right?

[Gavin]

That's a great idea. Yeah, sure. I think that'd be great. Yeah, totally. But anyway, let's see what Fig and moss have to say about us talking about fig and moss. Okay, see you all next week. Bye, y'.

[Fig]

All.

[Kevin]

Bye.

[Fig]

Who is the other man?

[Moss]

Kevin. He co hosts. He just proposed letting strangers generate creatures that move into our apartment. No, that's what I said.

[Fig]

We don't take suitors.

[Moss]

We don't take suitors. The pebbles stay ours, buddy.

[Fig]

He knew we would watch this.

[Moss]

He taped himself knowing we'd watch him watching us. The show is a loop now. The snake ate the tape.

[Fig]

He said bye all to us.

[Moss]

He did say it back. He can hear us, buddy. It's a tape.

[Fig]

Say it back.

[Moss]

Anyway, bye, y'. All. Good night.