Google's AI-First Laptop, Meta's Spy Games, AI Monks in Middle America

Transcript

The tech workers of the world are sitting at home while their agents are running, whispering to themselves all day? [SPEAKER_04] Yes. [SPEAKER_04] And those that do work in person now have these coworking spaces where everyone's whispering to their computer. [SPEAKER_03] And it's super weird. [SPEAKER_03] Oh my God, guys, this is so weird. [SPEAKER_01] Your hands can only type so fast and it's just more context per second. [SPEAKER_03] Guys, I hope we still talk to each other. Do we have podcasts for that? [SPEAKER_00] We'll debate the tech that's best when we get more or less. Dave and Britt plus Sam and Jess put it all right to the test. More or less. Why hello, friends. Welcome to More or Less. The gang's all here. Hi, everybody. [SPEAKER_02] Hey, Jess. We're back. Jess, you and Britt are both looking very nice right now. [SPEAKER_04] Thank you. I can't handle how nice Sam is these days. What's going on, Sam? Is it peptides or something? Something changing in the diet? [SPEAKER_02] I got a lot of supplements and I got a NeuroPod. [SPEAKER_04] What's a NeuroPod? [SPEAKER_02] It's awesome. It's this thing that I was marketed on Instagram that clips on your ear and stimulates your vagal nerve. [SPEAKER_01] Oh, yes. It's wild. I've tried one of those. They're pretty wild. [SPEAKER_03] What is the outcome of the NeuroPod? Oh, I'm glad you asked, Britt, because it is tracked on the Oura Ring. I was super jet lagged. I was up at three in the morning. Let's do a hit of the NeuroPod. And so I tried it and it's wild. My Oura Ring, my HRV went through the roof. It doubled. My heart rate's dropped. [SPEAKER_01] Because it's terrified because you're electrocuting yourself? [SPEAKER_01] Yes. [SPEAKER_04] Yes. HRV is supposed to be low though. [SPEAKER_01] You want—No, no, no. You want high heart rate variability. [SPEAKER_01] Yeah. [SPEAKER_01] You want high HRV, Britt. [SPEAKER_03] I want low heart rate, high HRV. [SPEAKER_01] High HRV. Okay. [SPEAKER_03] I'm going to step back and sum this up because this actually was topic number one. We've got Google hardware. We've got IEO. We've got a company called Microsoft we never talk about on the pod. I have a lot going on. But I do want actually to hear the greater public's opinion on whether Sam should be stimulating his vagal nerve every night to diminish his stress receptivity. But it seems appealing. It seems appealing. [SPEAKER_01] It is a well-trodden path to regulate one's nervous system. Jess, you're the wife. Is it working? [SPEAKER_04] I just need to know what the outcome is there. [SPEAKER_01] Yeah. Do you feel better? Do you feel un-jet lagged? Yeah. I felt great waking up. [SPEAKER_03] Really? Sam's heart rate—I mean, you can see it in the charts. My second topic of conversation is my new Oura ring, which it turns out is not a fitness tracker, folks. [SPEAKER_01] But your heart rate tanked. Did you just upgrade? [SPEAKER_03] Yes. His heart rate tanked and his variability spiked exactly when he was done doing it. And he said he felt great. I just want to know how much is too much. When do we get into One Flew Over the Cuckoo's Nest territory? [SPEAKER_04] At least Sam can be our guinea pig and we'll just do the things that work after that. Some vagal nerve stimulation is good for everyone. You can do it in many different ways. [SPEAKER_03] Is Sam doing it 10 minutes a day? Dave, what's some here? Yeah, I need a recipe. [SPEAKER_01] I don't know what the actual recipe is, but parasympathetic and sympathetic nervous system stuff is regulated through the vagal nerve. And it's one way to cold plunging. There's all these different ways that you can shock the system. [SPEAKER_02] I got it all. I got the cold plunge. I love the cold plunge. I got the vagal nerve. I got the red light. I just ordered. I ordered so many supplements while I was on the airplane. [SPEAKER_04] Sam is like Brian Johnson 2.0. I do respect Brian Johnson. I once was very proud. He published his score on the Brian Johnson test having done all this stuff. [SPEAKER_03] I knew this would get here. Yeah. And I took his test and I was half a point off him. And I messaged him. I was like, look, I do nothing. And yet here I am half a point behind you. And then he was like, no, no, no. The public things are old. I'm much lower. [SPEAKER_04] By the way, I'd like to say there's a white space opportunity because there's no real female version of Brian Johnson yet. I think there will be. That's going to happen soon. The women that are doing the biohacking to the level he is are all women. [SPEAKER_01] Isn't his girlfriend doing that? [SPEAKER_04] No, but she's not—most women I know couldn't name who that is. There's no one with the popularity. Women are naturally good biohackers. You have to be, right? You don't need to be mansplained to, Brett, about our longevity and our— [SPEAKER_03] Let's not be mansplained to, Brett, about our longevity and our— [SPEAKER_02] I'm down. I'm happy to keep going. It's not clear to me that—Do women want to experiment on themselves constantly? Yes. We do it with our faces a lot. [SPEAKER_02] Do you want to look like Brian Johnson? [SPEAKER_04] Nah, but she's not like most women I know couldn't name who that is. [SPEAKER_04] There's no one with the popularity. Women are naturally good biohackers. You have to be, right? [SPEAKER_01] You don't need to be led anywhere. [SPEAKER_03] Let's be mansplained too, Brett, about our longevity and our- [SPEAKER_02] I'm down. [SPEAKER_02] I'm happy to keep going. [SPEAKER_02] It's not clear to me that- [SPEAKER_02] Do women want to experiment on themselves constantly? [SPEAKER_04] Yes. [SPEAKER_04] We do it with our faces a lot. [SPEAKER_02] Do you want to look like Brian Johnson? [SPEAKER_04] We do all kinds of things to our faces. [SPEAKER_04] And I think women, this is the tip of the iceberg. [SPEAKER_01] They probably do want Brian Johnson's skin. I would guess he's got the ultimate skin, apparently. Yeah. Skin is the way into women. Cellular rejuvenation is the thing that's happening and I'm really excited about it. [SPEAKER_02] The thing is, do women want to live forever or be young forever? [SPEAKER_02] Because Brian Johnson just wants to live forever. [SPEAKER_04] Women don't want wrinkles. [SPEAKER_02] Because wrinkles are a sign of poverty. [SPEAKER_03] And we don't want saggy stuff. [SPEAKER_02] Should we talk about tech? Well, yeah. Okay. So guys, I don't know if you- The people wanted a different topic. [SPEAKER_01] Exactly. Everyone's like, please stop talking about what you talk about. So, all right. [SPEAKER_01] We'll talk about skin. I do want to say because I pulled the internet on whether I should get a Whoop or an Aura ring. And I got a lot of thoughtful feedback. And the dirty secret is I had both all along, but that would not have been a good pull if I hadn't. [SPEAKER_03] But I wasn't using both. [SPEAKER_03] Decided to go Aura ring. [SPEAKER_03] Got the rose gold. Doesn't bother me wearing it all the time. But I did a 30 minute advanced Peloton tread HIIT workout that it labeled housework. And said it was moderate activity. [SPEAKER_02] Is that because you're a woman? No, I don't know. It was like, oh, people are doing housework. [SPEAKER_03] That's what I was thinking. [SPEAKER_01] Is it trying to algorithmically encourage you to be a trad wife? [SPEAKER_01] Is that what it's doing? [SPEAKER_03] That's what it felt like. [SPEAKER_03] Because honestly, I was running. [SPEAKER_03] I'm not really. [SPEAKER_03] I was running at a nine on the right, which is not slow. [SPEAKER_03] That's really fast. [SPEAKER_01] Yeah. [SPEAKER_01] Housework. [SPEAKER_03] And it told me it was housework. [SPEAKER_03] And I had to walk 12 minutes briskly to meet my activity goal. [SPEAKER_04] I think your Peloton is trained on Sam Lesson's dataset. [SPEAKER_04] No, no. [SPEAKER_04] This is my Aura ring. [SPEAKER_04] This is my Aura ring. [SPEAKER_01] I love Aura ring. [SPEAKER_01] I was a seed investor. [SPEAKER_01] I'm up a hundred X on that. [SPEAKER_01] Just look. [SPEAKER_01] The cookie is the number two angel investor in the world. [SPEAKER_01] Guess what guys? Dave got demoted from number one to number two. [SPEAKER_01] Sam, I am out. [SPEAKER_01] I'm official a hundred X DPI on Aura. [SPEAKER_01] Personally. [SPEAKER_01] A hundred and twelve X. Let's burn some of that jet fuel. [SPEAKER_03] But Dave, were you going to say something about the state of their health fitness tracking? [SPEAKER_01] No, I just wanted to lick the cookie. [SPEAKER_03] I am going to put my Whoop on and start to track my activity with Whoop. [SPEAKER_03] I also love that Whoop is pro tennis, so I'm going to support them. [SPEAKER_03] But I also pre-ordered and here we get into the tech news of the week, folks. Google's new hardware because they launched a Whoop competitor. [SPEAKER_01] Jess told me about it, so I bought it, even though I'm not going to use it. [SPEAKER_03] I bought one too, and I'm excited. [SPEAKER_01] I did actually see that. [SPEAKER_01] I actually wanted to buy one. [SPEAKER_01] I thought it wasn't available yet. [SPEAKER_01] Yeah, pre-ordered. [SPEAKER_01] I pre-ordered. [SPEAKER_01] I pre-ordered. [SPEAKER_02] You don't do pre-orders? [SPEAKER_02] Don't you like an Apple fanboy? [SPEAKER_02] Don't you stand in line for Apple shit? [SPEAKER_03] No. [SPEAKER_03] No, Apple hand delivers. [SPEAKER_03] Tim Cook hand delivers Dave the latest Apple product with freshly baked cookies. [SPEAKER_03] Or John Ternus. [SPEAKER_03] Sorry, you got a new messenger. [SPEAKER_02] Did you send John Ternus your address? Because he might not have it in the transition. It's in the CRM. [SPEAKER_03] They've already pre-briefed everyone on the transition with Dave. Brit, we are on by my account, the ninth attempt to revive Google Health. So we should just at least notice that. [SPEAKER_04] Yeah, but here's what's happening under the hood. [SPEAKER_04] I have some Intel. It does look good. It looks good. I want to try it. [SPEAKER_04] Okay. [SPEAKER_04] Former Googler here. [SPEAKER_02] That's Brit's looking good. [SPEAKER_02] He's like, I made a hundred times my money on moving food. [SPEAKER_02] It's like, did you know I worked at Google? [SPEAKER_04] I just want to remind Sam, the man who never bought Google stock during the AI boom of the last two years. [SPEAKER_04] So there you go. [SPEAKER_01] But I did. [SPEAKER_04] And I'd like to remind everyone that I'm still longest on Google out of everyone in the ecosystem. [SPEAKER_04] Okay. [SPEAKER_04] Google Health was a thing when I worked there 20 years ago. [SPEAKER_04] It's gone up and down over the years. [SPEAKER_04] However, with the AI revolution that's happening right now, Google DeepMind has an enormous interest in health as do all the other frontier labs. [SPEAKER_04] And there's a lot of activity happening under the hood. [SPEAKER_04] Multi-billion dollar companies being created and invested in by DeepMind. [SPEAKER_02] Only single billion dollar companies. [SPEAKER_02] What is this? [SPEAKER_02] Farming? [SPEAKER_04] There are a lot of biotech things going on, drug discoveries, this Whoop band. [SPEAKER_04] Google's now, as we are going to talk about today, starting to launch new AI first hardware. [SPEAKER_04] Google Health was a thing when I worked there 20 years ago. [SPEAKER_04] It's gone up and down over the years. [SPEAKER_04] However, with the AI revolution that's happening right now, Google DeepMind has an enormous interest in health as do all the other frontier labs. [SPEAKER_04] And there's a lot of activity happening under the hood. [SPEAKER_04] Multi-billion dollar companies being created and invested in by DeepMind. [SPEAKER_02] Only single billion dollar companies. [SPEAKER_02] What is this, farming? There are a lot of biotech things going on, drug discoveries, this whoop band. Google's now, as we are going to talk about today, starting to launch new AI first hardware. [SPEAKER_04] So watch this space because I think this is going to get very interesting very soon. [SPEAKER_04] And right now, all the infrastructure is being laid by GDM to get there. [SPEAKER_04] GDM? [SPEAKER_04] Google DeepMind. [SPEAKER_04] That's how the cool kids say it, Jess. [SPEAKER_02] What's the deal with everyone being scared? [SPEAKER_02] I heard something from my trainer that apparently the net of the OpenAI Elon trial is that Sam Altman and Elon are terrified of DeepMind. [SPEAKER_02] That's the entire net of the trial. [SPEAKER_03] Well, yes, because it just relives history as documented in the information about that rivalry. [SPEAKER_03] So yes, that has been a theme exhumed in the old diary entries of Greg Brockman. Wait, they were terrified or they are currently terrified? [SPEAKER_01] I'm not understanding the narrative. [SPEAKER_03] Both, both. But it has come up mostly through the history. [SPEAKER_03] Because remember, this all started when Elon, Sam, and Greg were on the same team. And they were all trying to compete with DeepMind. [SPEAKER_02] Dun, dun, dun. Dun, dun, dun. And now they're fighting in court. [SPEAKER_03] By the way, after Altman's latest testimony, the odds that Elon wins this thing have tanked, according to Polymarket and the prediction market. The trial is focused on those conversations. [SPEAKER_03] So that's why it's in the past. [SPEAKER_03] But very safe to assume everyone's still terrified by Google and DeepMind. [SPEAKER_03] And yeah, Brett, I think the health, I mean, both in drug discovery, obviously AlphaFold was one of the first big protein folding revelations out of DeepMind a long time ago now in relative tech years. [SPEAKER_03] But I'm excited for these tracking things. [SPEAKER_03] But what else are we seeing on the AI hardware front? [SPEAKER_03] We're entering what I call developer conference season, or maybe other people call it that too. [SPEAKER_03] So we're going to see a lot of product. [SPEAKER_04] Yeah. [SPEAKER_04] I mean, this week, Google announced a new AI-first agentic computer called the Google Book instead of the Chrome Book. [SPEAKER_05] What? [SPEAKER_05] Yes, it was announced today. [SPEAKER_04] Oh, awesome. [SPEAKER_05] Built on top of Android. [SPEAKER_04] And made to not only leverage Gemini, of course, and all of its products, but to actually do all kinds of interesting things with local compute on the laptops themselves. [SPEAKER_01] I love it. [SPEAKER_04] Yeah, this feels like Dave and his Dell. [SPEAKER_01] Yep. [SPEAKER_01] This is the best. [SPEAKER_01] I'm very into this. [SPEAKER_04] Dell is one of the OEMs that is coming to market with the Google Book, as well as all the other Android OEMs. It's also creating this spark of activity amongst parents and schools nationwide, because Chromebooks have been now distributed to all these schools. And the question is, will Google Books be the successor? [SPEAKER_04] And that creates this whole wormhole of whether parents want their kids using AI in schools, and all that, which we can go down that rabbit hole. [SPEAKER_04] It's causing a lot of chatter so far on the internet this week. [SPEAKER_04] What makes it AI hardware? [SPEAKER_01] So I can answer that question generally, Jess. [SPEAKER_01] We'll see what the actual specs are. [SPEAKER_01] I haven't read it, but what makes it local AI hardware is usually some combination of a CPU with a GPU or a neural chip, and then a ton of RAM. [SPEAKER_01] You need to be able to hold very large models in memory. [SPEAKER_01] So you want at minimum 128 gigs of RAM, usually bigger. [SPEAKER_01] Apple's Mac Studio, for example, at 512 gigs sold out immediately when OpenClaw launched and has never come back into stock because you want to be able to load an entire model into memory while you're doing the inference. [SPEAKER_01] So having some kind of a chip, it could be a GPU or a neural chip. [SPEAKER_01] AMD's got a lot of new tech around this. Close to the huge amount of RAM locally is what you want. Right. So that's saying that some models may have up to 128 gigabytes of RAM. They're using Intel's Wildcat Lake, which is the Core Series 300, and ARM-based chips from Qualcomm and MediaTek. They're running an operating system that combines Android and Chrome OS. [SPEAKER_03] I was wondering about this too, because it's been a minute since we've heard about the OpenAI hardware project. [SPEAKER_03] Unless Dave, you have any intel that you'd like to share on the pod? [SPEAKER_01] That's not really what that project is about. [SPEAKER_01] That project is much more consumer, it seems. [SPEAKER_01] It's much more focused on the ChatGPT consumer and the 30 million people a week that are using ChatGPT. [SPEAKER_01] It's only 30 million a week? [SPEAKER_01] Yeah, that's my understanding is their weekly actives are around 30 million. [SPEAKER_01] That's not good. [SPEAKER_01] The question you're asking is a different one, Jess, which is, if you're going to run models locally, why are computers configured the way that they are? [SPEAKER_01] And the answer is large RAM, fast ability to access that RAM and switch between the CPU and the GPU to do the different operations. [SPEAKER_01] And that's actually one of the reasons why Apple Silicon is extremely efficient for this use case, because they carry the GPU or the neural chip and the CPU on the same chip rather than separate. Just to give context, a standard Chromebook today has 8 gigabytes of RAM. So the new Google book having 128 gigabytes of RAM is a big deal, 15 times. [SPEAKER_03] And the benefit to running models locally is speed. [SPEAKER_01] Speed and cost. [SPEAKER_01] You don't have latency and you don't have to pay for the tokens, right? [SPEAKER_01] My favorite joke of the week is, hey guys, we used to be able to write code for free. [SPEAKER_01] Because everybody has to pay for Anthropic luxury tokens now in order to write code, unless you've got a computer that can do it for you locally. [SPEAKER_04] Sam Altman today tweeted that enterprises who want to move to Codex can get their first two months of tokens free. [SPEAKER_04] So the new Google book having 128 gigabytes of RAM is a big deal. [SPEAKER_03] And the benefit to running models locally is speed. Speed and cost. [SPEAKER_01] You don't have latency and you don't have to pay for the tokens, right? [SPEAKER_01] My favorite joke of the week is, hey guys, we used to be able to write code for free. Because everybody has to pay for Anthropic luxury tokens now in order to write code, unless you've got a computer that can do it for you locally. [SPEAKER_04] Sam Altman today tweeted that enterprises who want to move to Codex can get their first two months of tokens free. [SPEAKER_04] A little competitive warfare happening there. [SPEAKER_01] It's not just competitive warfare. [SPEAKER_01] It's just that they have the compute and nobody else does. [SPEAKER_01] They can use it as a strategic weapon. I know. I thought it was a smart move. I liked it. Guys, Anthropic is getting the compute. They did a deal with Akamai. [SPEAKER_02] Yeah, it has compute. Do you guys remember Akamai? [SPEAKER_02] Yeah, of course. [SPEAKER_02] But I'm happy for them to each just be like two months free, two months free forever. [SPEAKER_02] Yeah. [SPEAKER_02] It's great. It's going to be like the gig economy where you could get your free on-demand dinner, get your free massage, and then it all collapsed because none of it worked. [SPEAKER_02] I mean, this is the whole generation of kids who grew up in the VC subsidized lifestyle. [SPEAKER_03] Yes, subsidized. [SPEAKER_02] And then you had the clean girl aesthetic that came after that because they couldn't afford makeup anymore. [SPEAKER_04] No, that's not what happened. [SPEAKER_04] Women got excited about skincare. [SPEAKER_02] I don't know. [SPEAKER_02] I think they just ran out of money for Soul Cycle. Full circle to the beginning of our conversation. [SPEAKER_03] So this is really interesting and obviously not related, but Apple and Intel are striking. I mean, Intel might be crawling back from the dead. It doesn't matter how you slice it. [SPEAKER_01] You need new types of chips that are very efficient at what we were just talking about and the capacity to make them. And so it makes sense to me that ARM, all of the memory providers, Intel, anyone who knows how to make this stuff is in a really good position right now in the ecosystem. And so I think you'll just see more of it. [SPEAKER_03] So while a different category of AI hardware, the consumer side, I am waiting here. [SPEAKER_03] Do we think, and again, if the information is not going to report it, we would have reported it. [SPEAKER_03] We've reported a lot on the various roadmaps for different devices that OpenAI has been cooking up, but somewhat unclear what we're going to see when at this point. [SPEAKER_01] I think it's also unclear what is consumer and what is developer and where we are in the cycle, to some extent. [SPEAKER_01] One of the things I think that OpenAI and agentic year that is 2026 has become has shown us is that there was a totally different way of thinking about computers and what computers should be and how to use them. [SPEAKER_01] And that was something that wasn't in the zeitgeist at all last year until we saw this new way to think about it. [SPEAKER_01] And so I think a lot of the hardware projects that you saw in the prior two years were things like Rabbit. [SPEAKER_01] And I mean, if you go back, even in this podcast, we were talking about Rabbit. [SPEAKER_01] We were talking about Humane. [SPEAKER_01] We were talking about. What's that cookie, Dave? [SPEAKER_04] Was ahead of its time, I will say. But that was really trying to go consumer, I think almost too fast in a way. And a lot of this innovation that we're talking about here is developer innovation. It's how do you get a whole model into memory and then do stuff with it? I still think two, three, four or five years away from being consumer today. And so I think it's hard to actually answer that question because I don't know that we've figured out how to make computers that are even usable for developers yet, let alone consumers. And I think we're a ways off still. So I don't know. It's exciting to me. [SPEAKER_03] Is there a device, Britt, or something? [SPEAKER_03] I mean, you're always tinkering and thinking that you feel you crave. [SPEAKER_01] The Mac Mini. [SPEAKER_01] The Mac Mini is the breakout AI device of the year. [SPEAKER_03] But also on the consumer side, I feel recording is the killer input of the AI era, which isn't that a lot of things record. [SPEAKER_01] Actually, yeah. [SPEAKER_01] Here, let me get it. [SPEAKER_04] What's funny about recording that a bunch of people have started to bring up to us is that remote work actually works best for an AI brain within a company because everything is recorded and you have to record all your meetings and transcripts. [SPEAKER_04] But right now, the culture of in-person work is such that you don't want to be recording every conversation or every hallway run-in with your coworker or things like that. [SPEAKER_04] And so there's a lot of context that starts becoming missing. [SPEAKER_04] So I'm starting to see a lot of founders who are actually second-guessing in-person versus remote. No, don't do it. In-person work. [SPEAKER_04] Don't do it. [SPEAKER_04] Founders, don't do it. Several times in the last couple of weeks, Jess. You, I, okay. [SPEAKER_03] You're hearing it because it's summer. [SPEAKER_03] You're not hearing it because. No, I think there's a reality to that. I understand it. [SPEAKER_02] Look, I'm remote and I have everything recorded. [SPEAKER_02] It's awesome. [SPEAKER_03] I'm in person and I have everything recorded. Here's my answer, Jess. This thing, this is the DJI microphones. So you can use super noise canceling microphones with Whisper flow. [SPEAKER_01] I think that's actually the hardware device that matters. [SPEAKER_01] I don't actually think it's the ones that record things. I don't think those are good businesses because you can just use your phone if you want to do that. But this whole thing, which is can I whisper or talk directly to the AI to give it dramatically more context? The number of people I see carrying these things has dramatically increased this year. Okay. So for those who are listening, Dave just showed very large, enlarged AirPods with very fuzzy. [SPEAKER_01] So you can, there are super noise canceling microphones that you can use with whisper flow. [SPEAKER_01] I think that's actually the hardware devices that matter. [SPEAKER_01] I don't actually think it's the ones that record things. [SPEAKER_01] I don't think those are good businesses because you can just use your phone if you want to do that. [SPEAKER_01] But this whole thing, which is can I whisper or talk directly to the AI to give it dramatically more context? [SPEAKER_01] The number of people I see carrying these things has dramatically increased this year. [SPEAKER_03] Okay. So for those who are listening, Dave just showed very large, enlarged AirPods with very fuzzy. [SPEAKER_03] No, it's a microphone. [SPEAKER_01] It's a lapel mic. [SPEAKER_01] It's a microphone. [SPEAKER_01] This is a microphone that you wear. [SPEAKER_01] You put it on. [SPEAKER_03] Okay. I see. Yes. I've seen the influencers do this. [SPEAKER_03] It's Brit's broach that she wears. [SPEAKER_03] Stick it on. [SPEAKER_03] You can stick it on your clip. [SPEAKER_03] I mean, the thing about this though, Dave, is first of all, you're going to be very excited with QAI. But Dave, so walk me through. [SPEAKER_02] So why do you need this? Grinnell records all my meetings. Many of them are in person, but so why do I need this? It's about putting more context into the computer so that the AI has a much larger context window to help you with whatever it is you're trying to create. So you finish a meeting and then come back to your desk and you're like, that was a great meeting. I really love how so-and-so stepped up. [SPEAKER_03] No, it's you're prompting. [SPEAKER_03] It's instead of prompting it with one sentence, you can prompt it with three minutes of talking, which is many paragraphs. [SPEAKER_04] And it actually is going to get you a much better answer because it has so much context of what you are trying to get out of it. [SPEAKER_04] Why do I need a fancy microphone? [SPEAKER_04] You could do it that way too, but these are designed for whispering and for extremely high fidelity capturing. [SPEAKER_03] And so you can just sit there and talk to your computer and it gives it tons more context. [SPEAKER_01] I made my own app for this, which is great, that I used to use. What are the features of it? [SPEAKER_01] It literally is just click a button and it opens directly to a recording, that's live. [SPEAKER_02] And then when I hit enter, it sends it up, transcribes it and dumps it into my bot, into a table. [SPEAKER_02] But it's way better. All I do is this. I literally have built an iOS app that just does this for me. [SPEAKER_01] And so I don't need any special for it. [SPEAKER_02] By the way, it took me two seconds to code it and deploy it in clouds. [SPEAKER_02] I have my own. I also built a version. [SPEAKER_02] I just got sick of clicking all the buttons for them. [SPEAKER_02] So now I just have a composer that I pull up and it's its own apps. [SPEAKER_02] You one tap it and then you can talk or write and just one tap it to your bot. [SPEAKER_02] So I don't think it's that important. [SPEAKER_02] But you tell me the tech workers of the world are sitting at home while their agents are running, whispering to themselves all day. [SPEAKER_02] Yes. [SPEAKER_03] And those that do work in person now have these co-working spaces where everyone's whispering to their lap, to their computer. [SPEAKER_04] And it's super weird. [SPEAKER_03] Oh my God, guys, this is so weird. [SPEAKER_04] It just comes down to how much data can you get into the computer? [SPEAKER_04] I mean, it is, but it also isn't. Your hands can only type so fast. [SPEAKER_03] And it's just more context per second, right. [SPEAKER_01] And I think that's the big change here is that the models want more context to do a better job for you. And so we are shifting the way that we're interfacing with the computer to match the capability of the computer. That is actually a big change. It's something that's happening in real time right now, but people on the frontier, almost everyone I know is doing this. Guys, I hope we still talk to each other. Do we have podcasts for that? [SPEAKER_03] Yeah, I think so. [SPEAKER_03] Guys, social is the new tech. [SPEAKER_03] No, but everyone's just mumbling under their breath to their AI with their fancy microphones to their context window. [SPEAKER_04] Guys. Maybe that means the AI can get more done and we can all hang out with each other more and have better social lives. Because that's always how it works. That's always how it works. So far I'm more stressed than ever before in the AI era. [SPEAKER_04] I'm sleeping less and I'm socially less. [SPEAKER_04] Okay. [SPEAKER_04] We've burned through AI. [SPEAKER_04] What else is happening? Speaking of AI, what do you guys think about this keystroke logging, tracking that the Meta employees aren't too happy about it? [SPEAKER_03] We called that. [SPEAKER_03] Meta is so good. [SPEAKER_03] Meta is so good at being a heel. [SPEAKER_03] Oh, Dave. [SPEAKER_01] Oh, gosh. [SPEAKER_01] It's true. [SPEAKER_03] It's of course this was going to be the outcome. [SPEAKER_03] The irony is we just shouldn't have told anyone. [SPEAKER_01] It's they probably were thinking I'm going to do this. We should just tell people, right? [SPEAKER_02] Because on the, and look, just log the keystrokes. [SPEAKER_02] It's fine. [SPEAKER_02] No. [SPEAKER_02] But I'm wondering if they intentionally did this. [SPEAKER_02] Isn't it? [SPEAKER_02] It's like they might as well. [SPEAKER_01] People always react negatively. [SPEAKER_01] So maybe it's just, it's good they played it this way. [SPEAKER_01] Maybe it's a honeypot. [SPEAKER_01] Like, oh, the people who reacted negatively to this, we should just fire them. We should tell people, right? [SPEAKER_02] Because that's—do you, on the—and look, just log the keystrokes. [SPEAKER_02] It's fine. [SPEAKER_02] No. [SPEAKER_02] But I'm wondering if they intentionally did this. [SPEAKER_02] Isn't it? [SPEAKER_02] It's they might as well. People always react negatively. So maybe it's good they played it this way. [SPEAKER_01] Maybe it's a honeypot. [SPEAKER_01] Oh, the people who reacted negatively to this, we should fire them. [SPEAKER_01] Clearly, they're not the AI people who get it. [SPEAKER_02] So it's a trap. [SPEAKER_02] It's a trap. [SPEAKER_02] Do you think—I should have set this up better. [SPEAKER_02] I think this is a little bit old, but Meta has been keystroke logging employees to train AI. [SPEAKER_02] Their AI is not so popular. Last week we did. I'm hearing even crazier things, which is that people are building ad agencies or marketing agencies that are filled with people who used to work in ad agencies. And it's not a real agency, but they're recording everything that they're doing in order to get the dataset that is ad agency. But of all the people who got fired, you're recording, it's how to be a bad ad agency. [SPEAKER_01] So they—wait, are they play acting their former lives in an ad? [SPEAKER_01] Yes, they're play acting their former life. This is a great spinoff of The Office. [SPEAKER_03] Wait, Dave, explain this in one coherent soundbite, because this is funny. [SPEAKER_01] The soundbite is that there are AI companies—I don't—doesn't matter which ones—that are hiring the former employees of ad agencies to do their job, to play act their old job, and then recording all of it in order to get the training data to learn marketing agency. [SPEAKER_04] That's very funny and makes total sense. Let's call up Don Draper. [SPEAKER_01] Yeah, it's basically Mad Men. I guess they could have just trained on Mad Men. [SPEAKER_03] Maybe. [SPEAKER_01] Yeah. [SPEAKER_01] It can be a creative agency because AI can't do creative ideas. [SPEAKER_01] You can't train AI to do that. [SPEAKER_01] So look, here's the thing. The funny irony of all this in terms of Sam Lesson is always right but wrong, but lick the cookie is—look. I need to lick a cookie before this episode's over. [SPEAKER_02] We were running too. [SPEAKER_02] You have said twice now as mentioned by the information. [SPEAKER_03] That—I mean, that's just fact. That's just historical reference. [SPEAKER_04] Listen, eight years ago, Cortina and I were running on the order of 20,000 customer service agents on the Finnautics platform across a lot of very big unicorns. [SPEAKER_03] And recording the entire work stream click flow through our clone plugins to build a dataset to automate work. [SPEAKER_03] We were eight years ahead on this and had a huge dataset. It was a mistake only to ever get rid of it because we had the full click stream history of—it was awesome. Right? It was exact. But this was exactly the use case. [SPEAKER_02] I have a question. Did you sell Finn to Intercom? Isn't that funny? [SPEAKER_01] No. [SPEAKER_01] Is that Finn? No. [SPEAKER_01] Here's the funny thing. [SPEAKER_01] Everyone's—Finn. [SPEAKER_01] I was—Sam, did you sell Finn to Intercom? No. So here's a quick story. We were deep in partnership conversation with Intercom for Finn Analytics, which was—we started out with the assistant service, which honestly is a lot of the interfaces I've rebuilt for myself now with AI—just not going to book restaurants. Then we pivoted it to Finn Analytics because we realized that while the assistant business was difficult, building the dataset and finding the errors in operations work was what we got good at. [SPEAKER_02] Because we basically had the full click stream of humans completing tasks. [SPEAKER_02] And they were—okay, this is where this gets messed up, where you're losing all your time. [SPEAKER_02] And we sold this as a service to a lot of the biggest companies in the world and had tens of thousands of operations agents of these companies. [SPEAKER_02] Literally, where we were in their browser, we had a Chrome program, we're getting the full click stream history of every single thing they were doing, everything they were interacting with, time spent, time series, everything. [SPEAKER_02] And we were building up this catalog of how to efficiently complete tasks and then finding all the errors and patterns where people were off. [SPEAKER_02] We also were finding places you could automate because you'd be—okay, these four steps always happen in a row this way. [SPEAKER_02] Automate it. [SPEAKER_02] Right. [SPEAKER_02] So we were literally building this dataset and doing this probably six years too early, which is hilarious to me. [SPEAKER_02] But this, we're going to see more. [SPEAKER_02] I mean, the fact that we're seeing these ad agencies, I also think you're going to see more. [SPEAKER_02] I assume all the AI labs are doing a variant of what Meta's doing, although probably not as intrusively. [SPEAKER_03] I mean, I don't. [SPEAKER_03] I for sure think Anthropic's doing this amongst all of their employees. [SPEAKER_03] Anthropic has somehow convinced all of the CTOs to join Anthropic so that they can train on all of the CTOs to eliminate them. It's pretty wild. [SPEAKER_04] I just think it's really funny. [SPEAKER_01] What's this point with OpenAI and Anthropic? [SPEAKER_02] I love the whole thing about them starting these companies to send people out into the field to convert them to users. [SPEAKER_02] This is literally exactly what religions do. [SPEAKER_02] They've sent all these monks into the field. [SPEAKER_02] Oh, that's funny. [SPEAKER_02] I hadn't thought of it that way. [SPEAKER_02] I thought you were saying it's exactly what Palantir does. [SPEAKER_02] But no. [SPEAKER_02] I thought you were going a different way with this, which is that it's counter to their entire message. [SPEAKER_03] The entire message is that AI is so good. [SPEAKER_03] It's going to AGI and take care of itself. [SPEAKER_01] And if that's true, then starting a deployment company is the worst idea you could possibly have because it will be deploying itself in two years. [SPEAKER_01] The thing is, the irony is Palantir did actually the opposite, right? [SPEAKER_01] Which is the great trick of Palantir—they started an only world deployment. [SPEAKER_01] There was no central cloud to it. [SPEAKER_02] There was no—they were a consulting firm that had everyone in the field with very, very high multiple. [SPEAKER_02] They've been trying to go backwards and stitch it into a central religion of Palantir. [SPEAKER_02] Whereas ironically, OpenAI and Anthropic have Rome, right? [SPEAKER_02] Or now I guess the new Rome is San Francisco. And if that's true, then starting a deployment company is the worst idea you could possibly have because it will be deploying itself in two years. [SPEAKER_01] The thing is, so the irony is Palantir did actually the opposite, right? [SPEAKER_01] Which is the great trick of Palantir is they started an only world deployment. There was no central cloud to it. [SPEAKER_02] There was no, they were a consulting firm that had everyone in the field just with very, very high multiple. They've been trying to go backwards and stitch it into a central religion of Palantir. Whereas ironically, OpenAI and Anthropic have Rome, right? Or now I guess the new Rome is San Francisco. And they are now sending out their monks into the world to go sit in with the heathens in their companies and convert them to the word of God, right? It's hilarious. Now what happens in proselytizing with private equity backers? This is also private equity salivating at the companies they can raid, right? [SPEAKER_02] Well, it's also just, maybe, but they don't want OpenAI to send their monk in and fix it. [SPEAKER_03] They want it to be broken and take it over. It's really funny because the history of monks, I mean, monks were crazy. [SPEAKER_02] Totally nuts. [SPEAKER_02] Yeah. How else do you think that they can talk to God and walk on water and all that and fly and all the things they could do? [SPEAKER_01] It's they go to some dramatic town that worships a tree and they, you're they worship a tree for hundreds of years. [SPEAKER_01] And the monk would come in and just cut the tree down and be, I am the word of God. [SPEAKER_01] You're going to get killed. So what's going to happen to these OpenAI engineers? [SPEAKER_02] They're going to go to Flint, Michigan and bring the AI, which is so popular. [SPEAKER_02] And I'm a little worried about they're getting crucified. [SPEAKER_02] That was our pod last week. [SPEAKER_02] Yeah. [SPEAKER_02] So for what it's worth, you are right. [SPEAKER_03] And one of the things that's happening to me on the front lines of OpenClaw is we have a lot of people asking us to help them, forward deploy, deploy multi-agent systems. [SPEAKER_03] So I'm in a lot of conversations related to this and the resistance on the front lines is absolutely enormous. [SPEAKER_01] People do not want, and I've actually only really in the last two weeks really come around to having a lot of first party conversations about this. It's a pretty serious problem actually out there that people do not want to deploy this stuff. [SPEAKER_01] I got to tell you, our creator fun day, hackathon was the opposite. [SPEAKER_01] I've never seen such a hungry group of people excited to engage with using this to build their empires. So, I think it depends, but yes, I agree. [SPEAKER_02] Just as the monks would go to Germanic tribes and just get literally killed for cutting down sacred trees, the anthropic engineers will go to middle America and they're going to need a large security force. [SPEAKER_02] Super interesting. [SPEAKER_02] Did you guys see the UCF commencement speaker this week? [SPEAKER_02] So in Florida, they did a graduation ceremony and the commencement speaker, I don't remember who it was, said one line. [SPEAKER_02] It was just, we all know we're entering the next revolution. [SPEAKER_04] We had the industrial revolution, now it's the AI revolution. The students went bananas, starting booing, shutting it down. And the whole commencement ceremony went sideways just from that one line. [SPEAKER_04] Right afterwards, they all opened their phones and started swiping TikTok. [SPEAKER_04] They're, oh, I didn't know I was already using AI or only. [SPEAKER_04] They're all on TikTok and Instagram using AI all day, every day. [SPEAKER_01] The only way that they learned that AI is bad is through the AI that's feeding them the content that they're watching. So the great television show known as Hacks has also taken on this topic. [SPEAKER_01] So that's how you know. [SPEAKER_01] I no longer know where this is going to go. [SPEAKER_03] It's a very interesting conversation. [SPEAKER_03] Okay. [SPEAKER_01] We're not going to go there, though, because we went there last week and the people don't want us to go. [SPEAKER_01] And we're going to. [SPEAKER_03] Can I do the sexiest pivot ever and ask you guys about Microsoft? [SPEAKER_03] Oh, God. [SPEAKER_03] I was hoping it would actually be sexy. [SPEAKER_03] Bummer. [SPEAKER_03] Bummer. [SPEAKER_01] Well, you want to talk about the OnlyFans sale? How about Janitor AI? Do you guys really not want to talk about the third most valuable company in the world? Here's my only line on Microsoft because I don't care about Microsoft. [SPEAKER_03] Wait, what are we? [SPEAKER_03] But what is the question before we get into it? [SPEAKER_01] There is a reason I put this on the agenda, which is not because there are activist investors. [SPEAKER_01] One of my columnists argued there might be. [SPEAKER_01] So we should be very careful about that. [SPEAKER_03] But no, Microsoft is the only of the big tech companies that has been consistently down this year. [SPEAKER_03] And investors are really worried about Microsoft. [SPEAKER_03] And all the things we talk about day in and day out. [SPEAKER_03] And yes, there have been ups and downs to tech stocks this year. [SPEAKER_03] This isn't about that. [SPEAKER_03] We're talking about Google, Amazon, all the big guys, the 40% growth of their cloud business, this whole AI boom. [SPEAKER_03] And here we have Microsoft. [SPEAKER_03] Two years ago, Satya was probably the most popular person at any founder conference from here to New York. [SPEAKER_03] And the internet just loved him and founders loved him. [SPEAKER_03] Ted, what a genius and all that. [SPEAKER_03] And now today there are LinkedIn layoffs. They've faced a lot of heat over the gaming business. So they're really Asha, the new CEO of Xbox. And investors are not buying the story. So I guess my question is, what do you guys think? [SPEAKER_03] And what should Microsoft do next? [SPEAKER_03] Right? [SPEAKER_03] So I think that is actually an interesting question. [SPEAKER_03] Why did they fold on OpenAI? [SPEAKER_03] That's something they don't understand. [SPEAKER_03] They gave up $300 billion in revenue or something. Basically, they ended up as documented in the Information. Ding, ding, ding. There were some key inflection points around compute. [SPEAKER_03] Call it 12 months ago, give or take. I could be off by six months. Who knows? [SPEAKER_03] Microsoft had a right of first refusal on all new compute from OpenAI. [SPEAKER_03] And Sam was saying, we need more. [SPEAKER_03] We need more. [SPEAKER_03] We need more. [SPEAKER_03] Why did they fold on OpenAI? [SPEAKER_03] That's something they don't understand. [SPEAKER_03] They gave up $300 billion in due revenue or some shit. They ended up as documented in the information. Ding, ding, ding. There were some key inflection points around compute. [SPEAKER_03] Call it 12 months ago, give or take. [SPEAKER_03] I could be off by six months. [SPEAKER_03] Who knows? [SPEAKER_03] Microsoft had a right of first refusal on all new compute from OpenAI. And Sam was saying, we need more. [SPEAKER_03] We need more. [SPEAKER_03] We need more. [SPEAKER_03] And Amy Hood, the CFO of Microsoft and Satya were, we're not going to give you more right now. [SPEAKER_03] And they underestimated their needs and let them pursue other partnerships. [SPEAKER_03] And it turned out that may have been the right decision from a CapEx perspective for Microsoft. [SPEAKER_03] But it opened the door for them to go and build these very major partnerships with others, including, of course, now Amazon. [SPEAKER_03] And so this has been a source of huge tension between OpenAI and Microsoft. [SPEAKER_03] And very recently, they evolved what could have been heading towards a lawsuit by clearing the way for AWS to have more access to the APIs while still being the primary compute provider, which is meaningless. [SPEAKER_03] So they still have a very deep partnership. [SPEAKER_03] But the short answer to your question, Sam, is the amount of CapEx need foisted them into the hands of other people. [SPEAKER_03] Didn't they also do a bad job contracting the definition of AGI or some bullshit? [SPEAKER_03] Yeah, well, they wrote the partnership in a way. [SPEAKER_03] And I don't think you can fault them for this, to be honest, because who would have predicted back whenever this partnership was struck what form it would have taken? [SPEAKER_02] And so I think it's cool to Microsoft's credit. They were willing to fly by the seat of their pants a little bit in negotiating these things. [SPEAKER_03] A huge percentage of Azure is OpenAI. [SPEAKER_03] One of the things that came out in the trial, as well as other reporting this week from us, is how much they've also benefited from this partnership, including owning a very sizable stake in OpenAI. [SPEAKER_03] So the idea, maybe when it was struck, the idea that there would have been some exclusivity held water, but the scale of everything and especially the compute. [SPEAKER_03] And there was a specific point at which Sam and Amy and Sasha just didn't see eye to eye on the compute needs. [SPEAKER_03] And so they went elsewhere and that opened the door. [SPEAKER_03] Is it the only big tech company that bet $100 billion on someone else's religion? [SPEAKER_03] And now that religion is starting its own church. [SPEAKER_03] And so they're, oh, I don't know how I feel about this. [SPEAKER_01] Well, maybe, and history will tell whether because they had that deal, they didn't invest enough in their own stuff, which they've tried to do retroactively with Mustafa and buying Inflection. [SPEAKER_03] That has gone sideways. [SPEAKER_03] But the flip side, what Microsoft comes, tells me every time I tweet something to the contrary, is name a bigger incumbent suite of productivity apps today. [SPEAKER_03] They are still so dominant in so many key services. [SPEAKER_03] As long as they're not that far behind on adding agents to them over time, they'll be fine. [SPEAKER_03] But I think that's what you're actually talking about, Jess, is that this was a great financial strategy for Microsoft. [SPEAKER_03] It wasn't a great product strategy. [SPEAKER_01] They executed a massive trade, the trade looks like it's working out. [SPEAKER_01] But the product outcome that they wanted or that they perhaps needed hasn't materialized because they didn't build the church inside their house. They looked really, really smart for doing this thing. [SPEAKER_01] And it's financially been great for them. You don't see products materializing out of this. You see them putting ChatGPT into the operating system and Copilot this and Copilot that, but none of it is really both Microsoft and Meta have had this same problem where they're making a lot of smart financial moves, but not making good product moves and paying the price for it. [SPEAKER_01] And I don't know. [SPEAKER_01] We'll see. Right. Maybe there's going to be a second wave here after this big infrastructure investment that they'll go and do it. [SPEAKER_01] Can't you argue that Microsoft has just been trying to be learning on the data of Anthropic and OpenAI while they're off now going to build their own models and use their scale to come in over the top over the next five years? [SPEAKER_04] Well, they've been trying unsuccessfully. [SPEAKER_04] It's just not that big, Britt. [SPEAKER_04] It's 20 million paid seats or something in Copilot. [SPEAKER_01] I'm not talking about Copilot. I'm talking about Microsoft's base at large. If over the next few years, they're able to take what they've learned and aren't they doing this? [SPEAKER_04] Mustafa, Mustafa Suleiman is the person that Satya has put in charge to pursue superintelligence. [SPEAKER_04] They actually restructured and pushed him aside. [SPEAKER_04] They've had many reorgs. [SPEAKER_04] It seems like they're trying to build their own. [SPEAKER_03] They're the only one whose entire AI strategy depends on a partner that's actively becoming a competitor is another way to put it. [SPEAKER_03] They don't have a frontier strategy. Yes. [SPEAKER_01] And I think to be fair, you could say the same thing about Amazon. [SPEAKER_01] I mean, Amazon is Nova. And I think that. For sure. To be clear. I'm having, because this is how fun my life is, a running debate over whether you could say the same about XAI. Right. And is this question, is Elon going to give up on developing XAI? I don't think he is. I don't think you can say that about XAI. No, I don't think you can yet either. [SPEAKER_01] XAI did produce. [SPEAKER_01] I mean, I don't think you can say it about XAI or Meta because both XAI and Meta, despite the fact that their models are not performing at the highest level, they at least produced them. [SPEAKER_01] Right. [SPEAKER_01] They went and they built out big infrastructure. [SPEAKER_01] They trained models. [SPEAKER_01] They built the teams. [SPEAKER_01] Whereas it's hard. [SPEAKER_01] It's fuzzier to see that in the Amazon, Microsoft world. [SPEAKER_01] I also think there's a question, what is the XAI strategy going forward? No, I don't think you can yet either. [SPEAKER_03] XAI did produce. [SPEAKER_01] I don't think you can say it about XAI or Meta because both XAI and Meta, despite the fact that their models are not performing at the highest level, they at least produced them. [SPEAKER_01] Right. [SPEAKER_01] They went and they built out big infrastructure. [SPEAKER_01] They trained models. [SPEAKER_01] They built the teams. [SPEAKER_01] Whereas it's hard. It's fuzzier to see that in the Amazon, Microsoft world. [SPEAKER_01] I also think there's a question, what is the XAI strategy going forward? [SPEAKER_01] Elon's gutted the team, but I think he's taken it to the studs to build it up in the way I like. But that. The strategy is to be Anthropic's landlord. I don't know. [SPEAKER_01] Well, guys, let's. Here's some other quick news things we should weigh in on. Anyone have opinions? Anyone going in? You VCs on the $1 trillion Anthropic fundraise that's around the corner here? There's a new fundraise every two weeks. I can't keep up with this. It's not a trillion dollars. [SPEAKER_04] What is it? [SPEAKER_04] It's $800 billion. It rounds up. It's fine. [SPEAKER_04] Is that through a first, second or third derivative SPV that people are entering? [SPEAKER_04] And didn't OpenAI allow employees to do a stock sale of up to $30 million each? [SPEAKER_04] And now all of San Francisco real estate is totally imploding right now? [SPEAKER_04] Yes. [SPEAKER_04] Those are all true facts. [SPEAKER_04] But this is also net valuable for our home prices on paper, correct? [SPEAKER_04] I don't know that it's good. [SPEAKER_04] I think we got a I love this map somebody made of all the abandoned buildings in San Francisco yesterday. [SPEAKER_01] Somebody AI created a very sophisticated map of how little building is actually going on in San Francisco. [SPEAKER_01] We also need to keep the building going so that the prices don't go absolutely through the roof. [SPEAKER_01] The prices are going to go absolutely through the roof, but we can also keep the building. [SPEAKER_01] There are different segments of the market. [SPEAKER_01] I heard an anecdote about someone who had a house on the market and then they decided they were not going to take any offers until SpaceX IPOs. Until the Anthropic fundraising closed. Smart. And then they would take. But yes, we're a month out from the SpaceX IPO, guys. The big news out here in Marin is that somebody has been trying to sell their house for it. They'll only sell it for Anthropic shares, which is ridiculous. I can tell Sam's bored, guys, because my email is full of confirmation codes for things that he's signing up for. Why do you get the confirmation codes? [SPEAKER_03] Guys, today's Jackson Hole. [SPEAKER_03] Today's Jackson Hole season pass. [SPEAKER_02] I got our ski passes. [SPEAKER_02] I already got them. [SPEAKER_02] Don't worry. [SPEAKER_02] Good. [SPEAKER_04] There's a whole new system. [SPEAKER_04] It's the worst. [SPEAKER_02] I have to figure out all the kids' birthdays again. [SPEAKER_02] Guys, get your agents to do things for you. [SPEAKER_02] You guys are so behind. [SPEAKER_04] I would actually not trust my agent with this task. [SPEAKER_04] This is also, by the way, Jess, where you need to whisper a lot because the ski calendars with three kids and all the weekends, that's a lot. [SPEAKER_03] You need to get your agent on that ASAP. [SPEAKER_04] We've got a great, so OpenClaw has a new computer use tool called Peekaboo. [SPEAKER_04] That's incredible. [SPEAKER_04] It will use your computer. It's so weird to watch it do that. What's the thing that OpenClaw is built on? Pi. [SPEAKER_02] Pi. [SPEAKER_02] Pi. [SPEAKER_02] What's your view on Pi? [SPEAKER_02] Pi is great. [SPEAKER_02] It's another harness. [SPEAKER_02] It's a TypeScript-based harness. It's really easy to modify. You can customize it in a lot of ways. [SPEAKER_01] The reason Peter used it was that it's built in TypeScript, and it's customizable, and it's very proactive. [SPEAKER_01] I was just talking to the founder of Pi two days ago. [SPEAKER_02] Mario? [SPEAKER_02] Yeah. [SPEAKER_02] Nice guy. [SPEAKER_02] What's the deal with where does that go? [SPEAKER_02] What is it? [SPEAKER_02] It's a language, a framework. [SPEAKER_02] What is Pi? [SPEAKER_03] It's a harness. [SPEAKER_03] Okay. [SPEAKER_03] I don't know what a harness is. [SPEAKER_02] Sorry. Really? Jess, come on. [SPEAKER_03] No! [SPEAKER_03] I know what a horse harness is. [SPEAKER_03] Harness is the new word as of two months ago. [SPEAKER_01] You have to know what a harness is. [SPEAKER_04] A harness is the software built on top of a model to make the model useful. [SPEAKER_04] The models are not useful by themselves. [SPEAKER_04] They need to be given software that helps them think in certain patterns, repeatedly run loops to accomplish tasks. [SPEAKER_01] And there's a lot of different ways that you can build those harnesses. [SPEAKER_01] And OpenClaw is a harness. [SPEAKER_01] Pi is a harness. [SPEAKER_01] Codex is a harness. [SPEAKER_01] Clawed code is a harness. [SPEAKER_01] Got it. So it's harnesses on harnesses on harnesses. Yes. [SPEAKER_02] Do you double harness? In fact, a lot of people actually are really loving swapping Pi out with Codex under OpenClaw currently. I still love Pi, but some people prefer using Codex instead. Dave, I guess this is a question. A long time ago we had this conversation on what is this? What is OpenClaw? [SPEAKER_02] And I go back to this question again with well, OpenClaw is built on Pi. How do you define what these things are? [SPEAKER_02] I think of them as the apps or the interfaces that help people use AI models more effectively. [SPEAKER_02] Do you double harness? In fact, yeah, a lot of people actually are really loving swapping Pi out with Codex under OpenClaw currently. [SPEAKER_01] I still love Pi, but some people like using Codex instead. [SPEAKER_01] But what, Dave, I guess this is a question. [SPEAKER_01] A long time ago we had this conversation on what is this? [SPEAKER_01] What is OpenClaw? [SPEAKER_02] And I go back to this question again, well, OpenClaw is built on Pi. How do you define what these things are? [SPEAKER_02] I think of them as the apps or the interfaces that help people use AI models more effectively for whatever it is that they want to use it for. [SPEAKER_02] The operating system that sits on top of all these low level things. [SPEAKER_01] We're talking, we're all using Mac OS right now. [SPEAKER_01] Underneath the windows and the things that make up the screen that you're looking at are a bunch of different tools that technically don't matter to the end user. [SPEAKER_01] And Apple has brilliantly put it all together in a user interface that makes it possible to take advantage of the power of the computer. [SPEAKER_01] And so I think that's what harnesses are in this era. [SPEAKER_01] They make it possible to take advantage of AI models, types of AI models, different kinds of models to accomplish different things that you might want to do. [SPEAKER_01] And they make the user interface for doing that easier. [SPEAKER_01] And there's going to be lots more of them. [SPEAKER_01] And we're already seeing this at OpenClaw where people customize it in a hundred different ways to do the use case that they need from it. [SPEAKER_01] Dave, I'm going to tell you that your wife is telling me you have to wrap. [SPEAKER_01] So can you wrap? [SPEAKER_01] I'm producing behind the scenes with you. [SPEAKER_03] Well, there's a flight we're about to miss. Yeah. Before we do bounce, though, I've got a plug. I am excited for an interview I'm doing next week at the Commonwealth Club in San Francisco with Joanna Stern. [SPEAKER_03] I don't know if you guys know Joanna. She was the Wall Street Journal's wonderful tech video columnist who left the journal to start her own media company. [SPEAKER_03] Can't imagine where she got that idea. [SPEAKER_03] She's wonderful. [SPEAKER_03] She has a new book out. [SPEAKER_03] And so if you like such things, come see us in San Francisco at the Commonwealth Club next Wednesday. [SPEAKER_03] And with that, Moran's go do what you do. [SPEAKER_03] And I'll plug that I was able to finish buying our Jackson Hole ski passes just before this episode ended. [SPEAKER_03] San's agent. [SPEAKER_02] Good job, San. [SPEAKER_02] I do. [SPEAKER_04] I do my own hands. [SPEAKER_04] Since your last year of purchasing without an agent. [SPEAKER_02] Well, I don't know how to land all these planes at once. [SPEAKER_02] So I will just say a big thank you. [SPEAKER_04] I think we covered some good ground, folks. Fresh topics, old topics. We're happy to be here with all our listeners and viewers. Thanks for listening. And we'll see you back here next week for another episode of More or Less. [SPEAKER_03] Bye. [SPEAKER_03] Bye, guys. See you guys later. If you enjoyed this show, please leave us a virtual high five by rating it and reviewing it on Apple Podcasts, Spotify, YouTube, or wherever you get your podcasts. [SPEAKER_01] Find more information about each episode in the show notes and follow us on social media by searching for at more or less at Dave Morin at lesson at Jay lesson. [SPEAKER_05] And as for me, I'm at Brit. [SPEAKER_05] See you guys next time. Don't worry. Good. There's like a whole new system. It's the worst. I have to figure out all the kids' birthdays again. Guys, get your agents to do things for you. You guys are so behind. I can't believe it. I would actually not trust my agent with this task. This is also, by the way, Jess, where you need to whisper a lot because the ski calendars with three kids and all the weekends, like, yeah, that's a lot. You need to get your agent on that ASAP. We've got a great, so OpenClaw has a new computer use tool called Peekaboo. That's incredible. Like, it will use your computer. It's so weird to watch it do that. What's the thing that OpenClaw is built on? Pi. Pi. Pi. What's your view on Pi? Pi is great. It's another harness. It's a TypeScript-based harness. It's really easy to modify. You can customize it in a lot of ways. The reason Peter used it was that it's built in TypeScript, and it's so customizable, and it's very proactive. So what, like, what's the, I was just talking to the founder of Pi, like, two days ago. Like, what? Mario? Yeah. Nice guy. What's the deal with, like, where does that go? What is it? It's a language, a framework. What is Pi? It's a harness. Okay. I don't know what a harness is. Sorry. Really? Jess, come on. No! I know what a horse harness is. Harness is the new word as of two months ago. You have to know what a harness is. A harness is the software built on top of a model to make the model useful. The models are not useful by themselves. They need to be given software that helps them think in certain patterns, repeatedly run loops to accomplish tasks. And there's a lot of different ways that you can build those harnesses. And OpenClaw is a harness. Pi is a harness. Codex is a harness. Clawed code is a harness. Got it. So it's just harnesses on harnesses on harnesses. Yes. Do you double harness? In fact, yeah, no, I mean, a lot of people actually are really loving swapping Pi out with Codex under OpenClaw currently. I still love Pi, but some people like using Codex instead. But what, like, Dave, I guess this is a question. I go, a long time ago we had this conversation on what is this? Like, what is OpenClaw? And like, I go back to this question again with like, well, OpenClaw is built on Pi is like, how do you define what these things are? I think of them as the apps or the interfaces that help people use AI models more effectively for whatever it is that they want to use it for. Like the operating system that sits on top of all these low level things. Like, you know, we're talking, we're all using Mac OS right now. Underneath the windows and the things that make up the screen that you're looking at are a bunch of different tools that technically don't matter to the end user. And Apple has brilliantly put it all together in a user interface that makes it possible to take advantage of the power of the computer. And so I think that's what harnesses are in this era. They make it possible to take advantage of AI models, types of AI models, different kinds of models to accomplish different things that you might want to do. And they make the user interface for doing that easier. And there's going to be, I think, lots more of them, right? And we're already seeing this at OpenClaw where people customize it in a hundred different ways to do the use case that they need from it. Dave, I'm going to tell you that your wife is telling me you have to wrap. So can you wrap? I'm producing behind the scenes with you. Well, there's a flight we're about to miss. Yeah. Before we do bounce, though, I've got a plug. I am excited for an interview I'm doing next week at the Commonwealth Club in San Francisco with Joanna Stern. I don't know if you guys know Joanna, she was the Wall Street Journal's wonderful tech video columnist who left the journal to start her own media company. Can't imagine where she got that idea. She's wonderful. She has a new book out. And so if you like such things, come see us in San Francisco at the Commonwealth Club next Wednesday. And with that, Moran's go do what you do. And I'll plug it that I was able to finish buying our Jackson Hole ski passes just before this episode ended. San's agent. Good job, San. I do. I do my own hands. Since your last year of purchasing without an agent. Well, I don't know how to land all these planes at once. So I will just say a big thank you. I think we covered some good ground, folks. Fresh topics, old topics. We're happy to be here with all our listeners and viewers. Thanks for listening. And we'll see you back here next week for another episode of More or Less. Bye. Bye, guys. See you guys later. If you enjoyed this show, please leave us a virtual high five by rating it and reviewing it on Apple Podcasts, Spotify, YouTube, or wherever you get your podcasts. Find more information about each episode in the show notes and follow us on social media by searching for at more or less at Dave Morin at lesson at Jay lesson. And as for me, I'm at Brit. See you guys next time.