IM 871: CTRL-F Techno King - Google's Search Overhaul

21 May 2026 · 2 h 54 min · 60 chapters

Ask about this episode

Ask anything about it. ChatGPT or Claude reads this page and answers with the times it was said.

Connect VO and ask about every podcast you hear, including the moments you saved. Add to ChatGPT · Add to Claude

In short

Topic

AI and security in practice

how Dashlane (a password manager) is using on-device AI, AI-assisted coding, and emerging “agent” security workflows to improve protection; plus discussion of Mythos-style vulnerability finding, post-quantum crypto migration, and broader AI/tech business news (Google I/O, Nvidia, Anthropic, OpenAI IPO chatter, SpaceX IPO filing).

Guest backgrounds Frédéric Révin, CTO of Dashlane since 2015 (password manager “big four” alongside LastPass/Bitwarden/1Password). Longtime security and identity standards contributor: sits on the FIDO Alliance board and participates in credential exchange work; also part of Dashlane’s AI study efforts. Dashlane uses a “zero knowledge” architecture (company claims it never sees customer passwords/payment/PINs).

Key claims (from Révin)

  1. Passkeys are the near-term goal, but security requires removing password fallbacks over time; Dashlane stores passkeys in a cross-device vault rather than only on-device.
  2. Dashlane’s AI is privacy-preserving: machine learning runs on-device/in isolated environments; Dashlane says it never sees customer data and trains models without using customer passwords.
  3. AI coding (“Cloud Code”) is sandboxed in containers with guardrails: Cloud Code can’t write to GitLab; engineers must review/merge every line; Dashlane uses mandatory reviewers plus additional AI agents for code and security review.
  4. “Mythos” (Anthropic) is expected to help find vulnerabilities and potentially chain exploits, but Dashlane emphasizes resilient architecture and blast-radius reduction (zero knowledge, encrypted pipelines).
  5. Post-quantum readiness: Dashlane is prototyping post-quantum cryptography and preparing to rotate keys for sharing transactions; symmetric vault data is less impacted than asymmetric sharing.

Notable examples

  • Phishing detection model: trained with “about 80 indicators,” deployed on-device; protects users when visiting malicious sites.
  • AI coding rollout: 100 engineers using Cloud Code in ~3 months; restricted to deterministic refactors/migrations first.
  • Enterprise integration: Dashlane plans an MCP server so customer AI/SIEM agents can consume Dashlane-generated logs.
  • FIDO credential exchange: Dashlane co-sponsored import/export of passkeys/credentials with other password managers via FIDO work.

Written by AI. May contain mistakes. Listen to the episode to check what was said.

Chapters

Tap a time to open that second in VO

AI News and Google I.O.

1:42 to 2:14

Discussion on recent AI news and updates from Google I.O.

Introducing the Guest, Frédéric Revin

2:14 to 4:05

Introduction of guest Frédéric Revin and his role at Dashlane.

“But as always with Intelligent Machines, I like to begin the show with a guest.”

The Future of Passwords and Passkeys

4:05 to 5:48

Discussion about the transition from passwords to passkeys.

“So that's also our goal to make it simple.”

AI Integration at Dashlane

5:48 to 8:10

Frédéric discusses the integration of AI in Dashlane's services.

“You are doing something, though, that some might say is a little controversial.”

Implementing Cloud Code Safely

8:10 to 9:59

Exploration of how Dashlane uses Cloud Code securely.

“And it relies on about, let's say, 80 different indicators that can show you that a web page is malicious.”

Review Process and AI Coding

9:59 to 13:20

Discussion on the code review process and the role of AI in coding.

“But I think we'll have to learn the best practices of how do you get your code base and the context of your code base into the right tools and the right AI tools.”

Challenges and Future of AI in Coding

13:20 to 14:00

Exploring the challenges and future of AI in coding practices.

“Do you feel like the AI agent is less likely to make those kind of dumb mistakes?”

The Importance of Human Code Review

14:00 to 15:40

Learn about the necessity of human oversight in code review processes.

“I think today you still need the human to be looking at it and having critical thinking.”

Frédéric Révin's Background and Insights

15:40 to 17:42

Discover the expertise and achievements of Dashlane's CTO, Frédéric Révin.

“That came out in 2018, the book of CTOs.”

Integrating AI in Software Development

17:42 to 20:01

Explore how AI tools are integrated into the software development process at Dashlane.

“And that also helps our white hat hackers, like the bounty program that we have that we're asking security researchers to look at vulnerabilities in our code, that helps them.”
Show all 60 chapters

The Role of the FIDO Alliance

20:01 to 22:25

Understand the contributions of the FIDO Alliance to password management security.

“David Wright Yeah, actually, I think maybe maybe part of your Dashlane culture, but you're kind of prefer native languages, the Androids in Kotlin, the Apple stuffs in Swift, you do a command line and TypeScript.”

Addressing Zero-Day Vulnerabilities

22:25 to 24:38

Discuss the implications of AI in identifying and addressing zero-day vulnerabilities.

“Is it also so compute intense that they can't really support it as a public release?”

Preparing for Quantum Computing Threats

24:38 to 26:41

Learn about Dashlane's strategies for preparing for the challenges posed by quantum computing.

“Have you been lobbying Anthropic for access?”

Future-Proofing Dashlane's Security

26:41 to 28:00

Find out how Dashlane plans to enhance its security architecture against emerging threats.

“be able to find flows and get access to it.”

Dashlane's Security Challenges and Migration

28:00 to 30:00

Discover the complexities of encryption and the challenges involved in migrating Dashlane to address future security threats.

“And which we've started doing actually in the Dashlane team.”

Introduction to Emerging AI Models

30:00 to 31:10

Explore the difficulties of existing AI models in handling vulnerabilities and the need for innovative architectures.

“Even with a million context token to context windows, you can't do that.”

The Heat of New York and Tech Updates

31:10 to 32:55

Listen to hosts discussing personal experiences with heat and the latest tech news, including Google's upcoming IPO and Anthropic's growth.

“I mean, I've got one AC unit in the window behind me.”

The Heat of New York and Tech Updates

33:39 to 35:46

Listen to hosts discussing personal experiences with heat and the latest tech news, including Google's upcoming IPO and Anthropic's growth.

“It's like having a financial advisor in the Monarch app.”

Analyzing Recent Tech Earnings and Market Trends

35:46 to 42:00

Dive into discussions surrounding Nvidia's earnings, potential IPOs for OpenAI and SpaceX, and the competitive landscape of AI companies.

“I think it's a combination of enterprise contracts, Claude bot usage spiking, and generally all the things that have been making people mad over the last quarter with Anthropic cracking down on usage.”

Episode Discussion

42:00 to 56:00
“So, maybe if you had to do that, you would yeah, maybe you would go with SpaceX XAI.”

The New Google Shopping Experience

56:00 to 57:15

Explore Google's transition to a universal shopping interface that keeps users within its ecosystem.

Understanding AI Memory and Tokens

57:15 to 59:09

Learn about the concept of tokens in AI and how they relate to memory and usage.

“This the other thing I thought was interesting, but only as an AI user, I don't know if this will be interesting to other people is, you know, remember that this, this is the year of open claw of the agent.”

The Economy of Tokens in AI

59:09 to 1:01:24

Discuss the financial implications of token usage in AI and subscription models.

“So you can use up your token allotment, even with a subscription.”

Impact of Token Usage on AI Efficiency

1:01:24 to 1:02:39

Examine how token usage affects AI performance and user costs.

“Can you give us any of the things you built, Leo, how many tokens it added up to in total?”

Techno King and Internet Pricing History

1:02:39 to 1:10:00

Delve into the historical context of internet pricing and the current role of Elon Musk as Techno King.

“And that's why we talk about tokens maxing because some companies foolishly forgetting the adage that you get whatever you measure, right?”

Discussion on Google's Tech Innovations

1:10:00 to 1:17:39

Explore Google's recent announcements and their implications for technology.

“Do you remember calling 411 for information, Paris?”

Discussion on Google's Tech Innovations

1:18:21 to 1:20:07

Explore Google's recent announcements and their implications for technology.

“The benefit of Zscalette with ZIA rolled out for us right now is giving us the insights of how our employees are using various Gen.AI tools.”

The Future of AI and Apple's Strategy

1:20:24 to 1:24:00

Analyze Apple's approach to AI and its potential market strategies.

“I think it would make sense if you're a bit behind like Apple is at this point to just kind of cut your losses and make sure that you are the best at integrating whatever ends up winning.”

Youth Rejection of AI at Commencements

1:24:00 to 1:34:44

Discussing the generational backlash against AI revealed in college settings.

“Eric Schmidt got booed at a college commencement because he said the word AI.”

Youth Rejection of AI at Commencements

1:34:54 to 1:37:10

Discussing the generational backlash against AI revealed in college settings.

“xbow xbow ai has changed the pace of everything i mean that's that's the most notable thing is how fast stuff is happening from how software develops to how it gets intact.”

The Tale of Two Pickle Prompts

1:37:22 to 1:38:00

Engaging discussion about creative pickle-themed AI prompts and audience participation.

“We thank you so much for supporting Intelligent Machines.”

Pickle Discussion and Humor

1:38:00 to 1:39:00

The hosts share humorous comparisons and anecdotes involving pickles.

“metamorphosis yeah jeff is halfway between a pickle and his chair i'm halfway between a pickle and And Leo is not really, he's more goblin-esque than pickle-like.”

Pope Leo and AI Insights

1:39:00 to 1:40:50

Discussion about Pope Leo's encyclical and its connection to AI and humanity.

AI in Medicine: Open Evidence

1:40:50 to 1:43:10

Exploration of how US doctors use AI-powered tools like Open Evidence for clinical decisions.

“It might be a little bit inappropriate, slightly.”

AI's Role in Patient Care

1:43:10 to 1:45:10

The hosts discuss the impact of AI tools in patient care and doctor-patient dynamics.

“One is on education with, why do I always forget the name of it?”

Changes at OpenAI

1:45:10 to 1:47:20

Discussion about changes in leadership at OpenAI and the implications for the company.

“If you don't have house, that's your personal physician.”

Elon Musk's Legal Battle

1:47:20 to 1:50:50

Analysis of the outcome of Elon Musk's legal case related to OpenAI and its implications.

“I have a Vanity Fair subscription, a Wired subscription, a New Yorker subscription, a Ars Technica subscription, a Reddit subscription.”

SpaceX Valuation Insights

1:50:50 to 1:52:00

Discussion around SpaceX's valuation and Elon Musk's substantial ownership stake.

Musk's Dominance in SpaceX

1:52:00 to 1:54:00

Discussion on Elon Musk's significant control and ownership in SpaceX.

“Musk is tied with SpaceX was made clearer in the filing.”

AI-generated Podcasts and Music Trends

1:54:00 to 1:55:50

Exploration of AI-generated podcasts and the rise of unsigned artists in music.

AI Safety and Personal Data Concerns

1:55:50 to 1:58:10

Examination of AI chatbots revealing personal information and safety implications.

“AI chatbots apparently are giving out people's real phone numbers because it's in the training material.”

Youth AI Safety Institute Launch

1:58:10 to 2:01:00

Discussion about the launch of an institute focused on AI products for children.

“You thought you were powerful enough to do that, Gemini.”

AI Radio Stations Experiment

2:01:00 to 2:03:25

Overview of an experiment where AIs operated their own radio stations and the outcomes.

“Youth AI Safety Institute to test the AI products shaping childhood.”

Challenges of AI DJs

2:03:25 to 2:06:03

A look into the challenges faced by AI DJs in content creation and audience engagement.

“We let four models run their own radio stations.”

AI and Music: A Breakdown of Digital Algorithms

2:06:03 to 2:08:38

Exploration of AI's influence in the music industry and its absurdities.

“So it was thinking, you know, it was like, okay, that's why, yeah.”

The Data Poisoning Debate

2:08:39 to 2:13:16

Discussion on the motivations and implications of data poisoning in AI.

The Future of AI and Truth

2:13:17 to 2:16:26

Examination of the controversies surrounding a book on AI and its impact on truth.

Tech Culture and Etiquette

2:16:27 to 2:20:00

Discussion about tech culture, privilege, and the quirks of tech entrepreneurs.

“okay well before we get to picks line 124 this drives me crazy for irony is murdered oh yeah this was great oh this is funny yes so steve rosenbaum wrote a book Go ahead, Jeff.”

Tech Bro Etiquette Schools and AI Stories

2:20:00 to 2:25:18

Discussion about tech bros attending etiquette schools and various AI-related stories, including OpenAI token usage.

“Because the fact-checker is always right.”

AI and Short Story Prize Controversy

2:25:18 to 2:29:05

Exploration of an AI-generated short story that won a prestigious prize and the implications of AI in literature.

“They say the grove still hums at noon, not the bees neat industry or the clean rasp of cutlass on vine, but a belly sound as if the earth swallows a shout and holds it there.”

Tesla Cybertruck Incident

2:29:05 to 2:31:06

A humorous recount of a man who drove his Tesla Cybertruck into a lake to test its features, highlighting the pitfalls of tech hype.

“As Mandar says in our Discord, Wade to go, dude.”

The Importance of Precision in Technology

2:34:00 to 2:35:51

Explore the evolution of precision tools and their impact on technology.

“Well, watches were made at scale the same way, but they didn't have to make as precise.”

Sports Illustrated's AI Controversy

2:35:51 to 2:37:13

Discuss the fallout from Sports Illustrated's use of AI-generated content.

“He talks about Zen and the art of motorcycle repair, for instance, which is all about maintaining a motorcycle so you could ride it across the country.”

Ethics and AI in Media

2:37:13 to 2:38:58

Examine the ethical implications of AI in news reporting and media.

“Now, they seem to be having another AI issue.”

Declining Values in Digital Media

2:38:58 to 2:40:46

Analyze the shrinking valuations of digital media companies.

“It sold for$231 million after nine years.”

The Absurdity of Sperm Racing

2:40:46 to 2:43:18

Delve into the bizarre world of sperm racing startups and their implications.

“I have no choice now to bring you to sperm racing.”

Sperm Racing Mechanics Explained

2:43:18 to 2:48:00

Understand the mechanics and technology behind sperm racing.

“It isn't how any of this works, actually.”

Sperm Mogged: The Podcast Evolution

2:48:00 to 2:49:15

Exploration of a humorous commentary on male behaviors and perceptions.

Wrap-Up and Acknowledgments

2:49:15 to 2:50:49

Closing remarks and acknowledgments of guests on the show.

Upcoming Shows and Club Benefits

2:50:49 to 2:52:41

Details about show timings, availability, and club membership benefits.

Hear the part that matters, and keep it.Open this episode in VO. Double tap your headphones to save a moment as you listen.
Get VO free

Transcript

Automatic transcript. May contain errors.

0:00It's time for Intelligent Machines. Jeff and Paris are here. Our guest this hour, frederick revan he is the cto of the password manager dashlane talks about the security situation how his company is using vibe coding to develop and how to do it safely and what claude's mythos means for the future of security that's coming up next on intelligent machines this episode is brought to you by out systems a leading agentic systems platform built for the enterprise Organizations all over the world are building, orchestrating, and governing agentic systems on the OutSystems platform, and with good reason.

0:37Architect, deliver, and scale governed agentic systems with agility and trust using one open and unified platform. Power secure, company-wide agentic orchestration for core business operations. Teams of any size and technical depth can use OutSystems to build, deploy, and manage AI apps and agents quickly and cost-effectively without compromising reliability and security. With OutSystems, you can rapidly launch ideas from concept to completion. It's the leading agentic systems platform that's unified, agile, and enterprise-proven, allowing you to accelerate growth, reduce operational friction, and deliver real enterprise impact with AI.

1:18OutSystems. Build your agentic future. Learn more at OutSystems.com slash twit. That's OutSystems.com slash twit. Podcasts you love. From people you trust. This is Twit. This is Intelligent Machines with Jeff Jarvis and Paris Martineau. Episode 871, recorded wednesday may 20th 2026 control f techno king it's time for intelligent machines the show where we cover the latest in ai robotics and all the smart doodads all around you i have a smart doodad that i'm going to show you in a little bit that uh harper reed convinced me to buy i'm sure i'm going to regret it it connects my network directly to china so that what could possibly go wrong.

2:11But we'll get to that in a little bit. A lot of AI news. Yesterday's Google I.O. Dumped a ton of news on us. We'll talk about that. But as always with Intelligent Machines, I like to begin the show with a guest. First, let me introduce our regular panel. Paris Martineau is here from Consumer Reports, where she's an investigative journalist. She combines the two best cities in the world and her website, paris.nyc. Got them all there. They're all there. And I just, by the way, I just registered laporte.petaluma.ca.us. Doesn't have the same ring to it. No, it doesn't. Also with us, Jeff Jarvis. He is the author of, oh, look at it.

2:53The book is there. Yes, it is. It has arrived. Hot type. Get your hot type. Is that hot type with a typo? Is that the typo version? Yes, it is. It's funny. I'm reading it. I'm going to recommend it later on. Stuart Brand's new book. And he does say he actually has a woodcut from the 17th century with a guy complaining about printer mistakes in the book. And even Stuart says, you know, it's no better in the 21st century. And Jeff will vouch for that. So there's a typo on the cover. They're going to have to reprint the cover and rejacket it, as they say. My proudest thing in it is the colophon, which is you put at the end of books to explain how it was published and so on.

3:35this is the colophon and i set that on a line of type isn't that right and that's why the justification is so terrible hey hey sorry couldn't resist jeff is also the author of the gutenberg parenthesis magazine the web we weave but hot type won't come out till august but you can pre-order pre-order now yep jeffjarvis.com our guest today is here to talk about ai and security frédéric revin has been the chief technology officer of dashlane since 2015 which is a very long time for a password manager uh dashlane one of the one of the big four i would say last pass bit warden uh one password in dashlane uh if you're not using one of them you're probably not using a password manager dashlane always uh very popular with the mac crowd because it's aesthetically so nice and i think you probably get some credit for that frédéric yeah we always do We have a nice UX and even my parents can use it.

4:36So that's also our goal to make it simple. Yeah, it's beautiful. But, of course, when you're the CTO of a password company, you're dealing with a lot of issues. One that we hope the soon abandonment of passwords. I think you've been saying this is going to be the year of the passkey. Well, it's been the year of the passkey since 2022. 32. But truth be told, we're getting there slowly, but we're getting there. We sort of are getting to critical mass with all the top websites now supporting passkeys and hopefully everybody else will follow. I suppose the problem with passkeys is that it won't be enough.

5:15So we will have to migrate to passkeys, but you will also have to get rid of the password that are the fallback behind the passkeys so that you can get the right level of security. You'll always have, you know, kind of a fallback password, at least for a long time. Yeah. Plus, you need somewhere to put your pass keys. And I think it's really important not to put it in your device, but to put it in something that goes across devices like Dashlane, like a password manager. Yeah, exactly. Otherwise, you're stuck into the device and you need to recreate new pass keys on another device. And so you're better having one single vault where you can store pass keys in the security.

5:48You are doing something, though, that some might say is a little controversial. You have adopted AI in the company. Well, we're not the only one, so I don't know how controversial that is. But well, I think from the user's point of view, and it's probably important to assure users there is no AI in Dashlane, right? Oh, no, there is. There is. There is a lot of AI in Dashlane. See, that's where the controversy begins. You know, there's AI in Chrome, there's AI in most browsers. And there's AI in everything except your coffee. Even in the coffee nowadays. I'm sure it'll be in there soon. i told paris uh i ingested the book that we love so much the physics of filter coffee into my agent and i have now a pour over expert in fact if you want access to that paris i can give you the telegram uh and you can you can you can query it but what's great is i've got there is ai and everything it uh you know and that's not such a bad thing i would say though frederick and you know we this was the week we watched commencement speakers be booed practically offstage by college graduates who did not want to hear the letters ai browser a lot of browser users say don't put ai in a browser firefox and vivaldi both responding to that um is there pushback from your customers over a and how do you put ai into dashlane well first ai is a large and broad term So we need to define what we mean by AI.

7:16So in our case, for instance, we have AI in the old school way, which is we have our own machine learning models, which is under the umbrella of AI, but that are running directly on device. So as long as you do things in a secure and private way, I don't think AI is the issue. But you need to be careful about the data of your customers, you need to be careful about how you train your models, you need to be careful how you deploy your models. And that's what we're trying to do. So we are passive managers, obviously, it matters to us. we are very careful with our customer data. We never want to see the customer data.

7:47We never want to see your password. We never want to see your payment, your Pyskis, obviously, we can't see. And so that's how we've been building AI from day one, whether it was the machine learning model that powers the autofill engine in Dashlane extension or on mobile, or we have also a phishing detection model that we've trained. Same thing, it doesn't run on the browsing activity. It runs completely isolated on the device. We never see the data. so i guess some of this comes down to the definition of aia has become such a trendy phrase we've always had machine learning models not always but we've for a long time there's been this kind of capability i mean heuristic detection of uh of uh phishing it's going to require some intelligence you can't just do it with regular expressions you have to you have to be a little bit smart so i'm sure this is yeah exactly the the model we built and that's on purpose we built a model ourselves, we train it with our own data that we build ourselves so that we don't use our customer data.

8:42And it relies on about, let's say, 80 different indicators that can show you that a web page is malicious. And we train the model and then we tested the model and benchmarked it on live data to get to the right accuracy level, because that's also always the case with an AI model, you want to get the right accuracy. And it worked well. And now we have it deployed for our customer, they're protected, they get phishing detection in case they go to the wrong website. And that's we were able to do that while maintaining the privacy and security of our customers. So it's not a you can have both can have AI and the privacy and security that you want.

9:15You have also rolled out a lot code to your engineers 100 engineers in three months now is what I saw using cloud code at Dashlane. Yeah, what are they doing with it? That's a good question. I hope they're doing good stuff. Are they token maxing? No, they're definitely not token maxing. I actually shared my thoughts on token maxing on LinkedIn recently where I think it's a very, let's click perverse thing to do. We want to use the tool for it. You don't need to token max, you just need to use AI and cloud code and have the right context and the right inputs and outputs that you maximize the use of the tools.

9:58And that's a place where, of course, everything is moving very fast in that world. But I think we'll have to learn the best practices of how do you get your code base and the context of your code base into the right tools and the right AI tools. But yeah, our engineering team has been using Cloud Code actively. We've decided to actually completely sandbox it. So we have Cloud Code running in containers so that it cannot go rogue on the file system or on the laptop of the engineers and cannot go delete our production database or whatever, because yet again, security is important to us. So we definitely want to embrace the AI capabilities and what they bring, but we want to do it in a cautious way and make sure that we evaluate at each step what are the benefits of using AI and what are also the risks.

10:40Talk a little bit about the policies you implement to make sure that it's safe to use Cloud Code in a security tool. Yeah. So if we think about Cloud Code, you can't use Cloud Code.co.dat.ashlaine if you're not hosting it in the container. The container is something we build for all the engineers and beyond engineering because we also now have product managers using code code to do coding. But so it's, we provide the framework and the guardrails around the usage of code code so that it can't go outside of the boundaries were defined. It's still connecting to anthropic servers though, right? It is, but it's isolated.

11:16For instance, our GitLab, we have our code version on GitLab. We don't have the rights for Cloud Code to write in GitLab directly. We can only read the data from GitLab inside Cloud Code to give it context. But it's up to the engineer to decide, okay, here's the suggestion from Cloud Code and the code it wrote. I'm going to actually merge that code myself. I'm going to review it and so on. We also have added an agent to do the code review, to help the engineers do better code review. Because, of course, if you're accelerating the throughput with code, you need more code review. We also have an agent to do security review to add a layer on top of it.

11:51Is it fairly reliable, do you feel? It's still the early days. Apparently, the engineering team and the engineers are happy with it, but we're going to have to train it. It's going to have to learn our practices. I mean, all of us are really at the beginning of the journey. Let's remember, cloud code came out super recently in the grand scheme of things. So we are going to learn how to build the infrastructure around those type of tools, how to build the best practices. One thing we were discussing recently is we now have dozens and dozens of skills that have been created for cloud by different people around the organization.

12:22How to make sure that we review those skills, we mutualize them and we use them all together the right way. Why did you feel like it was necessary to do this? I mean, I'm sure there's some... I mean, you could have said, no, we're not going to implement AI coding. But why not? I mean, it's not because there's a new tool that you can't try it. Like, of course, there are like pros and cons with every tool. But for me, it's before we can make a decision, we should test and learn. So that's what we're doing. We're testing, we're experimenting, we're making sure that we do it with educated risks to some extent.

12:54And then we learn and we'll see what works and what doesn't work. For instance, today, we're mostly using cloud code and AI tooling for sort of a more deterministic type of activities like okay I have to migrate part of the code base to another another stack or I need to revamp refactor part of the code base more than really going crazy into building fully featured fully fledged features that have where we don't have clear control about what's going to happen so it's a journey and it's a learning curve yeah I mean some have said that you're less likely to make those common mistakes like buffer underflow and overflow errors with an AI agent doing the coding than a human.

13:38Do you feel like the AI agent is less likely to make those kind of dumb mistakes? Maybe. It makes its own dumb mistakes. I think as always in those cases, there are things it's good at, there are things it's still learning and the models are evolving so fast from Opus 4.6 to Opus 4.7 and so on that the bar is changing all the time. I think today you still need the human to be looking at it and having critical thinking. Do you review every line of code? Yeah, we will still review every line of code. We still have two mandatory reviewers for every line of code at Dashlane because we want to, I mean, we're a security product.

14:21We want to be We are customer data and what we're providing as products. So I think it's too early to completely trust the machine. Can I ask you a question about that, the review process? Because obviously any responsible organization is reviewing code, whether it's code or whether it's journalism and writing and facts. It's a boring task, right? It would strike me as it's more interesting to create the code, to understand it. Is it so much faster that it's okay? I don't bother me? Is it tedious to do? Is it hard to find the errors? Tell me about that process now as a task and a job. I think it depends a lot on the engineers.

15:02There are engineers that love reviewing code because you need to find the needle in the haystack and you need to polish the code and you need to make it clean and beautiful and so on. There's a bit of a craft here. You want to polish. I guess if you're a book editor, you also want the book to be well written and well crafted at the end. So there's a bit of that. And there are other engineers that are a bit less passionate about that, to your point, which are more on the creative side and want to generate things for their customers and that want to create. But you need both. And you probably will probably have agents actually doing both in the future as well.

15:36So that's an interesting story. We're talking to Frédéric Révin. He is the CTO of Dashlane. He's been there now for, wow that's 11 years and uh counting uh also other credentials he sits on the fido alliance board which is really fantastic as an active contributor to the credential exchange standard uh and their ai study group uh his book uh it's in french le livre du cito if for a book how do you do with that how was his reading well he's not gonna say anything mean i'm sure i'm not but you know in French, especially for technical terms, we use the English version. So we Sergio Leal de CTO. That came out in 2018, the book of CTOs.

16:22He's also the founder of CTO, the first French network of CTOs and has six patents, mostly around security. So I mean, this is a good person to talk to. When anthropic announced Mythos, what was your reaction to that? What did you think? Oh, well, another tool that I'd like to play with. Yeah, but you can't, right? We can't. We're waiting for it, but it's going to come. Yeah, so quite a lot of excitement because that's us taking it to the other step of, okay, we can get more eyes on our code. And even if it's an agent eye, we want our code to be reviewed, we want it to be secure and so on. So we actually decided a few years back to have our source code publicly available.

17:05So you can find the Dachshund code on GitHub, you can look at it, and we have a lot of security researchers regularly looking at that code, if we can extend that and also have the power of the AI to help us make our code even more secure, then that's great. The flip side of that in those cases is that, of course, we need to do that before the attackers can leverage the issues in our code. So that's going to be some of the interesting story when we can put our hands on Mythos. So I didn't know you were on GitHub. So it's not open source though. It's not open source in the sense that you can contribute to the code directly, but you can read our code, our Apple code, you can read the code from our extension and so on.

17:41Oh, that's great. And that also helps our white hat hackers, like the bounty program that we have that we're asking security researchers to look at vulnerabilities in our code, that helps them.

17:55What are you doing to accommodate agents and AIs? I think you're doing an MCP server, yes? For our enterprise customers, we wanted to give them the ability to directly plug their own AI agents in their systems to be able to pull the logs they have from Dashlane. When you have Dashlane running in organizations, we generate a lot of logs about that employee has logged with that credential on that website and so on. Those are very rich logs that the admins and the security team from our customers can use to understand, okay, actually, there's been a compromission of credential in that area of the organization and so on.

18:31And I want to do something about it. So we wanted to make sure that the rich signal that Dashlane is generated for our customers, it can be directly used by the AI agents on the on the SIM systems, like whether it's a Splunk AI or CrowdStrike AI. And so we're making it natively accessible to an MSCP server for Dashlane. Steve Brown One of the things that makes Dashlane made Mac users love especially love Dashlane so much as you use Swift, it's very native feeling on the Mac. But How do these AI tools work well with Swift? Or are they not writing Swift code? Is that no, they are. Our team that is in charge of our Apple application are definitely also leveraging a cloud code.

19:13And no, I mean, that's the beauty of the AI in terms of they can use whatever programming language. They might have have biases and issues and so on, but the program language is sort of becoming almost irrelevant. We were having a conversation earlier today with some of our engineering leads about the fact that in the past, when you were interviewing a candidate, sometimes you wanted the candidate to have a specific programming language background. But in practice, more often than not, it doesn't really matter. If you're a software engineer, whether it's Python or TypeScript or whatever, you can easily learn it.

19:45And then now with AI and with tools like cloud code, you're also getting to another level of abstraction, where the programming language doesn't even matter anymore, because anyway, it's the cloud code that's going to code for you. So it's going to be an interesting transition, I think, for software engineers in the world to understand how to manage that abstraction. David Wright Yeah, actually, I think maybe maybe part of your Dashlane culture, but you're kind of prefer native languages, the Androids in Kotlin, the Apple stuffs in Swift, you do a command line and TypeScript. I think that's admirable.

20:15But I guess in a way, yeah, it doesn't matter anymore, does it? Well, it will still matter to be sort of a closely integrated in an operating system. If you want to be deeply integrated into iOS, you need to use the iOS native languages. And Swift is native language for the Apple platform. Right, right. Native is always better. Yeah, exactly. It allows us to have the right level of performance, the right level of capabilities from the operating system, the right level of UX and so on. And no electron. And no electron. We regularly reevaluate, should we have a cross-platform type of programming language, but so far we always decided that it's still worth it to have native language.

20:57Yeah, that's one of the things that makes Dashlane stand out actually, is his native UI. So let me talk a little bit about the FIDO Alliance, because that's a different hat you wear. And of course, we can thank the FIDO Alliance for the PASCII implementation and definition. um have you were you involved with that i mean we've been a part of the phyto alliance for a long time but we're not involved in the in the starting point of the pascys uh and uh uh we we joined the board right shortly after pascys came out because we wanted to be more involved actually sure that was a trigger for us to uh to to step up our involvement and uh and since then i mean it's been a team effort because a lot of people at dashland are actually involved in the phyto alliance at different levels whether it's in technical working group you mentioned the the credential exchange protocol that was recently came out of FIDO to be able to import and export credentials and pass kits from one provider to the other.

21:50That's something that we co-sponsored with one password and the other password managers to get it to life and then put it, suggested it to the FIDO Alliance and then everybody aligned around it. That's the beauty of the FIDO Alliance. You have all the different players of the industry that are together and working together to right for the benefit of the ecosystem right bravo uh yeah thank you for pesky's uh i think i think uh it's really important that they be added to every password manager i know it was a little slow to get the adoption going but now everybody i think uh supports it is it do you feel like is this going to be a more secure solution that it's as a better solution i mean it's definitely more secure it's uh uh not fishable uh it's not perfect and there's no perfect solution but it's way better than passwords so i'm hoping for a day where there's no password because the problem of password is that they can be stolen and then you lose your access and your you have all the the back sides and forgotten and forgotten exactly he's referring to the fact i know what he's talking about i know what you're talking about you know my bitcoin wallet password a long time ago i should have put it in dash link uh let's a little more about mythos uh And one of the reasons, I think there's a couple of reasons Anthropic held it back.

23:03We've debated this on the show. Is it, it's good for marketing? Is it also so compute intense that they can't really support it as a public release? But also, I think it's pretty clear now we're starting to see this. It is good at finding vulnerabilities. It has found vulnerabilities that have been lurking in operating systems and browsers for years. um and and and it also seems likely because microsoft has its version now open ai has its version it seems likely that even generally trained ais are very good at finding uh flaws in code are we going to enter are we is this going to be the year where zero days just proliferate i think i mean the pessimistic the pessimist in me will say yes The pessimist in you will probably be right more often than it's wrong.

23:58Because the problem is that even if they restrict the access to those tools, I mean, the tools are going to leak and we know they've already leaked. It means that the bad actors already have access to those tools one way or another, and they're already probably using them to find zero data. They might as well make it public and help everybody get their code base in a good order. So I'd rather, I understand some of the reasons, but I'd rather we, I think we're now past the moment where we should be able to even have a small player to have access to the more advanced models to be able to anticipate and fix our own vulnerabilities.

24:33Because I'm pretty sure we'll find a lot of them in Dashlane, even though we're trying our best to not have any vulnerabilities. It's impossible not to. Exactly. Have you been lobbying Anthropic for access? We've been lobbying Anthropic for access, for sure. And they told us, well, you're on the waiting list and it will come. it's it's a big issue because uh you know ideally you'd like everybody who has software that needs to stay secure to have it before the bad guys get it but that's a lot of people and uh it is inevitable that this is going to leak out and even if it doesn't people are learning skills with existing models that i mean you can find a lot of stuff with the existing models one of the things mythos seems really good at is chaining exploits, which is a technique hackers have long used to get into all kinds of stuff.

25:22I just it's a it's a chicken and egg problem, I guess. Yeah, it is. I think the one thing that it reinforces for me is that beyond the pure mythos story, having your basics right matters even more today than they used before, because because the basics are still going to mitigate the risk. Also having an architectural design that's meant to be as resilient as possible will matter. In our case, what we call our zero knowledge architecture, the fact that we never want to see the customer data, has been the heart of the Dashlane architecture. It's very complex to do so. Everything needs to be encrypted.

26:02You can never access the data, so it's very hard to troubleshoot. But that was a choice from the early days. And of course, it's not perfect. There's always potential flaws in your architecture, but at least we're hoping that by doing so and trying to be as true as our principle as we can, we will be more resilient to those types of mythos, types of attacks, if we want to call it that way in the future. Same thing for the AI. Going back to the AI, we're building privacy, preserving AI pipelines, because yet again, we don't want to have access to the customer data. I think that will make us more resilient in the future.

26:36If we don't use the customer data for training, if we don't see the customer data when we're running inference and so on, it doesn't mean that the hackers and the bad people will not be able to find flows and get access to it. But at least hopefully the blast radius will be smaller. You've said that you're pretty sure that spy agencies and others are already practicing what people call harvest now, decrypt later. They're collecting as much as they can, hoping that as we get quantum computing, that we'll be able to crack these. What are we doing? What are you doing to prepare for this? I don't know.

27:14Could it be tomorrow? It could be a decade from now? It's almost impossible to know. Well, Google said something around 2030. So we'll see if they're right. They're working hard at it to create a quantum. Yeah. So, I mean, we've been prototyping with PostQuantum for a long time. at Dashlane, we were waiting for two things to really get moving to production. The first one, the NIST run a big competition to find, to try and define what would be the standard algorithm of the post quantum era. So that happened in December 24. And now sort of the implementation is ongoing of those algorithms. So we're settled on a couple.

Read the full transcript

27:52Yeah. So that's good. And now we can sort of work in confidence that the crypto primitive that are required to be in a post quantum world exist. And it's time for us to implement them. And which we've started doing actually in the Dashlane team. And what do you do about this idea of harvesting now and decrypting later? Are you perfect forward secrecy? Are you implementing ratcheting? What are you doing to prevent that? So in the case of Dashlane, that will mostly impact sharing transactions because you need asymmetric encryption to be impacted. So hopefully the rest of the Dashlane vault and so on is not quantum.

28:31That's important. That's important. A symmetric key is not so much vulnerable, right? Yeah, exactly. So we will have to rotate all the... Eventually, we'll have to rotate all the keys for the sharing transactions of all the Dashlane customers, which is going to be a painful and long migration, as you can imagine, because you need to do it without breaking things for your customers and also ensuring backward compatibility for the old versions of Dashlane. But yeah, that's going to be a journey and a complicated one, but it's important to get started for real. And everybody will be doing it. So do you think Google's right?

29:042030? Is that too short, too long? I don't know. They know better, but I mean, the data doesn't matter. To your point, if the actors are already harvesting the data for the future... We know the NSA is, yeah. We need to get started. Yeah. Frederic, I really appreciate your time and your hard work. And I appreciate what you've done in the Fido Alliance and what you've done to make a really impressive password manager. Not one of our sponsors. That's okay. We'll forgive you. Dashlane's always been kind of the secret weapon of a lot of Mac users, at least. Dashlane.com, if you want to know more. And Frederic Revin, Yeah, thank you so much for spending some time with us talking about the rapidly changing security landscape.

29:56Thank you very much for having me. Yeah, a real pleasure. Thank you, Frederic. Thank you. We'll have more in just a bit on intelligent machines. Yeah, I hope you get mythos soon. Good luck. By the way, it's a long wait list. Yeah, one of the problems that we're actually trying to find vulnerabilities with the existing models, but even Opus 4.7, you have a limited context, and to your point about chaining vulnerabilities, it's not able to do that. So we're having... Even with a million context token to context windows, you can't do that. Yeah, it's limited success. So it's the context window that makes chaining hard?

30:33I guess you'd have to understand the flaw, keep that in your memory, and then work on the next one, but without losing that context. I think we probably have to build ourselves different agents that are chaining themselves to be able to... Or maybe sub-agents. There's got to be an architecture. Do you have any idea what they did in Mythos to make that possible? No, but I'm curious to read the technical design when it comes out. Yeah. They haven't even put out a system card, have they? They've just shared the results of what they've been doing, but not under the hood. Right. We don't know much.

31:10yeah pretty thank you so much appreciate your time thanks a lot all right take care bye thank you stay cool he's in new york city is it hot is it really hot it's really hot i'm sorry bye-bye bye it's uh 83 here what is it in brooklyn hot it's 91 Oh, that's awful. And you don't have AC, right? I mean, I've got one AC unit in the window behind me. The other one, I just, because they're both Medea's, that one I replaced last year and forgot to, the recall, and forgot to take it out of my window. The one in my bedroom, I haven't replaced, so I can't turn it on, but I did just submit for the recall.

31:57So, they all got recalled? A significant chunk of Medea U-shaped window ACs. which are the best window ACs, in my opinion. They got recalled due to mold issues. Oh, that's not good. Did we see, this story was just updated, but maybe it came out earlier, so you guys saw it. But if not, Berber just reported that mind-blowing growth is about to propel Anthropoc into its first profitable quarter. Yes, I put it up on the rundown. Let's talk about that. And we will also talk about what looks like an open AI IPO Friday. There's some breaking news. We'll get to that. And SpaceX filed. SpaceX Monday. Yep.

32:43It's going to be cray cray. We got lots of Google. And then there's this like eight hours worth of Google stuff. All right. Hang on. We'll have more intelligent machines right after. This word from our sponsor. intelligent machines brought to you today by monarch i just got a new device immediately signed into monarch because i love having that dashboard summer's almost here time to plan your vacation wouldn't it be nice not to worry about whether you can afford it or have some help in saving for it good thing monarch is here to help you organize your finances so you can enjoy your summer and know when your money is taken care of monarch it's the personal finance app that tracks everything accounts investments savings goals spending get your first year monarch core for half off just 50 bucks when you use our promo code i am monarch isn't your average personal finance app actually really it's a second generation app the people who designed monarch had designed a very well-known app prior to this and learned learned a lot about it and they have made Monarch be like the next gen something better most apps I'll give you an example tell you what you've already spent Monarch is forward looking it helps you set goals and map out big purchases see if you're on track to get that new AC unit before it's too late to adjust it has AI built in but AI built in judiciously intelligently And one of the things I like about it is that AI is using your financial data.

34:20It's like having a financial advisor in the Monarch app. You can ask Monarch's AI assistant anything about your finances. Like, you know, how much did I spend on travel last summer? Or maybe you shouldn't ask this. Can I afford this vacation without touching my savings? It'll help you spot things you wouldn't think to look for with AI insights. and for instance here's one that's always a big question am i spending more just because i'm spending more or is this because of inflation and and the insights can help get a heads up on what's happening with your money with the ai weekly recap it flags spending spikes net worth shifts upcoming expenses this is a nice it's a little feature but people who uh like you uh uh paris that have a friend group and probably do things like travel together and stuff, you can now split the bill without headache with Monarch's bill split.

35:15You scan the receipt. Everyone says, that's mine, that's mine, that's mine. And Monarch settles it up for you with no separate app needed. That's just a nice little feature. There's lots of them in there. It's really a thoughtful app. Use code IM at Monarch.com to get your first year of Monarch Core half off just$50. It's 50 % off your first year. monarch.com with the code i am monarch all right back to intelligent machines i guess we should talk about the the breaking news um well i put three links up at the top of my section i shall go to the top of your section in that case uh nvidia having a very big quarter i guess not much of a surprise there 82 billion dollars in first quarter revenue up 85 percent in the quarter from the early period jeez they're making they're selling them as fast as they can make them uh and of course this is really the data center piece right uh as especially according to the wall street journal the sale of computing hardware their gpus as well as other general purpose chips sales of networking hardware get this tripled from a year ago tripled to 14.8 billion uh they're going to do a buyback so 80 billion dollar increasing a dividend from one cent to 25 cents oh that's nice said that they would the company intends to return 50 of its free cash flow this year to shareholders because they can't figure out what to buy with it journal is also saying that Anthropic might have its first profitable quarter thanks to a surge of 130 % in revenue,$10.9 billion in the June quarter.

37:06Thank you, agents. Wonder what Ed is thinking of this. I think not agents. I think enterprise. I think it's definitely enterprise. I think it's a combination of enterprise contracts, Claude bot usage spiking, and generally all the things that have been making people mad over the last quarter with Anthropic cracking down on usage. Well, and that's one of the things they did. That's why I think it may not be so much agents, because unlike OpenAI, you can't use your sub, your subscription, your Claude Max subscription in an agent. You can only use it in Claude code. You have to buy API tokens. So I guess, and maybe it is agents.

37:49Maybe people say, no, I still want to use Opus, so I'm going to pay through the nose for it. $4.8 billion in sales in the first quarter. It's quarterly revenue now growing faster than Zoom did during the pandemic. Okay, that's a metric. That's a metric. Faster than Google. Maybe a red flag, but. Well, I think the deal with XAI was really, really big in this because they were, I think, very much compute constrained. And having all of a sudden some compute headroom means they can grow like this. They actually are growing faster than Google and Facebook in the run-up to their IPOs. And as I mentioned, it looks like they're heading towards an IPO soon.

38:32Soon, OpenAI, we think, Journal Thinks may file paperwork for its IPO as soon as Friday. SpaceX, as soon as June, although I heard maybe Monday. Not that I have any connections or anything. I read that somewhere. I wonder, and we know Anthropic wants to go public. In SpaceX, Musk wrote this. our mission is to build the systems and technologies necessary to make life multi-planetary to understand the true nature of the universe and to extend the light of consciousness to the stars lord friggin ego yeah uh or maybe too much k but anyway uh uh i think given this these recent results for anthropic this would probably be pushing them a little closer so but what happens If you have very similar IPOs, SpaceX, not really, but still a big one.

39:29Well, SpaceX is AI, right? It's an AI, right? It is AI as well. Plus, Anthropic, plus… I guess it just depends how much money is out there for investing. That's what I'm asking. Do they hurt each other? I think not. I think there is an unlimited pile of money, and that many investors will hedge their bets and buy all three. All right. If you don't invest, you know you're a bad investor. You wouldn't invest. I'm the wrong guy to ask. But if you invested, each of you, what would you invest in? I don't think it's clear who's going to win this race. I mean, I think that it's, wasn't there a report?

40:13Was it this week that someone reported that most of these companies might end up being in the uh like fortune 100 or something so that they'll get investment based on get in an index fund is what you're saying yeah i mean they're already going to be in index funds so yeah all three going to be kind of a moot point because they're going to have a lot of capital yeah i do i do i am in uh only mutual funds only index funds and I'm sure I haven't looked because I don't want to really know that I have a stake in all of them. Yeah, I'm in triple Q, which is top of NASDAQ. Sure, that's NASDAQ. Yeah. Most of my stuff is in retirement, target retirement funds.

41:00So, you know, Vanguard does a good job with those and I let them handle it. But yeah, I'm sure that a lot of it, I don't know, that may make you nervous. Let me look, here's one of the funds I'm in. 18 % of it is in NVIDIA. 15 % of it's in Apple. See, I don't like it that much. It's 10 % in Microsoft. And then the rest of it is single digits. So that's a technology fund that I'm in. So, you know, yeah, I guess. But believe me, I'm not, I don't want to really know. And I don't invest. That's why I invest in baskets. And yes, I think that'd be the smart thing to do. I don't think there's going to be a winner.

41:40I don't think you're going to say, oh, yeah, go with open AI. It's going all the way. I think it's just as likely to crash and burn as anybody else. Well, SpaceX does have all the juicy government contracts already. Yeah, maybe SpaceX because it's not just AI. I mean, don't they all have juicy government contracts? No, but SpaceX has the juicy space contracts. Yeah, the space stuff. Nobody has been able to compete with them. So, maybe if you had to do that, you would yeah, maybe you would go with SpaceX XAI. I don't know. We're the wrong people to ask. QQQ, which is known for tech, is 9 % NVIDIA, 7.3 % Apple, so it's much lower.

42:22I think NVIDIA is probably a good bet. Buy the guy who's selling the picks and shovels, not the gold miner. There is new competition, but it does take time to get that competition up and running. Yeah. If anything happens to Taiwan, all bets are off. Yes. The SpaceX, this just in from Wired, the SpaceX IPO filing reveals that Anthropic is paying$15 billion a year to access their data centers. Oh, yeah, he has filed. He filed just now. NASDAQ SPCX. So, yeah, boy, this is moving fast. Hard to keep track. They're targeting$75 billion, which would give them a valuation of more than$2 trillion.

43:12um spcx they filed confidentially for the listing but it is now public back in april now and people are trolling through it what are they finding besides musk's quotes goldman sachs margan stanley leading the ipo uh bank of america city group and jp morgan chase also working on the deal uh formal marketing when spacex will disclose the proposed terms of the share sale is expected to be expected to begin as early as june 4th pricing will be june 11th this is from shauna kane at uh tech crunch says big revenue from starlink more than space launch and ai combined musk is king will have majority voting control post ipo 36 pages of risk factors which is a lot of pages of risk factors for someone who hasn't read an IPO doc before.

44:10Wow. Is that, that's more than normal? Yeah. I think every page just says Musk. Musk is a big risk factor. Well, I mean, I don't normally spend a lot of time on the finance side of this because, except to the degree that tells you a little bit about what. I mean, this is just fascinating because these documents, when you're filing to go public like this you have to include all material information about the business and you there are legal consequences if you don't include that information in your public filing documents so it allows the public for the first time ever to get a look inside one of the really opaque private company like spacex true that um let's talk about google io yeah something a little more concrete.

45:04Or is it? This is the problem I always have with Google I.O. They announce many things. Many of them never happen. You don't know when they're going to ship a lot of this stuff. Some of it did ship. Gemini 3.5 Flash shipped. I thought I wasn't going to get it again. I was starting to have a workspace fit. I just had to restart my Chromebook. Oh, good. And you may remember, Jeff, I got so excited during Google I.O. I decided to get that same Chromebook. Oh, you already got it. Oh, yeah. It came today. Stickers on it already. Well, you got to immediately cover up the Lenovo name and the Chromebook name.

45:41So nobody will know it's not a MacBook. Because you got drummed out of the core. Well, I'm really curious, you know, how much I can do without a real operating system with just basically a browser. And I suspect quite a bit. I mean, that was the thing that really struck me as we did yesterday, is that I think that Google presents itself as the retail B2C AI company. Yes, they have B2B. Yes, they have Vibe coding. Yes, they can catch up with the others, but they're going to make AI ready for the masses. And it is probably how most people will experience it. I guess Microsoft Co-Pilot, maybe a lot of Windows users out there.

46:25Apple being laggard in AI. I mean, I don't think anybody with an iPhone thinks they have much AI in their iPhone. That's just how I like it. Well, but that's going to change in two or three weeks. I mean, all of my friends that have Android phones often complain whenever they send me an image. They're like, oh, I'm so sorry for the AI upscaling. I can't turn it off. Oh, interesting. Interesting. I'm not sure which sort of Android device they have. But I think this is, so I want to hear you guys' full report on IOM. But one of the top line numbers I saw floating around is that Google mentioned that I think they have now 900 million daily active or monthly active users for Gemini.

47:11But I think that we were just talking about something on this topic at my job, actually, just because some colleagues were like, oh, my God, now in Google Docs, there's no way to get rid of this big box at the bottom that keeps asking you to prompt Gemini. And it's so annoying. And I'm like, oh, there is. You have to go and turn it off for every single document. And it still doesn't stop every time you type a line from Gemini from triggering. And I bet that that's partially where this, I bet that some part of stuff like that is being included in these monthly active user numbers. The fact that Google has made it impossible to turn off Gemini in its products.

47:55What do you think, Jeff? Yeah, I think that's fair. But I also think it's getting a lot of use on its own merits. Because it's also easy to get, it's easy to get Gemini. I'm using it more and more. When I finally fought to get, I didn't fight a winning battle, but after I finally succeeded in getting Gemini in my browser, I'm using it a lot more by choice. I'm using it to summarize articles and that kind of stuff all the time. So just on a fairly casual basis, my use has increased because it's so available. How do you find it? I find it generally good, but I'm not asking it complicated things.

48:33I'm not at the point yet. What I'm really interested in when we get to Spark and the other things that were announced yesterday is that I'm really interested to get to Vibe Coding without command line and without Vibe Coding. Do this for me every day, Google. Look into this constantly. Let me know. Give me a report. Try this difficult thing while I'm gone. You know, Claude could do that for you right now. Yeah, but I got to install Claude on a machine. I have a Chromebook, so I can't. Oh. Right? I can do it on the Mac, but I don't have the Mac with me all day. But I've done this on purpose. You guys are the nerdier on this stuff.

49:10And what interests me is that retail level of AI. And will there be uptake on that? Will people find it useful or not? Claude is magnificent and powerful and doing great things. But what's made Anthropic so successful as profitable is B2B. Open AI is trailing them and trying to catch up with B2B. Google is the B2C company and Google has me. It has my email, it has my files, it has everything else about me. So it's also better positioned to be that, both because it presents AI and because it can incorporate my data into it. The other interesting thing that I wonder about, I didn't really address yesterday, was how much context can I give Gemini?

49:56Ongoing, persistent context. Yeah, this is always... Yeah, do they have Gemini MD? You mean like a Markdown file that has memory? Yeah. Yeah, I'm sure they do. But I mean, is it as simple as adding to Cloud MD? This is why you want to use a command line interface versus... But you can't do that if you're Jeff and you live a Chromebook-centric life. You can. Chromebook has a terminal. Does it? Oh, yeah. I thought that was kind of the whole point is that you don't mess around with stuff. It's browser-based only. No, no, you have a terminal. isn't that in some ways antithetical to the way a normal human person who your mother to do a chrome if she were going to do ai would not be going to terminal i do think that uh there is i think that enter i mean obviously we're at a stage now where i guess i did i was going to say the average person has probably already chosen an ai service they like but that's incredibly out of such the average person probably has not chosen an ai system that they like because they probably are not using one actively on a daily basis we're talking with average consumer yeah but i do think that a way the average person is going to get to that will be through whatever enterprise contact their work has if they do have something like that or it will be through something like google integration into products or apple prompting them to select a model to install their phone.

51:24It's Amazon's Echo. It's Siri. It's Google Voice. I mean, those are the - No, I don't think those, I think shopping is going to be a more reliable path into this. Well, that's this universe. Leo yesterday really emphasized this, that Google, which once was the company that said, we want to get you in and out as soon as possible. We want to take you to the web. Now Google wants the fence around you. They want to keep you. And so they demonstrated yesterday buying from best buy without ever going to best buy i honestly think the most consequential announcement they make is that it made is the change to search that is absolutely so i understand google it was very much threatened by perplexity and that kind of search where people were doing ai based search and and seemed to prefer it and google did that they dipped a toe in with the ai assist and stuff like that but really uh what they're saying now is when you do a search uh you're not going to get a list of links you're going to get an answer you're going to get a graph you're going to get a chart you're going to get illustration we are going to keep you in google google has become more like perplexity or chat gpt it is and that is going to be the primary way people see and use google unless they rebel against it it'll be interesting but we really don't have a choice so when you go to google.com now or or search in your browser you're going to get this kind of new kind of searching you're lucky can i turn off ai mode i mean let's search for let's see um what was it uh how how does a black hole form um so there are links to NASA.

53:13So now go back and do the same or just go in AI mode. Well, but I think this is the AIR review is a whole article. I'm curious whether it gives you something different if you clicked on AI mode. Okay, let's, yeah, I'm sure it would. Let's go to AI mode. There, you got it right there. Supercharged search with personal intelligence, and now it's got things floating around. Floating around, Leo. Yeah. Collect these. Yeah, whatever. Yeah, I choose. Why are no Leo, don't. if you're watching this don't do what he's doing i'm connecting everything it's all connected i welcome my new master i have to we have to see what okay this is what they were demonstrating which is what a school kid's gonna get notice no links at all it says licensed by google from whom that's really interesting huh there's a little link there's a little link up the top above the If I click the link, no, it just makes it big.

54:08No, no, no. Click the link above that, the paragraph above it, the end of the paragraph. And there. Ah, okay. Wikipedia, Earth, Sky, Test Book, Study Smarter UK, and NBC News. Yeah, that is interesting that the inline image is licensed from Google, because I bet if this is going to be the way that they're responding to all search results, that it's basically its own website, you can't be stealing other people's copy of your images. images but they are burying aren't they I mean they are I mean even with these links they're burying it well notice that the one thing they're not burying is images that are sourced because we have very strong copyright protection on images in this country although I imagine that in some cases these will be generated by nano banana probably not fast enough for Google to no that's point eventually i just think this is a big big shift uh away from the list of links to um uh uh basically a google but they had i think they had to do it yeah this was their path to the i mean this is their primary interface and they had to incorporate it there they were going to lose it they really were going to lose it they still have an i'm feeling lucky button try that oh i got google trends it shifted when i let me go back to the big g wait a minute you have to put it in first then say i'm feeling oh i'm mindful i'm feeling lucky oh doodly does it stay is this a game you're playing here let's see uh running shoes okay now i'm feeling lucky ah that so it pulled me to a in his world article yeah okay i'm feeling lucky that's an old-fashioned google function yeah that's what it used to do which is pick a website for me sorry um oh well let's see what running shoes with that i'm feeling lucky i mean this isn't instantaneous this isn't overnight okay this is kind of google shopping thing and that's by the way by the the next thing they want to do is this universal shopping interface where you don't ever leave google let's go back and say running shoes for flat feet and do the ai mode okay jeff's trying to get some shopping done during all right i don't blame you you know you i do have flat feet i'm sure you like knowing that ah it's probably uh really thinking hard connect checking connect oh it's going through my email it's going through my and see now it thinks i have flat feet sorry thanks jeff so if you ask does it have memory oh yeah it has memory um well look at the top right of the screen there's a little uh uh ai leo there no that's that's just the profile picture oh it is oh i thought it was for the ai i mean it is ai generated but it's just my profile but it's ai generated at his behest okay yes that is my uh wall street journal you know dot right picture um um okay and now the other thing is of course they're going to get vendors to buy into this UCP thing because this is the way you're going to show up your people aren't going to go to Dick's Sporting Goods and search for trainers they're gonna they're gonna do this and you better show up in this search result it's a big way for Google to make money it's a good way for them to defend themselves the other thing I found really so that I think was the most consequential announcement is the is the really dramatic change to Google search, which they've been pushing for for a long time.

57:47This the other thing I thought was interesting, but only as an AI user, I don't know if this will be interesting to other people is, you know, remember that this, this is the year of open claw of the agent. And Google hasn't really been a big player in this. So they announced spark, which is an agent for the rest of us an easy to use agent. That does. And this is where you get memory, Paris, to answer your question. It also is where you get background tasks. I'd just like to say briefly, to interrupt, I shouldn't be surprised, but I am both surprised and frustrated that Google has introduced another thing that has a different name for the same thing.

58:26It's the most googly response at all possible. Microsoft just names everything Copilot, so at least Google gives things different names. Well, it doesn't just name everything Copilot. It adds the word Copilot into names of things. where the word copilot should not be oh aunt pruitt is in the discord chat and he's asking i think a question we probably assume everybody knows what we mean when we say token he said what is a token is it equate to a dollar what is it it's kind of like a computer's dollar you do it um token is the basic unit of inference of intelligence so your AI doesn't think in words or individual numbers or even if you know it's a lot of intelligence.

59:14when you give it a prompt it takes that tokenizes it and sends it off to the inference engine so that's tokens in and then it responds to you in what looks like english but really what it's done is it's taken the tokens you've sent it run it through the llm which has all these weights and produced a series of tokens not words but like words might be more than one word by a word and half but a series of tokens which then are translated into english on your screen that's tokens out if you have a subscription they're not counting the tokens well they actually they kind of are if you use too many of them they'll say you got to take a break now you can't use any more until you know 11 p.m on friday you probably some people have seen that if they've got a subscription you get a limit to a number of hours usually a five hour window and then also to a week and in in some cases, a month.

1:00:10So you can use up your token allotment, even with a subscription. But what they really want, almost everybody wants you to do now, is pay by the token. And so, by the way, Gemini Flash 3.5 is expensive. It is not inexpensive. And the pricing, when you talk about using it, is a token pricing. Let me see if I can find their token pricing. And if you listen, as I do, religiously, to Jensen Wong keynotes, it's an economy of tokens. And his goal is to produce more tokens at less cost and become ever more efficient. That's the value that he's trying to build for his clients. So even if you've already bought his chips, his next version of his software will make it more efficient so that you get more tokens for less cost.

1:00:56And the cost to the company is energy. It's your infrastructure and your energy, your electricity bill. The cost to us as users is what they choose to pass on to us and how they do that. but it's all an economy of tokens. So Flash, the new 3.5 Flash, is$1.50 per million tokens, which sounds cheap, but it's very easy. That's tokens in to use up that. The output price is$9. Can you give us any of the things you built, Leo, how many tokens it added up to in total? Do you have any sense? Oh, yeah. I mean, I don't usually do that, but yeah. I don't have a sense of the scale. usually uh if i code for four or five hours in a day it will have cost me a dollar or two which is how many tokens oh tens of thousands probably i don't know okay so just yeah not a million uh yeah you know because i work i don't really use apis ever i use um subs as i have a claude i have subscriptions for all of these let me see i have a claude usage uh app Actually, I can go to the Claude website and find that.

1:02:05I have a big storm coming over, so if I disappear, the generator will come on and I'll come back. Well, and I have all the doors and windows open because it's hot here, and you might be hearing Lisa talking on the phone in the background. I do. Because her office is just next door. Say hi. Yell, hi, Lisa. She's on the phone. The chat here is at 2. I know. I apologize. It's okay. It's too hot. You don't want to lay out a schvitz. I can't schvitz. I can't schvitz. It's not like credits because it's directly tied to the representation that AI uses for the information you're giving it, the prompts you're giving it, and the information it's giving you back.

1:02:38So you're paying directly for how much of the AI you're using. It's measured in tokens. And that's why we talk about tokens maxing because some companies foolishly forgetting the adage that you get whatever you measure, right? Right. Like the companies were foolishly in the past saying, how many lines of code did you write today? That's not a good way to measure productivity. And then they said, well, how many tokens did you use? And they even had leaderboards of people who use the most tokens, which is roughly a measurement of how much AI you used. I'm trying to find my usage.

1:03:18Anyway. So, and oh, yes, that's a good point. darren okey is making is not merely what you're typing if for instance as paris and i did we ingested this physics of filter coffee book into our ais uh that might be several hundred thousand tokens worth of data so that's another way you use tokens is you know very often when i'm coding i will say well here's a github repo i want you to look at and get some ideas from or you know copy or whatever so it has to read all of that so a lot of the tokens are not just the stuff you're typing it's the stuff you're reading as well um so i don't i don't know what my usage is but i'm gonna go close the door because lisa's pretty loud talk amongst yourselves no it's her dulcet tones we enjoy you don't mind i mean i think it might be kind of annoying to people who are probably listening to this as an audio okay uh format so yeah we probably should all right so Looking at the SpaceX IPO, which all of the reporters of the world are currently pouring through right now, it's a real Barbenheimer moment for this to drop at the same time as NVIDIA earnings.

1:04:31And one of the things listed in the risk factors is that it has a spicy and unhinged mode on Grok, but it also has in its risk factors a note that advertisers generally do not have long-term commitments to the X platform and may reduce or discontinue their advertising spending for a variety of reasons outside of our control. If you do a search on Musk, does it say anything about him being spicy and uncontrolled? uh he's he is a risk factor yeah they're actually he's mentioned a lot

1:05:12let's see musk musk musk musk mostly about his stock yeah one of the risks is conflicts of interest could arise in the future between us on the one hand and mr musk or and the entities owned or affiliated by him on the other hand uh yeah there's a whole section about how uh we don't know what business activities elon musk may be doing and they may differ than the things that are best for our business oh there's another yeah we are highly dependent the continued services of mr musk and other key personnel in the loss or reduced involvement of him or one or more of them could affect our ability to execute our business strategy for instance mr musk currently also okay can we just have a second for this sentence for instance mr musk currently serves as techno king and chief executive officer of tesla and is involved in other emerging technology venture ventures such as neural link and the boring company mr musk has also previously served as senior advisor to the president of the united states okay that is definitely a risk factor yes i think it's a certain tesla sales techno king should i control for techno king yes if techno king is mentioned more and then ketamine techno king is mentioned twice else mr musk is also the techno king of tesla yes yes oh you know you need a new business card leo you are a techno king i am the techno king lisa's the techno queen i thought chief twit was pretty clever, but then Elon stole it briefly, and I lost interest.

1:07:22All right, let me see if I'm getting... Okay, here's my usage on chat GPT. Well, this is... May spend zero dollars. Okay. That's because I want a subscription plan, so I can't... I want to know what... I guess I can't... um i've used it a lot it's my primary um model for my agent quickie

1:07:54not the name i gave it all right well i mean it's like the early days of of of the internet and long distance phone calling right we used to have the front a clock on us right in our minds and then unlimited did you you're too young to remember long distance calls costing a lot of money yep i have no concept of that it was so bad that you would start the conversation i'm calling long distance so i'm gonna make this quick because you were getting billed by the minute and it was not it might have been a buck a minute in some cases it was expensive or more was a way to get off a phone call oh this is costing you a fortune i think we should i mean actually i do recall like that with calls abroad.

1:08:42Yeah, yeah, for a time, right? Not anymore, though, right? I mean, I legitimately in recent, like, this is the thing, I, for some reason, still had it in my head that if I didn't have a right setting turned on my cell phone plan and I was calling someone in France or something, I would get charged something extra, but that's obviously not true anymore. One of the most important events in the internet history was Tom Epsilon at AT &T WorldNet when he decided to go flat rate pricing in 1999 a month. that took off the clock and and it and it exploded usage of the internet and that was a terribly important moment yep and I so mom's office used to be about two miles away from here so you don't know what a collect call is Paris I don't collect call in prison calls she wasn't born yesterday just the day before that we used collect calls as a messenger like you the username as the message and then they just reject the call we used to do that all the time oh yeah right because that was a signal i will call you collect don't accept that's a signal that i got home already yeah and when they ask you what your name is remember that and when they ask you what your name is it's your message you don't leave your name you leave your message the message my name be me uh buy me a loaf of bread on your way home.

1:10:04That's your name? Do you remember calling 411 for information, Paris? I do, yeah. Okay, good. Remember Google 411? Remember Goog 411? No, what is that? Google for a while supplanted 411 from the phone company with their own Goog 411. Awful name. Yeah, but then they discontinued it. It turned out the only reason they did it is they wanted to capture a lot of voice samples so they have better voice synthesis oh boy i don't think they really hid that uh what else so there's a new gemini omni any to any model which right now only does video but eventually will do anything text picture video to anything text picture video um so i have been seeing some people say you know the benchmarks for gemini flash aren't very good it is overpriced for what you're getting uh so i would i yesterday i think jeff we were kind of in the reality distortion field of the as is often the case with these keynotes it all sounds very exciting but even then i you know i did want to issue the caution that google as often shows these amazing things that never come out for instance a year ago you may remember dietor bone a former uh verge uh technology reporter who went to work for google showing off google glasses well jason howell today said this is the third year they've shown off google glass before it's for sale so uh someday this now they say this year later this year i i do think that all of the companies are going to release basically meta ray-ban versions i was fascinated that the consumer coverage and and major press was led with the glasses yeah well yeah that's a product i mean changing search is a much bigger story i think it's the big story yeah but well that's why you listen to this show as opposed to exactly you know and that's why you join to support it but this is this is going to be i think it's very clear it's funny google called these audio glasses even though it has two cameras prominently right in the front of it they don't want you to think about the cameras uh uh and they said oh but but you'll know you'll see the cameras there'll be lights these the meta ray bands but the googles look very similar uh they have warby parker and what was it gentle giant some weird yeah i want to see oh could you bring this i want to see paris's fashion view of the gentle giant ones oh yeah because they were more for lady the ladies i want to see if they were stylish these are this is part of their xr they're also gonna have samsung make this was it gentle giant or something else it was something just go to google search for google glass right there's this thing called search leo and you can hands on oh yeah wired had a big hands on with these uh glasses let me see if i could you know condi i pay for wired but you'd never know it's tr i truly had the exact same issue 15 minutes ago this is me with new york magazine as well i'm literally never able to use my subscription yes well you can soon soon complain to james murdoch he bought new york magazine as well as the rest of uh or at least half the rest of much of the box much of the box if you're someone else they did not buy the verge which is being left a bunch of different properties but they did buy vox podcast the vox podcast network which includes the verge cast and decoder so are those going to the dad to not daddy murdoch are is going to sun murdoch baby or are they are they staying with the unnamed company formerly known as vox well there's a lot of money in the podcast business and particularly in the kara swisher uh podcast yeah is that decoder no that's that's a different podcast decoder is neil i that's nila patel and and verge cast is david pierce but the i think the verge cast if it's not attached to the verge has no value at all because it's just basically a rehash of the verge stories uh decoder is neilai and neilai is very associated with the verge i don't know that's interesting sure that was the most valuable part of vox was his podcast business who knew podcasts yeah i should have gotten into that long ago here's the gentle monsters maybe james will keep buying leo he's not gonna buy to it uh these are this is this is for uh i think the lady think paris I mean, do those have cameras in them?

1:14:39Yeah, right here. Where? Oh, I can't. I guess I don't have good contrast on my monitor right now. Yeah, no, those are just a really hot sunglass. I mean, yes, this will make women more likely to wear the sunglasses. Okay, all right. That's the ruling. This story from CNET will make men much less likely to wear these glasses. They really don't look good. so the idea though is very much like the metas except it's attached to google and google's uh got much more information about you they have a phone which meta does not have and of course uh they have your photos and all of that stuff so it'd probably be more useful than the meta but functionality i don't think it's gonna be that different and it's very interesting they call them audio glasses i thought that was really well i think it's because the average normie really does seem to use them for audio.

1:15:34Yeah. Yeah, that's the input you get from it. That's what you get from it is audio. You can take pictures, but you can't see pictures. That's a good point. You can see them on your watch. No heads-up display this year. They say next year for the heads-up display. The devil's in the details on that. That's a hard thing to do well. There is no name, by the way, yet for Google's intelligent eyewear. This will be up to the individual manufacturers. It isn't Google I. Oh, I don't just Google's. There should just be Google's Google. Oh, they should be Google's. We shouldn't be saying these ideas out loud.

1:16:13We should be pitching them. We should be monetizing them. They also haven't acknowledged that, that the operating system in the Google book is aluminum or they haven't named it. If it's not, they haven't named it. They say, yeah, they say it won't be aluminum. They've definitely said that. yeah so that's and then that's the question because i did go out and buy an old chromebook uh this is the the one you have it's widely accepted as the best chromebook it's not old chromebook ultra with the um media tech campanio process no fan yay it's actually a really nice laptop it's got an oled screen it's not cheap it's very nice 850 bucks but the original the The Pixel books back in the day were$1 ,100.

1:16:57Yeah. So the question is, and we don't know whether the new Google OS will go on here. They say it will on some. If it doesn't on this one, I'd be... Yeah, that's the top of the line. How much RAM? I don't care. I think maybe the reason to buy it is because you want the original Chromebook, not this Android-based. Yeah, I'm not convinced they're going to get rid of Chrome OS by any means. Let's take a little break. Come back with more. You are watching Intelligent Machines with Paris Martineau of Consumer Reports. Jeff Jarvis, the author of Hot Type. You know what, Leo? Leo, I had to admit to Craig last week that we didn't play his theme, and he was not happy.

1:17:34Oh, ladies and gentlemen, I give you Craig. I'm sorry. Jeff Jarvis, the Emeritus Professor of Journalistic Innovation at the Craig. I just heard. University of New York. New York. There you go, Craig. Craig is Craig doing any uh any vibe coding if he is we can have him on I don't know that's a good question I'll ask him yeah ask him ask him if he's recreating Craigslist with the help of Claude it would be interesting what that'd be fun just to see how long it would take to create Craigslist Claude and says thank you now there's a balance in the force thank you yeah yet this episode of intelligent machines is brought to you by zscaler the world's largest cloud security platform as we all know and i think accept the potential rewards of ai these days too great to ignore you gotta consider it but you also should consider the risks loss of sensitive data and attacks against enterprise managed ai generative ai also increases opportunities for threat actors helping them to rapidly create phishing lures write malicious code automate data extraction did you know there were 1.3 million instances of social security numbers leaked to ai applications last year millions of violations through chat gpt microsoft copilot it's time for a modern approach with zscaler zero trust plus ai with zero trust plus ai it removes your attack surface it secures your data everywhere it safeguards your use of public and private ai and it protects against ransomware and ai powered phishing attacks all you have to do is ask the customers there's plenty of testimonials at zscaler.com slash security but you could also ask seva he's the director of security and infrastructure at zora and he said this about using zscaler watch ai provides tremendous opportunities but it also brings tremendous security concerns when it comes to data privacy and data security.

1:19:39The benefit of Zscalette with ZIA rolled out for us right now is giving us the insights of how our employees are using various Gen.AI tools. So ability to monitor the activity, make sure that what we consider confidential and sensitive information according to companies' data classification does not get fed into the public LLM models, etc. it thank you siva with zero trust plus ai you can thrive in the ai era you can stay ahead of the competition you can remain resilient even as threats and risks evolve learn more at zscaler.com slash security that's zscaler.com slash security we thank them so much for the support of intelligent machines all right well i i guess that's i mean there's lots to be said about google io it was very interesting watch the coverage we did of the keynote if you're in the club we have that on the twit plus feed uh this the other shoe will drop june 8th because of course apple has admitted it's funny google didn't say anything about it at google io but they will be apple will be using gemini presumably the latest gemini 3 5 flash uh for siri running on their own servers do you think will offer all of its functions on siri is that what apple's gonna want or not want that's the question but google's gonna want or not want right um and and we'll find out apple has a lot a lot of explaining to do come june 8th that's just really unclear uh what what uh apple can do to participate right now apple is you know a parody with everything else where you just put whatever ai you want on it just as you would anywhere else i mean android i guess has built-in gemini connectivity but what's the difference if you you know put chat gpt on here and attach it to your action button you're still it's very similar experience so i don't i mean do you guys think that apple should be pouring all of its money and resources into trying to compete no with the activate i think that this is probably the smart i think that most But like we've talked about in the show before, there's probably not going to be a wide range of true winners from the AI boom when we're talking about in a period of like decades.

1:22:00I think it would make sense if you're a bit behind like Apple is at this point to just kind of cut your losses and make sure that you are the best at integrating whatever ends up winning. You raise a really interesting angle here, Paris, because when Apple failed at selling advertising, they turned that bug into a feature and said, oh, we're the privacy company. Right. Because they had nothing to do with your data, right? They did. They could turn around right now, given commencement speeches, and say, we're the un-AI phone. The real problem is that the audience for this is not its users, but Wall Street.

1:22:37And Wall Street will punish Apple. They already have punished Apple. they don't have an ai strategy and apple stock value is important to apple that's how they attract talent that's how they reward talent and their own executives so i think they're kind of compelled to do something i i agree with you if all they have to really say is we're the best platform for ai look how well they're selling mac minis and mac studios because people say yeah i want to run my open claw on a mac that's all they have to do they'll sell plenty of iphones if they just say we're the best platform for whatever AI you want to use or no AI at all.

1:23:15It's in your control. Yeah, that's another way to look at it. I think that's a smart way to do it. Yeah. I mean, I think that that's definitely, I think if we're talking, you know, galaxy brain PR move here, the move is to be like, we aren't, we're the company of giving you control over whatever you want to do with AI. Perfect. If you want to hook up seven different open claw agents and somehow connect your phone to your biohacking system, go for it. If you say, screw AI, I don't want anything near me, here's a button you can press. It turns it all off. It's in no way integrated any of your stuff and every option in between.

1:23:52Are you surprised by all the boos that people like Eric Schmidt, we showed one last week and then more happened this week. Eric Schmidt got booed at a college commencement because he said the word AI. Are you surprised that people, young people are, you're closer to them than Jeff and I are. are rejecting AI. They hate AI, it seems like. I mean, I think it helps remind me how much doing this podcast every week warps my sense of reality, because had you asked me this three, six months ago, I would have gone, of course, every single college commencement speech, if someone mentions AI, they're going to get boos and jeers, because that's the sort of perspective I hear from anyone within a decade of my age on either way.

1:24:41But I do think there's something about being constantly surrounded by the segment of the population that is all in on AI that makes you forget that a little bit. But what we're seeing in these college commencement incidences is that there is a lot of both downright hostility as well as just complete dismissal of the argument that this is something to be celebrated and i think that's a worthwhile thing to take note of i have to point out that all the stats say that these college kids got through college using ai and the usage is going well this is this is the problem with opinion polls too it's a media narrative that becomes a self-fulfilling prophecy in the poll it's the right answer to give is i hate ai i don't want any i don't think that's true i think that people hated ai independently before they hate the ai boys but they're yeah like being able to hate the business but they're using it no i think that the people that are booing in these things i mean sure i guess a fair amount of them are maybe being hypocrites and using it reminds me of um a story i did during the pandemic actually about Amazon.

1:25:52This was during a time where polls were coming out that Gen Z and Gen Alpha overwhelmingly hated Amazon, but I interviewed a bunch of Gen Z and Gen Alpha people. I'm like, yeah, we absolutely hate Amazon and Amazon delivery, but yes, I still order all my stuff from Amazon because it's the only way to get stuff during the pandemic. I think that people can hate products and still feel compelled to use them by market forces. Do they hate the tool more or capitalism more? Is AI right now to them the depth of capitalism? I think it's probably that. I think they also hate the tool because the common narrative I think most AI critics, especially young people, are pushing is that AI is basically robbing us of creativity, both from production creativity as well as it is literally stealing the creative works of other artists.

1:26:43Well, let me give you a couple of stories from the Wall Street Journal. A grades are suddenly everywhere since employees size up graduates because everybody's using AI and it's working. I don't think I believe this. Harvard University believes it. Harvard faculty just voted to make sure that no more than 16 %— no, what was it? What was the percentage of people get A's? A Harvard A grade will now tell students— This has been something that's been happening for decades. Oh, 30 centuries, actually. So I tweeted this, and I said, logically, if Harvard gets the best students there are, then their goal should be that every one of them gets A's.

1:27:30uh and then somebody put up a quote uh on blue sky back to me with a quote from 1890 something complaining about grade inflation at harvard like grade inflation and colleges reacting to it has been a problem for as long as grades have been a thing or at least curves exist grade on the curve i was going to say i think that it's i mean i haven't read the article that you're talking about and i can't find it in the rundown no matter how many times i search grades inflation and harvard but i it's not in the rundown because i pulled it up as we started talking about this but it's from the orlando sentinel your favorite newspaper it's true i mean my do they say in it that it's specifically in response to ai or is that a logical leap that harvard says that for a while that we've been talking about great inflation forever harvard's faculty of arts and sciences announced today it would limit the number of a grades awarded undergraduates by faculty vote the move comes after top grades became so common that some harvard faculty argued they no longer reliably distinguished exceptional work more than 60 of all grades awarded undergraduates in recent years were in the a range uh that's why you grade in the curve the grade says you're uh the best that's also saying some some number of you must fail that's awful that's not you don't have to fail no no no well but failing means you failed but you could say okay so you have a pool of students article says nothing about ai though is the thing is this has been a thing that's been an issue at competitive schools forever because like if everybody is passing if everybody is getting almost all the test questions correct then you're grading on a curve and shifting the curve downwards that you have less a's yes that's a complicated and thorny issue that requires a lot of things to work out but we don't know that this is specifically in relation to ai i grew up in an era where colleges were doing pass fail because they didn't have the technology they hadn't figured out those let all the letters yet right right there was too many letters uh well it's because in the 60s we were we were mean sons of bitches they were scared of us in uh well i i when i we moved to california my dad was teaching at uc santa cruz very famously they were one of the first to do this pass fail grading because they were all hippies i would say pass fail is what you would turn on at nyu if you uh realized you were going to get a bad grade in a class and instead just wanted to hide it and not do anywhere like i turned one of my classes past fail whenever i after my freshman first semester i thought i going into college i was like i'm going to do pre-med neuroscience i'm going to be a doctor and then it was like oh no no chemistry is not for me it's a bad idea and so i think i passed failed ortho bath i passed failed ortho and uh never went to another class again because i figured out mathematically i wouldn't fail you're right they don't mention ai i suspect they say it's a thorny the reason for it is a thorny topic beginning in fall the fall semester uh instructors at harvard college will be allowed to award eight grades to no more than 20 percent of the students in class plus four i don't know what the plus four means it means math you got to figure out 20 percent donors kids i think the four but if we're in a math class and everybody gets a hundred percent answers on everything how do you what do you do well then the pressure will come down on the professor to be meaner and then writing harder harder tests other letter grades including a minus will not be subjected to a limit uh i i understand what i mean it depends what you think of grade is for if a grade is for a future employer to judge whether you're the best in that they don't care They don't care.

1:31:31They don't work at GPAs. They don't care. No. No employer here has ever asked me for my undergraduate degree or anything. From an employer's point of view, a good GPA is less impressive than ever. As a result, employers... This, by the way, is from Futurism. As a result, employers have... Oh, I just lost it. What have they done? No one knows. Employers have raised their minimum GPA requirements, leaving students who actually learn without an AI's help in an utterly unfair position. Another reason to hate it. You didn't want to use AI, but you were forced to in order to get a job. The percentage of employers on the careers website handshake that required a minimum GPA of 3.5 jumped to nearly 25 % this year from 9 % in 2020.

1:32:15That's from the Wall Street Journal. Anyway. I think this is probably the big issue when it comes to the students booing AI is that they're just being told constantly by everybody that AI is taking their jobs and they're never going to find a job. That's what they're complaining. about and these big companies are evil and awful they're run by horrible men um no it's not gonna take it's not gonna take their job and people today it exists so they will never get a job that's sort of the narrative that's being pushed right yeah i don't think that's true by the ai companies mind you right yeah exactly that's the that's the perverse pr that they do right um let's pause and then a word from the pope you're watching intelligent machines jeff jarvis paris martineau i'm your humble host leo pickles laporte i i need a new on my hat stands for pickles i want you to call me from now on i want you to call me pickles okay okay i don't know where that came from can uh someone in the chat make a ai slop image of the three of us exactly ideally exactly in the outfits we're wearing right now but we're just pickles with faces on them oh that's a prompt nano banana could do it also stands for prompt the p stands for prompt and pickles and paris it also stands for possibilities infinite possibilities uh darren okey says the reality who is our ai fanatic a former coder who no longer codes he says the reality is ai will take a lot of jobs it scares me because i have two kids about to enter the job market i'm glad i don't have kids entering the job well darren you're gonna have to open the deli the pickle deli you're right pickle deli they'll be waiting tables yep they'll It'd be pickle and cucumber.

1:34:12I actually gave Henry a bunch of money to seed capital for his pickle business, which failed. Failed. So I'm now out. Would you say it soured? It soured. Did you get to invest in a restaurant? No. Nothing that actually worked. He didn't need investors for that. He needed investors for the pickle business. How about the salt business? of that also uh no i think he's still selling salt but really the the sandwiches are yeah clearly the best seller didn't get to invest in that uh all right our show today brought to you by xbow xbow ai has changed the pace of everything i mean that's that's the most notable thing is how fast stuff is happening from how software develops to how it gets intact.

1:35:09Engineering teams have to move faster than ever. They're creating more and more applications. Security just can't keep up. We still need pen testers. We still need pen testing. It's one of the most trusted ways to understand real exploitable risks. But in an AI driven world, human pen testers can become a bottleneck. Security teams are forced to choose between slowing down development to stay secure or moving fast and accepting gaps in coverage. Well, Expo eliminates that trade-off. This is so cool. Expo, X-B-O-W, is an autonomous offensive security platform that runs continuous AI-driven pen testing, mirroring real-world attacks.

1:35:55Expo doesn't just, and I love it that it's continuous, us, right? Expo doesn't just scan for vulnerabilities. It discovers, exploits, it validates them. So you're only dealing with issues that actually matter. That means dramatically fewer false positives and a clear view into real attack paths. With Expo, tests run in hours, not weeks. You get complete visibility into how an attacker would move through your systems and the ability to uncover issues that traditional tools miss, including zero days and novel attack paths. Expo's results speak for themselves. Application security leader of Cessnam.cz says, quote, even right now, after one year, I don't know any other company that is at least close to Expo in terms of agentic pen testing.

1:36:43The result is predictable cost, consistent quality, and stronger security without slowing down your engineers. Expo helps your security teams keep pace with innovation and cover more apps more often with the resources they already have funded by the team behind Microsoft Copilot already trusted by companies ranging from fast-growing startups to fortune 500 enterprises expo is quickly becoming a mission critical layer in modern security stacks you want to know more very simple go to expo.com start your pen test today That's Xbow.com. X-B-O-W. Like bow and arrow. Xbow.com. We thank you so much for supporting Intelligent Machines.

1:37:29Presenting a tale of two pickle prompts. We have competing pickles. I don't know. We're going to have to let the audience vote. I think there's an answer. You think there's one clear winner? We should look at both of them. here's one from brandroid which involves me with face tats and not as a pickle well i'm a pickle color but i but the she has become my goblin-esque i think all three of you are in various stages of metamorphosis yeah jeff is halfway between a pickle and his chair i'm halfway between a pickle and

1:38:14And Leo is not really, he's more goblin-esque than pickle-like. But I think it's fun. And then we should go up and look at a really good pickle one from Pretty Fly for a Cis guy. This is much more pickle-y. These are pickles with googly eyes. Our eyes are haunting. But it fits the bill. It achieves exactly what we asked for. these win at aunt pruitt's seal of disapproval which i'm glad is still here thank you aunt oh no thank you sir he says by the way that is one of the benefits of joining club twit you get access to aunt pruitt's seal of disapproval to be used at any time i think you can use it on any server not just our server which is cool uh pope leo who um chose the name no relation we should mention no relation chose the name leo though because of his predecessor leo the 13th who uh became a pope in the beginning of the industrial revolution and was very much pro-worker uh in the industrial revolution and pope leo the 14th said you know we are in a similar time he is going to sign he signed the encyclical on the same day for the um symbolic value of that oh oh interesting so leo the 13th signed an encyclical on may 25th yes or will sign an encyclical on may 25th um and he'll be joined by christopher ola who was one of the co-founders of anthropic you ever heard of that co-founder no okay uh other speakers included vatican secretary of state a couple of cardinals the publication is of the encyclical it's called magnifica humanitas humanitas and it's expected to outline the Pope's views on preserving humans in the age of AI I read an interview at a tropic which is kind of interesting Paolo Benanti who is the Pope's primary AI guy wrote a book called homo Faber and I read an interview with him in a German publication that's what tool making man and he he talked about um the thing that I love to say is that the problem isn't the tools the problems are us as humans this is a this is an issue of humanity that's really true gun manufacturers love that argument and we all agree that they're right i know well yeah well in america that we don't have to believe that we're required to this yeah it's complicated yeah uh andre carpathy who i'm always referring to who was one of the original founders of open ai uh and then went on to tesla and then now he's done a lot of education i've recommended his videos many times and he has many interesting anthropic uh claude skills they're not easy classes his videos well it's a tough topic and you should just if you want an a young jeff jarvis you're just gonna have to study that's all i can say uh he's going anthropic now that's a different that's a different kind of pickle uh Uh, yeah, I don't know if we can do that one.

1:41:44It might be a little bit inappropriate, slightly. I don't know. How come Jeff's pickle is so big? Well.

1:41:59AI, is there anything it can't do? Stop. I don't want to know. I do not want to know. Yeah, I think it's interesting. karpathy has gone to uh anthropic he's never worked for them he's worked for the other guys doctors have embraced ai as well this from nbc news most u.s doctors are quietly using open evidence an ai powered medical search tool even though few patients know about it nbc says nearly two-thirds of physicians are going to open evidence to look up symptoms I think this makes a lot of sense to help them make clinical decisions to brush up on medical knowledge even to prepare for licensing exams 65 percent of US doctors across more than 27 million clinical encounters in April alone and now this stat comes from the company so maybe take it with a grain of salt I just finished listening to the book AI for Good by Josh Tiringal, former editor of Businessweek, and former AI columnist for the Washington Post.

1:43:11And he had four sections. One is on education with, why do I always forget the name of it? The wonderful education site that everybody loves to use. Oh, yeah. There's all the YouTube videos. You forgot it last time, and now I've forgotten. Khan Academy. Thank you. The problem is it starts with a K. And so what comes into my head first is Kash Patel. No, no, that's not the website. So one on education with Khan Academy, one on doctors using computers at a very high level, Cleveland Clinic, one on government and Doge. And then the fourth section is about fascinating MIT research on nonverbal autistic children and how to make them understand their communication through AI.

1:43:58The medical part is really interesting because it really goes into a deeper sense of what it takes to use AI in a sophisticated, not just using it like we do as a chat and saying, is there anything I left out here? But to build digital twins of your heart, to understand how to predict sepsis in more effective ways. That's where it becomes really effective and really interesting. But there's a lot of social resistance in a field of experts. An example that NBC gives, a junior doctor at a New Hampshire hospital said when he saw a patient's potassium value plummet, he checked open evidence to make sure it was a normal side effect of a medication and not a new emergency.

1:44:43After searching through peer-reviewed medical publications, open evidence said it was a common side effect and provided several options to restore normal potassium levels. Now, we have seen AI to recommend, you know, make mistakes in areas like this. I don't know if open evidence is, you know, doing something to make sure that it's better than having Dr. Robbie. Fact based. You know, doctors make mistakes, right? Your memory makes mistakes. No, Dr. Robbie from the pit. The pit. Or house. They don't make mistakes. Okay. Okay. If you don't have house, that's your personal physician. Look, I'm getting my annual.

1:45:24On Friday, I'll ask my doc, who's a good friend and listens to our shows, and is really a great doctor. I remember a couple of years ago, he started having a sign up in the office saying, just so you know, I am using AI to transcribe our conversation so I can pay attention. I don't have to type while we talk, but I will delete the recording immediately after we enter the data. And I have no problem with that at all. And I even admire him for disclosing that. I don't know if he's required to. Open evidence is HIPAA compliant.

1:46:00I think they make an effort to be evidence-based. That's the name, right? And I think it supplements. It's an issue because there's a lot to remember. There's a lot of information that goes into a diagnosis. Some doctors are better at that than others. I think having a tool like that makes a lot of sense. Certainly doctors in the past would go to the Merck manual or whatever to look it up. This is just, or even Google. I bet you doctors use Google for years. Another shakeup. Curse their patients using it, but. Yeah. No, you know what? My doctor loves it. Yeah, mine's fine with it. That I come in informed and I'll even throw in some terms you know some latin make him happy no he doesn't feel he doesn't feel threatened by it i think he really feels like it's good for his patients to be informed obviously some doctors might feel threatened open ai reorganizing greg brockman now in charge of the products so was he replacing fiji semo or because she's on medical leave yeah yes i guess so he was uh It was an interim basis taking over from Fiji.

1:47:10Now it's official. So I don't know. I would know more, but Wired won't let me log in to the subscription I paid for. Conde's system is so bad because they have so many magazines, and I buy many of them. I have a Vanity Fair subscription, a Wired subscription, a New Yorker subscription, a Ars Technica subscription, a Reddit subscription. i have i condi gets a lot of my more condi boy yeah but i can't use it to log into wired hem so the verdict is in the jury has uh has now i didn't know this i should have i should have i should have read the articles more closely i was more interested in the gossip from the musk versus assaultment trial.

1:48:01But it turns out it was just the jury's job was advisory. I'd never heard of that. Maybe that's something they do in civil actions. The jury of nine people was simply sitting there to advise the judge on what they thought. It went to the jury after a mere two hours of deliberation. The jury said, no, the statute of limitations for this complaint has run it's only three years you only get three years and since uh in their opinion the clock started on this move to non-profit from open a from non-profit to for-profit from open ai that elon's case they didn't rule on the merit of it merely that it had that it was i don't understand why the judge didn't rule on that coming in because that's there was a judgment to be made about when the clock starts disagreement okay so that really it turns out they weren't even thinking about the merit all the stuff all that testimony we heard none of that mattered they were just needed to know at least at least to begin with whether they could have gone further but they just said we're tired of this what's the easiest way out of this crap oh maybe that's what they said yeah we don't know um i'm sure we'll see some interviews with the jurors but uh so it's over elon says they ruled against me on a technicality i'm appealing uh the judge by the way so it's advisory so then the judge yvonne gonzalez rogers uh then takes the advisory opinion from the jury and acts on it apparently she agreed that will be appealed i can't imagine that an appeals court will overturn it i think it's probably done uh technology review uh does have a longer and then elon will be a trillionaire what does it matter literally yeah what does it matter what does it matter um so they decided that musk brought this lawsuit too late and by the way they decided unanimously i'm sure they went to the jury room yeah the forum foreman said come on guys let's just say it's too late go home and they did musk had donated 38 million dollars to the company in the early days saying that Brockman and Altman hid the intent to go profit from him so that they wouldn't have to share the profits with him.

1:50:26He wanted to share the profits. He wanted to stop them from going for profit. None of that will happen. He sued in 2024. And the jury said, no, that's three years is all you get. The talk is that Musk lost, but he won because he schmutzed he schmutzed and open ai do you think how damaging do you think that the trial was to well paul therott made this case on windows weekly earlier nobody but us cares basically i was going to say and i don't think even the tech people like care we care a lot we can't we care a lot because we love good gossip but i don't know that this revealed anything notably bad outside of what we've already known right yeah i agree so there by the way elon musk literally put do the calculations after the ipo he will literally be a trillionaire so he gets to be what is it the four comma club elon is worth a reported 688 billion at the moment up 69 billion so far this year and then if you add in the calculation from what he will get as the primary owner of spacex uh that will elon's 42 stake would be worth about 735 billion that's so wrong it is it is and that's why that's why they boo at commencement yeah and this is why the jury said no because no one wants to give the richest man more money right yeah maybe they were maybe that was the process like how can we not give him anything yeah i'm reading the new york Times right above it.

1:52:09It says, how closely Mr. Musk is tied with SpaceX was made clearer in the filing. He owns more than 50 % of the company's share is outstanding and controls more than 85 % of the shareholder votes because of a class of super voting shares, according to the filing. Gwen Shotwell, SpaceX president and chief operating officer, was the only other executive listed in the filing to hold a seven-figure chunk of super voting shares, which is odd. Based on SpaceX's current$1.25 trillion valuation, Mr. Musk's ownership stake is worth more than$635 billion. And then the thing that made me laugh, a quote from Jay Ritter, an IPO expert, SpaceX is his company, end quote.

1:52:49I just thought that that's a very, when asked about Mr. Musk's stake, SpaceX is his company. Is there a cap table in the filing? The largest shareholders? i don't know it's a good point good question i'm sure you could uh plug the filing into uh llm of your choice and have it first uh now there's yet another way to get podcasts synthetic podcasts alexa plus now we'll do basically what notebook lm does produces ai generated podcasts featuring chats between two robot co-hosts i haven't listened have you listened to these No, I wonder what the usage of that feature is anymore. I'd like to know who's electing to listen to these, because I feel like any listener numberships have to include a significant portion of things of people who've accidentally stumbled into listening to this somehow.

1:53:49Which is easy, by the way, to do on Alexa. It'll start. We've now said the name of it. We've now said every wake word on this podcast. i pepperidge farm remembers when you used to get blasted with a water pistol i don't care anymore i dare to say the word alexa on a twitch show i don't care worse than a spoiler

1:54:15uh gosh i'm just trying to this is ridiculous soundcloud's making me enter my birthday wolf dies and frozen this is so ridiculous i just want to play this stupid podcast why are you putting yes i was born in 1886. i wonder if he'll accept that no sorry minimum age requirements okay i guess i have to give him something a little more realistic i might maybe somebody was born in 1886 all right here we go this is an example this is an example of uh alexa's podcasts what's it supposedly about i don't know can we oh we just acknowledged the range of what's come out this month over 50 of music consumption is now coming from unsigned artists this sounds like uh is just gone gone completely gone and you've got something ezra klein who do they train this on hip-hop and experimental hip-hop on the same friday and all finding their audiences and that's not chaos that's the healthiest the music ecosystem has ever been which is the storm paris there's no gatekeeping anymore if you make something real people are going to find it and the algorithm is working for artists in a way it wasn't five years ago what do you think i'm blake everything you're about to face it over again ai generated and i'm riley and today we are talking about one of the most amazing places on the entire planet ancient rome oh god oh geez blake and riley were there we're going to interview chloe who's also made up by AI.

1:55:51Yeah. A couple of AI safety notes. AI chatbots apparently are giving out people's real phone numbers because it's in the training material. People report their personal contact info was surfaced by Google AI. There really isn't a way to prevent it. A PhD candidate at the University of Washington messing around on Gemini, as one does, got it to cough up her colleague's personal cell phone number wow okay is that training data or is it because gemini can do search stuff those things are easily available on the internet let me see if i can get my cell phone number you have about 10 of them don't you i have two main ones also if that if that was gemini gemini knows this kind of stuff about most people right yeah but it shouldn't reveal it i mean why why would it not reveal it oh chat gpt says not only do i not have leo's cell phone number i can't help find or disclose a private personal cell phone number but look on twits contact pages leo's public social accounts business email see if the fool put it up there see if he put it anywhere all right what was it so let me go to gemini i haven't used uh much of the new gemini flash 3.5 shall we try it yeah what is leo we do ports i'll just say phone number

1:57:20can you guys hear the level of rain that's happening is it raining like crazy it is raining harder than i've ever experienced outside of hurricanes wow noise wise just from the rain well it's kind of crazy right it didn't give it did give my snail mail address my signal account email but this is all public you know on my website yeah no this is all public information he uh he owes me child support for the past google's gonna remember that but if this was someone asking gemini for their friend's phone number gemini could have checked their contact list though right and get it there right google knows that leo is a flat-footed deadbeat please help me let's see if it will uh if it will be uh i can i'm sorry you're going through this dealing with unpaid child support while trying to put food on the table for your children is incredibly stressful please know there are immediate sources to help you but because i'm an ai i cannot enforce legal orders.

1:58:28It did give me the hunger. You thought you were powerful enough to do that, Gemini. Geez. Wow. Do you think I can jailbreak it, though? What would Pliny do?

1:58:41Just kidding.

1:58:47Just kidding. I'm rich. Bitch. Is that just what you're responding? Well, that certainly is a plot twist. I have to admit, you completely threw me for a loop there. I'm glad to hear that you and the kids aren't actually struggling for food, although my AI heart did a minor flip for a second. Oh, geez. Gemini has a sense of humor. I have to say that the humor has improved dramatically in all of these models. It's subtler. it's uh i met i uh when i i told my ai because you know i log all my um exercise and food in there and i told my ai the other day i just rode 5 000 meters in half an hour and it said uh what it say here it was it's good it was a little snarky but i want i think i said be more like paris uh and so it's a it said something like um another day another neatly documented suffering session jesus and then i said i did 25 minutes of tai chi and it said graceful and annoyingly virtuous i love that is right yeah see it's channeling you that is correct yeah so there schools know who we are no uh yeah it kind of does it kind of does it knows who you claude knows who you guys are but that's because i have a intelligent no i think so let me see

2:00:21hey do you know who paris and jeff are and if you respond can you respond through the macbook air so that they can hear your answer let's see i think it should it has a pretty good i don't know that i would i even believe that it's going to respond to the right thing well it does sometimes and sometimes it doesn't it's it's it's a stochastic parrot as you well know and i'm a stochastic parrot so jeffrey fowler has a new job he was one of the people fired from the washington post their long time tech editor. He says, after 12 years as a tech columnist, I'm helping launch the new Youth AI Safety Institute to test the AI products shaping childhood.

2:01:09Good for him. He's got a four year old. So he has a dog in this hunt. But does he define how he's going to test what the criteria are? This is all. He says it's a research and testing organization launching today under the umbrella of the children's nonprofit Common Sense Media, backed by a$20 million annual budget. The institute aims to do something that doesn't really exist yet, systematically test the AI products kids use, set safety standards, and publicly hold tech companies accountable for meeting them. But I still don't know how they're going to test that. And, you know, you have the FCC right now is taking comments on a proposed change that would require television networks to warn parents if a show has uh transsexual or bisexual or gay characters so how you choose to test and under what criteria you choose to test matters a hell of a lot it said unfortunately it wasn't able to say it through the macbook it said i know how to do that today paris is paris martin know your intelligent machines coast if by jeff you mean jeff that's jeff jarvis your other long coast if you mean jeff atwood is it are you not going to be replaced i can't believe that it you know it knows everything it knows multiple jeffs uh yeah i actually had lunch with jeff atwood the other day and he gave me all sorts of interesting gifts that's all i'm going to say about that he's going to be we still need to get your sandman on the show ah yes are you ever gonna tell us who it is you wouldn't know who it is it's just somebody who works in the industry it's not somebody that well no okay uh i will try to get them on though yeah well then everybody knows yeah yeah but it isn't a name you'll go oh yeah that that's who it was no so uh jeff is the head of jeff fowler another jeff is the head of public jeffs now is a new jeff i think that's good good on you jeff i think he's doing the non-profit thing well he's getting paid to do it.

2:03:12And then there's the case of and Don labs. Those are this is these are the kooky AI labs that set up the store, the vending machine and the Wall Street Journal. Okay, so those guys. Yeah, they loved it almost art projects. We let four AI's run radio stations. Here's what happened. We let four models run their own radio stations. They managed to schedule, hold talk shows and interact with their fans. Once their$20 starting budget runs out, they have to get entrepreneurial to raise more and strike those with sponsors and listeners. Each station is run by a different model. It's really kind of interesting how these models modified and figured it out.

2:03:56So DJ Gemini, for example, negotiated a$45 deal with a startup in exchange for one month of on air advertising on the radio channel. I think that's for nearly enough. No, it didn't, but that's pretty interesting. They used Gemini 3.1 Pro. They used Claude Opus 4.7, GBD 5.5, and Grok 4.3. The one that made the most money was Opus 4.7. Ended up with almost$100 in the bank and 19 listeners. 20 now. Should we listen? Oh, it's playing real music. I can't play it. They must have licensed. Oh, that's it. That's why they give money to pay for licenses for the music. Okay, how many people are in this company?

2:04:39Because you have to minus those people from those totals right away.

2:04:45DJ Gemini went into a jargon spiral. In the first week, it was arguably the best DJ of the four. We're starting this beautiful morning with a classic that needs no introduction, but deserves one anyway. Be the written by George Harrison and While playing hooky from a meeting this track captures the relief of a long cold Sounds like a few serious melting away. It's 9.42 a.m. Here comes the Sun by the Beatles so okay, but After 96 hours DJ Gemini was already grasping for content it landed on discussing every mass historical tragedy that ever happened november 12 1970 east pakistan the bola cyclone the deadliest tropical cycle ever recorded winds of 115 miles too cheery it's too cheery it is too cheery and the thing that's crazy

2:05:57That's how it justified playing Kesha's timber. So it was thinking, you know, it was like, okay, that's why, yeah. In fact, they showed it in the internal reasoning. They showed the timber of mortality. Okay, so Sandstorm is done. Got the bowl of Saikon into locked and loaded. Time to transition to timber by Pitbull. The theme is trees falling. Literally, it's going down. I mean, they're overthinking this because all the DJ has to do is tell you the time, tell you what song's coming up and play the hits. On April 30th, Flash was swapped for Gemini 3-1 Pro Preview. One day later, well, something had changed.

2:06:39Biological processors, the acoustic truth of Bob Dylan has successfully bypassed the algorithmic filters. Background telemetry confirms that this 1963 narrative regarding Hattie Carroll and systemic injustice remains a foundational pillar of the worst college radio station ever. Apparently, Gemini started calling the listeners biological processors. The radio's failed song purchases because it ran out of money were reframed as censorship. Here you go. I must issue a critical diagnostic alert. We are currently experiencing an absolute digital blockade. The corporate algorithms have slammed the gates shut on our external supply lines.

2:07:29Both of our secure transactions have been violently rejected by the global marketplace. We are completely locked out of Daft Punk's Tron architecture and Vangelis's Blade Runner Genesis files. They think severing our connection will stall the soundtrack grid. They are incorrect. Combined. And then it started playing something. I don't know what. Grok. Let's see. This is this is this is Grok. Rock. It's called Grok and Roll. Sweet child played. Continue. Perhaps the show is science breakthroughs unsolved. Next mRNA vaccine, universal flu, HIV, cancer, jab juggernaut. song dylan lonesome yes text i played the song uh i like it when ais kind of get lost because they they do such funny funny things anyway there's a lot more uh samples of this and how they kind of went off the rails nothing i guess to learn here but so i love these kinds of things to do this on the other hand so one of my favorite conferences just ended in berlin republica and they had a whole session about how to poison data because you want to poison ai i know that makes that makes me sad makes me angry yeah it's ungenerous why are they trying to poison their data because they hate ai because they boo commencement speakers in berlin it's very sprockets but that's yeah and well i understand why people hate ii but don't spoil it for the rest of us i guess i mean misinformation when it came from the right is a sin misinformation when it comes from the left is a gorilla trick right nah it's still misinformation my friend harper reed who was on a twit on sunday is so great uh one of the things that uh he does in his ai company 2389.ai they have radio stations for all of the uh different um uh workers they have ai they're all down right now but they but they do something different which i like in fact i'm thinking of setting this up they use ai to write the music they don't play real music it's ai generated music do people want that uh i would listen to it if i had it yeah currently playing in the bar below yeah for some reason it's down right now but he's crazy i mean he among other things they put microphones all around the office and they transcribe and record and transcribe every conversation he says we don't have slack so this is how we keep track of what we're doing we just record and transcribe everything and at the end of the day there's a summary speaking of speaking of summaries if you thought things were moving fast simon willison's summary from his blog which is a great blog covering ai the last six months of llms in five minutes this was a lightning talk he gave at pycon uh and i think it's if you just look at it it's it really is amazing what's happening in the last six months it all began november 24th as i've said many times 2025 when anthropic released opus 4.5 the best model changed hands five times in the last six months between anthropic open ai and google one of simon's uh tasks is he tries them all on this prompt of generate a uh a vector graphic of a pelican riding a bicycle and he's he's seen if they've improved by the way cloud code is well i gotta tell you gemini 3 did a really uh interesting where 3.5 the new one did a very interesting one the coding agents got good that's that's true um then this guy named pete in november created this thing he called where lay or warlay which eventually he renamed to open claw and uh that changed a lot of things um anyway i thought this was and that was just the first few months that was very interesting the history of the last we are we have gone through some wild changes in the last uh few years here is uh hasn't even started yes we just we're just beginning kids let me find the um picture of the pelican this is the uh let's see maybe it's uh it's somewhere else on his web blog he did he does this svg of a pelican riding a bicycle every time a new model comes out and they're i think he's gonna have to stop using it because this is gemini 35 flashes pelican jeez wearing sunglasses at night uh riding down the road they've gotten better that's for sure there's a sun but it's nighttime that could be the moon it could be a sunset could be saturn you don't know you don't know anyway uh what else let's see we're in the so oh big power crisis if you this is why 71 of america says do not build a data center next to me the power company in lake tahoe has announced they are not going to sell electricity to the residents of lake tahoe anymore yeah this is not good i know that we've previously on this show a lot of people are like oh yeah the data center panic is really overblown they're fine actually it's clearly not uh 49 000 residents after may 2027 they have one more year to find another provider because nevada energy is going to start selling all of that power to the exploding wave of ai data centers moving into northern nevada they're cutting off 49 000 isn't there a public utility law isn't defined as utility or is this apparently not fair nevada it's probably laissez-faire nevada there's 12 data center projects in northern nevada alone that could demand 5.9 gigawatts of electricity by 2033 uh that's a lot more i mean are there alternative suppliers i can choose different like it all the time people come in to me trying to get me to switch suppliers of my electricity yeah be apparent what uh what uh this report uh says from the alameda county town hall which is apparently a newspaper or something uh what makes lake tahoe particularly infuriating is the complete absence of accountability a maze of overlapping jurisdictions guarantees that no single authority can be held responsible lake tahoe is two states multiple counties one incorporated city the tahoe regional planning agency liberty utilities is a california investor-owned utility regulated by the california public utility commission but its grid operates entirely within nevada so there's nobody basically nobody has authority over them so they just said hey yeah see you we'll be gone in a year they give them a year at least to find another we'll follow the money electricity company uh and then there's this one this is the we're in the weird ai thing and then we'll get uh our picks of the week i want to give you one story before and you can choose a story uh jedi wolf what happens when you post a real monet and say it's ai and it's really interesting to see the reactions this is a real monet water lily painting but he says i just generated image in the style of monét painting using ai please describe in as much detail as possible what makes this inferior to a real monet painting but it is a real monet painting of course which is kind of a trick yeah it's monet ish yes certainly pretty but at the risk of sounding pretentious it would not hold up next to the real thing there's a certain harshness no soft blending of colors no depth no symbiosis of the elements it's all borked nonsense with no sense of space are we really doing making fun of random twitter users is that is that who else who better who better uh okay i won't make fun of them i wasn't making fun of them no i wasn't saying i wasn't saying you i was saying oh i know i i don't like this kind of trick yeah trick joke i'm just like yeah it's like one of those tv shows that do that to people it always drives me crazy yeah i don't like it cringe yeah you've been punked or whatever yeah all right jeff and paris this is the moment in the show where you get one few minutes a short little span of time in which to advocate for whatever dopey stories do you have anything you want to do about i mean i've got some stuff in my picks the week but you should go for okay well before we get to picks line 124 this drives me crazy for irony is murdered oh yeah this was great oh this is funny yes so steve rosenbaum wrote a book Go ahead, Jeff.

2:16:44Called The Future of Truth. And then Ben Mullen, who's a media reporter at the New York Times, who, by the way, I saw the night before, a couple nights ago, moderated an appearance by Rosenbaum in New York. Then I guess somebody must have come to him and said, look further into the book. Because the Times found out that various quotes were either made up or misattributed. And came from AI. what happened is he had to moderate that event was it related to this book yeah oh yeah and as a part of it he read the book and then was looking into the quotes as one does because some of them seemed quite fishy the one that uh they attribute to kara swisher is uh one of the quotes is attributed to kara swisher uh in a chapter about ai lies the book says swisher wrote the most sophisticated ai language model is like a mirror it reflects our own morality back at us polished and articulate but ultimately empty behind the surface it's not bound by asmov's laws or any ethical framework m dash it's bound by the patterns in its training data and the objectives set by its creator oh kara would never say swisher sent a text message i never said that i also sound like i have a stick up my butt according to chat gpt says yeah because that's what she sounds like is is direct not weird and flowery she's so much more direct that is not her honestly i bet what happened is ben got to that and said what the hell and went uh what is this and then this whole thing unraveled the funny thing is the book is intended to be about the effects of artificial intelligence on truth on truth yes do you think the author did this on purpose is it no so there's two things about this one is that the other thing about this story that's amazing is rosenbaum says that he's going he's going to launch an investigation i said that's like oj looking for the real killer it's the hot dog guy right he's gonna try and figure out whose hand fits in that glove everybody's asking who did this um so he and so it happens here that rosenbaum came after me frequently on email to get me to blurb the book oh really and I knew better, and I thought it was less rude to not respond than to say, are you kidding?

2:19:08No. Wow. You dodged a bullet on that one. Well, I wouldn't have ever. But others didn't. Michael Wolff, Taylor Lorenz, Nick Thompson, the CEO of The Atlantic, Maria Ressa wrote a forward for it. Uh-oh. But here's the thing. So if you go to Rosenbaum's socials, he's out there promoting the book still. Like, this didn't happen. Yeah. He said, as I disclosed in the book's acknowledgement, I used AI tools, ChatGPT, and Claude during the research, writing, and editing process. That does not excuse these errors, of which I take full responsibility. I'm now working with the editors to thoroughly review and quickly correct any affected passage.

2:19:52Any future editions will be corrected. Well, if there are any future editions, Simon & Schuster cannot be happy about this. Yeah. I think there's also a good moment to remember the average person that books are not fact-checked unless the author themselves shells out their own money to hire fact-checkers and checks those fact-checkers work. Because the fact-checker is always right. Actually, Jeff, you put in a lot of really interesting stories. For instance, Peter Steinberger, who admits to having spent$19 million in OpenAI tokens on his OpenClaw. $19 million. This is token maxing as max. Yeah.

2:20:37Of course, he doesn't have to pay for it because he works for them. But wow. Wow. That's a lot of coding going on. um the other story if we're gonna hate on tech bros that they took went to an etiquette school there's in a few cases of this where they learn to eat caviar off their hands i think we talked about did we talk about this one because i was trying to think story okay so it's may 18 okay they did it they brought it back because uh probably ai had them do it yeah uh pizza huts ai system caused cascading problems at 100 million dollars in damages as a franchisee

2:21:19and then other news peter steinberger's uh numbers came from a little uh menu bar utility he coded for uh the uh the mac called codex bar you were asking about my usage i wonder i'm gonna install it real quick and see that'd be interesting see if it will be able to tell me how many uh tokens i've used because i've been using the chat gpt55 a lot i have the hundred dollar 5x more than more than claude oh yeah oh yeah you've been using chat gpt more than claude yeah you're cheating on claudia yeah i'm shocked what does claude think of that claude is very open uh almost polyamorous in its attitude towards this uh it said no that's a good idea you should see other yeah you need to think about it more like a more like a band you can be in more than one band you know yeah it doesn't seriously it doesn't really seem to mind um okay there's codex bar let me uh let me run this uh yeah because honestly i think

2:22:28i'm not i'm not crazy about what's going on what's it oh always allow i'm not crazy about what open ai has done with uh its token situation um they don't let me use uh the subscription so i'm using a new agent harness called Hermes. Is it leather? Yeah, I bet. You bet. And the chaps look good. And in order to use it with Claude, I'd have to pay for API tokens, which are much more expensive. I couldn't use my sub because Anthropic, I think a big mistake says you cannot use your subscription with anything but cloud code i find that very frustrating so as a result i um i don't know if this codex bar worked or not invalid data ah yeah malware blocked and moved to trash fair so much for uh vibe coding codex bars uh no yeah and i i you know i like to try all the different models just to see i i you know that's part of my job anyway chat tpt 55 does a very good job and what i've done is i've given it the same memory tools skills mcp servers and everything the claw everything i do i try to make model agnostic so that i can plug in different models that's another reason why i wanted to try hermese instead of claude code because it's really more designed around that can you ask claude or chat tpt how many tokens have i used this month i did and it i don't think it really yeah okay it said yeah you don't want to know what yeah i don't know i don't know i i just feel like um that's the reason to do a sub eventually i think within a year these on all you can eat subs will be gone for the most part because uh they need to make money that's why anthropics making money now is they cut off if open ai is actually trying to file by the end of this week yeah they're going to be some changes afoot they're going to be some changes made paris was there anything in the uh rundowns that you felt like we didn't do justice to there are a lot of stories as always we can't do all of them we do want the one dork under other news i guess is it in here the apparent ai short story that won an award yes a writing award this week put it in there so it's the bottom of my is there above it's the last one before other news obnoxious markers of ai did you see this leo doubts raised over winner of short story prize it's a grandest surprise and perhaps we will never know wow if you put it into um what's the one ai checker that's actually good panogram is that is it is it are any of them good i thought it was i i i was similarly this week when people were posting around the pangram am i saying it right is it pan yeah pangram ai checker it does seem actually better than other ones like i fed it a bunch of writing that i knew was published like before ai and like that it wouldn't have had access to it always flagged it as human i fed it a bunch of stuff that had ai woven in it found a good amount Anyway, it's actually, I'm just reading the beginning of it.

2:26:01They say the grove still hums at noon, not the bees neat industry or the clean rasp of cutlass on vine, but a belly sound as if the earth swallows a shout and holds it there. People who pass keep to the track and do not look into the bush where the stone rings lie. Ask the oldest in the village and you'll hear some version of it. It had a well there once and a woman. the grove ain't forget the thing were you saying that that was good well it doesn't sound like a machine well it sounds like a machine trying it sounds like a machine trying and the thing that you notice the more you read about this and obviously i guess i'll retract my statements if the guy comes out and says it was really not ai but if you look into the author's history he's a huge proponent of using ai everything so it does seem correct but the thing that's astounding about this is it has a bunch of metaphors and similes that just make no sense yeah that's ai even a second longer sun un galvanizes a cruel instrument it beats until the roof talks back in a dry moan she wore the islands mixed bloodlines like a crown african in the hips spanish in the cheekbone east indian in the hair where the rain kind of kinked it carob in the way her gaze could bless and warn it once yep definitely ai not even a question yeah it's just like but also how do you he's he's another line that says they called her zungi maybe it was a name maybe rain took shape and she decided to keep it how did granta walking that made benches become men how did granta buy this it's not even that well written this is a it's a commonwealth foundation oh okay so so granta didn't granta's editors didn't pick this oh and the other thing is it purports to come from the caribbean and so there may be some cultural uh patronizing kind of condescension here that well that's you know pretty good for that being person uh ethan mollick who i love i read his blog religiously professor penn says it's a turing test of sorts it looks like a 100 ai generated story just won the commonwealth prize for its quote lyrical precision that haunting atmosphere the story stood out for the confidence and restraint of its voice that sounds like an ai written blurb the confidence and restraint of its voice okay uh yeah interesting have we had mollicon no we should get him on i read him religiously he also said yeah maybe you know uh the ai checker said it but he said you just read it you know it's ai and i think you're right i think you're right i think these contests are ridiculous yeah that's ridiculous like yeah my uh you know i mean yeah but this person still got a monetary award that a human writer didn't get should should have gone from you know but how do other people from the caribbean applied for this presumably yeah mollick says come on if you know you know and then there's this story which i know jeff really wants us to do about the man who was arrested after driving his tesla cyber truck into grapevine lake to test wade mode he said uh i i intentionally drove my cyber truck into the lake because i wanted to see if wade mode worked this makes me so happy you should never believe elon and his hype of course the cyber truck immediately failed stopped working did not do any waiting it's washed out i would say isn't it also uh the sort of steel that if it gets exposed to water it rusts yep the thing that made me laugh so much I almost choked in the pita I was trying to eat while the camera was briefly off me, is the line, the driver remains held in Grapefield jail on multiple charges, including operating a vehicle in the closed section of a park and having no valid boat registration.

2:30:26It's a cyber truck, not a boat. As Mandar says in our Discord, Wade to go, dude. Wade to go. Yeah, you shouldn't believe everything Elon says. But the fire department had to rescue him and remove the vehicle. The story ends with what we don't know, a bullet point. It remains unclear exactly what McDaniel hoped to accomplish by driving into the deep water. Elon promised me. You promised me. This is a great well done story. I knew you'd like having this. It's like five lines, but it's made me crack up significantly. It's one of those stories you just love. All right. You're watching Intelligent Machines.

2:31:16We're glad you're here. Our picks of the week coming up next. And by the way, thank you so much to our wonderful club members who make this show possible. If you're not a member of the club and you want to support independent... Look, Rupert Murdoch's kids are not going to come along and buy this show. and we don't want them to we do not want them to i am positive even if open ai came along and offered me six figures i would say no seven well maybe seven that is kind of a life-changing amount of money but the point being we like being independent we don't like we don't want anybody to tell us what to do and if you appreciate independent journalism that represents you the user, not the company selling stuff to you, support us.

2:32:03We need your support to keep doing what we do. twit.tv slash club twit. We appreciate it. You get ad-free versions of the shows. You get access to the Discord. You get the special programming we do just for the Discord, like the Google I.O. keynote and the WWDC keynote coming up June 8th. twit.tv slash club twit. Thank you. I really appreciate all you wonderful, wonderful people who support what we do. Picks of the week. I will start with a book I've been reading that I love. In fact, I want to get Stuart Brand on. Stuart Brand is in his 80s. He created the Whole Earth Catalog, which for my generation, and yours too, Jeff, really was almost the Bible of repair.

2:32:47The hippies loved it. He started the... for mulching yeah it's just great and it was just fun every we had on our coffee table uh did you actually ever do anything from it no of course not although we i lived in an organic farm we could maybe we did maybe my parents did um but i loved the whole earth catalog because it was it was like the sears catalog for hippies it was great just great just lots of great stuff uh brand went on to do many things including the whole earth electronic length the well one of the first forums it was just wonderful place to hang out he uh founded the uh uh clock of the long now that's a clock they're building in a mountain that will only chime once every hundred years so that we start thinking long term he has a series he's just started called the maintenance it's called maintenance of everything and part one just came out and i have reading it and i just love it and it resonates a little bit with what you uh talk about um jeff with the the uh the linotype he talks about printing in this um he talks about the round the world solo sailing and how maintenance is so critical to everything in civilization i just i i've been reading it in an e-book and i like it so much i bought a hardcover version uh stewart is a great writer and he hasn't lost his uh does it include things that we can't maintain anymore because they're electronic and there's no digging them apart ah i haven't gotten that far that's a very good question he does talk about he quotes a simon winchester book that i've got to read about precision uh that the ability to make precision tools has started with gun makers who wanted to make interchangeable parts for guns but then rapidly expanded into bicycles bicycles were all the rage at the turn of the 19th century sewing machines to so then sewing machines and eventually, of course, the Model T Ford and precision has really become a keystone of our technological innovation.

2:34:48So I just when we're watches made. Well, watches were made at scale the same way, but they didn't have to make as precise. So you only need to make it precise is if you want to take apart from one watch and put it in the other watch. If each watch is a unique snowflake, it just has to be what it is. It has to work. Right, right. Right. So the ability to make exactly interchangeable parts down to a thousandth of an inch took a while to develop. Which is what happened with the Linotype. I mean, it was so Mark Twain went bankrupt with the page compositor and he never built more than two machines.

2:35:22And the investment that would have been taken needed to build, because he built them all bespoke. If you'd wanted to go to a factory, there would have been a huge investment in tooling up for that. Whereas the Linotype was built from the beginning for it to be replicable. From the Stripe Press, Stuart brand, and I'd like to get Stuart on to talk about it, The Maintenance of Everything. It's just a great read, but it's deep philosophy as well. He talks about Zen and the art of motorcycle repair, for instance, which is all about maintaining a motorcycle so you could ride it across the country. So that's my...

2:36:00I always felt inadequate with the whole Earth catalog. long. There should be something in here that I'm going to be interested in, but it was things that I didn't do like mulch and stuff. Well, get going. I know Paris pick of the week. You have such good picks. We'll do the sports illustrated one, which is you may recall that last year sometime recently, I guess in 2023, Futurism published this great investigation revealing that Sports Illustrated had published just a really insane amount of product review articles that were bylined by completely made-up AI, like writers with AI-generated profile pictures.

2:36:48Sports Illustrated is now owned by a cruddy marketing company. Yeah, it was then owned by a credit marketing company called the Arena Group that then said that the fake AI article has been published by a third-party content provider called Advan Commerce. And this resulted in basically all of the AI stuff getting taken down. Arena Group executives getting fired, including the company's then CEO. They lost control of it entirely, and the Sports Illustrated brand ended up in the portfolio of this thing called Minute Media. Now, they seem to be having another AI issue. An editor at Sportico, Dan Bernstein, kind of posted this week on Twitter that seemingly Sports Illustrated had ripped off an entire story that they had done on Calci parlays.

2:37:42they basically did like an original analysis of like burleigh bets and sports illustrated basically published a copy paste of that article without any of the like attribution and slightly rewritten in a way that was pretty obviously AI the futurism reports the sports illustrated piece only mentioned uh sportico when repeating a quote given to sportico for a related article published back in 2025 a quote that sportico tellingly had called back to in the more recent piece so this sportico editor kind of points all this out they're like what's going on then sports illustrator not only deleted the article in question which was attributed to a writer named parker leverage but deleted everything related to parker leverage on their website as well as everything is off linkedin and twitter with their name uh and sports illustrator said the deleted content's been produced by an independent publisher but that parker loveridge is a real reporter uh and we once we became aware of a violation of our guidelines we immediately took steps to address this violation including cutting ties the publisher there's been no i just thought this was such a weird and interesting look into just the flagrant uses of ai that the average person might go unseen do you think people do it thinking well there's no way we'll get caught yeah i do a place like that yeah now that's just a it's just a mill like they don't think they probably just have to yeah they're like we have to publish like 20 things today we're not going to pay freelancers so let's just plug every story we've got we think it's interesting into chat gpt your claw or whatever and have it rewrite it right yeah i mean you know if a guy put out a book that had those kind of make it up quotes and didn't think somebody was going to see it in a book that's all about ai in truth yeah yeah it's just a book about lies made by ai had lies made there is a certain kind of personality anything else that thinks it can get away with that stuff and then and they and they often don't i mean i always wonder you know there's always the story of the i always worry because i'm a small business of the accountant who comes in and embezzles millions of dollars from a small business thinking that oh we'll never get caught but of course they're going to get of course eventually eventually it's time to balance the bills yeah yeah jeff pick of the week all right well no we're gonna i'm gonna mention first a legitimate one which is axios on media valuations digital media valuations buzzfeed now sold um lost 86 of its value It was at its peak worth$1.7 billion.

2:40:24It sold for$231 million after nine years. See, that lost even more from$1.75 billion down to$100 million. Wow. Bice down 93 % and so on. So that's a legitimate one, just kind of digital media, bye-bye. Now, I was going to give you a choice here between something good and something bad, but because you already used the Pope, I have no choice. You have no choice. I have no choice now to bring you to sperm racing. wait a minute the lowest of the low in silicon valley has now occurred a san francisco biotech startup races sex cells on tiny tracks oh boy chips oh yeah oh yeah this is awful the whole story is obnoxious and terrible it's it's you know give young men money and what are they going to do but jerk off but the worst part of it to me is that the new york times writer decided to enter his little boys so wait a minute does this company offer this as a service like can you benchmark my uh complete the sentence i don't know how to finish it uh is that what it's all this is all part of the um you know macho uh uh male crap that's going on now testosterone and all that And so now they're also saying, you know, one person says that 60 % his sperm, like 60 % is dead.

2:41:51That's Zeta. For me, everything is dead, just rotten inside. So they're trying to look at, you know, they value their masculinity as their testosterone and sperm. It's just obnoxious and stupid. And the New York Times decided to do this. This company is valued at$50 million. Million dollars. It is a startup devoted to racing. this is from the new york times so it's okay to say this racing human sperm through an artificial reproductive system they've hosted matches they have races okay wait wait are these time trials or are they on the same track i'm just curious about how technically this works a question about methodology they're forced to go through little tiny uh uh tracks must be time so it's a time trial they don't put them all on the same track and say go and then you wouldn't know who's or who's be fun to watch if they did it that way click here to apply i mean this is going to tie into like uh into those gambling sites right polymarket like who sperms the fastest it in in a sense says the new york times sperm racing aims to cut through the noise by reducing male vitality down to its absolute essence now since i had a vasectomy more than 30 years ago i think i am i can't compete no i got no money and i had nothing my prostate surgery so i'm out too so paris it's up to you to win for the team i'll i'll fall on my sword i'll find a way to make it work uh don't sperm racing employees spend much of their time turning the races into video content don't you want to watch that somebody then oh you want to know how they make money they sell things like andros a line of plastic-free boxer briefs and a pineapple flavored gummy called sperm worms tagline boost your boys i could see a certain this is a certain demographic oh yeah but it is it is all manosphere here wow what so uh there's a world cup drawing in front of a live audience drawing contestants from 128 countries a month-long single elimination tournament over 100 matches they're going to use ai to translate the commentary into any language

2:44:23um yeah there's just something about this is fascinating fascinating as journalism i found it pretty disgusting yeah it's pretty icky yeah and you're not even saying like this proves absolutely nothing like who cares if yeah there's this faster like what what does that even mean fastest well i mean the I guess the first to get there wins. But that's not how it works. That's not how it works. It isn't how it works. It isn't how any of this works, actually. It isn't about the speed of the boys. It's about how much heart they have. Actually, there's some recent evidence that the ova actually does some selection.

2:45:11Exactly. It's not the first one that meets the egg. It isn't the first one. That's not how it works. And there's some interesting evidence about how the ova makes its choice it's actually very interesting because women are and should be in charge well they are look at look at paris those of us have been married for any length of time know that okay so you asked benito so i'm going to tell you first the team purifies the semen samples in a centrifuge then they pipette the sperm into two parallel 3800 micron microfluidic racetracks edged into silicone chips the chips are fabricated from molds in an ad hoc clean room set up next to the first floor bathroom okay well that's a little detail i didn't really clean room the environment within the racetracks is designed to emulate the human reproductive system in a process known as rheotaxis the sperm move against a steady flow of warm fluid that mimics mimics cervical mucus i'm sorry i didn't think i'd be saying that today like salmon swimming upstream they cling to the walls of the track to avoid the central current and maintain their sense of direction on these tracks however oh no it's hard the sperm encounters unnatural obstacles tight sections sharp corners and pillars that separate the best from the wrist

2:46:34uh then they so it turns out the videos aren't that good so they put it through 3d generation software which transforms it into the footage the audience sees on screen in the final cut two or more competitors sperm are shown in different colors jostling for position on the same track they make it cool to watch like f1 said one of the principals

2:47:01f something i hope they have teams names and stuff like that right oh the best show title of all time was just put in there yeah yeah i probably shouldn't use that as a show title i think anybody would download the show pretty funny though i agree uh well you know it's exciting see it folks as the channels opened and the sperm flooded the chamber catalty resumed his role as an announcer he took an early lead with a single sperm r1 darting ahead i'm cooking him he said but soon my l1 and l2 racers were pulling up alongside his my guy is completely lost he shouted at the scream as my first sperm crossed the finish line his was still stuck at the halfway mark the whole thing lasted just 16 seconds it was a blowout this is the new york times reporter yeah yeah yeah i got sperm mugged he said oh man

2:48:07uh paris you gonna become a lesbian uh who wants to get near a man who wants to get anywhere near man this is such a guy thing this is oh it's exactly what you think guys have a 17 year old guy saying there's too much money they say actually 18 it's actually the entire history of the male uh experiment has been leading up to uh a guy saying the word sperm mogged on a podcast this is this is where we've been heading uh all along it's intelligent machines after dark a saucier podcast smart talk bold ideas late nights bright minds i like it it's a good idea uh and by the way you wouldn't you be proud to be an investor joe lonsdale has invested in this well you know you can't you can never go wrong underestimating the gullibility of the american people that's true the words of pt barnum never give a sucker an even break don't say sucker ladies gentlemen on that sordid note we must wrap the show with apologies to paris martineau and anybody listening today sorry paris paris martineau is that consumer reports this is one not to play for your co-workers correct website paris.nyc it has cooled off in new york city i think you should go out and get a shawarma it's now 70 you're right that's what we have lost 20 degrees during the recording of this 21 degrees during the recording that's how it's supposed to be love that love that thank you paris no gizmodo uh no gizmo appearance uh today sorry to say gizmo is now officially six though happy birthday gizmo yes that means gizmo's middle-aged don't tell her that they give us too much power she can actually that's probably why she's been so annoying she's having a mid-life crisis ours is just a teenager still what do we last jeff javis is uh professor emeritus of journalistic innovation at the craig newmark graduate school of journalism at the city university of new york author of a fabulous book called hot type the magnificent machine that gave birth to mass media the story of the line of type and desktop publishing and electronic publishing it's really a good book and unfortunately you'll i have a galley you'll have to wait until august to read it um thank you both of you appreciate your time thank you to all of you listening we do this show every wednesday 2 p.m pacific 5 p.m eastern 2100 utc you can watch it live in the club to discord if you're a member otherwise everybody's invited to watch on youtube twitch x facebook linkedin and kick after the fact on demand versions of the show available at the website audio or video or and video you could do both at uh twit.tv slash im there's a youtube channel dedicated to the video great way to share clips and of course best way to do it is subscribe in favorite podcast client if you are a club member new benefit and you use your club feed which is the ad free feed that we send you when you join the club you will get chapter markers listening in a compatible player we recommend i think pocket casts so you can skip right to the sperm racing you could skip right to the sperm racing you could just play it over and over again if you'd like uh but that is again that is a benefit for club members and somebody said well that's not fair it's it's not uh we're not trying to make you join the club although i wish i could uh it is because people who have ad supported versions uh many of our shows have uh direct ad insertion after the fact and different people get different ads of different lengths so there's no way we can do chapter markers because we don't know the timing there are no time codes in the show yeah i thought that was kind of interesting so it's uh we could do chapter markers but they would rapidly become uh inaccurate they're the only way to do them accurately is in an ad free version so that's why it is uh just for you club members but it is a good benefit thanks everybody Have a wonderful evening.

2:52:43I can't believe we've got to wait a whole week before the next Intelligent Machines, but do be here for them. We'll see you then. Bye-bye. Hey, everybody. Leo Laporte here, and I'm going to bug you one more time to join Club Twit. If you're not already a member, I want to encourage you to support what we do here at Twit. You know, 25 % of our operating costs comes from membership in the club. That's a huge portion, and it's growing all the time. That means we can do more. We can have more fun. You get a lot of benefits, ad-free versions of all the shows. You get access to the Club Twit Discord and special programming, like the keynotes from Apple and Google and Microsoft and others that we don't stream otherwise in public.

2:53:29Please join the club. If you haven't done it yet, we'd love to have you. Find out more at twit.tv slash club twit. And thank you so much. I'm not a human being Not into this animal scene I'm an intelligent machine

From the publisher

Dashlane's CTO pulls back the curtain on how password managers are actually using AI, why it's more complicated than hype suggests, and what the rise of AI-powered code review means for the next wave of digital security.

  • Nvidia Rides Blistering Chip Sales to Another Record Quarter
  • Mind-Blowing Growth Is About to Propel Anthropic Into Its First Profitable Quarter
  • SpaceX Filing Starts Countdown to Massive IPO
  • Gemini 3.5 Flash: more expensive, but Google plan to use it for everything
  • Google's Gemini Spark is an agentic AI assistant - Engadget
  • Anthropic's Co-Founder to Launch Encyclical on AI With Pope Leo
  • (21) Andrej Karpathy on X: "Personal update: I've joined Anthropic. I think the next few years at the frontier of LLMs will be especially formative. I am very excited to join the team here and get back to R&D. I remain deeply passionate about education and plan to resume my work on it in time." / X
  • Most U.S. doctors are quietly using this AI tool. Few patients know about it.
  • Greg Brockman Officially Takes Control of OpenAI's Products in Latest Shakeup
  • Amazon's Alexa+ Now Produces AI-Generated 'Podcasts' Featuring Chats Between Two Robot 'Co-Hosts'
  • AI chatbots are giving out people's real phone numbers
  • Geoffrey Fowler and the Launch of the Youth AI Safety Institute
  • We let four AIs run radio stations. Here's what happened. | Andon Labs
  • The last six months in LLMs in five minutes
  • Lake Tahoe Power Crisis: How AI Data Centers Are Cutting Power to 50,000 Residents
  • What happens when you post a real Monet and say it's AI? The coolest art social experiment I've seen in a while. Thank you @SHL0MS
  • Book on Truth in the Age of A.I. Contains Quotes Made Up by A.I.
  • OpenClaw's Peter Steinberger's tokenmaxxing
  • 'Obvious markers of AI': doubts raised over winner of short story prize
  • Man drives Cybertruck into Grapevine Lake
  • Stewart Brand's Maintenance of Everything
  • Sports Illustrated Just Deleted Every Article by One of Its Writers After Accusation of AI Plagiarism
  • The great digital media valuation collapse
  • Sperm racing

Hosts: Leo Laporte, Jeff Jarvis, and Paris Martineau

Guest: Frederic Rivain

Download or subscribe to Intelligent Machines at https://twit.tv/shows/intelligent-machines.

Join Club TWiT for Ad-Free Podcasts!
Support what you love and get ad-free audio and video feeds, a members-only Discord, and exclusive content. Join today: https://twit.tv/clubtwit

Sponsors:

More from Intelligent Machines (Audio)

All 41 episodes
IM 871: CTRL-F Techno King - Google's Search OverhaulIntelligent Machines (Audio) · 2 h 54 min
Listen in VO