← All Episodes
AI for Humans
Google Is Cooking Again. The I/O Leaks Are Wild.
Thanks to @HPInc & Intel for sponsoring us! More on the Zbook Fury https://bit.ly/4uapNHs Google I/O is next week and the AI leaks are pouring out: a new Spark agent, Veo 4 Omni, Gemini 3.2 Flash that's reportedly 20x cheaper than GPT-5.5. This week on AI For Humans, Google is cooking again and th
Show Notes
Thanks to @HPInc & Intel for sponsoring us! More on the Zbook Fury https://bit.ly/4uapNHs
Google I/O is next week and the AI leaks are pouring out: a new Spark agent, Veo 4 Omni, Gemini 3.2 Flash that's reportedly 20x cheaper than GPT-5.5.
This week on AI For Humans, Google is cooking again and the I/O leaks are stacking up. We dig into Google Spark, a new Gemini agent that may have access to your entire digital life. Veo 4 Omni model leaks suggest deeper reasoning and character consistency, and the model gets math right. Gemini 3.2 Flash is rumored to deliver 90% of GPT-5.5's capability at a fraction of the cost and dramatically faster speeds. There's a new GoogleBook with Gemini built in. And Google is reinventing the mouse cursor, the input device that's been largely unchanged since 1968, with voice AI.
Plus, Thinking Machines dropped voice interactivity demos that feel a lot like ChatGPT Voice from two years ago. OpenAI is reportedly already working on GPT-5.6, and Sam Altman is giving away two free months of Codex to companies to drive adoption. Gavin's been experimenting with local open-source LLMs and shares his setup.
AND…we get into the data center sickness conversation: infrasound from data centers may be causing cortisol spikes in nearby communities. Figure 03's package sorting livestream proved the robot is autonomous after skeptics accused it of being teleoperated. Unitree dropped a transformable robot.
AI KEEPING US UP AT NIGHT. NO MATTER. WE COOK.
// Show Links //
Google Spark: Gemini's Agent With Access To Your Life
https://x.com/kimmonismus/status/2054855742247584231?s=20
Veo 4 Omni Model Leaks: Gets Math Right
https://x.com/TomLikesRobots/status/2053845600051798065?s=20
More Veo 4 Omni Examples
https://x.com/testingcatalog/status/2053718756799467735?s=20
Omni Model Added To Gemini Web Build
https://x.com/testingcatalog/status/2054196983523393857?s=20
Gemini 3.2 Flash At 90% Of GPT-5.5 For Way Less
https://x.com/kimmonismus/status/2054887891222802633?s=20
New GoogleBook With Gemini Built In
https://x.com/Google/status/2054270454467121187?s=20
Google DeepMind: Rethinking The Mouse Cursor With Voice AI
https://deepmind.google/blog/ai-pointer
Thinking Machines Voice Interactivity Demos
https://thinkingmachines.ai/blog/interaction-models/
Sam Altman: Two Months Of Free Codex For Companies
https://x.com/sama/status/2054626219858293128?s=20
Data Center Sickness: Ben Jordan's Video On Infrasound
Figure 03 Package Sorting Livestream
https://www.youtube.com/live/luU57hMhkak?si=KZHwUdYUwY4SIRUp
Brett Adcock: Figure 03 Was Not Teleoperated
https://x.com/adcock_brett/status/2054737974710169840?s=20
Unitree Transformable Robot
https://x.com/UnitreeRobotics/status/2054067819634159622?s=20
Transcript
AIForHumansGoogleGeminiLeaksIO
===
Gavin Purcell: [00:00:00] Google's big I/O event is next week, and rumors are leaking like ooze from the goo.
Kevin Purcell: Mm, lovely visual. We've got new video model samples which show deep reasoning and character consistency. We have a new flash model that's quicker and perhaps up to 20 times cheaper than GPT 5.5, and a new Spark agent that might know everything about your life.
Everything.
Gavin Purcell: And they're using AI to reinvent something that you use daily and has remained mostly unchanged since 1968.
Kevin Purcell: Now see, that's a juicy tease. Can we get a little hint,
Gavin Purcell: Gav? Uh, no spoilers, but it's the mouse cursor.
Kevin Purcell: Okay, I don't think you know how sp- I don't think you know how spoil- Oh, uh,
Gavin Purcell: oh, Kevin, Figure 03 is sorting packages longer than you can stay awake.
Kevin Purcell: Well, I will be sleeping soundly inside my mecha Unitree as we stomp around and terrorize entire cities, Gavin , so, you know.
Gavin Purcell: All of that and more on this episode of AI for Humans.
Stomp, stomp,
stomp[00:01:00]
Welcome, everybody, to AI for Humans, your twice a week guide to the wonderful world of AI. We are here- Not so menos. Yeah, yeah. So every once in a while. Yeah, there you go. Uh, eh, this week, Kevin, we have some big leakage. There's leakage from the goo. Yeah. Ugh. We got some ooze leaking from the goo. We are always talking, uh, a little bit ahead of things that are happening here, and one of the fascinating things about the AI space is you do hear bubbles of news coming out of large events.
And next week is Google's big I/O event, and we know that we have been following this for a couple years. You might remember a couple years ago there was the famous AI, AI, AI Sundar Pichai video.
AI and AI, AI, AI, generative AI, generative AI,
generative AI. Last year we got a really cool look at v03, and Kevin, there are some really big things that they are teasing, well, not officially teasing, that are leaking about next week's event.
Kevin Purcell: Yeah, grab your wooden flutes. Let's go to the ren fair. Bard is- Ooh ... coming back, baby. [00:02:00] Oh, no . No, no, sorry. That's for- That's what they should use
Gavin Purcell: as their slogan. I know. Grab your wooden flutes- ... and join Bard on the AI march to win the world.
Kevin Purcell: Hey, don't, don't close the browser tab yet. We do have actual news.
We'll get to it right now. Google Spark is the rumored agent, Gavin. Yes. People were seeing pop-ups. It might have, uh, leaked a little too soon, but Google Spark is their agent that has access to all of the things, and it makes complete sense because w- every other agent seems to have access to all of your Google.
Yes. So if you want an agent that can work in the cloud 24/7 that has access to your Gmail, your calendar, your Google Docs, et cetera, that's what this agent is promising.
Gavin Purcell: Yeah, so I think one thing that's interesting about all these leaks is they are coming out of kind of like these people that just suddenly opened their Gemini and it was like, "Oh, there's a thing there."
And we're gonna talk about the omni model coming up as well too. Kevin, to me, Google Spark is a thing that Google should be able to own, right? So we've been talking for the last couple months about the movement towards personalized AI agents. [00:03:00] Obviously, OpenClaw kind of began this conversation in a big way.
I think a lot of people have, have been struggling with OpenClaw, which sometimes can be the case with open source software when especially- Yeah ... it's on the bleeding edge. What Gemini has the opportunity to do here is to create something that feels all-encompassing of your life. Because you and I have talked about this before, uh, G Docs is how I do my writing and work.
I often have, um, Sheets up for Google Excel. Yeah. Version, a version of Excel. Gmail is where all of my... In fact, I have, like, six Gmail accounts now because I'm doing a couple other jobs. So if there really is something that can bring all these things together, that feels like it's a pretty significant place for Google to live.
Kevin Purcell: Look, a, a lot of people use the Claude desktop app, and even now- Yeah ... you know, Codex or ChatGPT, they have their own connectors or plugins that will let you access a lot of this Google stuff, but you have to download that separate app. You usually have to pay, you have to authenticate, you gotta click through, you gotta authorize a bunch of stuff.
It feels weird. It still requires multiple clicks. This is a chance for [00:04:00] Google to provide agents to people where they don't have to know what a command line is, they don't have to know what a plugin or a connector is. They could just- Yeah ... click the button and ask, and then there is all of their info, um, put in there.
Mm. I wonder how opt-in it will be, how opt-out it will be. I wonder if you will have to pay for an Ultra plan for it, or if they're just gonna- Oh, yeah ... you know, just go for it and subsidize the cost of this thing like so many companies are.
Gavin Purcell: I mean, for any company that could subsidize cost, Google should be subsidizing cost.
Yes. 'Cause Google and Meta are the two mon- still the two money printers of the internet business, and, like- Google has the opportunity here to, no pun intended, claw back a lot of actual AI usage. There's a couple other things we should talk about that are kind of leaking out ahead of this event, and I want to apologize ahead of time to the Google IO team because I, I, I said I could attend Google IO.
I will not be attending Google IO, as much as I wish I could. Well, this is the breaking news, Gavin. Yes. Let's
Kevin Purcell: get to it. This is, yes. This is the story people- This is the story ... clicked on our YouTube video for. They want, they want to know- Give them the exclusive ... Gavin's
Gavin Purcell: life. Give them the exclusive. I will not, I will be [00:05:00] traveling down to Los Angeles during that time, but thank you to Google.
But first of all, Kevin, the other big thing that I'm very excited about that is going to come out next week, we think. Now, again, these are just things that have flitted into the world. This is the rumor mill,
Kevin Purcell: yeah. Um,
Gavin Purcell: yes, this is the rumor mill. And people, it's more than rumors because we assume that these are things that are popping up, and we've seen some things.
There is a new video model coming from Google that we think is going to be Veo Omni. Now, we don't know that's the name of it, but it seems like that's the case. And again, somebody has opened their Gemini, and supposedly there's a thing here. And Kevin, what this is, is a new video model that does a lot of reasoning behind the scenes.
Now, we've talked about reasoning video models before, and obviously Seedance 2 is what I still think, I've, I've been kind of toying with it. I'm trying to think of some things that I can do with it, but it is still by far the best at action and other things and physics in the world. Something that Seedance 2 fails at all the time is writing.
So if you have a scene- Mm ... where you're doing, like, a big active scene, and you put in the background you're in a town, or you say you're in Paris or wherever you are, [00:06:00] um, what you'll see on the signs in the background is gibberish, like you would in kind of old school AI image gen. What this, the leaks of this model are showing is that maybe they have focused on making all of that make sense, and there's a video that I want you to play a little bit of here that we'll listen to, but for the viewers on YouTube will be able to see what's interesting about this.
Play this video about the math professor that's doing, uh, uh, an equation on the board.
AI Video: We start with the fundamental identity, sine squared plus cosine squared equals one. Now, if we divide every term by cosine squared, we arrive at the identity for tangent.
Gavin Purcell: Okay. Okay. Now- Hold on. So Kevin, this is way... I understand all this stuff, so I wanna make sure that you're understanding what's going on here in this math video.
Well, I was just gonna say I don't want a
Kevin Purcell: weird flex because people know that I, you know, I was a bit of a mathlete in my day. You were, yes. You
Gavin Purcell: were, yes.
Um,
I was. So I do wanna be, yeah. I was.
Kevin Purcell: So I don't wanna flex on the math. We don't know if the math on the board is correct. Well, no, we do. We do. We are both too lazy.
Oh, we do? Because
Gavin Purcell: there are, there are people out there who have confirmed the math on the board is correct. So just to explain to- What? Just- Yes. [00:07:00] Yes, yes. So I'm not saying I confirmed it. Oh, we, we can confirm it's real- We're not journalists, Kevin ... because some random bot on X- No. No, no, no, no ... said that the math checks out?
More th- more than random bots. These are, these are multiple people have said the math checks out. And here- ... here's what I wanna say.
Okay. The thing that's
going on in this video, there are small problems with it. When you watch it, you'll see the chalk at one point kinda jumps to the next thing. But- Mm-hmm
yes, according to the world at large, and this is not just one random person, multiple people have said, it is doing the math right on the board. And you can imagine a world where if it can get the right text and the right math formulas right, it opens the door to a much larger understanding of the world.
And Kevin, we know that with, uh, Gemini, and Veo, and also Genie, don't forget about Genie, which is their world model, that Google has been thinking a lot about this idea of like a unified model that can see the world. Yeah. So I will say like, there's another clip which we can show here, uh, on video, which is two men kind of sitting down and eating some spaghetti together.
Th- supposedly this was somebody trying to recreate the Will Smith spaghetti thing, but it didn't allow him to use the [00:08:00] word Will Smith, so this is a version of it. I think it's fine. It's definitely better than Veo 3, but if you just look at the raw video, I don't think it's as good as Seedance 2. No. Like, there's things in it that don't feel as strong to me.
Kevin Purcell: No, someone in the thread actually posted a version of the same prompt generated by Seedance, and it kind of blows it away. Um- Yes ... but that said, you know, listen, okay, the, the whole point of the Omni model is that you won't be going to, uh, a drop-down and selecting generate video. Yes. You won't be going to a separate tool to make audio.
You won't be going to a separate tool to do the reasoning of the math. This is supposedly an all-in-one model, and so that in and of itself is impressive. Um, uh, you know, I'm looking at like, uh, the, again, the goalposts have moved because y- Yes ... some people are going, "Well, the-" Hugely moved. Hugely moved, yeah.
Like you said, "Oh, the chalk jumped a little bit," and the, the, the chalk itself isn't winnowing down properly when- Yeah ... he's using it. I'm like, okay, this is where we're at now with these models. Um, crazy, crazy impressive, and I'm w- again, curious to see if they will [00:09:00] subsidize the living heck out of this.
Gavin Purcell: Yeah, so I think we have to look at a couple quick things with this world that Google's kind of putting together. To the point of their, uh, Gemini agent Spark, like that idea that like they can bring all this together, to the Omni model, again, they're trying to bring all this together. The other thing that they're trying to do, I think, Kevin, is there's, there are a few rumors floating out there that we may not be the new Gemini, that they're gonna drop Gemini probably 3.2 Pro at this event, may not be like pushing the edge.
But I don't think that's what they're trying to do. The more interesting rumor, again, that's floating out there is Gemini 3.2 Flash. And what's interesting about this is- By far the more interesting rumor, yeah. Yes, yeah. So Ge- Gemini 3.2 Flash is their cheaper, faster model, and you and I both know that the Gemma models from Google have been also been very good for small models.
There are rumors that are floating around about Gemini Flash that it is way lower than GPT 5.5 costs at about 90% of the capacity. And so far, according to the people that have like tried it, quote, unquote, it is also much faster. And you and I have been talking about this forever. When we first tried Cerebras or we tried these fast [00:10:00] models, part of what is the tr- the tricky thing with AI is how long it takes to go back and forth- Yeah
with it. If you get near instant answer, if you get near instant answers in a world where you're getting this sort of thing, that is a huge, huge, huge deal with good intelligence.
Kevin Purcell: You will take 90% of the quality of response or accuracy of the response. You will take that because you can generate multiple responses or have a mixture of experts distill their responses, and then take 10 of them and smash them- Yeah
all together in the time that it would normally take for one generation. And when you, again, rumor mill, but when you cut costs by, like, 20%, you make it 20% cheaper, suddenly this becomes... Like, uh, if you don't have enough RAM to run a capable model locally, this could become your daily driver. This could be the model- Yeah
that powers, uh, that spark agent in the cloud, right? Or maybe this will be the model that powers the new Google Book, which is clearly- Yes ... trying to, a rebranding of the Chromebook- ... that is gonna run a lot of AI locally, um, because now the operating system is [00:11:00] serving up AI as opposed to just launching apps and managing memory.
Gavin Purcell: Did you ever try owning a Chromebook? Did you ever have that experience?
Kevin Purcell: Um, I worked on a Chromebook once, and that was- Yes ... fine. It was fine for the time. The cloud apps are good. Yeah. But, you know, I, I re- when I used it, it was still tough to get, like, files off of an SD card at one point.
Gavin Purcell: Yes. Yes. Yes. I was gonna say, it's been a while, and I, I have to say, like, the Chromebook always felt to me like it didn't feel powerful enough to do the thing I wanted it to do, and it kind of felt like, I guess, I...
I know a lot of kids in schools get them, right? They were handed them for a long time. Mm-hmm. Like, that's a cheap computer to hand out to kids. Still, this is exciting, right? We're starting to see hardware designed with AI built in. Now, a lot of hardware manufacturers are kind of, like, rushing to throw AI, uh, things into their actual devices, and I'm, I'm more interested in, like, more powerful onboard processing than I am necessarily, like, Gemini intelligence.
But it is cool to see the idea that there will be a computer that's built from the ground up. And Kevin, this also transitions into a pers- kind [00:12:00] of weird piece of big news that Google also dropped, and it came from Google DeepMind, which is always interesting to me when a Google DeepMind thing drops. And this is- Yeah
a team that's been working on rethinking how you use your mouse cursor with AI. And you may hear that and you, you may, like, be like, "Get AI out of the dumbest thing in the world." "I don't wanna hear how to rethink my mouse cursor. I know that. My fingers understand how to use a mouse." Right. But really, this is kind of interesting.
You wanna kinda tell people what the basics of this are?
Kevin Purcell: Yeah. It's, it... Imagine attaching an AI assistant to your mouse cursor so that in success, it is contextually aware of what it is you're trying to do on the screen at any time and can offer AI-powered suggestions. What's interesting is that that DeepMind teaser came out, right?
Like, "Oh, hey, we're experimenting with this thing." And then I think less than 48 hours- Yeah ... it was announced as a feature as part of this new Google Book. So you just kinda shake your mouse, which if you do that on a Mac, sometimes it makes your cursor giant if you have that installed. Um, you'll shake your mouse, and it will- I did.
I did.
Gavin Purcell: I've never done that before, Kevin. There you go. You've brought life [00:13:00] to me. I just shook my mouse.
Kevin Purcell: Oh, look at that. Yeah, it, it'll, it will help you find your cursor. But here it will turn your cursor into sort of an AI enhanced cursor- What does that mean? That means you can select things on the screen, and it will contextually offer you some things that you can do with it.
So if you shake it over an image, it could offer you editing for that image, or you could grab that image and drop it on another image and say, "Stitch these two things together." Or select a to-do list and say, "Create a widget to track this and share it with the family." So it's, again, it's putting AI exactly where your cursor is- Yeah
and trying to make it contextually aware. I think it's... I mean, well, look, the proof will be in the proverbial pudding. Oh. I think it's really interesting. I think- The go-
Gavin Purcell: the Gooding? The G- the Gooding,
is that what's Google Pudding called? Google , I don't know. I just tried to say something and it just-
garbled out of my mouth. Did you hear that? Like, I could barely get a word out. The Gemini
Kevin Purcell: Jello, Gavin. Oh, there you go. That's pretty good. But you can use AI, you can use AI to, again, do anything with that cursor. You can create dynamic widgets, which they have examples of. I'm gonna [00:14:00] keep talking. I'm not gonna let you get in- Keep talking, keep talking
with whatever this is. Jello is Jello- And, and also- ... it's
Gavin Purcell: just Jello with a G. Jello with a G, Kevin. It's just Jello with a G. Let's keep going. That's right.
Kevin Purcell: He did it. He did it, everyone. I did it, I did it. And he didn't need AI to get there. Remember that. Um, and also, last thing about these Google Books is that, you know, uh, what's, what's interesting is that they're kind of doing the mirroring on an iPhone thing.
Yeah. You can mirror your phone elegantly and quickly, and you can run Android apps on it. So I don't know, I think the Google Book is, is suddenly pretty interesting for that, if they're going after- Yeah ... like, it's a budget, uh, budget notebook, but it has all these AI features in it. Totally. I think it's interesting, and we've said that AI is gonna make its way to the OS level.
Yes. Here it is.
Gavin Purcell: Yes. Here it is. And also, the other thing that AI, I think, speaking of the, the thinking, the pointer, is that you can talk to it, right? And I think the other thing that- Yeah ... that happened earlier this week that really got quite a bit of coverage, but also was an interesting story, is that Thinking Machines, which is Mira Murati's startup, released basically a series of videos on the new set of voice models that they are working on.
And Kevin, the funniest thing about this, I don't know if you caught this or not, the first guy in [00:15:00] these videos is actually the original guy from the GPT-4o, uh, launch of the voice models, where it's the same guy who had- Yeah ... kind of a, it seemed like he was getting kind of attracted to the voice he was talking to.
But this guy's working for Mira Murati now. But what's cool about this, Kevin, and I think you and I with our work with AI voice really understand this, is that what they're trying to do is make- The feeling of an interaction with an AI voice model feel way more natural and normal, so you can interrupt- I got, dude-
it's listening and doing stuff in the background and- I went
Kevin Purcell: so, I went so in the weeds on this model. I went so- Oh, you did? ... in the weeds on it. I was fascinated by it. Yeah, so basically- Did you, it's downloadable,
Gavin Purcell: is it? You can't use it, I don't think.
Kevin Purcell: No, no, no. No, no, yeah. But I read all the papers- Yeah, yeah
and I even looked at, like, the, the, the, the, the tech that inspired their papers. Yeah, yeah, yeah. Like, I went really deep on it, 'cause I was, I was fascinated with it. It is- Yeah ... um, truly bidirectional. Like, it do- Yep ... most, most of these models, they take your input, they usually convert your audio into a transcript.
They feed- Yes ... that into the machine. Yes. You get the output. It then waits. It's doing input and output simultaneously. Yes. Every 200 [00:16:00] milliseconds there's a tick, and it's doing it. Um- Which is amazing when
Gavin Purcell: you're working with AI audio, because that is a thing that really slows you down if that doesn't happen.
Kevin Purcell: Yes. Yes. And it can interrupt the user. It can talk over you, which is kind of important. It can- Yeah ... it can wait. It can lay back in the cut until it's activated. Um, it is multimodal, so it can process, um, video as well as audio at the same time. Um, it uses, like, an interactive model as the main model- Yeah
to handle that interaction, but then it also can access a smarter model in the background for deep reasoning and things like that. But the audio, uh, I didn't know this, it makes a heat map of your audio, Gav. Oh, wow. So it literally- I didn't know that either. Yeah. That's
Gavin Purcell: really interesting. So
Kevin Purcell: imagine, like, up or down is the frequency, and left and right is sort of like the, uh, like the, um...
What is it? Tone? Like the, the, the volume or, or, or like- Yeah, for sure. So, oh, sorry, sorry I, I did... And so time goes left to right, and then the frequency is top to bottom, but then there's a brightness or a color, the intensity- Oh, interesting. Interesting ... is, like, is there a lot there or not? And it couples that with the transcript, so it can actually [00:17:00] tell the performance, and that's why it can emote.
It can tell if you're happy or sad. It can shift. It's actually looking at, like, a heat map of the transcription and the audio. I think it's super cool. This is the small version, the one that they demoed. Yeah. It's the small version of their model. We don't know how capable the full thing is, but an interesting first shot out of their cannon.
Gavin Purcell: Yeah, and we also don't know when it's gonna be available. I looked into that a little bit. Right. You know, Thinking Machines has had some struggles where people have left. In fact, there was a story this week where, like, a bunch more people just left because their shares vested after a year, so they, like- Yeah
jumped over for a bit, and then they left. But I am excited to see how this goes because I do think this kind of ol- overall omni model idea and the Google Book and all this stuff, this idea that, like, you'll interact and be part of your computer in a different way than has happened before- Mm-hmm ... is big.
And I think, you know, something that just, right, that's happening right now, I think, um, there may be a Codex u- update today that's something to kind of be aware of. There's been teasing, a new OpenAI's been teasing a Codex update for today. But Codex is one of those things that I have seen a lot more people starting to dive into and [00:18:00] use.
And I do wanna just quickly mention that, like Sam has been talking about, Sam Altman has been talking about this idea that they're trying to bring Codex to a ton more people in the world. And in fact, Kev, I thought this was crazy, and if you're out there and you're a owner of a business or you're a small team, they're giving teams now two months of Codex use for free, and all that compute for free to try to onboard more people right now.
So this all kind of just fits into this much larger idea that we are kind of entering a stage where, like, this is just gonna be part of our lives. It's just gonna be kind of, uh, omniversal around us at any given time.
Kevin Purcell: Ooh, I like that. Well, Gavin, what if you're one of those people out there that doesn't want to upload all of their data to the old cloud and rely on one of these massive companies that can giveth and then taketh away?
Gavin Purcell: Well, you know, Kevin, there's a really great series of open source models that if you want to use local AI, you can do that on your own, and I'll tell you how right here. Are you-
Kevin Purcell: Oh, are you setting up a clip? I thought you were- I am. I was just asking you a question. It sounded like you were tossing to a
Gavin Purcell: clip.
This is a clip. I'm about to do [00:19:00] something I would never do with a cloud AI. That's feed it my personal data and let it yell at me. And I can do that because I've got this HP ZBook Fury cranking away at my desk. I mean, cranking on my lap. There are powerful laptops, and then there's this. Big thanks to HP and Intel for sponsoring AI for Humans this week and hooking us up with this massive, amazing machine.
Powered by a Core Ultra 9 V processor with 256 gigabytes of RAM and an Nvidia RTX Pro 5000 Blackwell GPU, this thing handles AI like almost nothing I've ever seen. Today, I'm running Ollama with Qwen3 8B, which is one of the best open source local models out there right now, entirely on this machine. No cloud, no API calls, no data leaving my desktop.
Here's why that matters. If you wanna run a private LLM of your own stuff, your journals, your contacts, your finances, you really don't want any of that sitting on someone else's server. And with the ZBook Fury and local AI models, it doesn't have to. The model runs directly on this computer. Everything works locally, and no one ever has to phone home.
And because I'm me, I did [00:20:00] not use this for something sensible. I fine-tuned Qwen3 8B to be the most aggressive, deeply unhelpful motivational coach possible. I call him Big Tony. Big Tony has read my calendar, and Big Tony does not approve of my calendar. Again, the wild part is that none of this actually leaves my desktop.
Big Tony lives here. Big Tony cannot be subpoenaed. Big Tony is mine. Local LLMs, private data, zero cloud, that is the workflow. Check out the link in our description below if you wanna build your own ZBook Fury, and again, thank you to HP and Intel for sponsoring AI for Humans. Thank you, everybody, and also go follow us.
That's right, like and subscribe to our YouTube channel. We have a Patreon. We have a newsletter. We have all that stuff. Uh, go find us. Oh, also hype this video. We've been talking about this for a little while, and everybody in the bass is like... Somebody said, "Don't ask for people to comment. Ask for hypes."
Okay, your job today- Oh ... uh, AI for Humans watcher, is go hype this video. I don't know how to do that. You are smart. You can find that out. Go for it.
Kevin Purcell: Yeah, and if you have trouble hyping it, just leave a comment and let us know. Sure, yes. But I think it involves, uh, waving a white towel around your head like a helicopter.
Just shake
Gavin Purcell: your mouse cursor [00:21:00] really fast and see what happens.
Kevin Purcell: That's right. Kevin, we gotta talk about this computer. Gavin, listen, I, I love running local models. I love running local models, but I do have a complaint, if I may. Sure. They don't make my neighbor ill
Gavin Purcell: Oh, well that's, that's good, isn't it, if they don't make your neighbor ill?
Kevin Purcell: Okay, okay. Someone doesn't like data centers. All right. Okay. I really interested this- Uh, teacher, you forgot to assign homework. What a dork
Gavin Purcell: You went deep on this story, and I am very interested in this- I did ... 'cause this is like, it is probably the biggest story outside of the models in AI right now. I don't know if you caught this, but, uh, Bernie Sanders and, uh, AOC just introduced- Yeah
a, a bill to try to ban data center development, uh, countrywide. I think this is a bad idea. I'll be very clear about that. Um, I think we need to see- Whatever ... the AI race. But- Whatever. I'm
Kevin Purcell: gonna take the monorail to get around Kevin O'Leary's data center- Oh, wow ... which got approved in Utah. There you go, okay.
And it's bigger than Manhattan. Look, there's a lot of data center hate going around. A lot of it is justified. Yes. Communities don't want it. Sure. It can pollute the water. Yep. It can drain a lot of energy. Uh, there's a lot of, like, council members [00:22:00] basically taking bribes and driving around in souped-up Humvees, uh, in order to get these data centers in their areas.
But here's an interesting reason to hate them as well: data center sickness. And I wanted to just- Yes ... quickly shout out Ben Jordan. Uh, I went deep on his YouTube channel. He has some phenomenal videos, like, really well put together stuff, uh, about technology, a lot of times leaning on the darker side of tech.
Um, but Ben has one on something called infrasound, and I wasn't even- Mm ... aware of infrasound. I mean, I think I was spiritually, emotionally- Sure ... subconsciously. But infrasound is whenever there's a, a sound with a frequency lower than 20 hertz, lower than most humans can hear, it's like a, it can cause rumblings, and it can literally shake the walls.
You can see it- Oh, wow ... in, like, glasses of water. Think Jurassic Park stomping around- Interesting ... like the dinosaur. Um, but this low frequency sound, though you can't hear it, you can absolutely feel it. Right. And it can cause cortisol spikes. It can cause restlessness. Oh. Interesting. It can cause anxiety.
It can cause a whole lot of stuff. And he's done some amazing work exposing infrasound pollution, [00:23:00] because I think that's a real thing. Yeah. He's measuring it. He's interviewing people, like, boots on the ground, whose lives are getting upended by it. And guess what, Gavin? Let's play a little guessing game.
I'm gonna give you one guess. It's got one... It's called One and Done. It's a brand-new game show. Play the sound effect. I love it. One and Done. Let's see it. Here we go, Gavin. One and Done. Okay, great. Guess, guess where a lot of infrasound pollution is being generated.
Gavin Purcell: Kevin's, uh, bathroom.
Kevin Purcell: Ooh, I'm sorry.
You're not a winner on One and Done today. No, everybody can hear that. Yeah, okay. It spans the entire audible spectrum, unfortunately, and it does shake the foundation. But we were looking for- I'm sure, yes ... data centers, Gavin. Of course, yes. Okay. Unfortunately, data centers. The, the, the, it happens around crypto mining, a lot of- Yeah
you know, it's the same stuff, right? It's big generators, uh, kicking in- Yeah ... and, and, and, uh, y- diesel gen and whatnot to, to provide enough power. So it's yet another problem. It's one that we probably should be regulating and solving. I don't see a lot of efforts happening there, but I- Can I ask a question?
Please. I
Gavin Purcell: have a question about this, 'cause I did not go deep on this story, and [00:24:00] obviously the data center story is going to be, I think, the largest public-facing AI story- Yes ... in the next couple years. Yeah. Here's my biggest question about this, and I, and I understand, I'm not, I'm not saying this is fake, and also I- obviously the world at large, we understand that things that don't seem believable on the top of your brain, like of course these things really happen.
Yeah. My question is, like, when you go back to something like the Industrial Revolution- Mm-hmm ... was this the same sort of thing that like, say, a factory does? Or is, how is it differ from, like, a normal factory or a manufacturing plant? Because when I was, uh, 18 I worked in a place called the Longview Fiber Mill, and the Longview Fiber Mill was a paper pulp plant.
It was a job I had after school. Hmm. Very hard job. Mm-hmm. My grandfather got it for me because he's like, "Hey, you do this, you'll understand why I want to go to college," and I did. Ah. But my question is, how does that, how does da- do data centers differ in this from a normal factory, and does that, is this, is it a different sort of thing than, than, uh, what a normal factory would be?
That, that would be my big question. I'm not saying... I, I also believe that, like, data centers should probably be put in remote places. I mean, we [00:25:00] have a lot of land on this world. Like, it makes sense. Right. I know they have to build the facilities up, but like, it doesn't d- it doesn't... Like, I just wanna make sure I have a good sense of that, I think, is what I would ask.
I
Kevin Purcell: think that's great. Yeah, I'm, I'm, I'm barely qualified to, uh- ... be on YouTube, Gavin, so I don't know that I'm qualified to- Okay. Uh, j- just sincerely. To, to- Maybe somebody in your audience does to current infrastructure? I don't- Yeah ... you know, I don't know. But, I mean, if you look back at, like, the air and water and the pollution and urban filth of the Industrial Revolution- Yes
and then you, and then you extrapolate, oh, the scale with which people are talking about- Yes ... these projects, right? Yes. You look at even- Yeah ... the pollution that the, the Colossus data center had, right? Yeah. Running all these turbines that were already against what little- Of course ... regulations we have. I, uh, I, I'm sure it's, I'm sure there are parallels, but I don't know.
It's just the scale and the scope with which- What's that? ... we talk about these d- Oh, uh, Kevin.
Gavin Purcell: Oh. Big data center's here. They have something to say to me, so give me one second. Oh. Okay, I just get one second. No, not
Kevin Purcell: a problem. And hey, by the way- What's that? What's that? ... if big data wants to sponsor the podcast, please let them know.
What's that? What's that? We will edit this out. Yeah.
Gavin Purcell: Oh, okay. They said, they said that they will only sponsor the podcast if we stop talking about this right now. So are you ready to stop- Oh ... [00:26:00] talking about this? Hey, you know what? Okay, great. Great. Listen. Big data, come on
Kevin Purcell: in. Infrasound, infrasound might be killing you and your pets, yes.
But is it worth it? Sure. Did you see the package sorting stream from Figure 03, Gavin?
Gavin Purcell: That's exactly right. What do you know? That's a perfect transition, Kevin, to talk about- Vibrate me, daddy.
Kevin Purcell: The
Gavin Purcell: 24-hour now, uh, package- It's gonna be rough ... sorting stream from Figure 03. So Figure 03, we have talked about here before, is a humanoid robot company that comes out of America.
It's an American, uh, company, and I think it is live still right now. What you are watching is one of their robots sort through packages, and this may look very boring, but according to Figure 03, this is completely autonomous, what's happening. Yeah. There's been some controversy, which we can get into that, but it's g- autonomous, and you're watching it flip packages so that the labels are, I think, down, right?
'Cause it's trying to get the packages right. Yeah. 'Cause it's gonna go- This- ... over a barcode
Kevin Purcell: scanner- Yeah ... at some point or some type of scanner. And this is a job that,
Gavin Purcell: like, humans have done. It is not a great human job, so it's not a- Yeah ... terrible thing to replace this job. But this is the beginning stages of seeing what it looks like to run a [00:27:00] robotic factory because- Not only did they do it for eight hours originally, but then they said, "Speed it up.
We're doing it for 24." Right. These things don't get tired. They just, they do the same thing over and over again, and it is fascinating to watch a live stream of it happening.
Kevin Purcell: Yeah, especially when, like, the robots, as they get closer to, like, running out of battery, they will- Yeah ... like, s- step back, and there's an interchange, a supposedly fully autonomous one where the- Yeah
next robot will sort of tap in and go. Um, I love that they've added name tags. I'm watching Frank right now go through packages. Yeah. The, the best is, is hanging out with, with a live chat as they lose their minds over a box flip- Yes ... because the box flips are the best, right? It's, they- Yeah ... they'll do the, kind of a two-handed flip to get the label down.
Sometimes they drop packages. Yeah. Sometimes they actually don't have the barcode in the right area. So it's not flawless, and watching- Not flawless ... chat pop off when that happens is just a bizarre, bizarre thing. But the teleoperation, real quick, there were rumors that these things were actually being teleoperated because there's a clip where the robot...
Yeah, that's it. That's [00:28:00] it. It looks like the robot's reaching for its visor, which is one of those things- Yes ... that a teleoperator may do at the end of their shift. Some people were saying it's, it's from the training data. Others are saying it's, it's clearly a lie. Um, Brett Adcock, the CEO of Figure, came out and said, "No, the robot's lifting its arm so it doesn't hit the metal chute- Yeah, yeah
that's to the side of it." And, and I don't know if they... And I, like, I'm watching now, I'm watching the, the live stream as we're talking. Yeah. And it several times has made that motion. I don't know- Interesting ... if they told the robot to go ahead and do that a little bit more- Do it, do it more ... to prove the point, but yeah.
Like, but... Oh, we just got a box flip, Gavin. We're 29,000 packages in. Oh. We're seeing so many box flips. Oh, we did it. Fantastic.
Gavin Purcell: Fantastic.
Uh, anyway, so this is a really, just a good example of, like, we are moving quickly into the robot world. Um, there was some conversation ear- earlier online about the idea of, like, all these manufacturing plants.
Speaking of manufacturing, for, like, car plants, like that Ford stock just shot up because maybe there's a world that Ford as a, as a company that can manufacture things quickly and has lines to manufacture stuff might be an AI stock too which is another- Mm-hmm ... whole big AI [00:29:00] stock bubble question, which is a big thing brewing right now.
But Kevin, speaking of AI robots, there is a very cool new robot from Unitree, and Unitree has been building forever, but they have finally done it. They have built what is a walkable Mecca that you can sit in, in a cage. Yeah. But not only- Yeah ... do you sit in it when it's walking, it can transform into a four-legged walker.
Yeah. And the funniest thing to me about this is... So if you're not watching this on video, it is about probably, like, 12 to 13 feet tall. It is very much like the first stage of, like, the big mech, uh, you would see in anime. But when it transforms, what it does is it kinda makes you go backwards, so you're sitting back and your head's looking up.
And then you hear it clomp. Maybe you can just, maybe you can just play the clomps for people- Sure, yeah ... so they can just hear the clomping.
Kevin Purcell: Uh, hey, royalty-free music rock heads, get ready.
Gavin Purcell: So that clomping you hear is actually the [00:30:00] clomping of its feet going up and down as it tries to walk around. Anyway, this is just a crazy, cool thing to see Unitree doing. And again, uh, you may have seen the President of the United States is in China right now. There's been a lot of really interesting stories about, like, where Chinese robotics are.
There's a great clip of Fox News's Bret Baier asking for a sausage, which we can't play here. For a sausage? He's asking for a sausage- Oh ... a- from a robot. But- Oh, I am bummed that we can't play that ... we can't play that here 'cause we will get demonetized. But anyway, very cool robot news.
Kevin Purcell: You know what we can end with, Gavin?
Gavin Purcell: What's that?
Kevin Purcell: A robot asking Bret Baier for a sausage.
Gavin Purcell: Oh, you have that?
Kevin Purcell: No, but I mean, we could Seedance it.
Gavin Purcell: Okay, fair enough. I'm out of credits, but- We'll see you all, we'll see you all next week. Thanks, everybody.
AI Video: Bye-bye. Bye. Good evening. Excuse me, sir. Do you have a sausage?