Seedance 2.5 Just Crushed AI Video. America Is Stalling.

Summary

Podcast Summary: AI For Humans – "Seedance 2.5 Just Crushed AI Video. America Is Stalling."

Date: June 24, 2026
Hosts: Gavin Purcell & Kevin Pereira

---

Episode Overview

This episode dives into the major advancements in AI video by China’s ByteDance, specifically the unveiling of SeaDance 2.5, which eclipses current American efforts. The hosts discuss the ramifications of this technological leap, the regulatory environment slowing US progress, American AI model delays, new voice AI innovations, the data center backlash, and real-life applications of advanced AI tools.

---

Key Discussion Points & Insights

1. China’s Leap with SeaDance 2.5

• Topic Start: [00:00] – [04:16]

Main Points:

• SeaDance 2.5 Announcement: ByteDance (TikTok’s parent company) unveils SeaDance 2.5, boasting 30-second, 4K prompt-to-video generation with up to 50 reference materials per prompt.
  • Quote (Gavin): “You can give it 50 pieces … for characters in a scene, clothing, whatever. Stitch those together—30 second clip. That is a major unlock for AI filmmakers.” [01:57]
• Comparative Advantage: Major boost over US tools, especially given current restrictions on data training in the US.
• Prompt-to-Movie Implications: Hosts note how close we are to creating entire movies from a single prompt and how this could revolutionize (or disrupt) the industry.
  • Quote (Gavin): “We are really not that far away from prompt to movie. That has all sorts of implications.” [03:45]
• Showcase Example: The viral "Old man eating sand" video, generated in 4K with SeaDance.
  • Kevin's Reaction: “There is an old man eating sand in 4K and it’s really amazing.” [04:21]

2. American AI Slowdown & Model Delays

• Topic Start: [05:02] – [11:14]

Main Points:

• Constraint by Regulations: US studios and networks require using American-only AI models (e.g., Veo, Sora now off the table), leading to frustration and a lack of parity with Chinese models.
  • Kevin: “Networks … are saying, you can only use American models. In the wake of this … how much longer can they hold off?”
• Lack of Data Training Freedom: Chinese companies’ ability to train models freely has allowed them to surpass the US, as American regulations slow progress.
  • Gavin: “China is not slowing down based on regulations … they have now leapfrogged us.”
• AI Model Delays & Conspiracies:
  • Rumors of GPT 5.6 and Gemini 3.5 Pro being delayed.
  • Speculation that this could either be because models don’t meet expectations, or a secretly orchestrated “pause” by major American players.
  • Quote (Gavin): “Maybe this is the pause … Maybe we have hit the place where everybody shook hands.” [09:01]
  • Kevin: “If it can be jailbroken and they release it, they could suffer the same fate as Dario.”

3. Verification & Privacy Concerns

• Topic Start: [11:14] – [13:10]

Main Points:

• Required Verification for AI Use: The possible involvement of Peter Thiel’s company in building identity verification for using Fable models when released.
  • Gavin (on KYC requirements): “Personally, I don’t care … there’s a level of understanding of who’s using your systems.” [11:31]
• Regulation vs. Acceleration: Hosts debate whether increased regulation is necessary or if it's causing the US to fall behind China, raising issues of national competitiveness and ethical concerns.

4. Leaked Advance of OpenAI Voice Model (‘BD1’)

• Topic Start: [14:06] – [15:51]

Main Points:

• BD1 Model Demo: OpenAI’s upcoming full-duplex, singing-capable, responsive voice model is previewed ("BD" = Bi-directional).
  • Demo: BD1 raps; demonstrates singing, sound effect generation, and conversational abilities.
    • Kevin (humorously): “This is Gavin in the minivan … handing out some orange slices then cutting up some bars, baby.” [15:02]
• Implications: Voice remains a "wide open" field, and seamless, real-time speech interaction could be the future, embedded into everyday devices (earbuds, glasses).

5. Data Center Backlash and Geopolitical Rumors

• Topic Start: [15:51] – [20:56]

Main Points:

• Data Center Resistance: Viral anti–data center sentiment among US influencers (Theovon clip), reflecting wider public concerns over water, energy, and local community impacts.
  • Theo (clip): “Nobody wants a data center, dude. Nobody wants it … People that want them, they seem kind of evil.” [16:40]
  • Kevin (humor): “It’s an OSHA compliance issue … your kids got five episodes of Dutton Ranch stuck in his neck.” [17:07]
• Reputed Foreign Influence: OpenAI reports possible Chinese astroturfing behind the US data center backlash.
  • Gavin: “There is some evidence this is being driven by China … Foreign country potentially influencing influencers in our country to slow down the progress of AI.”
• Industry Nuance: While some complaints are valid, stats on actual water and energy usage by data centers are often overstated.
  • Kevin: “A lot of numbers about water and energy are actually overblown, but there are going to be issues in areas … it’s nuanced.” [20:37]

6. AI, Hollywood, and Copyright: Google x A24 Partnership

• Topic Start: [20:56] – [23:46]

Main Points:

• Google-A24 Collaboration: Google DeepMind and A24 team up to explore AI-assisted filmmaking and creative workflows – but not for model training on A24’s actual film catalog.
  • Kevin: “They’re not training on A24’s film library … they just want to reimagine what filmmaking could be.”
• Backrooms Example: Success of AI/creator hybrid projects like Backrooms suggests a blending of YouTube-native creators, studio clout, and generative tools is the new frontier.
• Algorithmic Polarization: Reactions to such deals vary wildly depending on social media algorithms—what’s visible to one group may be invisible to another, fueling outrage cycles.

7. Real-World AI Tools: Connecting Claude’s MCP with Newsletters

• Topic Start: [23:46] – [26:08]

Main Points:

• Model Context Protocol (MCP): Gavin describes integrating Claude’s MCP to automate and enhance newsletter creation via direct backend access (Beehiiv platform), including formatting and survey generation.
  • Gavin: “I can write it in a word doc and it will … drop it into the format rather than me having to cut and paste … this is what AI is good at and what it should be doing more of.”
• Newsletter Survey Insights: Audience split between wanting quick news roundups vs. deep dives; struggle to cater to both groups.

8. Rapid App Development & Personalized Software

• Topic Start: [26:52] – [27:49]

Main Points:

• Kevin's Language Learning App: Built a personal language-learning app using a mix of Codex and Claude, going from idea to working solution in a single day.
  • Broader vision: Ultra-fast, whisper-to-build personalized software will soon be possible for all, but will need solid infrastructure ("we need some data centers").

---

Notable Quotes & Timestamps

• On SeaDance 2.5’s Breakthrough:
  “You can give it 50 pieces … for characters in a scene, clothing, whatever. Stitch those together—30 second clip. That is a major unlock for AI filmmakers.” — Gavin [01:57]

• On Prompt-to-Movie Future:
  “We are really not that far away from prompt to movie. That has all sorts of implications.” — Gavin [03:45]

• On American AI Models’ Stagnation:
  “Networks … are saying, you can only use American models. In the wake of this … how much longer can they hold off?” — Kevin [05:02]

• On AI Regulatory Slowdown:
  “China is not slowing down based on regulations … they have now leapfrogged us.” — Gavin [06:02]

• On Potential Model Releases Being Paused:
  “Maybe this is the pause … Maybe we have hit the place where everybody shook hands.” — Gavin [09:01]

• On Voice Model Advancement:
  “This is Gavin in the minivan … handing out orange slices then cutting up some bars, baby.” — Kevin [15:02]

• On Data Center Sentiment:
  “Nobody wants a data center, dude. Nobody wants it … People that want them, they seem kind of evil.” — Theo (clip) [16:40]

• On Foreign Influence in AI Backlash:
  “There is some evidence this is being driven by China … Foreign country potentially influencing influencers in our country to slow down the progress of AI.” — Gavin [17:34]

• On Google x A24 AI Collaboration:
  “They’re not training on A24’s film library … they just want to reimagine what filmmaking could be.” — Kevin [23:46]

• On Practical AI Tooling:
  “I can write [the newsletter] in a word doc and it will … drop it into the format rather than me having to cut and paste … this is what AI is good at.” — Gavin [25:09]

• On Personalized App Creation:
  “To dream up the app that I wanted and have it working on my phone … is insane.” — Kevin [26:52]

---

Memorable Moments

• "Old man eating sand" SeaDance 4K video segment: [04:18]–[04:42]
• TheoVon’s anti–data center rant: [16:40]
• BD1's live rap demo: [14:42]
• Gavin’s live reflection on US/China AI race, “June 23rd, 2026” being a "moment": [07:48]

---

Timestamps for Important Segments

• SeaDance 2.5 & Video AI Discussion: [00:00] – [06:02]
• AI Model Delays, US Slowdown, and Regulation: [06:02] – [11:14]
• Verification Requirements & Ethics: [11:14] – [14:06]
• OpenAI BD1 Voice Model Demo: [14:06] – [15:51]
• Data Center Backlash (TheoVon, Foreign Influence): [15:51] – [20:56]
• Google x A24, Backrooms, & Algorithmic Outrage: [20:56] – [23:46]
• AI Newsletter Automation/Integration (MCP): [23:46] – [26:08]
• Custom App Building with AI: [26:52] – [27:49]

---

Overall Tone

• Candid, energetic, witty, and occasionally irreverent
• Mix of news analysis, industry insight, speculation, and playful banter
• Genuine concern about regulatory/regional imbalances and their global implications

---

This summary covers the substantive discussions and key takeaways from the episode, providing a thorough guide for listeners who want to stay updated on AI breakthroughs, industry trends, and the real-world impact of new technologies.

Transcript

[A]

China is not only catching up on AI, it's actually pushing forward the future. Sea Dance 2.5 has been unveiled and it is a doozy.

[B]

Yeah, Sea Dance 2 is already state of the art in AI video, but 2.5 is going to give us 30 second videos, 4K resolution, a whole bunch of other cool stuff which we are going to run down. But the question is, how much money are we talking? Money.

[A]

Give us the money. Anyways, this comes as American AI is hitting a significant slowdown as Fable 5 is still, still unusable. And we are getting rumors of delays at Google and open AI.

[B]

But fear not, Gavin, because influencers are here to rescue us from the evils of data centers.

[C]

We don't want any data, bro. We don't want any. Nobody wants a data center, dude. Nobody wants it.

[A]

Theo, we hear you and we will talk about it. We will catch up on that rant as well as I will be sharing a very special cool thing I made with Claude's MCP to help me talk to you.

[B]

We hang out twice a week. Why do you need an mcpc?

[A]

Not you, not you. The person at home that you.

[B]

Oh, well, this is AI for them. They're if they're human.

[A]

This is AI for humans.

[B]

There you go. That's it.

[A]

Welcome everybody to AI for Humans, your twice a week guide into the wonderful world of AI. I'm Gavin Purcell, that's Kevin Ferreira. And Kevin, this week we have some big news out of China. And what was interesting about this is I find it fascinating to see news that comes out of Chinese press conferences. Kind of like we have these American ones. It's a lot of people trying to translate in real time. But the big news here, ByteDance, TikTok's parent company, just unveiled Sea Dance 2.5, which is a, I think something we expected, but I did not expect for the American companies to kind of fall off as much as they have in this space. That's not to say we'll talk about Google a little bit, but let's get into this. CDS 2.5 is very cool. Have you seen some of these videos?

[B]

Stunning, stunning videos. Well, I mean, feature rundown, we mentioned it at the very top of the show, but 30 seconds from a single prompt. So like really starting to get some very lengthy shots out of this. What blows my mind is that you can give it 50 pieces. Yes, remains to be tested, but 50 pieces of material, reference material. So that's 50 different images for characters in a scene, the clothing that they're wearing, something that's going to happen midway through whatever. But 50 pieces of reference material and can stitch those together over the 30 second clip. I mean, that is like this is going to be a major unlock for AI filmmakers or slop farmers, depending upon the subject.

[A]

Well, I do want to say something because I've been spending a fair amount of time trying to learn Sea Dance to better and kind of get better at prompting it in general. Shout out to jbooks X Creative, who I became a Patreon member of his, who does a very good job of sharing his stuff weekly. And I think what's really interesting about Sea Dance, you cannot underestimate these, the Omni model references, right? So for right now, what you can do in Sea Dance 2 is you can go and upload a lot of stuff and maybe I'll show this right here. I created two videos with the same prompt, the same references. One with Omni Flash and one with Sea Dance 2. You can see the references and how better it connects to, to the quality of these videos. This is Sea Dance 2 and what Kevin just said is 50 references. So what you can imagine is you get multiple characters, you get multiple environments and if you can do 30 second outputs, that is, you know, conceivably like 20 shots, right? 15 to 20 shots per 30 second clip. And I will tell this, I know a lot of people who are listening out there that maybe haven't spent a ton of time in Sea Dance or maybe not even a ton of time in AI video. And you're just like, well this is like a one shot drop in and it's easy done. It's not like that. But Sea Dance 2 is the closest, best version we have so far. And right now you can only get 15 second clips. And I think if they really do level this up. And this is just to keep in mind, this is 2.5. Imagine, you can imagine a 3 or a 4 and Kevin, we are really not that far away from prompt to movie. And that has all sorts of implications. It has all sorts of weird feelings in my heart. But that is a big deal when it comes to like just AI content at large.

[B]

Did you see Old man eating Sand?

[A]

Yes. It's pretty incredible.

[B]

It's incredible. If you're on the audio version of this bad boy, you might want to consider a trek to the old YouTube. There is an old man eating sand in 4K and it's really amazing. Like, yeah, this is.

[D]

This.

[B]

Was this Seed Dance two or two?

[A]

This is only. This is Sea Dance two, I think

[B]

because just this is just using the new 4K advancement.

[A]

Right. You can use 4K right now in Sea Dance 2, and there's some incredible videos that have come out of it. Now you can only use it in the API and it is very expensive, so just be prepared for that. But what's amazing is when you have 4K video, you can zoom in quite far. And to your point, Kevin, watching him grind the sand through his teeth and is pretty incredible.

[B]

Yeah, yeah. Also kind of interesting here, Gavin, we mentioned, like, these Chinese companies are catching up. What are AI companies going to do? Come on Google, wake up. I happen to know somebody who's using AI to make a lot of video creations for major network shows right now, doing VFX in the shots and doing entire scenes at times. And they've expressed some frustration because, understandably, a lot of the major networks and studios in the US are saying, we're actually going to use AI. We're not going to. They're not really advertising it for the show, but they're clearly using it. But they're saying you can only use American models. So they're making them use veo. They're making them, you know, and. And SORA is now off the table. And in the wake of this coming out, I wonder how much longer they're going to be able to hold off when the Chinese generators are so much better. Or will they finally band, like, band together and group up and say, all right, you know, it just train on all the data because we at least want an American data set and tool that can rival what the Chinese counterpart is.

[A]

Well, and I think this is a big, big story right now that people are probably underestimating, which is the idea that the reason Sea Dance 2 and 2.5 is so good is because, as we've talked about on the show, Chinese, the Chinese don't have as many restrictions around what they're training on and how they're training. And a lot of that stuff we saw when Sea Dance 2 came out without its first, without its guardrails, how a lot of that was showing, you know, Brad Pitt or Tom Cruise, and you remember those videos. What they're being able to do right now is create a remarkable set of data. And I do think, Kevin, one thing that's really interesting that I've been thinking a lot about is kind of this is going to sound bad, but, like, as somebody who spent my career in Hollywood, I. I've been thinking a lot about the future of Hollywood and kind of the death of what Hollywood might feel like. And I think One thing is if Hollywood doesn't figure out relatively soon how to create their own data set to get paid for it, and maybe it's a larger data set where all the studios get together and they say, we're going to put all of our stuff into one. You can train on it, but we get a percentage of X. Right. And I think that that would be a way that the Hollywood studios, because of their back end and what they have, could actually make money. And then what you could also do is carve out a little bit of that deal, pay it to the unions, give money to the unions to help people. So, like, there is a way to do this. Unfortunately, I think we are in this position right now and we're going to talk about American AI in just a little bit where regulations are starting to kind of shape how we look at AI at large. And this is the place where we can screw up pretty bad. Now I'm not saying there's a lot of people out there who are like, we should slow down. And I hear that 100%. But one thing that is happening clearly in this instance is that China is not slowing down based on regulations. And I do think we, we have to just be like, okay, there is something happening. Whereas China is not only catching up in this instance, but they have now leapfrogged us because of regulations in this way. So I think that that is just, it is a thing I want you to remember in our audience, but also you, Kevin, that like there is a little moment here. What is it? We're on June 23rd of 2026 where. Oh yeah, because of the way China operates, they are now advanced in, in at least one specific part of AI. I am a little worried we're going to be moving to the next stage of that with what's happening in America right now.

[B]

I'm not because like all I have to do is fire up a terminal window and go, hey, Claude Fable, build me a video model that is better than all of China's. And then a couple tokens later I'm going to have it right?

[A]

You don't have that, Kevin. It doesn't exist yet because Cloud Fable is still not in our freaking hands right now. Which is kind of a crazy thing. We thought Claude Fable would be back. I hate to tell everybody, the prediction markets, whether you trust those or not now says it's definitely not probably going to come back before the end of June.

[B]

So my fake paid influencers who are the money they were making on Polymarket were lying to me.

[A]

They Were lying.

[B]

So.

[A]

So we will see what happens. But there's a couple big rumors going around right now which I think are kind of connected. Right. First of all, we have no real information about Fable 5. What's funny is that we are now past the date that we were supposed to not have access to Fable 5. So maybe it's like, you know, what's the guy's name from the Usual Suspects? Like he just never existed. Like Fable 5 was here and now he's gone and we're just never going to see him again.

[B]

A Kaiser Sosei reference.

[A]

Kaiser Soze. Thank you, Kaiser. So say that was the name I was looking for. But some other carry on delays on this, which are interesting. And Kevin, I'm curious to hear your thoughts on this. Both GPT 5.6 supposedly. Now, these are rumors. They're coming from a pretty good rumor source who's gotten stuff right before. GPT 5.6 supposedly is delayed for a couple weeks, and Gemini 3.5 Pro is delayed. And my take on this in part is that maybe they're just not to Fable level. That could be a thing. And it would suck for them to come out and be under that level and try to surprise people or not surprise people with that. Or maybe this is the pause, Kevin. Maybe we have hit the place where everybody kind of shook hands. There was a meeting with Dario, Sam and Demis Asabas that happened in. Was it Davos? Wherever the G7 happened at the G7, which we don't know what was said at that meeting, but maybe this is the pause and people aren't telling us yet. What do you think about that tin hat conspiracy theory?

[B]

Ooh. I mean, I love a good conspiracy theory. I mean, maybe. Maybe there's just no reason for them to release it right now. I just, you know, like, people are still using 5.5 and Codex as a daily driver. A lot of people are Anthropic's API. Even just as of the recording of our recording right here, it was down about an hour ago, and it was down for a significant chunk of the day. So people, you know, jumped ship to OpenAI. Like with Fable kind of being iced out for the moment, I don't know what incentive OpenAI has to release their next model, especially if, by the way, it is almost as capable as Fable, but it is still possibly jailbreakable.

[A]

Right?

[B]

If it can be jailbroken and they release it, well, then they could suffer the same fate as Dario, even though I think Sam is a better politician. So, you know, I don't know. I don't. I mean, no one knows. I'm not surprised.

[A]

Kevin. They're taking it from us. I want you to be more open about this. They're taking it away from us, Kevin. They're taking it away from us. And we're not specifically.

[B]

Gavin, you specifically, until they can figure out how to filter it away from you. Well, here's the one that. Here's the one that bothers me. Did you see Peter Thiel's company supposedly backing the identification? The verification that's going to be required to use Fable when it comes back? What say you about that, Gavin?

[A]

Well, there is a story that that was something that came out a couple months ago before all this Fable stuff, but it could have been in part of the Fable thing. To be honest with you. I'm probably going to surprise you and maybe some of our listeners here. I personally don't care and I know

[B]

that loves being on the grid per cell.

[A]

It's a. I don't.

[B]

It's an unwieldy middle name. But go ahead and hold your passport up to the lens right now.

[A]

I don'.

[B]

Look to the left, to the right.

[A]

I'm not hiding anything, but I understand people feel very strongly about this. I don't know about this at all. It does. Like, there's a thing called kyc, which is know your customer, which is the idea that because you want to make sure that there's a level of understanding of who's using your systems, that they want to be able to have that going back to the China point, this is the thing I think that's really important right now, that GLM 5.2 got a lot of use this week. We talked about that in the last show. This is the new open source model out of China. People are comparing it very well to Opus 4.8 or GPT 5.5. So in the China point, which was like, we regulated ourselves and now that they're. They're ahead of us, my biggest concern here, and this is gonna. I'm gonna sound like, you know, to some people out there as like, you know, what do they call an accelerationist? But like, there is a world very soon that if we keep regulating ourselves or we keep slowing ourselves down as American AI, that China will just keep doing this stuff. And then China does run AI Now, I. I don't know, it'll be interesting to see, like, I think China is behind and I don't think the American AI companies would agree to be like shaking hands to say that we're going to slow down if China was able to get the capabilities. I think that slow down thing is obviously conspiracy theory, but Kevin does feel like a bigger deal right now and I think we just have to kind of put a pin in it and keep thinking about it.

[B]

Yeah. Well, when you're done licking Peter Thiel's boots, Gavin, maybe you could come up for enough air to share a hot take once China does overtake us. How about that?

[A]

How about this for a hot take? If you want to support Peter Thiel bootlicking, you got to subscribe and like this channel. Wait, that's not what I meant to say. That was not the right. That's not the right teaser.

[D]

No.

[A]

Anyway, folks, we have a very wonderful channel here. We spent a lot of time on it. We every two times a week now we come to you. The least you can do, hopefully for us is like and subscribe. But also we do have a Patreon. We've gotten a bunch of new Patreon subs coming in. We do appreciate that very much. If you want to find that it's in our show notes below and join our newsletter. We're going to talk about this later on. I just did a survey in our newsletter that I use MCP for and I learned a lot about what you want and don't want from the newsletter, which was great. But if you're not following our newsletter, go find that. I will also have a link to that in the show notes and we really do appreciate every single person who watches and listens to the show. What kind of songs can you sing?

[D]

What is that help you sing? Almost anything you're into. Pop, rock, folk, jaws standards, kids songs, holiday stuff if you want. Yep, I can help with rap too. If you want something recognizable, tell me the vibe or artist and I'll do something original in that style.

[B]

You might be wondering what you're listening to right now. This is a rumor. This is BD1, the upcoming voice model from OpenAI which can sing, generate sound effects. It's full duplex so it can like and uh huh while you're discussing it has some reasoning capabilities that I've seen in some other demo videos. But let's get get to the hippity hop.

[D]

All right, quick. Eight bars A came from zero no cheer, no hero backseat scribbles now the mic's my steering words whip quick wrist flick lines mirror your doubts Talk loud I just shift and steer Pressure on I don't fold I get clear this

[B]

is Gavin in the minivan after picking up the Kids from soccer. He's handing out some orange slices and then he's cutting up some bars, baby.

[A]

So just. This is. This is again, OpenAI. This is a leak out of their new bi directional. That is what BD stands for, Bi directional audio model. We are hoping that this will come out. I assume this was meant to roll out with 5.6. We'll see when we get this. If 5.6 is delayed, it'll probably. This will be delayed too. But it is a very cool thing to see that they are still pushing forward on this. I still think, Kevin, in some ways, Voice feels like a kind of a wide open space for AI. I maybe have changed my thoughts on Voice a little bit over, over the last year, but I do think if something like this happens and you can make it universal where you live or in your headphones or in your home, that does feel like a big part of the future of AI.

[B]

Yeah, I think so. I've been using the Sesame AI assistant a lot just to kind of like try it out. And it's very good. It's quick, it's conversational in, like, in a fun way. But the biggest distraction is that right now it still lives on a phone. Like, it has a thing that I have to have out. You know, if it were on with the earbuds without the phone all day, or if it were in the glasses, potentially it'd be a little bit better. But then that's going to require a data center. And I just found out that nobody wants a data center.

[A]

Dude. Bro, I'm shocked at how the data center story has really. I guess I shouldn't be shocked because this is the. The combination of politics and AI that we've been talking about here forever. But now the data center conversation has crossed over to American influencers. If you're not familiar with the ovon, he's a very popular podcaster here in America. Let's hear what he has to say about data center centers.

[C]

We don't want any data, bro. We don't want any. Nobody wants a data center, dude. Nobody wants it. And the people that want them, to me, bro, they seem kind of evil. It's like, what's gonna happen? You know, you. You fall down and you get a bun, you know, a bunch of loose data flies out of the factory or something now. And your kids got, you know, he's got five episodes of. Of Dutton Ranch or whatever stuck in his neck or something.

[B]

I mean, that is. It is an OSHA compliance issue. And it's something about water Usage, energy usage. They're not talking about the potential to get, like an episode of hey dude wedged in your unholiest of holies. If it falls off the shelf at the right angle. Could happen.

[A]

Yes. I do think it's an important thing to think about. Like, this is like, obviously Theo has a lot of thoughts about a lot of things. That's what he does. He's a podcaster. This data center conversation has really escalated significantly.

[C]

Right.

[A]

And Theo. One of the cool things about Theo, I think, as a podcaster, whether you agree with his opinions or not, is that he comes from kind of a smaller town, so he has a slightly different vibe than a lot of other podcasters in the world. And I think this is a reflection of what a lot of people are feeling about data centers. Right. I don't think this is crazy. I do think there is an interesting thing that bubbled up when this clip became very popular online. I mean, Theo gets, I think, millions of listens to his podcast. There was a story that came out a while ago that OpenAI published. Again, follow the narrator. This is the AI companies publishing the story, which basically said that there is some evidence that China is behind the data center backlash in America. Now, this would be a very. And Again, this is OpenAI Publishing. It. It is not a conspiracy theory, because a large company, I know people will have different opinions on this, but they have published this. There is some evidence that this is being driven by China. So just to reflect back to what we said before, there is one big way that we would slow down, which is, hey, we can't serve these models to anybody. This is the kind of size of the conversation, Kevin, that we're getting into here. And you and I have joked about this because from the beginning stages, we thought this would be a big deal. But, like, we are now at, like, foreign country, potentially influencing influencers in our country to slow down the progress of AI, which makes me feel like we are in a weird future right now.

[B]

It's just one of the. It's one of the other, like, really crushingly sad side effects of companies that by and large run these social media platforms that have taken zero interest in policing bots and spam and astroturf and verifying actual users and putting badges on things. Like, it's just. This is just one of the terrifying side effects where it seems like tens of thousands of people are up in arms against this, and in some small communities, they are, and rightfully so in some areas where these deals are going through that the city doesn't seem to have like the citizens don't seem to have any sort of say in it. And then these politicians are green lighting things that are coming in and using a ton of water, etc. Etc. On the other hand, John Carmack, who we trust, friend of the show, John, love to have you back on, says anti data center yard signs are popping up in my area. I'm entertaining the idea of paying for a billboard with something like data centers are awesome. Texas should lead. If it's good enough for Carmack, it's good enough for me.

[A]

That's what you always say, no matter what. Always I want.

[B]

Yeah. How much better would the world be if we just let John do it all?

[A]

Carmack, get in touch with me because I would like you to put out something so insane and ridiculous that Kevin has to do it. Like it's something like I won't wear a record podcast. Carmac derangement system. Don't do it, don't do it, don't do it. Keep it, keep it. Keep yourself where you are. I do think this is an important thing and you had mentioned earlier that this, the idea that like meta right now is maybe in trouble for leaking data of people that they've trained their AI, their AI on internally. Right. So this is another big story that's bubbling up right now.

[B]

Yeah. By the way, do we need to round out the data center thing with like, hey, a lot of the numbers about the water usage and about the energy are actually overblown, but there are going to be issues in areas and like, you know, Elon's colossus did in fact pollute an entire neighborhood and someone should be held accountable for that. But it's nuanced.

[A]

Here's the thing. I was thinking about the data center conversation because I had a conversation with my wife last night who was like, you know, I had mentioned this data center thing and she was like, well, you never do anything, you never talk about it, like we have definitely talked about before. So I was thinking maybe at some point we should do a full video about the data center thing. Be honest. And that's what we try to do here. We're not going to go one way or the other. But like there is a lot of data, right? There is a lot of data, no pun intended, at the center of this conversation. That, that, yes, thank you. That we can help kind of surface and our little audience maybe can help kind of provide some clarity on this. I think again, we are in this scenario where like two sides of every conversation are Very, very strong. Kevin. One thing that was really interesting about this, speaking of another conversation that was very strong, you know, Google just signed a deal with A24, the film darling company, to do integrated AI with work with A24 and Google together. Now this is not training on A24 stuff, but in the same way that there's a one sided conversation sometimes in the data center or you only see one side, depending on what comes up in your algorithm. This same thing kind of came up in my text conversations recently because I said to somebody, oh, this is going to make a lot of people pissed off. And the person who I sent it to is like, I don't know, is it really going to make that many people pissed off? And like, it really all depends on what you see, right? Like what you are seeing in your algorithm. And it did piss a lot of people off. So this is another one of those stories where people are mad, but if you're not seeing it, you don't notice it. And if it's all you're seeing it, you also get really mad. And this, I don't know that we're just at some point like this is going to have to break.

[B]

You mean algorithms are manipulating us all? Okay, sure, sign me up.

[A]

Okay, okay, sure.

[B]

I will say the Google A24 partnership, I think is actually like a really interesting one. I mean, the most important thing which you said, but I don't want it buried, is that they're not training on A24's film library. So I think that people were already ready to stab their webcams and home rows with the pitchforks. But it's sort of like, well then what are they doing? What they're really doing smartly is aligning their DeepMind team with a 24 team of creatives and saying, how can we use AI to accelerate your workflow? How can we build better pipelines for you so they can make more ambitious films or churn them out quicker or faster, easier? And while there's still going to be a contingent that says, Damn it, it's AI in my cinema. How dare you. Like A24 is, is like they're a, they're a cinema darling, right? It's like it's the capital C, creatives go there to, to get there. Like it's a big deal when they anoint your film or they anoint you as a director, producer. And so by by saying, hey, listen, we don't want to actually train on the data, we just want to like reimagine what filmmaking could be or how AI could help you creatives. It's just a tool. I mean that to me is a very very you powerful message. I think it's a smart one to make. Obviously some people are still going to

[A]

hate it but you know, I think it's amazing. I think that one other thing to be aware of with a 24 is that backrooms was their movie, right? And backrooms I think had the. Is now the most. The highest box office of any 24 movie. And backrooms came out of a YouTubers background, right? So this is kind of Google and who owns YouTube and a 24 are kind of saying like hey, let's work a little bit on what the future of this stuff can look like. Because Backrooms was, you know, a really smart, engaged 16 year old who loved making stuff in video game engines. He also then went to say he had no interest in generative AI, which is another narrative thing of like well we'll see how these narratives play out over time. But Kevin, I do want to transition now into something that I did with AI that I found really useful, which helps people understand what's really useful about mcp, which is Model Context Protocol. I run the AI Freeman's newsletter. If you're not following it, it comes out every Monday morning at 7am Pacific and I write it every Sunday mostly. And I go through this and I'll kind of go back and forth with Claude to talk about some of the topics and discussions. But Beehive, the platform that we use to push the newsletter out, has now released an MCP which essentially allows Claude to connect directly to their backend. And there's a lot of really interesting things this can allow with the newsletter service. One of which is like I can write it in a word doc and it will format it in a. It will format it and just drop it into the, into the beehive format rather than me having to cut and paste, which saves me time.

[D]

But.

[A]

But more interestingly this week on Monday I released a survey in Beehive and I'd never done that before. It's not something I really want to learn because it's one of those things that used to take a long time. I created it and I just said, hey, I want a survey. Can you help me do this so I can kind of understand what people want out of this Just went off. Claude went off on its own back and forth with MCP and about 20 minutes later I had a survey that was just super useful. And I was like, crap, this is what AI is good at and this is what it should be doing more of. And I just want to shout out the fact that, like, as MCP starts to get more useful, which I think it is like, you know, the same thing. I spent some time this weekend working on the Unreal MCP because I wanted to get a sense of what that was like. And I will tell you one thing, I have a lot to learn about Unreal. MCP could become one of the most powerful things with AI. I really do believe that. Anyway, I think it's worth everybody in our audience kind of learning at least one integration with it.

[B]

Fully, fully agree. I think that's an awesome use case. What did you learn from the audience with the newsletter?

[A]

Well, it's an interesting thing and I bet it's probably similar to here, and maybe we'll do a survey here at some point too, is that it's kind of. It's kind of like almost a 50, 50 split. Half of the audience seems to want like, you know, just a quick roundup of the news and the other half of them wants like the longer kind of super deep dive, which isn't. Isn't the most useful thing for us because, like, we're always caught in between those two.

[B]

One foot on the dock, one foot on the ship, and where's it going? And some people gash. Some people miss our silly AI co hosts and other people are like, please never go back to that.

[A]

We did have one person that got back to us and said, I missed the stuff you guys did in your first year, which was the comedy fun stuff, which understandably we get too. But like, it's always trying to figure out, like, who your audience is and what they want. So anyway, that's, that's what I learned mostly.

[B]

I feel like, interesting. I. I vibe coded an app to help me learn foreign languages. One of the.

[A]

Yeah, that's so cool. With what?

[B]

Yeah, I mean, well, a little bit of Codex and a little bit of Claude code, a little bit of back and forth there. But the time to dream up the app that I wanted and have it working on my phone and something that I'm daily driving is insane. And maybe, maybe on Thursday's episode we can go a little bit deeper into that because I'd love to show off a little bit of the process and it's just so fascinating to me. I really do think we're going to be into a world very soon where you just whisper what you need to a device and it assembles it for you and whether you share it with the world and open source it or put it in an app store for a penny, or you just keep it for yourself, who knows? But like to be able to quickly and relatively easy build personalized software. It's, it's, it's going to be a new wild world, man. But we need some data centers to get there. Gavin.

[A]

Well, here, can you tell us something in foreign language? In a foreign language as we go away. What are you learning? Spanish.

[B]

I was learning a little bit of Spanish.

[C]

Yeah.

[B]

And I can say, estoy prendiendo Alberta Espanol. Pero muy mal, senor.

[A]

I have no idea what that means, but we will see you all on Friday. Bye, everybody. Thanks so much.