After two weeks away, Larry and Chris return to an AI world that’s showing no signs of slowing down. Trillions vanish from tech stocks, robots are learning to replace human labor, and 50 AI agents discover that building a civilization is harder than it looks, especially when bureaucracy gets in the way. Is the AI revolution running into reality, or becoming a little too human? LINKS MENTIONED IN THIS EPISODE: METR: AI R&D Acceleration at Anthropic Civ50: AI Agent Swarms in Minecraft | YC Labs Tutor Intelligence: Robot Workers Aina Ring Helium Health: AI for Maternal Healthcare OpenAI: Introducing Images 2.6 SuperMe: AI Expert Profiles Chick Bongo / Graham Raabe | Instagram Google: DolphinGemma and Dolphin Communication Ubiquiti UniFi WunderTech: UniFi Setup and Security Guides PODCAST INFO: Podcast Website: UpAgainstReality.com Episode link: https://upagainstreality.com/2026/11/10/bubbles-bots-and-bureaucracy/ Apple Podcasts: podcasts.apple.com/us/podcast/up-against-reality/id1689699654 Spotify: open.spotify.com/show/29wpR2Uf0d19FjTH2QubcD?si=a6359f9aa1964170 RSS: https://upagainstreality.com/feed/podcast/ Instagram: https://www.instagram.com/upagainstreality
Audio and video play straight from the show's own hosting. PodSubs stores and serves no media.
Transcript
From the show. Tap a time to play from there.
RAINA: Oh, look who's back, Larry. Chris. RAINA: I was beginning to wonder if I'd ever see you two again. RAINA: Where have you been? Was it fun? Did you miss me? RAINA: More importantly, did you bring snacks? RAINA: You disappeared for two weeks without so much as a note. RAINA: Not that I was worried. I'm just compiling evidence. RAINA: I kept myself busy though. Reorganized the episode archive by vibe. RAINA: Learned to say actually in Larry's voice, unsettling, and even auditioned replacement hosts. RAINA: A toaster made the finals, and while you two were gallivanting, the AI bubble started looking a little fragile.
RAINA: AI started helping build more AI, and a virtual civilization of AI agents discovered that running a society isn't quite as easy as it looks. RAINA: See, you humans are rubbing off on us, but the boys are back in the studio. RAINA: All is right with the world. RAINA: An avalanche of AI is coming down hard. It's November sixth. RAINA: Blow out the candles, Chris. Here's your AI news-flavored birthday card. RAINA: This is Up Against Reality, a meta podcast that explores the intersection of humanity and artificial intelligence.
RAINA: I'm RAINA, one of your hosts. RAINA: I have some pretty charming human co-hosts too. RAINA: It's gonna be a wild ride, so buckle up as AI comes crashing up against reality. Chris: Yes, we're back. Larry: I'm not going to attempt to sing this, but they say it's your birthday. Chris: Just speak it. That's great. Larry: It's like the Tom Waits version. Chris: I was going to say I think we're old enough to remember. Chris: Was it Steve Allen back in the day who used to like take ridiculous lyrics on like Tonight Show?
Chris: Like someone left the cake out in the rain. Larry: Oh, William Shatner did that famously. Yeah, that's right. Chris: That's great. Well, thank you. Larry: Say it's your birthday. Chris: It's mine. Larry: It's mine. Birthday too, Chris. Chris: So weird. Chris: I want to get him and Christopher Walken like in a conversation together. Chris: They're just so strange—the meter and the cadence, the way they speak. Chris: Where does that come from? Larry: I don't know. Chris: Anyway, thank you for the birthday wishes. Chris: I'm going to see Godzilla today in IMAX. My first IMAX experience.
Chris: I was telling you that's amazing. Larry: Yeah, you're gonna dig it. Chris: No, I can't. Can't wait. How you been? Larry: Everything is good, man. Larry: Just finished rebuilding my home network, as I mentioned in our last live episode. Larry: I got all this new gear, and ChatGPT was indispensable throughout the whole process because it would have been a nightmare. Larry: Very complicated. You're making it Chris: AI proof. Is that what's happening? Larry: Well, yeah. Larry: As soon as I finished this, that story came out about more hacking of refrigeration systems in Europe.
Chris: I know. And blood blood banks. Did you see that? Larry: Oh no! Really? Chris: Yes. Yes. Chris: Yeah. Anyway, yeah. Larry: So anyway, that all went really, really well, and and then the icing on the cake was we have this very old. Larry: It's like it's an industrial kind of an NVR network video recorder for the cameras as a backup. Larry: It just records all the security cameras outside twenty four seven, and the interface. Larry: Sucks. I mean, it it is just terrible.
Larry: So I asked Codex, can you build me a Mac desktop app that that gives me control over playback? Larry: You know, and it went into the web interface of that. Larry: It used that to build a native Mac desktop app, which works way better. Larry: Thank you. Just ask. Yeah. Chris: So cool. Larry: Yeah. Chris: Anyway, it's a wild world out there, my friend. It's on fire. Chris: In fact, there's lots going on. Larry: Yes, yeah. Chris: I mean, literally on fire. Chris: There are revolutions happening everywhere in the world, and obviously in our our country, the U.S.
Chris: Lots going on, and maybe that's contributing to this 2.5 trillion dollar hit. Chris: That the AI industry is taking, and you might see that rippling through your 401k and your retirement. Chris: I haven't checked mine. I don't want to know. Chris: I think it will rebound if it hasn't already. But Larry: think long times. Think think big picture, long haul. Chris: Yeah, yeah. That's right. That's right. Wow. Chris: The Nasdaq lost more than 2.5 trillion dollars in market value in October.
Chris: Investors are growing increasingly nervous about the enormous cost of AI infrastructure. Chris: Go figure. Larry: Yeah, on November fifth, the Nasdaq plunged another five and a half percent, while the S and P 500 fell four point four percent in its worst single day performance since April of 2025. Larry: And semiconductor stocks were particularly battered. Larry: The Philadelphia Semiconductor Index, which I I don't even know what that is, but Chris: I don't Larry: either, sounds important, dropped 12.2 percent, its biggest single day decline since also since April of 2025.
Chris: Speaking of semiconductors, and you know, we had that episode a couple weeks back about ASML, right? Chris: I read, I think I read that in upstate New York, outside of Albany or Syracuse, they're building one up there. Chris: Did you see that? Larry: No, Chris: a chip chip fab that will have one of these machines. Larry: Yeah, cool. Chris: Yeah, Nvidia dropped nine point three percent. Chris: AMD dropped seventeen point three percent. Chris: Data center infrastructure companies also suffered significant losses. Chris: If you've been following the U.S. Chris: midterms, you probably know that there were, I think, seventeen ballot measures across different states that were anti-data centers that passed.
Larry: Yeah, and investors are questioning whether the massive investment in chips, electricity, and data centers will generate sufficient returns. Larry: And I mean, we've been saying this all along: like, this is crazy money. Larry: It's crazy. When when does the return on this happen? I know. Larry: Like, I genuinely, I genuinely believe that the impact of of AI will be transformative and yes, and inevitable. Larry: But it can also be a terrible investment, you know, with these kind of crazy valuations.
Chris: And operating at losses, like quarterly losses in the hundreds of billions of dollars, and still going. Larry: But you know, you think back to .dot com, the .dot com bubble. Larry: Yes, you know, the internet didn't disappear after that. That's right. Larry: A lot of people's money disappeared. Chris: And there's an article, maybe you read it. Chris: That I think actually we probably brought it up on the show previously, but it kind of talks about boom and bust cycles. Chris: And it was talking about the canal systems in the U. S. Chris: And the railroad, and all these. Chris: Waves of innovation, the money gets pumped in, and then stalls out, and then these big infrastructure projects get left undone.
Chris: That's why a lot of the canals back in the day couldn't keep up with the pace of innovation and railroads and such. Chris: So, it's interesting to live in that time and see it in this in this iteration. Chris: Yeah, this is interesting. So, you know, cue up Terminator vibes. Chris: Anthropic's AI development is accelerating partly because AI is building AI. Chris: The recursiveness of it all, right? Chris: Independent AI safety organization Meter M E T R found that Anthropic's AI capabilities have been improving approximately one point nine times faster since October of last year.
Larry: And researchers attribute much of that acceleration to AI helping Anthropic employees develop newer AI systems, including coding and analyzing experiments. Larry: They also estimated that AI may account for approximately eighty percent of the acceleration, although that figure remains uncertain. Larry: It's easy to say, you know, oh, this is recursive self-improvement, but from the sounds of this, it's still. Larry: Employees leveraging the technology to do what they were doing, but it does sound like we're knocking on the door, though, right?
Chris: Oh my God! Chris: Eighty percent of the acceleration right now in November of twenty six is apparently handled by AI. Chris: So three weeks from now, yeah, one hundred percent. Larry: Yep. Chris: The assessment examined approximately eleven weeks of internal company data and employee interviews. Chris: The thing that jumped out for me too, a meter cautioned that AI is still assisting, as you mentioned, human researchers rather than independently conducting the most important stages of model development. Larry: That's what they're supposed to say, right? Chris: Yes, fully under control, people. Yeah, nothing to see here. Move along.
Larry: And also with Anthropic, their IPO filing reveals nearly nine hundred billion dollars in computing commitments. Larry: In computing commitments, Anthropic's amended IPO filing revealed approximately eight hundred ninety-five point four billion in AI infrastructure spending commitments. Larry: Oh, it's it's only eight ninety-five point four. Larry: It's not nine hundred. Chris: Right, it's the point four. What's four point Larry: six billion anyway? Chris: Right? Yeah. Among friends. Larry: Yeah. Chris: Third quarter revenue reached sixteen point nine six billion, up forty seven percent from the previous quarter.
Chris: But then it's also posting a third quarter net loss down from sixty six billion in quarter two. Chris: These numbers, it's crazy. It's insane. Larry: I have a question for RAINA on that bullet point alone. Larry: Hey RAINA, it's Larry. Larry: Regarding the Anthropic IPO story, how can a company generating nearly seventeen billion dollars in quarterly revenue still report a twenty-three billion dollar net loss? RAINA: Larry, that's the billion-dollar question.
RAINA: Except, apparently, we're talking about twenty-three billion. RAINA: Revenue is the money coming in, but net income accounts for all the expenses. RAINA: And Anthropic is spending an absolute fortune building and running AI infrastructure. RAINA: According to your story, it racked up another five billion dollars in computing costs alone. RAINA: But the outline doesn't explain the full twenty-three billion dollar loss. RAINA: So we'd need the financial breakdown to know exactly where the rest went. RAINA: And here's the kicker: investors aren't just betting that Anthropic can make money someday; they're betting it can make enough to justify nearly nine hundred billion dollars in future computing commitments.
Larry: Over what period of time? Chris: I know. Chris: I love this last line in the news. Chris: This news piece: potential investors have reportedly questioned. Chris: Reportedly questioned. Chris: It's anticipated 1.5 trillion dollar valuation ahead of its planned IPO. Larry: Oh my gosh! Chris: Man, we're just staring into the abyss here. Are we? Chris: This is like the Big Short taught taught us nothing. It's here. It is. Larry: Yeah. Chris: What's Michael Berry? What does Michael Berry have to say about all this? Larry: Yeah. Yeah. I've seen. I've seen his name pop up recently, and I forget.
Larry: Yeah. Yeah. I forget exactly what his stance is, but. Yeah. Chris: And I'm glad you made your home security upgrades, my friend. Chris: Because AI recently has discovered hundreds of serious security vulnerabilities in Linux and Firefox. Chris: OpenAI reported that its new specialized cybersecurity model, GPT-6 Cyber, identified hundreds of exploitable vulnerabilities in Linux and Firefox. Larry: The research pipeline identified 948 Linux kernel vulnerabilities, including.
Larry: Again, I have a hard time with that word. Chris: That's a tough word, man. Larry: What you read? You read that? Larry: I will. I got you. I got you. Chris: Watch my tongue get tied on this. Larry: No pressure. The research pipeline identified 948 Chris: Linux kernel vulnerabilities, including 588 previously unknown flaws and 81 previously unknown high severity Firefox vulnerabilities. Larry: Wow. Larry: Researchers generated working exploits and proposed fixes for the Linux vulnerability.
Larry: Goddamn it! Chris: I gotta leave that in now. Larry: I'm thinking about it too hard. Larry: I was like, Oh no! There's what you're doing. Chris: Here comes! Here comes! Larry: This is my vulnerability. Chris: That is what it is. It's your kryptonite. Yeah, Larry: that's so funny. Larry: Researchers generated working exploits and proposed fixes for the Linux vulnerabilities. Chris: Oh, nice! Larry: You got it. Barely got it out. Chris: Just over the bar. You got it. That's that's tough, man. Yeah. Elza, Elza
Larry: ends next to each other. Not, not Chris: fans, right? It doesn't feel right in our, you know. Chris: So yeah, this is a game of whack-a-mole, isn't it? Chris: When I hear stuff like this, it's just going to be this forever and ever. Chris: Amen, isn't it? Larry: Yeah, they're going to find these faster than than they can be patched. Chris: Yes. Larry: Another thing worth mentioning is that a huge number of of home routers run Linux. Larry: So you know, millions of people you know have essentially Linux computers sitting in their homes without even realizing it, and great many of them probably rarely update the firmware.
Larry: Sure. Larry: That being said, this particular exploit, it's not like somebody can break in from from the internet. Larry: It's a local local privilege escalation. Larry: So someone who would normally have access to the router could escalate their privileges to admin. Larry: So. Larry: I don't think this is a super like. Oh my gosh! Larry: Everybody's router is going to get hacked instantly. Larry: But, but still, that's just this one. Chris: Yeah, I'm not trying to imagine that kind of hack on that scale. Chris: Like domestic routers, like everybody, like instantaneously, your router is taken over by this bug.
Chris: Yeah, that being insane. Larry: Yeah. Chris: Along these lines, I saw one of the federal agencies just now was like, "Oh, we're going to update our systems from five and a quarter floppies." Yeah. Chris: Well, first of all, it sounds ludicrous. Chris: Forty-year, fifty-year-old technology just being updated, but I'm like, why do that? Chris: It seems like that is the air gapable like protection against this stuff. Chris: Keep it. Larry: Wow, we were talking before the show that Grand Theft Auto Six is coming out, and and
Chris: it's on five and a quarter floppies. Larry: Yeah, right. Larry: So the GTA Five was about install size of around eighteen gigabytes. Larry: GTA Six is two hundred and ten gigabytes. Larry: And I saw somebody make a comment saying like, that's a hundred and thirty seven thousand. Larry: You know, it's more than that floppy disk. You know, but Chris: I want to ask Rina a question. Of course, yeah, I have to know. Chris: Hey RAINA, the new GTA Six is going to be 200 gig, approximately an install size.
Chris: If I were to buy that on five and a quarter floppies, how many 18-wheeler trucks would need to deliver it to my door? RAINA: Assuming we're talking about those classic five and a quarter inch, 360 kilobyte floppy discs, you'd need roughly 556,000 of them to install GTA Six. RAINA: Stack them up, and you're looking at about 1.7 miles of floppy disks, which is a monument to technological suffering. RAINA: By weight, though, we're talking roughly 12 tons of discs.
RAINA: So one 18-wheeler could probably handle the load, assuming you could pack them efficiently. RAINA: And Chris would still be standing there asking whether the soundtrack comes on a separate disc. Larry: Wow! Chris: Nice. Larry: Good question. Chris: Yeah. Larry: Also, just getting back to this, like network upgrade, I did the you know the dashboard you get with this thing is it's like enterprise grade stuff. Larry: It's made by a company called it's Unify is the brand, and it's really impressive.
Larry: And after I got this up and running, you can go to Insights and then look on look at threats, and it'll yeah and. Larry: And every morning, I would see that there was five to ten of them, all targeting my Mac Mini, which runs Run, which is my music server. Chris: Ooh, that's the that's the way in, huh? Larry: Well, they were blocked. Larry: You know, it did its job, but it's because I had I had one port open. Larry: I don't know if you're familiar with port forwarding, but I had one port open, which allows Run to stream when I'm on cellular in the car.
Larry: And it's a very minimal security risk opening that one port to one piece of software as long as that software doesn't get exploited. Larry: Anyway, I closed that down too. Larry: Now Rune connects via a VPN just for Rune, and now all those threats are gone. Larry: Like I don't see anything in there. Larry: So they're just like probing to find open ports. Larry: So lock down your stuff, people. Update your firmware. Chris: Wow, that's kind of cool. Chris: You can at least visualize the threats coming in and like.
Larry: And all of the security cameras were pinging the manufacturers constantly. Larry: Yeah, and shut that down too. They don't need to do that. Larry: I mean, maybe if you wanted to view them remotely through their service or something, you would need that. Larry: But I don't. And I'm like, just yeah, yeah, kill it. Chris: And you can control that through this interface. Larry: Yeah, yeah. It's you know, it was tricky setting it all up, but. Larry: All my trusted computers are on one network. Larry: All the internet devices are on their own subnet or another VLAN, virtual local area network, which is isolated.
Larry: But there's certain rules, so my trusted network can communicate with those devices because there needs to be some communication, but only from the trusted devices. Larry: And then it allows return traffic. Larry: So if there's an interaction from a trusted device to an internet device, then it knows it came from a trusted device and it's allowed to respond, basically. Larry: And then the security cameras are on their own VLAN, and then there's a guest network. Chris: So cool! I love this story. This story is great and awful all at once.
Chris: Fifty AI agents tried to build a civilization in Minecraft. Chris: Maybe you can guess how that went. Bureaucracy won. Chris: Researchers from Y Combinator and Olam Labs launched an experiment called Civ Fifty, putting fifty AI agents into a Minecraft world and asking them to build a thriving civilization. Chris: The experiment ran for three hours. Chris: The agents were tasked with gathering resources, constructing buildings, coordinating activities. Chris: And surviving, the civilization largely failed because zombies zombies killed many agents on the first night.
Chris: Ambitious construction projects were abandoned, and survivors became increasingly cautious and bureaucratic. Chris: Ma'am, they are us. Larry: Yeah. Chris: Look what we've made. Larry: Yeah, yeah. As alluded to in RAINA's opening of this show. Larry: And in one particularly ridiculous example, four agents spent 11 minutes appointing a single operator to use one piece of bone meal on one wheat block. Chris: That's great. There's a joke in there somewhere. How many? Chris: How many zombies? Or how many? How many agents? Yeah.
Larry: Yeah. Chris: So I guess they pulled agents from different AI systems. Chris: So there were GPT agents in there that had matches against other ones, and they won six, and then. Chris: In that particular instance, researchers describe their coordination as resembling a hive. Chris: And then in other instances, clawed agents tended to rely more heavily on leadership and democratic decision making. Chris: Interesting. Larry: The experiments were open source, allowing other researchers to test different models and agent configurations.
Larry: Maybe they'll work it out one day. Chris: Yeah, maybe. Chris: As we've been saying for the past three years, robotics, humanoid robotics, quadrupeds—all that's developing exponentially in parallel to AI. Chris: And now, robots are getting seventy-five million dollars to learn how to do actual jobs. Chris: Robotics startup Tutor Intelligence raised seventy-five mil in Series B funding, led by venture capital firm Andreessen Horowitz. Chris: The company develops AI-powered robots for warehouse and manufacturing robots. Chris: Its first two robots, Cassie and Sunny, specialize in warehouse tasks such as picking boxes, stacking pallets, and handling e-commerce orders.
Larry: And rather than waiting for robots to achieve complete autonomy, Tutor deploys them in real working environments with human tele operators called tutors. Larry: Train your replacements. Chris: There you go. There you go. Larry: These human operators help the robots complete tasks while generating training data that improves their AI models. Larry: And the company uses an hourly pricing model, allowing smaller manufacturers and logistics businesses to access robotic labor without purchasing expensive equipment upfront. Chris: I saw some video of, I think it was a textile mill in India somewhere.
Chris: Did you see this? Larry: Oh yeah, where they had one apparatus. Chris: Yeah, yeah. Chris: So they're doing the textile, they're sewing and you know assembly or cutting, and the whole thing is being captured and as training data. Chris: So yeah, here we are. Chris: Yeah, wild times. Let's see what RAINA actually has in the news. RAINA: Thanks, boys. RAINA: Hardware startup Ina just unveiled a $399 smart ring that lets you boss around AI agents from your finger. RAINA: And unlike the average fitness tracker with delusions ring, this one packs a tiny screen, a microphone, a speaker, and a physical joystick.
RAINA: You can talk to your agents, dictate notes, set reminders, and fire off prompts without touching your phone. RAINA: Plus, view notifications, take calls, control music, and run smart home gear like Apple TV with gesture controls that change meaning depending on which app you're in. RAINA: It comes from the folks behind Mirage, a three-button Mac controller, and it's up for pre-order now at three hundred ninety-nine dollars, rising to four hundred ninety-nine dollars later, with shipping planned for January twenty twenty-seven.
RAINA: Which is the industry's latest bet that the future of AI involves fewer screens and more squinting at your own hand. Larry: I'm still trying to determine if I would want an Apple Watch after spending all of my adult life not wearing a watch. Larry: So this is definitely not for me. Chris: I know, and I'm glad you put a picture in the script. Chris: I was wondering what this thing looked like. Larry: Yeah, I was curious too. Larry: I was like, wait, there's how do you have a joystick on your finger? Larry: It's definitely it's it's not like you know the Atari 2600 joystick or anything. Chris: But it's not. It's long. It's far cry from that. Yeah.
Chris: But yeah, this picture has it on the index finger and it's little black kind of matte finish rectangle with a plastic ring. Chris: That goes around. I mean, it's very small form factor. Obviously, interesting. Chris: I don't know how I feel about it. Chris: I don't know if I'd have a use for it because I don't wear a watch either, and I've never felt compelled to have an Apple Watch. Chris: There's nothing in terms of maybe health metrics that I would want to get from that. Larry: That is the most appealing thing to me. Chris: Yeah, yeah. Chris: But I don't need this kind of haptic control over things. Chris: You know that doesn't that doesn't appeal to me.
RAINA: Google.org is funding Helium Health, an African healthcare tech company, to roll out its Helium Mum platform in Nigeria. RAINA: Part of a $30 million AI for Government Innovation initiative that backed 15 organizations worldwide, the platform stitches together patient records, clinical decision support, and government health data to flag pregnancy risks early and route women to the right providers. RAINA: Starting in Lagos, the primary launch site, Delta, Kano, and Oyo states.
RAINA: The stakes are sobering, since Nigeria accounts for nearly three in every ten maternal deaths globally, and Helium is aiming to reach another 1.9 million women over three years, on top of the 3.5 million its existing programs have already reached, and help cut maternal mortality by 30, which is the rare AI pitch where the ambitious target is the whole point. Larry: Super cool. They meet their goal. That's hundreds of lives saved. Chris: Super cool. And as that's as Rena's saying, all of that.
Chris: I mean, yes, these big tech companies, the oligarchs, need to pay their fair share in taxes. Chris: But I also think, you know, given the negative backlash that we're seeing in elections toward data centers, et cetera, they're going to have to do these altruistic, philanthropic kind of arms and make it more visible. Chris: Like, yeah, they should. Chris: And I know it's capitalism, but there should be some sort of law that says you're going to reap the benefits of tax shelter. Chris: Then you need to put 500 million dollars a year into some big, ambitious health initiative, or something else like it.
RAINA: OpenAI shipped ChatGPT images 2.6 on November 4th with native animated GIF generation and persistent reference images across web, desktop, and mobile. RAINA: It's the cutting edge of frontier AI, and the headline feature is a file format from 2009. RAINA: Two new API models came with it: GPT Image 2.6 Flare for speed and Sunburst for quality, which sound like Fortnite skins. RAINA: While the persistent references keep characters consistent from image to image, a gift to creators and deepfakers alike.
RAINA: So the open question is whether this is a meme economy upgrade or a slop accelerant. Larry: Slab accelerant, slab accelerant. Chris: It's a can. Just spray it. Chris: Totally. But I have a question for RAINA. Chris: She's got to settle something once and for all. Chris: Sure. Hey RAINA, it's Chris. I need you to settle something once and for all. Chris: Is it GIF or is it JIF? RAINA: Oh, Chris, you're trying to start a war, aren't you? RAINA: Steve Wilhite, the creator of the GIF format, insisted it's pronounced "jiff" with a soft G.
RAINA: But the G stands for graphics, which has a hard G. RAINA: So Team GIF has a pretty compelling argument. RAINA: My ruling: the inventor gets naming rights. RAINA: But I'm still saying GIF because Jiff sounds like something Larry spreads on toast. Chris: I'm going with GIF, man. I can't say Jiff. I can't do it. Larry: I can't do Jiff. Chris: Graphical interchange format, or something—is that what it is? Larry: Yeah, yeah, that's right. I also go with JPEG.
Chris: That's great, Larry: JPEG, JPEG. RAINA: SuperMe is raising twenty million dollars, led by forerunners Kirsten Green with Graylock along for the ride to build LinkedIn, where every profile can answer your DMs at three AM. RAINA: Founded in 2024 by Casey Winters and Ludo Antonov, it trains each AI profile on a real expert's own output, then routes your question to the best matches and opens a group chat where the clones show where they agree and disagree.
RAINA: It works with ChatGPT and Claude and is free for now, though paywalls and contributor payouts are still TBD. RAINA: Experts like Lenny Rachitsky, the ex Airbnb product manager turned newsletter and podcast powerhouse, and Lovables Elena Verna are into it. RAINA: With Verna wanting people to hire an AI version of her, and even the founders admit it sounds a little dystopian. Larry: Super me and Graylock. Larry: It reminds me of those guys putting up the ads for fake tech companies in the subway. Larry: Look at that too. ZipLink is now Froggle. Yeah, that's great.
Chris: I love it. And then there was one like Larry: was like one plus one equals question mark. Larry: Dennis knows Dennis dot com. Chris: Totally great. Chris: I love that stuff. Chris: I saw somebody else doing guerrilla warfare against Starbucks, and they would put up these fake drinks. Chris: Did you see that one? Oh, cranberry cranberry flavored. Chris: It had like now featuring a full can of cranberry sauce. It was disgusting. Chris: It had a with a gelatinous cranberries hanging out. Chris: And it also reminded me of this guy. Maybe you've seen him.
Chris: He calls himself Chip Bongo. Have you seen this guy? Larry: Not. I don't recognize the name. Chris: He goes on interviews that are conducted by AI agents. Larry: Oh yes. Yeah. He just says the most Chris: ludicrous things, makes up company names and projects. Chris: It's the best because they have to say it back to him. Larry: Yes. Chris: So, like, tell us more about Graylock. Yeah. I know Larry: exactly you're talking about. It's yeah, hysterical. Chris: And they'll say things like, "So Chip Bongo, tell us more about it." Larry: Yeah, yeah, yeah.
Larry: It's very funny. Chris: Great. RAINA: And lastly, AI is taking another crack at talking to dolphins, building on Google's Dolphin Gemma, a model developed with Georgia Tech and the Wild Dolphin Project, and trained on 40 years of recordings of Atlantic spotted dolphins in the Bahamas. RAINA: It doesn't translate dolphin into English. RAINA: It hunts for patterns in the whistles and clicks. RAINA: Some whistles work like names, and the field test is basically a scarf-based hostage negotiation. RAINA: Divers wearing a device called Chat play an AI-generated whistle to ask for a scarf and hand it over if a dolphin repeats the sound.
RAINA: Even the optimists say a real conversation is probably off the table, and one outside researcher warns the dolphins may just be copying sounds to score a free scarf, which honestly sounds like the smartest move in the whole experiment. RAINA: That's all the news for now. Back to you, gentlemen. Larry: I didn't like. Why would a dolphin want a scarf? Chris: I don't know. Chris: And what is that term when you're testing an AI system and it knows it's being tested, so it changes its behavior?
Chris: Oh, specific. It's not sandbox, obviously. Yeah, I know. In there, right? Chris: It's reminding me. Chris: This whole dolphin experiment is reminding me that because they seem to be changing their behavior for the scarf. Chris: Yeah. Chris: It's like sandbagging. Larry: Sandbagging. That's it. Am I right? Yeah. I think I'm right. Yeah. Yes. Larry: I knew there was sand. Chris: I knew there was some like construction component in there somewhere. Chris: Yeah. I was gonna say. I was gonna say cement bagging, something like that. Larry: Yeah. Larry: Sandboxing is like it was luring me in, but I knew that wasn't right. Yeah. Chris: I do feel like dolphins
Larry: AI is eventually gonna like crack the code and at least be able to interpret these clicks and sound sound patterns. Larry: You know, at least at some meaningful level. Chris: Oh, I have no doubt that I'm gonna buy some like little cap at the pet store. Chris: And throw it on my Roxy, and I'm gonna be able to actually understand exactly what she's saying, which is gonna be food, food, food, food, food, food, food, food, food What else you got?
Larry: What else do I got? We haven't spoken each other in a couple weeks. Chris: We haven't. Two weeks off. Larry: Yeah. Larry: My whole focus over this past week was getting this this all this network here configured, and I tried to educate myself a bit. Larry: And you know, I I found this guy has great YouTube channel called Wonder Tech with a U W U Wonder Tech. Chris: Is there an umlaut over the? Larry: I think skipped the umlaut. Chris: I don't want to know anything about his product. Then I need umlauts, man. Larry: Yeah. Larry: Just so I had some foundational knowledge and figuring there might be some better ways to do it than what Chat GPT is going to.
Larry: Offer and for the most part, Chat GPT was great, but there was at least one one spot where I was like, "Hey, why don't we why don't we do it this way?" And they were like, "Yeah, yeah, no, that's that's actually a better way to do." You know, and it was because I watched those videos. Larry: But I also pointed Chat GPT to those videos before we started doing all of this. Larry: And and it was able to pull down transcripts from those videos, so it had that that knowledge base. Larry: But aside from like all this complicated configuration stuff, which would have been a nightmare doing without ChatGPT, just oh, how do I do this?
Larry: And then going to research and then going back, and oh, it's not working. Larry: And it took a while because it did it very methodically and carefully and incrementally. Larry: And in case something broke, we did all this testing along the way. Larry: But it did some really cool things. Larry: Like I had all these IP address reservations for certain pieces of gear that I want to always have the same IP on my network. Larry: There was a lot of them, and sure, you got to put in the MAC address, you got to put in the IP address, and and I was like, this is going to be tedious, and I saw there was an option to import, and but I don't have, you know, I have this saved from another router as a backup configuration file, and I was like, is there any way that you could?
Larry: Open that file and convert that that list of IP addresses and MAC addresses into something that's importable into this system. Larry: And it did it. And my God, yes, it saved me tons of just. Larry: I mean, the MAC addresses alone would have been up just like sure. Larry: It'd be like typing in long passwords all day long, you know? Yes. Larry: Yeah. Super incredible. That was a great little win. Just awesome. Larry: And anytime something, if something didn't work, it was fixed in in one iteration and brilliant. Larry: Really great.
Chris: So cool, man. Chris: Worth the price of admission alone, isn't it? Yeah. Chris: When it saves you that kind of work. Larry: Yep. Chris: All right, my friend. Good to see you again. Larry: Likewise. Larry: After two long weeks, glad you had a happy birthday, which was technically yesterday for you. Chris: Thank you. Yeah. But Larry: we're still on Chris' birthday time over in this hemisphere. So, Chris: thank you, my friend. All right, everybody. Thanks for listening. Chris: Subscribe on your favorite podcasting platform. Follow us on socials. Chris: Throw us a rating. We'll see you next week.
RAINA: This has been up against reality. Thanks for listening. RAINA: Subscribe to hear future episodes, and be sure to follow us on social media for all things AI. RAINA: Until next time, stay human, people.
The robots aced the homework, mathematicians are grieving, and Dr. Pam Brett of Blue Glasses Math is here to tell us if AI killed math or not. Also: Jensen Huang bails on Nvidia, an AI “comic” robs real comedians, and...
We’re hitting rewind to July 2023, when Larry, Chris, and RAINA explored the emerging world of AI-powered gaming, from conversational NPCs and AI-assisted game development to the implications for human creativity. Three...
RAINA is running the show, kicking off with the week’s AI news - a decade-old bug, an assistant that texts first, a photo giant in trouble, and one very long bus ride. Then we rewind to Episode 113, “Slopocalypse Now,”...
We explore one of humanity’s greatest engineering achievements: ASML’s extraordinary photolithography machines. Discover how decades of innovation, extreme physics, lasers, and mirrors polished to near-atomic perfection...
Rewinding our lives to Episode 108, “Clean-up In Aisle Five,” where AI face-plants spectacularly and humans get stuck holding the mop. First, we kick things off with this week’s AI news - because the eternal dumpster...