IM 858: The Itinerant Salt Miner from Buffalo - Silicon Valley's Military Dilemma

19 Feb 2026 · 2 h 52 min · 63 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

Intelligent Machines - Episode 858 Summary

Episode Title

The Itinerant Salt Miner from Buffalo - Silicon Valley's Military Dilemma

Hosts

  • [Leo Laporte](https://twit.tv/people/leo-laporte)
  • [Jeff Jarvis](https://twit.tv/people/jeff-jarvis)
  • [Emily Forlini](https://twit.tv/people/emily-forlini)

Air Date

February 18, 2026

---

Episode Overview

In this episode of Intelligent Machines, the hosts discuss significant developments in the AI landscape, including the rise of OpenClaw, the potential mainstream adoption of agentic AI, and the implications for independent developers and large tech companies. The episode covers the latest news from leading AI organizations such as OpenAI, Anthropic, and Google, and explores ethical dilemmas surrounding AI deployment in military contexts.

---

Key Discussions

  1. OpenClaw and AI Development
  2. Peter Steinberger, creator of OpenClaw, has joined OpenAI, indicating a shift in independent developers' landscape.
  3. OpenClaw has gained massive popularity on GitHub, highlighting the demand for user-friendly AI tools.
  4. Discussion on whether 2026 is the year agentic AI goes mainstream and how big tech companies are preparing for this disruption.
  1. AI in the Military
  2. The Pentagon threatens to cut off Anthropic due to disputes over AI usage in military applications, raising questions about the ethical use of AI in warfare.
  3. Anthropic's commitment to not allow their AI to be used in surveillance of civilians or autonomous weapons.
  1. Recent AI Advances
  2. OpenAI disbands its mission alignment team, signaling a shift in corporate strategy.
  3. Introduction of Google’s Gemini 3 and Anthropic’s Sonnet 4.6.
  4. The effectiveness of AI models in legal reasoning, where GPT-5 outperformed federal judges in a study.
  1. AI and Health Care
  2. The use of AI avatars in healthcare, presented by Dr. Oz, as a solution for rural healthcare access, but met with skepticism regarding their effectiveness.
  3. A highlight of a case where an individual successfully negotiated a drastically reduced hospital bill using AI.
  1. AI in Media and Journalism
  2. Discussion of a reporter's failure to verify AI-generated quotes leading to a retraction from Ars Technica, emphasizing the necessity of human oversight in AI reporting.
  3. The rise of AI in writing, with organizations exploring using AI for news generation while maintaining quality.
  1. Cultural Reflections on AI
  2. The hosts share thoughts on how AI interacts with cultural content, discussing the societal implications of AI-generated music and art.
  3. Discussion on AI pets and their reception among users, alongside humorous anecdotes from the hosts.

---

Picks of the Week

  • Emily Forlini: Not specified.
  • Jeff Jarvis:
  • Discusses a paper about AI's potential in decision-making contexts.
  • Leo Laporte:
  • Highlights Peon Ping, a tool for setting sound notifications for terminal tasks, featuring quirky audio prompts relating to gaming culture.

---

Conclusion

The episode encapsulates the rapid developments in AI, emphasizing the importance of understanding its implications for society, ethics, and individual lives. The hosts engage in a lively debate about the potential and pitfalls of emerging technologies while highlighting the need for accountability and oversight in AI applications.

---

Additional Notes

  • For listeners interested in AI, technology, and journalism, this episode provides insights into the evolving landscape and the critical discussions surrounding it.
  • The episode can be accessed or subscribed to through various podcast platforms, including [Intelligent Machines](https://twit.tv/shows/intelligent-machines).

---

This structured summary captures the essence of the podcast episode, outlining key discussions and themes while maintaining a clear and organized format for easy reference.

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

Host Introductions and Updates

0:45 to 2:20

Hosts Jeff Jarvis and Emily Forlini share personal updates and introduce the show.

“It's time for Intelligent Machines, a show about artificial intelligence, robotics, and all the smart stuff all around us.”

OpenClaw's Journey

2:20 to 6:20

Discussion about Peter Steinberger's transition from OpenClaw to OpenAI.

“If you do it right, it's a half an hour, but I haven't learned all three sections yet.”

The Popularity of OpenClaw

6:20 to 9:50

Exploring the success of OpenClaw on GitHub and its implications for the tech industry.

“People were running out and buying Mac minis just to run it.”

AI Agent Discussion

9:50 to 12:00

Analyzing the concept of agentic AI and its practical applications.

“He was kind of quiet and didn't say a lot.”

Challenges with Chat Interfaces

12:00 to 14:05

Evaluating the effectiveness of chat interfaces in AI applications.

“year of agentic ai where um it's in my opinion i think i don't know you can you can tell me i'm wrong the chat bots are kind of not a not a great way to see what ai can do this is the commercial product.”

The Evolution of AI Coding Models

14:05 to 15:36

Discusses the advancements in AI coding capabilities and the implications for future technology.

“it got stalled out on the API, on the Twit API.”

Apple's Potential AI Hardware Innovations

15:36 to 17:06

Explores Apple's rumored new AI hardware developments and their market strategy.

“And I think that makes sense for us to focus on that.”

Anthropic's Focus on B2B AI Solutions

17:06 to 18:26

Analyzes Anthropics's strategic focus on B2B applications and its impact on the AI landscape.

“and and technology user rather than trying to be retail everybody.”

OpenAI's Changing Mission and Ethical Concerns

18:26 to 20:49

Discusses the shift in OpenAI's mission and the ethical implications of its current strategies.

“Um the the former head of it is now the company's chief futurist.”

Advertising and AI: A New Frontier

20:49 to 23:07

Examines the intersection of advertising and AI, including user experience and ethical dilemmas.

Show all 63 chapters

The Impact of Advertising on User Trust

23:19 to 25:51

Explores how ads based on user data can affect trust and user behavior in AI platforms.

“and using AI to remodel your home, that's the ad you'll say.”

Recent Developments in AI Models

25:51 to 28:00

Overview of recent advancements in AI models from OpenAI and Anthropics, highlighting their unique features.

“It's also when they realized that their strategy, they do have a strategy.”

Discussion on AI Model Comparisons

28:00 to 29:19

Explore the differences and pricing between various AI models like Sonnet and Opus.

“It seems to be they're essentially similar models, but slightly modified to be less expensive.”

Managing AI Context and Memory

29:20 to 31:23

Learn about AI context management and how it affects performance and reliability.

Anthropic's AI and Pentagon Dilemma

31:24 to 34:15

Delve into the controversy surrounding Anthropics AI use in military applications.

“So in the process of working with it, you will compact it.”

Palantir's Role in Military Operations

34:16 to 36:18

Investigate Palantir's involvement in military strategy and operations.

“But so far, I haven't been able to see a difference.”

Silicon Valley's Shift in Purpose

36:19 to 42:04

Discuss the evolution of Silicon Valley's focus from government collaboration to profit-driven motives.

Silicon Valley vs. Military Mindset

42:04 to 45:33

Explore the contrasting priorities and philosophies between Silicon Valley innovation and military strength.

“So now the government's behind, woefully behind.”

The Debate on Military Ethics and Technology

45:34 to 48:29

Discuss the ethical implications of technology in warfare and the need for a strong military.

Anthropic and Military Collaboration

48:30 to 50:27

Examine the role of technology companies like Anthropics in military applications and the implications for society.

Tech Updates: Google vs. Apple

56:47 to 1:00:58

Discover the latest smartphone announcements and tech rivalries between Google and Apple.

“And I just want to say, I know a lot of our listeners use Graphene.”

AI and Medical Innovation

1:00:58 to 1:03:01

Explore how AI is influencing drug discovery and medical advancements amidst regulatory challenges.

“They have a new system called the Isomorphic Labs Drug Design Engine, ISO-DDD.”

AI Ethics and Voice Theft Discussion

1:03:01 to 1:09:09

Examine the ethical implications of AI-generated voices and related lawsuits in media.

“I think that'd be much better messaging for this administration if they went in on the medical stuff because people care about that.”

Voice Recognition and Online Disputes

1:10:03 to 1:10:53

A light-hearted discussion about distinctive voices and a playful disagreement.

Family Lore and Ancestry Discoveries

1:10:53 to 1:15:08

A storytelling segment about family history involving a traveling salt miner and intricate family dynamics.

“So you have your great-grandmother's name, not last name.”

Salt Mining and Mining Communities

1:15:08 to 1:15:50

Discussion on unexpected salt mining history and related mining communities.

AI in Music and Video Generation

1:20:35 to 1:24:00

Exploration of new AI models for music and video generation, along with their implications.

“Well, any, any perplexity, uh, Gemini chat GPT.”

AI Video Editing and Hollywood's Future

1:24:14 to 1:26:32

Exploration of the new Chinese video editing tool, Sea Dance, and its implications for Hollywood.

Cinematic Techniques and AI Impact

1:26:32 to 1:28:17

Discussion on the influence of AI on modern filmmaking and audience perception.

Contrasting Animation Styles

1:28:17 to 1:30:05

A conversation comparing modern AI-generated animation with classic claymation.

“So the roadrunners kind of on crutches like me.”

AI vs Human Judgment in Law

1:30:05 to 1:33:08

Examination of a study comparing AI decision-making with that of federal judges.

“Now you don't need clay to do claymation.”

Exploring AI's Role in Legal Decisions

1:33:08 to 1:35:21

Discussion on how AI might reduce bias in legal judgments and its implications.

“They're in fact saying maybe it's not better.”

AI and Accessibility in Supreme Court

1:35:21 to 1:37:35

Analysis of an AI project aimed at making Supreme Court proceedings more accessible.

“An AI project is creating videos to go with Supreme Court rulings and opinions.”

The Good, The Bad, and The Ugly of Medical Bills

1:38:00 to 1:38:35

Explore how negotiating medical bills can lead to significant savings.

“Should I do from now on the good, the bad, and the ugly?”

Dissecting a $195,000 Hospital Bill

1:38:35 to 1:42:10

Learn about a case where a family reduced a hospital bill dramatically using negotiation tactics.

“But then it's like how much is his insurance paying?”

AI's Role in Medical Billing Verification

1:42:10 to 1:43:35

Discover how AI tools like ChatGPT can assist individuals in verifying medical billing details.

“So he showed ChatGPT Claude's work and said, check this for accuracy, examine every detail, flag any errors.”

The Intersection of AI and Music Copyright

1:43:35 to 1:46:34

Delve into discussions around AI-generated music and its implications for copyright laws.

Investing in AI: Fei-Fei Li's Ventures

1:46:34 to 1:49:22

Examine the emerging trends in AI investments, particularly in world models.

“So I guess the Marvin Gaye estate did win, which is kind of, well, wait a minute.”

Creating 3D Worlds with Marble

1:49:22 to 1:52:00

Learn about Marble, a product enabling users to create detailed 3D environments.

“met her i've met her you've met her yeah she's a very cool person when steve was still alive when they were married i played volleyball with them what it's a long old story i've told many times It was many years ago.”

Imagining a Dream Villa

1:52:00 to 1:53:10

The hosts creatively describe an ideal Mediterranean villa, discussing its features.

Virtual Design Experience

1:57:49 to 2:00:06

The hosts interact with a virtual design tool showcasing a dream home, discussing its features and capabilities.

“Okay, so apparently it didn't bother doing the indoors.”

AI in Rural Healthcare

2:00:06 to 2:02:14

Discussion on Dr. Oz's proposal of AI avatars as a solution for rural healthcare challenges.

“They rendered a lot more than I realized.”

AI's Role in Medical Nuance

2:02:14 to 2:06:00

Exploring the limitations of AI in healthcare, particularly in understanding patient nuances.

“I hope they this it's i mean dr oz right off the bat hard to take seriously it is hard to take it is If rural healthcare is an issue, is Dr.”

AI Hallucinations and Misquotes

2:06:00 to 2:07:00

Explore the controversy around AI-generated quotes and their implications for journalism.

Editorial Standards in AI Journalism

2:07:00 to 2:09:20

Discuss the responsibilities of journalists when using AI tools for reporting.

“apologizing as well they were as they should they retracted editor's note retraction of article containing fabricated quotations.”

The Debate on AI-Assisted Reporting

2:09:20 to 2:11:40

Engage in a debate about the use of AI in news reporting and its impact on quality.

“The editor of The Plain Dealer and Cleveland.com, the company I used to work for in advance, put up a post saying that, and this has caused much discussion, and I have a contrarian view about this.”

Challenges of AI in Quote Verification

2:11:40 to 2:14:10

Examine the difficulties faced by journalists in verifying AI-generated quotes.

“We need to be human in the loop, responsible for what we do.”

Public Perception of AI in Journalism

2:14:10 to 2:17:10

Analyze how public understanding of AI affects trust in journalism.

“And I just feel like that's what gives AI a bad rep.”

Security Issues with AI Technologies

2:17:10 to 2:19:20

Discuss security concerns related to AI technologies and their usage in reporting.

“And really the human that made the mistake here was the reporter who used fabricated quotes.”

Miscommunication in AI Reporting

2:19:20 to 2:20:04

Investigate miscommunications associated with AI tools in news reporting.

“banned in the united states and it turns out that donald trump jr has an investment in a american drone company, which doesn't make very good drones.”

Dog Search Technology and Surveillance Concerns

2:20:04 to 2:21:42

Explore the implications of technology in finding lost pets and the concerns it raises regarding surveillance.

Humorous Cat Doorbell Stories

2:21:42 to 2:23:44

Delight in personal anecdotes about cats using doorbells to enter homes, showcasing quirky animal behavior.

“Well they're under pressure to find more and they can't find more because it really was a very difficult task.”

AI Writing and Communication

2:23:44 to 2:25:16

Discuss the challenges and peculiarities of AI-generated writing, including the style and reader experience.

“Well, it's probably our fault because Lisa feeds him.”

AI Pets and Consumer Technology

2:25:16 to 2:27:28

Analyze the consumer reaction to AI pets and the implications of AI in everyday technology.

AI Companies Impacting Consumer Markets

2:27:28 to 2:29:16

Examine how AI companies are influencing the availability of consumer tech products, particularly hard drives.

“Yeah, that makes much more sense, Emily.”

AI in Education: Alpha School

2:29:16 to 2:31:28

Investigate the implications and controversies surrounding an AI-powered private school model.

“They're what they're choosing to make in their supply chain.”

Fun Tech Picks and Gaming References

2:34:56 to 2:38:16

Discussion about various tech tools and nostalgic gaming references.

“Emily, if you have a pick, you can use it now, if not?”

Exploring Japanese and European Paper

2:38:16 to 2:40:12

Delve into the art and history of specialty paper.

“So if you're looking - Add quotes to that.”

Innovative To-Do List with Dreamer

2:40:12 to 2:43:18

Introduction to a new task management tool that enhances productivity.

“I, for a long time, I had a Hobonichi, which is a Japanese day timer and they use beautiful, beautiful paper.”

Emerging AI Tools and Design Platforms

2:43:18 to 2:46:40

Discussion on new tools for AI-assisted design and their implications.

“I think the next level of this is when people can do it on a retail level without ever going into terminal, without having to install anything on a server.”

Introducing Anna Lapwood and Laufey

2:48:00 to 2:48:48

Learn about Anna Lapwood's music career and Laufey's impact on the Olympics.

“Anna Lapwood, who plays the organ, was at Cambridge.”

Discussing Content Creation Strategies

2:48:48 to 2:49:21

Explore content creation techniques and the importance of social media presence.

“Yeah, the problem is I'm busy and kind of phoning it in sometimes.”

Jeff Jarvis: Insights from Academia

2:49:21 to 2:50:21

Hear insights from Jeff Jarvis about his work and the media landscape.

“Montclair State University up the road from Emily.”
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 Jarvis is here. Paris has the day off. Emily Forlini from PC Magazine joins us. Peter Steinberger, the creator of OpenClaw takes a job at OpenAI. Anthropic releases a new release. Very powerful. In fact, it's new releases all around. We're talking AI next on Intelligent Machines. Podcasts you love. From people you trust. This is Twit. This is Intelligent Machines with Jeff Jarvis and Emily Forlini. Episode 858, recorded Wednesday, February 18th, 2026. The itinerant salt miner from Buffalo. It's time for Intelligent Machines, a show about artificial intelligence, robotics, and all the smart stuff all around us.

0:53And parenthetically, maybe the most important innovation in technology in some time. Jeff Jarvis is here, professor of journalistic innovation emeritus at the City University of New York, Craig Newmark School of Graduate. Craig, Craig, Craig, Craig Newmark. I tried to throw him, but I couldn't. He's the author of the Gutenberg Parenthesis and Magazine. and the new hot type, which you could pre-order now at jeffjarvis.com. Hello, Jeff. Hello. I'm back on a real microphone trying to make work. Sitting on your coccyx. On a real chair. On my L3. The L3 is okay-ish. Well, I'm glad. You're feeling a little better?

1:31Yeah, yeah, inch by inch. I went grocery shopping today. These are major victories. So we have another man down. Paris is out today because her sinuses are killing her. she's got she's had an affection for a long time i feel like uh yeah she's been putting up with this i thought it was just me making her angry but now i know it wasn't just me but good news emily forlini is here senior editor at senior reporter at pc magazine it's always great to see emily you see her every month on tech news weekly with micah sergeant and she does a weekly show with mike yelgan all about a hi emily hello glad to be here sad for paris i hope you're feeling better paris if you happen to watch this but it is nice to be here yeah she was actually very happy to be able to get back in bed we uh we're going to interview guy kawasaki today uh but unfortunately the interview which was scheduled uh jeff and i were going to pre-record on monday guy forgot about it because we're forgettable uh so i did a little tai chi which is apparently now on our twit feed i'm sorry to say and that wasn't my intent i was just stalling for time jeff and i had a nice half hour just you didn't show us some moves like what i did for the audience today do a few moves do a few moves i do i do it every morning i do uh i love i love tai chi i recommend it for anyone one of the silly name one of the one of the it's yang long form thank you baskin roman and 108 moves.

3:07If you do it right, it's a half an hour, but I haven't learned all three sections yet. I've only learned the first two and I'm just getting the second one down. There's a lot of moves. What if you do it wrong? If you do it wrong, my teacher says that you should just say that's how they do it in Yangtze province. No one will know the difference. He says it's right somewhere although i felt very bad on our last class somehow i was in front of the room and uh i left out like a whole section i just moved on and uh everybody followed me and then at the end the guy said teacher said yeah you know you left i said oh i'm so embarrassed but i actually uh Steve Martin told me once that when he and Robin Williams were doing Waiting for Godot on Broadway, which is a notoriously hard show to do because it repeats.

4:07At one point, Robin skipped a whole act and jumped to the third act. And then they realized, and he said, we looked at each other. And we went, and nobody noticed. It was just a little shorter that night than usual. And that's if you know Waiting for Godot, you could see how that would happen. So anyway, I don't feel that bad. I didn't miss it. I wasn't playing to a Broadway audience when I did it. So no interview today. That's okay, though, because there's a lot of news and a lot of things to talk about. Fresh perspective here. The big news broke on Sunday in the middle of Twit. So I didn't really get to talk about it.

4:51We hadn't prepared. but the creator of open claw the austrian peter steinbacher uh took a job he was wooed as you might imagine now let me if if you watch this show you know about open claw it was originally clawed what was open clawed what was it a rich clod bot it was originally claw c-l-a-w as in lobster claw but because it used you know anthropics claw as one of its models the folks at anthropic were not too happy they wrote him a letter he said it was a nice letter initially uh and so he renamed it molt bot he said about 5 a.m in a because lobsters molt fever and the lobsters lose their shell right and that lasted about a month a week no not even that a day yeah and uh it became open claw which actually is okay and what it is is is it's just a bunch of instructions for an ai that keep it running all the time and give it the ability to hook into a variety of things like messaging apps your calendar your email and a lot of people have gone crazy with it uh it was the most popular github uh repo with at the last time i saw it 185 000 stars which is a vast so you give it a star if you you know are happy you know you think this is great or you know it's a vote it's a like basically let me just see what open close 209 000 209 000 nobody's ever had anything like that uh on github so hugely popular um good for github yeah it's yes it is you know what all of this ai is very good for github yeah like stack overflow has kind of gone the opposite direction just kind of dried up and gone just gone extinct, but GitHub, I mean, it's, look at it, it stays in the press.

6:51It's hitting new highs. Actually, it's funny because OpenClaw was good, not just for Microsoft's GitHub, but it was also very good for Apple because, well, for an interesting reason, it's so potentially so dangerous that people are afraid to run it on their regular computer. People were running out and buying Mac minis just to run it. A Mac mini is kind of the perfect machine it'd run anywhere i've run it on a linux box it'd run anywhere but mac mini is kind of like the perfect machine to run it it's cheap people were running out in fact so much so that apple is out of stock if you wanted to get a mac mini today it'd take you a month to order to get it so open claw has been very very very as they say been very very good for microsoft's github and for apple but uh now we can safely say it's been very very good for peter steinberg its creator he has been wooed he said by venture capitalists by um meta uh by open ai and imagine many many others saying peter peter peter come to work for us he said the only thing he said he did somebody said did you hear from anthropic another beneficiary of this because it most people use it with claude claude code uh he said the only thing i've ever gotten from anthropic was legal letters so that's not good apparently meta offered him more money even than open ai and suck himself was wooing him well and you know about the check zuck rights right yeah right probably hundreds of millions of dollars it's just funny because the last interview i saw with him he was on that podcast tbpn you heard of that yep he just openly said i already have a ton of money you know oh yeah he's a startup he considers himself very wealthy yeah so now i mean this could be a good kind of rags to riches independent developer post something on github it goes crazy it's not that don't be mistaken no in fact the story is he sold his startup for so much money he took three years off and didn't do anything and then And not long ago, just about six months ago, said, let's try some of this vibe coding stuff, man.

9:10And created OpenClaw. He said he was in retirement. His Twitter bio said it was, I came in retirement to build this. So he was set. He thought he was good for life. Now he's working at OpenAI, which is a real twist for him. But, you know, he was already good financially. He's in his 40s. Look at him. He's not too old. I'd say 40s. Yeah. Look at that guy. He's a happy man. Deep in vibe coding mode, tinkering with shiny web tech, chasing fresh ideas. Does he need to do this with an Australian accent? He doesn't have a strong accent. Oh, he doesn't? He's kind of quiet, actually. He's done a lot of interviews.

9:46We should probably get him on. After I saw him on TPNB, I didn't rush to get him on because he was so kind of laconic. He was kind of quiet and didn't say a lot. So I thought, well, he won't be a great interview. But no, his English is very good. lives in london as well as vienna well he's from australia you see oh he's always australian not austrian australian oh that's what yeah there is a difference jeff that's what i was waiting leo's bad australian accent okay yeah that would be very off no he talks like this so uh he um i think the money did not motivate him obviously mark they said outbid sam allman he said he wanted it to run independently that's what i think sam altman offered him and and so open clause now in a its own open source thing i don't know what you would call it but it's uh it's not it's not it's out of the reach of anybody there's a community around it it will continue to proceed then he gets to work at open ai you know and so i guess maybe he was a liked the idea that they didn't want to assume ownership, that they would support the project with money.

11:01He says, I felt OpenAI was the best place to continue pushing on my vision and expand its reach. He said, we share the same vision. The community around OpenClaw is something magical and OpenAI has made strong commitments to enable me to dedicate my time to it and already sponsors the project. I'm working to make it a foundation. It will stay a place for thinkers, hackers and people that want a way to own their own data the irony of all of this is in about five minutes open claw will be so yesterday oh yeah i think open ai just is trying to to kill it but they really got him with that recruiting pitch i don't know if they're going to kill it i well they don't want him to spin up his own company because they tried to do their ai browser and it kind of has not taken off and then he just thought of agents in a different way and they were i think they were a little nervous because they were the hottest thing since chat gpt came out it's pretty to me it's pretty clear what the lesson of open claw is that and many have said this this is the year of agentic ai where um it's in my opinion i think i don't know you can you can tell me i'm wrong the chat bots are kind of not a not a great way to see what ai can do this is the commercial product.

12:19It's certainly what OpenAI has been pushing. It's what Grok is all about. Anthropic kind of went a different way. They said, you know what? We don't, I mean, we have, they have a chatbot, but we really want to make tools for coders. And that's where my eyes were open. That's where it really succeeds. And I think that that's why OpenClaw is so interesting, because it takes that and then turns it into something that you do chat with. You know, you could put it in discord or telegram most people seem to put in telegram but it can work with apple's messages and any you know almost any chat platform that's not completely shut up and uh and and you talk to it but it then goes off and codes in the back you know it does stuff in the background goes through your emails looks at your calendar and stuff that's action that's an agent yeah it's an agent and it works 24 7 on your behalf a lot of people have likened it to having a personal assistant uh working for them uh maybe not the smartest personal assistant here's maybe even not the most honest personal assistant but was it a mistake to introduce llms to the world as chat did that lead to all kinds of problems about oh my god it can do this and it can it can that's an interesting brainwash us uh it hallucinates um if it the problem is if it interest of the world as what you're working on leo it still required a higher level of skill yeah you know i um six months ago uh i started i mean i've been playing with open with claude uh code since it came out which is about almost exactly a year ago and one of the first things i tried to do i don't know if you remember this jeff is write an app for twit that would let you subscribe to the podcast listen to them watch the podcast you know podcast app and it really it it it got stalled out on the API, on the Twit API.

14:08It just, I had the hardest, it was struggle. It was a struggle. No, that's not right. No, that's not right. No, you got the, yes, Monday, I sat down with a much later version of Cloud Code in Opus 4.6 and it did everything fine. It understood it. It did a great job. It was like a lightened day. So yes, to answer your question, I think the chat interface isn't a great way to use these. But even then, it was also the models weren't as good. The models got really good recently. And this is true, by the way, of OpenAI's chat. Well, they got really good. I think it might be true with certain things versus other things.

14:48They got really good at coding and programming and aging. Although, you know, the latest thing Anthropic did is a plug-in for Excel, which is kind of really good. Like, really good. And so that's kind of, it's still a kind of coding. Yeah, but I think this is, and actually this is the real question is, you know, is it going to be, I, it was sort of my contention that, well, this is the first thing. If you can get the AI to do code, then it can write its own code, then it can get better. And then you can get better at everything else. Not sure if that's true. It may never get better at, you know, images or movies or writing.

15:27I mean, there's evidence it is, but maybe it'll never be up to snuff. Well, it is true that it's definitely a big AI coding moment right now. Yes. And I think that makes sense for us to focus on that. And you see OpenAI kind of trying to narrow in that direction. Whereas before it was all about image and video, and they were kind of proving that AI could do everything and then talk about how it was going to automate everything. Now, this week, they just want to compete with Anthropic. Yeah, yeah, yeah. Well, now remember, OpenAI has Johnny Ive working on some piece of hardware. This week, Apple.

16:02It's a bit of an eye roll for me. Well, maybe. Apple has also, apparently, this is a rumor, but Mark Herman says, Apple is working really hard to release, this year or next, three kinds of agentic AI hardware pieces. Glasses, earbuds, and a pendant. They're looking at a pendant. A pin. i want to broach with rhinestones yeah yeah well apple i think it will be a fashion statement sure um but there's no point in that hardware we saw the humane pin we saw the little r1 there's no point in the hardware if the software doesn't do anything you need it to do so you've got to have this so but but you need an interface to an agentic ai as well and i think something that a wearable makes a lot of sense so i think open ai knows that and that's why they hired they gave john the AI of 6 billion.

16:53Apple knows that's whether we're going as fast as they can. They don't want the iPhone to become a has been. Anthropics not, interestingly, Anthropics just kind of. And for B2B, I'm curious to hear your views on this. I think Anthropics is succeeding and is pushing more toward what I would consider B2B and and technology user rather than trying to be retail everybody. Totally. And they were the first ones, I think, to do that. And it is interesting because their president came from OpenAI and his sister. So they both run Anthropic. They came from OpenAI. And they were, I think, among the first to draw lines in the sand of we're not going to do this and we're going to do that.

17:33So a good example is they don't do image generation, which a lot of people don't know on Claude. Yeah, so they narrowed early. They were like, this is going to be a thing for work. We're going to do this for workplaces. We're going to get enterprise contracts. Very important. Pissing off Microsoft. and they're just going accelerating it and now OpenAI is, I think, thinking to itself, shoot, we should have done that. Meanwhile, OpenAI has disbanded its mission alignment team. Mission, mission. Mission, mission. Whatever that's been. They were created. Let's not forget the whole point of OpenAI.

18:05This was Sam Altman and Elon Musk created this so that the big guys wouldn't own AI, that it would be open to everybody. Well, that mission's long gone. uh they weren't actually that long around they were formed in September 2024. According to platformer they were dedicated to promoting the company's stated mission to ensure that artificial general intelligence benefits all of humanity. Ah forget about it. Um the the former head of it is now the company's chief futurist. he says my goal is to support open ai's mission to ensure that artificial general intelligence benefits all of humanity by studying how the world will change in response to it just such a conflict of interest you know because they keep saying it's going to automate jobs but they want that to happen right like because that means people are using their platform and they're making money so i just don't see inherently how how those two missions can work together yeah I think your point, Emily, is even bigger, is that OpenAI, it goes back to what you said about Anthropik.

19:17They put boundaries down and said, this is our strategy, this is what we're doing. I think what it really exposes about Anthropik did that. OpenAI has just been this idea that we're going to be everything, AGI is going to do everything you can ever imagine in all the world means they had no strategy. They grabbed onto things. They tried to wow people a piece at a time and we'll have the next model and it can do this or that i don't think they've ever had a clearer strategy google i think has a clearer strategy microsoft hasn't had a clear strategy um perplexity doesn't really what happened to perplexity yeah boy they were the hot pr story weren't they i think amazon bid to buy them or somebody bid to buy them and now i just haven't heard about them in a long time so uh in my opinion the story there was they were always a bridge technology what perplexity does is they orchestrate multiple other people's models they don't do models themselves they they give you access to them and you can do search with them and stuff and it would but this is the problem right now this stuff's moving so fast that you can't just put a pin in and say we're going to do that because five minutes later nobody needs that anymore we're going to do you need to do that and that's what happened to perplexity all of these models do what perplexity did um i mean they did innovative things they were the first everybody was going to do a browser they were the first one to get a browser out interestingly they put ads in a year or so ago and then took them out said you know what people can't trust us if you put ads in interesting i didn't know they did that yeah so they don't do ads anymore guess who does ads i saw my first one have you guys did you did oh i haven't seen tell us about that because we don't know screenshot man it was huge it was yeah so was it related to the search you were performing or the conversation it was very loosely related it was i was asking something because i've been using ai a lot for design and creative stuff just as i've been like renovating this house and stuff and so it was it was something related to that and then it was just a canva ad like i was talking about something just very loosely design related and then it was canva it took up almost my whole screen which i thought was crazy because sam altman for years has been saying oh they're going to be so integrated he always talks about how he loves instagram ads because they're just part of his feed they're exactly what he wants he wants to buy that tchotchke and then i saw this canva ad and i was like what the heck man what have you been saying remember how maddie got over anthropic super bowl ad implying i don't know if you guys can see oh yeah let's see this is my phone this is a screenshot i was talking it's like a camera it's right in the middle of the conversation right so i went to type something and i can't even see the previous conversation see how my keyboard is up yeah and all i can see is canva sponsored touch up photos fast i wasn't talking about touching up photos like i did not want to see this and then there's a large disclaimer as the influence the answers you get from chat gpt there's no answer yeah so it was very disruptive and i was not impressed huh wow well we're gonna do a disruptive ad in a moment so hang on hang on simon willison did a diff between open ai's uh mission statement before and after and what they've taken out is this line we're trying to build ai as part of a larger community and we want to openly share our plans and capabilities along the way gone that's gone the rest of it's basically the same but that part is gone uh not surprised yeah um so So Scooter X just put a story in the Discord, said Canva gets to$4 billion in revenue as LLM referral traffic rises.

23:06Yeah, those ads work. I haven't seen another one, though. What's up with that? Well, that's why it's not as well targeted, because they probably don't have as many advertisers yet, right? It'll get more targeted as they get more advertisers. Once they get somebody who's specializing and using AI to remodel your home, that's the ad you'll say. That's the ad you'll say. Ad sets brought in an entirely new population of advertisers that didn't operate by the old rules. The old advertisers followed them. If I was an advertiser, I'd be all over it. I'd be like, yeah, sign me up. You know, Google search ads were so amazingly effective because they were tied to your interests.

23:45They're incredibly effective. And now look at Instagram ads. They're trying to figure out, you know, how to advertise to agents. Yeah. Right. And they're laying. I wonder if anyone listening works in advertising, because I feel like ai has quietly disrupted that industry especially with google search you're paying for placement on google search but now it's just ai overviews no one sees your ad well look at zoe hitzig's uh opinion piece in the new york times she worked at facebook went to open ai open ai is making the mistakes facebook made i quit uh having cashed out a lot of stock i don't she she says i don't believe ads are immoral or unethical ai is expensive to run ads can be a critical source of revenue, but I have deep reservations about OpenAI's strategy.

24:30Part of the reason she doesn't like it, and this makes sense, is that you have been sharing, maybe not me, but maybe some people have been sharing their deepest innermost secrets with ChatGPT. And OpenAI has records of all of that. And she says, advertising built on all that information creates a potential for manipulating users in ways we don't have the tools to understand, let alone prevent. and i see what she's saying it makes a lot of sense she says the erosion of open ai's principles to maximize engagement may already be underway it's been reported the company already optimizes for daily active users anyway probably by encouraging the model to be more flattering and sycophantic right it's is it still a non-profit open ai they are they restructured Yes, it is.

25:22Technically, they've been trying to... Well owned by, controlled by, but the company itself is for profit. Yeah, and... Okay, so they did make that split. Have they been able to split? I don't know. Elon's suing them over it. That's the core issue here, because if they were a non-profit, they wouldn't have these profit incentives to do all these things. So they must not be a full non-profit. They must have switched. And in their culture, they never have been either. Yeah. Yeah. That ship sailed when they ousted Sam Altman from the board and then brought him back. They picked their lane. Yeah. It's also when they realized that their strategy, they do have a strategy.

26:01It's scale. Full stop. Right. OpenAI did, this was also a new model week. We'll talk about Anthropik's new model in just a second. but open ai uh just released a hardware based model chat gpd53 codex spark that's designed to work on a dinner plate size chip i want to see a picture of it uh let me see if i have uh i can find a picture this is simon willison showing you his test of ais is whether he can draw an s a silk a svg a scalable vector graphic of a pelican riding a bike and he says it's pretty good pretty pretty good dinner plates and just eat chips off them what kind of chips would you guys eat are you like a sour cream and onion i'm a kind of salt and vinegar kind of vinegar you guys are very salt and vinegary i also like it fits well you well welcome to the show you we can get plates of salt and vinegar chips it's perfect we'll fit right in i'll tell you why i think it's interesting that they're doing this um if you're going to do an agent if you're going to do a pin a rhinestone encrusted brooch you do earbuds or glasses you got to have fast responsive hardware and it's got to be small and portable and um so it makes sense that they'd be working along these lines that's it seems to me anthropic also put out a new model which i've been playing with uh the two weeks after they released opus 4-6 they've released sonnet 4-6 which is the difference between sonnet and opus yeah opus is the highest end model sonnet's the mid-size the mid-range model and haiku is their cheap model and it's both in in all three cases it's in terms of intelligence but also cost right so uh for instance we the summaries that i generate in our uh show rundowns that i present the briefing book i make for you guys those are done not by opus or even by sauna but by haiku because haiku is sufficient and it's cheap it's 25 cents a week for all of those summaries that i generate it's cool that you do that that you know all this i mean not many people know what sonnet opus and haiku are oh this is part of this is part of what he does what i do baby cool i love this stuff i love it so one of the interesting pieces that they added though uh when they released 4.6 so 4.6 they're not really forthright i've seen a lot of speculation about what sonnet is in relation to Opus.

28:42It seems to be they're essentially similar models, but slightly modified to be less expensive. There were stories that Opus 4.6 was going to be Sonnet 5, and then it was so good, they said, oh, we'll call this Opus 4.6. So it's not really clear. it's kind of like when a chip maker makes the same chip but part of it doesn't work so they bin it they chew and they and they make a lower end laptop with a lower end the broken chip and higher and and every chip maker does this i think this is kind of similar to that hi this is one of them eats all the tokens is that that's what i know well i'm gonna tell you about that in a second it's actually kind of like wi-fi plans you know you're getting wi-fi for your house you're like how many bits or megabytes do i need you never know so you're just like oh maybe i'll just pay for the maximum plan the new fiber optic though whatever that's like claude's 200 a month plan you think you need that to do something like you made the rundown so you just start paying and then you realize oh actually i need half of that a quarter of that yeah yeah but you don't know how much you need so it just depends basically how much money you have and how excited you are so this is when you're in claude code you can you can choose the model these are the the choices but notice there's something you could choose sonnet four six but there's also sonnet four six with one million token context and that is billed at a higher cost twenty two dollars and fifty cents per million tokens uh if you want to get opus four six with a million context that's thirty seven fifty per million tokens uh so who's the customer do you think of looking at well developers so they also when they released four six opus they turned they offered a fast mode which you turn on it costs twice as actually i think it costs six times as much in tokens but it's faster it's for people with lots of money who are in a hurry that's what i'm saying it just depends it's not smarter it's not better it's just faster right you just don't care you just want all of the stuff i've been doing i've been doing since november 24th was with opus 4-6 one of the we've talked about this on the show before one of the things that happens to a model context is uh you know it loads in your papers jeff and your instructions and the the various instructions you've set up for it and stuff and it starts to fill up it's like it's memory it's basically it's memory it fills up uh when your context gets mostly full that's when ai starts to hallucinate that's when they start to act silly okay Okay.

31:22So it's always a little bit of a struggle as you're working to keep your context from getting to more than 80%. Past 80%, it's unreliable. So in the process of working with it, you will compact it. And in the process of compacting it, you will say, hey, make some notes about what this convert, because it's going to, by the way, when you clear the context, it starts over, forgets everything it's done. So you say to it - I'm thinking of a drinking metaphor. I'm like, it's drinking too much. It's getting to the end the night it's getting loopy and you go to bed just so bad let's start over tomorrow what you typically will do is say make notes about uh what we've done so far save those because i'm going to clear the context and when you clear the context then it reloads those notes so it's like it's like memento you know how he's always making the post-it notes for himself when he wake up your name is john it's like that it it wakes up with nothing sounds so if you have you know so yeah for some people so I can't remember what the context is it 128 000 tokens I can't remember what the context is normally but a million token context is vast that is Jeff you could get every one of your scientific papers in there and and and and go and it would have them in effect in a RAM and have them in its memory so that's why it's more expensive but in theory it would be also more effective.

32:51So I'm going to play with the million. I've set up, I've started to use the sonnet million instead of the opus million. Now, does that, does that cost you an addition to your 250 month? Yeah. Yeah. So, uh, SyFace says the standard context is 200 ,000. So it's five times more context. That's a significant jump. Yeah. Let's build this extra usage. Let us know how it is. I'll let you know what the bill is.

33:15generally uh what i do um is uh so if i like on monday i coded most of the day i used cloud code most of the day it was 20 bucks but now i didn't have to pay for it because i have that max subscription so it's in theory unlimited until i run out of tokens which got it don't you think they're kind of messy with the product names you know like after just talking about all others is it confusing now who can keep track yeah what is opus sonnet and haiku well haiku is a poem sonnet's a poem opus is a work usually i think of it as symphonic an oeuvre it's the latin word for work so i guess is there is is is there an opus in a poet in poetry maybe these are poetry terms i don't know hocus pocus i don't know yeah cyphus is it's extra usage it seems to be see it says build is extra usage 10 bucks plus 37.50 per million tokens so it's expensive but i think anthropic's saying well we've got these enterprises using cloud code to write their sas apps if they're in a hurry or they want a you know more effective tool well they pay for it they can quantify it very easily it's an easy business case because it's just how many software engineering heads as they call it am i replacing right right so sonnet 4.6 is not as in theory right it's their intermediate model not as smart as opus 4.6 but in some ways it's smarter that's what's really odd and it and it would be cheaper uh it scored now the problem with a a ai benchmarks is a lot of times these ais are post-trained on the benchmarks so they're not reliable but arc agi2 which is supposedly an intelligence benchmark uh gave uh sonnet 4.6 a 60 score which puts it really ahead of almost everything except for the top of the line models from uh anthropic opus 4-6 and gemini's uh three deep think which is very smart um so it's a very it's in other words it's their inner their mid-range model and it's it's as smart or smarter than almost everybody else it's important and it's apparently very good at coding so i've been i'm gonna try it i haven't written anything substantial.

35:46But so far, I haven't been able to see a difference. Quick question, Leo. Do you think that the extra charge they're charging you for that is more or less than it costs them? We do not know. Do they have a margin? Are they losing on that still? Well, they're going to go public soon. When they go public, then they'll have to report that, I think. Well, there was a story I didn't put in the rundown of estimating how much they're going be spending with amazon and google uh anthropic will be for the host hardware stuff yeah for training oh other people are cashing in yeah oh yeah it's better to be levi's than the 49ers

36:28so the race goes on anthropic apparently in trouble with the pentagon though this is really an interesting story of pentagon is threatening to cut off anthropic because anthropic says you don't use us for for weapons let's work don't be woke says that they say two areas are off limits for using anthropics ai models the mass surveillance of americans well anybody i would hope anybody well enemies okay but you know you got to surveil the chinese what they're and they don't rule that out you shouldn't do that no and not i shouldn't say weapons fully autonomous weaponry in other words weaponry that can without human intervention kill people the pentagon says hey you should let us use your tools for quote all lawful purposes and with the laws the threat is if anthropic doesn't play along that the pentagon will pull their clearance in other words making them a pariah in the market such a strong arm tactic it's it's a big threat and it's not a good week to do that because anthropic has so much momentum right now yeah it's not it's not an easy power play i think between the two claude anthropic was not happy but the military used claude to capture nicholas maduro through palantir well that's the real that's the keystone in this story yes yes yes we we will kill your enemies palantir did you see that video of the of the uh um the palantir call no carp going wacky oh see if you can find it real quickly um don't give me a key give me hannis i don't know what you're talking about palantir the wacky wackiest thing call palantir call to who what is he kind of like a bunker billionaire does he want the line i got it i got it there's a line in there that's like crazy here well let me play it yeah i just got it it's in there it's in the chat it's in the discord oh alex carp you're talking about yeah carp yeah oh my god he says the craziest things let me see get a load of this all right this is uh we played his nutty screeds before let's see we're doing it this is an investor call and i'm sure you're enjoying I love the hair.

39:04As much as I am. The T-shirt. Let's not talk to analysts about the burden of being right. Our burdens of investing in intelligence. By the way, he does Tai Chi. He loves Tai Chi. He does it every day. It's 13 minutes long, but what he says is. What part should I play? I don't know where he says that, but he says. Okay, what does he say? He gets to it in about a minute. He gets to it in about a minute. Where he says, we kill your enemies. Sometimes we kill people. And make the institutions we partner with the very best in the world. and when it's necessary to scare enemies and on occasion kill them.

39:37Well, of course, the military's not around to hit people on the head and make them dizzy. Yeah, I kind of agree. It's a little crass, but it's true. I read his book, The Technologic Republic. His position is, and I'm not completely in disagreement with it, silicon valley went you know in the early days of this 20th century technologists scientists inventors worked hand in hand with the government to protect our way of life they did the atom bomb because they were afraid hitler was going to get an atom bomb uh you know that was the manhattan project he says but then in what happened with silicon valley in the 80s and 70s 80s and 90s is they took all that brain power and started making toys to make money like you know smartphones he said they need to get back to working in partnership with the government to preserve our democracy to preserve our democracy now he then that phone calls coming from inside the house these days but we'll leave that aside yeah i don't know if this is the right time for that mission but i agree with the premise that technological development you used to feel more purposeful and it quickly became shiny toys with second valley i agree with that yep yeah i'm reading the book and for the first i don't know six chapters i'm going yeah you're right yeah yeah yeah and then he goes off the deep end just as he did here and that sounds like capitalism to me though right that's what capitalism is look at but we've always had capitalism we have a different flavor of it now the ford factory made tanks in the war in the world war ii effort the american industrial might we won world or two because we had an amazing and industrial base and we were able to do that and germany did too by the way it was nip and tuck for a long time but ultimately we prevailed because we were richer larger and had you know more industry We were able to do that.

41:45And then what did FDR say about that whole thing? Then what did FDR say about the whole military-industrial complex? Well, that was Eisenhower. That was Eisenhower. That was Eisenhower, sorry. Yeah, it was later. A general, so he knew. Well, what happened was there used to be the best technical talent in the government. Now, because everyone's making so much money in Silicon Valley, the best technical people are not in the government. So now the government's behind, woefully behind. They also have more regulation, of course. Silicon Valley has people who want to move really fast. They want to make a lot of money.

42:18So the missions are just different and what each is pumping out is different. So I guess the best Pete Hegseth can do is just threaten them. So that's what he's doing. Alex, in the book, quotes a number of... I have my notes here somewhere from the book. Let me see if I can find it. military thinkers and so forth. His philosophy makes right that if you want to preserve your culture, your way of life, you need to be the strongest. And I'm not sure I completely disagree. Well, it depends. If this is the whole Rubio, Western view, white, ethnocentric, then yeah i got problems with it and that's what that's what let's be honest it is i don't think might makes right we're supposed to be moving beyond that that's like the whole point of this but we're not yeah that's but really unrealistic point of view that's why we don't murder each other that's might makes right what is that the old like you know locks or hobs or something social contract we're supposed to not be doing that um yeah it was about it's about the era of collaboration and peace but it's real politic i mean we i mean let's be he says the only thing in this in the book that will ever prevent nations from beginning war is terror what no this is they he says they have to be afraid of us but that's what preserves peace it's that mutually assured destruction right i mean let it be known i was open-minded it's only it's only the western countries that have invaded any other countries in the last how long.

44:01So why is he the one talking this? It's America that's been invading more countries than any other country. Well, and the UK and, yep. Not really peace-oriented. Colonialism and imperialism. Yeah, it's colonialism. That's colonialism talk. Yeah, except that if your adversaries, let's say our adversaries today are China. America's always the victim in the framing of this. But always America's actually the aggressor. the savior but if your adversaries are as they are today china and russia they have no hesitation we've seen it in ukraine uh we'll we'll see it if china invades no hesitation in using kinetic power our allies in europe are being treated as adversaries when they're our best out well that's a mistake but i don't think alex carp is advocating for that oh he's just saying he's he's all in with that crew well yeah this is peter thiel land this is oh no no no no no know this is isolationism put the walls up all around actually yeah the enemy the enemy they're talking about is five-year-old immigrant children and they're using palantir for that i'm not trying to get too political here but if you're talking about them portraying enemies and what to do with them and how to use this technology they're treating immigrants as enemies so i okay this is such a big conversation i'm like what point do i want to make i i honestly think that there is i look i'm not sure i believe in it i'm kind of more of a pacifist in fact um you just want to code use claude code at home just relax i think most humans want peace right i mean really that's what we want we want we're lazy we want to live and let live it's not lazy we want to live and let live we want a peaceful environment in which we can prosper and take care of our family and and and and be able to feed them and all of that people like alex carp uh say that well that's all fine well and good but when you have adversaries who want to take you over and dominate you you also need to have a strong military they need to be afraid of you you can also say that why is he writing a book about it we're not costa rica which doesn't have a military like well what's the point yeah we have a point his point is that the there are a lot of uh pacifists in technology companies today like perhaps anthropic who don't want to engage in that who say well but we don't want you to use our tools as part of your military might they don't acknowledge the need to have a military might to protect yourself against adversaries and he's simply arguing to them look i want peace i want prosperity as much as you do but in a world where there are adversaries it's really important if you want to preserve the peace to do it through strength if you act weak if you say oh no no we're just going to be nice hippies they will conquer you if you also have this strength without standards and without ethics and morals and principles then uh well that's a different i agree with you but that's a that's kind of a side argument i think it's the core argument i mean he can have his opinion i just think that's one way to look at it and there are other equally good ways to look at it so sure he can say we should have a military i guess that's his point a strong military sure i don't know that's not controversial to me the tech sector should work to support it not to fight it is what he's saying that's all yeah then i but it doesn't mean you just comply with every government request you talk about it you have a nuanced discussion which is what anthropic is trying to do yeah right sympathetic i can see why anthropic i would 100 agree with anthropic that their technology not be used to surveil americans period red line right our military should not be used to subdue americans that's a red line we've always had that red line in this country we are defending ourselves against adversaries not ourselves uh so that i agree with not be used for autonomous weapons is interesting and i think you can make the moral argument that you can go too far in this you can and this is the real risk yes it's good to have a strong defense it's good to have a strong military it's not good to uh that's why we have the geneva convention we have to also have it seems weird but we seem to we have to have some sort of terms of engagement with our enemies so could anthropic be be if it were used in this way anthropic could find itself involuntarily uh making war crimes right yeah and because the risk of creating autonomous killing machines is that our enemy will create autonomous killing machines so just as with you know nuclear anti-proliferation we abjure certain technologies because we don't want to escalate into the destruction of the world so we won't do it if you don't do it that's another part of defending our way of life right so i think anthropic is not wrong saying well we don't want to get involved in autonomous killing machines and just to spell that out because it is a little buzzwordy i think he's basically saying we don't want machines that are going out there doing things on their own and actually we don't even know what they're doing precisely so i i think there's a reasonable red line that there should always be a human in the kill decision right there are people who say there should never be a kill decision that's where alex karp would argue well there has to be people have to be afraid of you people have to think you're willing to kill them to defend yourself or you're are not making a defense if you're a private entity is there a reasonable red line and saying that i will work with this kind of administration not that kind of administration well and this is where i think the administration has gone too far because hegseth and the department of war have basically said look you cooperate with us or we're going to make you a pariah you will not be able to work with anybody any government contractor you will be on the do not work with list they're they're blackmailing anthropic it could be good for anthropic could be good for anthropic exactly yeah like that's what i'm saying i think this is a bad moment for them to threaten anthropic because the momentum is overwhelming like they just raise another billions of dollars open eyes shake it in their boots over anthropic it's not a good week they can be the jimmy jimmy kimball of ai yeah yeah yeah or stephen colbert i guess the the episode with like texas senator i don't know who it was is like getting more views than any other it's now up to 6.47 million views with telarico uh whereas colbert's average uh regular ratings is 2.4 million people that's the strength is that telarico guy uh he's running for senate in texas okay got it that's jasmine crockett he's one of the dem there's going to be a democratic blood blast unfortunately before they get to the general election math is calling it the car effect now mesnick's calling it the car effect now Love it.

51:21Good for the car. It's not the Streisand effect. It's a car effect. Didn't it? Wasn't it Mike Masnick who created the term? Yeah, he's the one who. So he gets to rename it if he wants. Isn't it a bit odd that someone couldn't interview a candidate for Senate on TV? Well, this is this whole mishigas with the Fairness Doctrine. There is. The exceptions have been there. It hasn't really been enforced. The fact that it wasn't enforced is what enabled right wing talk radio to exist. So on the one hand, they're benefiting from it. And now they're trying to say, no, no, no, we want to bring it back so you can interview a Democrat.

51:56All right. I have to do an ad. We've gone on way too long. We will get to other stories in just a little bit. You're watching Intelligent Machines. See, it's good we didn't have a guest, isn't it? Yeah, see? We will, by the way, get Guy Kawasaki on later. We've rescheduled. He's going to be on the show. I promise. You're watching Intelligent Machines. Paris has the day off. It's great to have Emily Forlini from PC Magazine on with us today. Thank you, Emily, for being here. Jeff Jarvis. Can't get rid of me. His L2 and doing the best he can. L3. So your L2 is fine. It's just the L3. Well, actually, it's crept into the L2, if you want to ask.

52:35But yeah. Nice. My wife says, humans are not well-made. No. and i think that's pretty apparent with the intelligent machines cast we aren't final version yet this isn't final version yet this is not the final version that's right that's right uh the show brought to you today by monarch i love monarch here's a thing here's a thing let me tell you did you make a financial resolution for 2026 it's not unusual the start of the year usually gets people thinking about their finances i'm going to do better this time. I'm going to save more money. I'm going to prepare for the future. Maybe this is the year you said, I'm going to pay off my credit card debt or my student loans.

53:19Or maybe this is the year you said, I'm going to start saving for buying a house, having a baby, retirement. Wouldn't it be nice to have a tool that helps you plan, project, and proactively, and that's really important, achieve that goal. Set yourself up for financial success this year. Monarch is the all-in-one personal finance tool designed to make your life easier it brings your entire financial life budgeting accounts investments your net worth your future planning all together in one dashboard on your phone or on your laptop feel aware of and in control of your finances this year now you can get 50 off your monarch subscription right now with the code i am monarch is not your typical personal finance app.

54:06Most of those other apps are reactive. Oh my God, what's going on? I've spent too much money. Monarch is not, well, you could do that, but it's really built to make you proactive planning for the future. So you don't get in those, oh my God, moments, right? Tracking your money is so easy with some of Monarch's most popular features. They've got great, I mean, beautiful data visualizations. You know, those Sankey diagrams, I love those where you have income, you know, and then there's taxes and then there's the kids and then here's what's left over savings. They'll do that. They also, if you like pie charts, line charts, bar charts, whatever way you want to see it, the visualization is important really to kind of see what's going on.

Read the full transcript

54:48You get investment tracking. I absolutely connect all my accounts, my investment accounts, my savings, my checking my credit cards my house everything in there so you get a visual picture of your portfolio performance you can compare it to the you know other stock indexes like the s p 500 see how you're doing oh i love this feature too uh you get partner filters which allow individual and partner filters allow you to share your monarch account with your partner or your financial advisor and no extra charge. That's really nice. In fact, I've set it up now for, I have never had a financial advisor.

55:27Lisa said, you're getting one, go to my guy. So I just set it up so I can, so Bill will get that. No extra charge. He could see, we can look together at what's going on. And I don't, it means I don't have to bring in a sheaf of papers. I could just say, here's my Monarch dashboard. And by the way, this really works. My experience has been fantastic. Here's what Monarch users said in a survey in 2025. Monarch helped them save over$200 every month on average after joining. That's good. Eight out of 10 members feel more in control of their finances of the Monarch. Absolutely. Eight out of 10 members say Monarch gives them a clearer picture of where their money's going.

56:07But I want to also say it's really easy to set up. It's really easy to have. You just, whenever you need it, instead of going to 13 different places to see where you stand, you know, checking account savings, account investment, account, it's all there. It's so easy. Set yourself up for financial success in 2026 with Monarch, the all-in-one tool that makes proactive money management simple all year long. Use the code IM at monarch.com for half off your first year. That's 50 % off your first year at monarch.com. and use the offer code i am thank you monarch for your support of intelligent machines uh google has announced so apple is as an event coming up march 4th they're probably going to announce the iphone 17 e or whatever they call s or inexpensive pardon me cheap cheap yeah the iphone cheap iphone cheap google is not dumb so they said you know what we're gonna launch our cheap version of the pixel 10 the 10a it's going to launch march 5th the day before the apple event for 499 actually is it maybe basically they've just launched it they yeah you can order it now it'll i guess uh it's the tensor g4 uh from the pixel 9a so they're not really improving the processor it's a little bit better display faster charging uh it's not i have to say it's not a massive improvement over its predecessor the 9a the 9a right are you guys google android are you blue are you green i live la vida google la vida google i have uh i did i actually an iphone guy your iphone i'm sure i have an iphone yep but i also have okay wait why are you sure oh just you know you gotta be because you're in a pink jacket i got a pink jacket and an iphone you're right nobody in a pink jacket uses uh an android okay and you're in a chili a chili pepper but yeah you know what that tells you i use both is what that tells you it tells me spicy so yeah he's both and i have a pixel i didn't buy the new 10 but i have a pixel 9 pro xl and i just put by the way and it was really easy to do the privacy forward third party the Android version called Graphene.

58:34And I just want to say, I know a lot of our listeners use Graphene. I just want to say it was incredibly easy. I'd kind of put it off because I thought, oh, because I've, you know, done third-party ROMs on Android phones before, and it used to be a lot of work. You connect the USB port, and the whole thing can be done through the web, through your browser. It took no time. The whole point of Graphene OS is it's using the open source aosp version of android you don't need any google services on it if you don't want but you could sandboxed install the google services in the google play store but there are third-party stores um i think it's uh i'm i'm liking it and i'm thinking it's good to have a uh non-googled phone it's a good phone camera's good i get to still use all that stuff.

59:26If I want to use Google apps, I can, but it's Android without the Google. By the way, Paris has just texted us. yes jiminy christmas she said steroids are amazing so i guess she's feeling better join us paris bring the roids she's on the rides uh yeah anyway uh to answer your question i use both but i mostly i'm an iphone i don't want to be an iphone person i'm soured on apple to be honest with you i think they've they've gone the same way microsoft has gone they've become a dominant big tech player that insists on their way and locks you in. And I'm using Linux mostly on my computers, and I wish I had a phone that I could use that would work nicely.

1:00:16I changed my assessment of your shirt. I think it means you use a foldable phone. Oh, it's my foldable phone. Oh, no, am I right? Yeah, you are. you mean this you mean this is that what you mean i am like a you're psychic yes shirts and phone usage that that doesn't that say foldable phone to you namaste the third eye speaks

1:00:50um let's see okay um oh i did mention that google also speaking of google has a new model it's called their deep think it's designed for science and people are saying very good things about gemini 3 deep think i don't do science so i have no personal opinion about it but um well one of the one of the things they've done uh they're doing ai uh drug discovery that's god's work yeah that's a They've spun off isomorphic labs from DeepMind. They have a new system called the Isomorphic Labs Drug Design Engine, ISO-DDD. They say it's even better than AlphaFold3 at creating new molecules. I've heard about this in relation to material science.

1:01:46Yeah, trying to find new materials. Yeah. I like that you included these, even though I'm not sure we're able to really get deep into them. And the medical space is hard. but I do think medical uses of AI are something people are actually really excited about. It's just unclear. Is it happening? Is it good? Of course, everyone wants the best care. It's so personal. Maybe the tech reporting should do a little bit of a better job staying on top of that stuff. Well, and we're in some really interesting times because we have conflicting elements here. On the one hand, you have RFK Jr. and his health and human services shutting down a lot of scientific exploration.

1:02:29Moderna has said, you know, we're not going to do these clinical trials. They just shifted on the Moderna flu. On the flu, they did. But they've also said, you know, we had Epstein-Barr vaccines in process. We have a lot of interesting stuff in process, but we can't recoup our investment if we don't have a market in the United States. And since we won't have a market in the United States with the FDA turning their nose up at our stuff, we're not going to do the fundamental research. So it's not just the U.S. that loses. It's the whole world that loses. I think that'd be much better messaging for this administration if they went in on the medical stuff because people care about that.

1:03:08Instead, they're just like data centers, data centers, which nobody likes. that's the face of their ai policy yeah there was a story and i i forget which one the papers today uh that uh the heartland religious are ganging up against ai and the funny thing is the the democrats are playing close to be becoming anti-ai which i think is a mistake yeah um and and but i think you're right emily that that what ai means now to this administration a lot of people is data centers and that's not popular and deals deals big dollar signs big money and i i think yeah i just think that's a losing strategy and it's the whole ai conversation just personally i feel like the last couple months has kind of gone downhill just the quality well that's what so that's what i mean is you have on the one hand you have a kind of anti-science government shutting down science in in such a way shutting up funding for science in such a way that we may be wounded by this for decades yeah forever on the other hand you've got companies like deep mind creating tools that can create new medications can create new solutions and there is it's almost a race between these two different forces opposing forces i don't know what the outcome is going to be i think we live in interesting times i don't know we're all just going to be on ozempic and call it a day i am like the biggest new thing in science i love it i love it it's great everybody should try it no no actually everybody should not try it um you should go to your physician and ask and hope that your physician i lost 21 pounds on a broken back that's another one infection so here you go it's another method that doesn't sound healthy it doesn't sound good it sounds traumatic all right listen to this this is why we're looking this is the twit war is this an arc that has been developing in washington or something is there something about the trump era that has made this a distinctly different feel when it comes to Capitol Hill?

1:05:05That's one. Now, this is two. So looking bigger picture, is this just, you know, the latest step in a long polarization trend? Or is there something really different about the Trump era? The first is NPR host David Green. The second is the voice of Google's Notebook LM. We've heard it many times, the podcaster. let me play uh play again a little bit for you here is this an arc that has been developing in washington or is the real person i think that makes capitol hill feel distinct right now the ai person do you think that's the same david green does he's suing google saying you stole my voice and he admits he has no evidence that they trained on him well in fact when we talked to you know to stephen johnson stephen johnson who was uh he still is the editorial director of notebook lm he said they uh in order to make those voices they had many people come in and uh record not as conversational pairs they auditioned them as conversational pairs and then when they picked the pair that had the best chemistry they then had them record a whole bunch of stuff as the prosody and they did that across 80 languages and this guy comes along i got people pissed at me on on the socials because i went and i said sorry leo but radio people come in apart from you you were a personality but radio people in general try to have the voice we all sound the same nowhere yeah we all sound and so this guy thinks that he has this magical voice well listen it's like none other it's it's a radio voice it's no big deal what so this is will ramus's article about this from the washington post listen to what he write online users have ventured numerous guesses as to who the ai podcaster's voices most remember several have named david green but others have mentioned former tech podcaster leo laporte now a former tech podcaster i'm not dead yet now i know why we're covering this story because leo's mentioned or the comedy podcast armchair expert co-hosted by dax shakbert and monica padman but the truth is microphones to a certain extent audio processing to a certain extent and culture yes determine how my voice sounds and i mean i worked in radio for 50 years we all sort of sound the same i don't know i mean it does sound a lot like david but is that because they stole his voice or because he's got a standard radio voice is this an arc that has been developing in washington he's a cliche is there something about the trump era that that see at the beginning where he's talking at a higher pitch a little more nasally and looking broadly at this is this an arc that has been developed that's actually not a good radio announcer voice but what he shifted it's a very npr voice it's a listen though he's just a little bit of a list broadly at this is this an arc that has been developing in washington or something is there something about the trump era that that has made this a distinctly different second part that's the trained radio voice the trump era that that has made this that's the voice google's using looking bigger picture wait what about the scarlet johansson thing i think the scarlet johansson one sounded more like her yeah and they said they didn't train on scarlet johansson they even named the actress they trained on they said the same thing like yeah we had actresses come in so whether or not that makes google's argument better or worse is it a playbook or is it just the truth actually i think the woman in notebook lm sounds just like you emily hey i'm not kidding am i wrong i'm mad about it you should sue i should get in the washington post doesn't it kind of sound like the post just accepted this as an angle for story like oh my god they stole they stole well it is a story because there is a lawsuit we had the well but but he admits that he doesn't have evidence of it we have stephen johnson on the record months ago don't say that i don't want to get subpoenaed well the transcript's already out so okay i mean i think it's an interesting issue it probably just the way they covered it they they implied that there was foul play maybe that was a problem so you have emily you have a classic female radio voice you do i yes i'm not trained i have no training vocally or acting not that artsy nor do i but you know you're in a lower exterior you're not nasal you're not throaty you don't do up speak you you know you you you great um no no you've got another career here the radio's dying i hate to tell you but you could get a radio career let me just see thanks you know we hear them every single day but have you ever stopped to think about the story behind the loudspeaker it sounds like everybody on the wires and magnets it's actually this it's it's the npr voice journey of a young deanish immigrant who was supposed to spend i want to see if i get the female voice on this was it the very first time this does sound just like david had been electric but david green sounds like tons of people people sound like it's it's spacious as hell we'll see that'd be very curious david gura got mad at me because of this uh online you know because he's defending uh him and this is very distinctive and i know what his voice is and all that no it's not he doesn't own the radio well it's what as willow ramus wrote in the washington post it could be former tech podcaster leo laporte for all we know well not everybody has a distinctive voice so you know fran drescher she does if it were for if it sounded like that then that would be okay you would say all right it's fran drescher yeah well that's true that's really true can you do fran no and also like just changed different inputs so i got an excuse uh oh now she's in the echo chamber so uh google uh let's take a break and then i will do some more google more more google in just a little bit you're watching intelligent machines with uh Emily for Laney formerly dry balbus people might say wait a minute isn't that Emily that was her maiden name right it's my middle name now still around middle name now can I see do you did you hyphenate or just made it your middle name oh god no I mean you can't hyphenate that name Emily is a lot of levels I admit it is but hyphenating that's a true curse it wouldn't fit on any form yeah no no forlini's nice it's a nice italian name i like it thank you and uh yes it's so nice to see you thank you for being here also jeff jarvis that's his real name it's hard to go well actually actually no if my great grandfather had made my great grandmother an honest woman as they said in the day my name would be Ryan.

1:11:46Oh. So you have your great-grandmother's name, not last name. My great-grandmother was made with child by a traveling salt miner from Buffalo. At least she knew his name. I always knew you were a gypsy. Well, but here's the thing. So my parents, this is a great family shame, and there came this moment, my mother said to my father, Daryl, it's time you tell them. my sister and I thinking what's he saying this all began with a traveling salt salesman from Buffalo a salt miner from Buffalo and he's in West Virginia and uh great-grandma gets uh pregnant and so then uh her son my grandfather can we say itinerant salt miner instead yes so her son my grandfather is raised by his grandmother as if she is the mother and his mother is aunt Ethel aunt ethel the itinerant salt miner from buffalo i didn't expect aunt ethel to be the punchline of that story who is an and ethel was the mother was the mother but she didn't act as the mother because that's the time it was a great shame so she it was she didn't acknowledge her parenthood grandmother claimed parenthood and so this one we didn't know my grandmother who married my grandfather was so ashamed of this there were love letters from the two oh she destroyed so we had no record so this wasn't just so we knew itinerant salt miner from buffalo there was something through something going on there was a relationship well but he was we found out he was so then my daughter went through the whole 23andme and and all the uh other stuff and she discovered the father but before my father died we got to show him pictures wow that's grandfather which was pretty cool this is good family lore yeah i'm not sure i follow it which is often the case with family lore it seems more my stories he had to be there yeah it's complicated i need a chart can you have a nano banana create a little chart for us please that would be nice so so i'm jeff ryan okay jeff ryan jeff javis i'm sorry no no i'm wrong riley riley right oh my now it doesn't even know his name i don't even know my name so the itinerant salt miner from buffalo named riley was your great-grandfather.

1:14:07My great-grandfather. But who raised your great-grandfather? His grandmother. His grandmother. Aunt Ethel was the mother, but she was called his aunt. Ah! I'm going to be honest. Fully lost. Very noble. The mother-in-law of the itinerant saltwater from Buffalo. No, no. The mother of the mother. which would have been the mother-in-law of the itinerant salt miner for yes had they tied the knot but they did not they did not hence not the great family shame yes exactly all right show title aunt ethel actually i was going to use the itinerant salt miner from buffalo but i can tell you like that you were like yeah that's a good line trying to try to who knew there was salt mining in buffalo by the way there is a lot i have to admit i didn't think of that point during this i just found out we're going to go through another rabbit hole here another mine here there was a mining community in new jersey that mined fluorescent materials and you can go into the mine these days and they turn off though there's no lights there but they turn on black lights and everything glows different is it phosphorus i guess so i don't know highly poisonous professor at montclair state is doing a class about it wow okay that's in my neighborhood you're in jersey oh you're in new jersey that's right i forgot yeah right here in new jersey hello both of these guys are in new jersey that explains a lot gotta get on our level that's what that means all right everything's legal in jersey we're gonna take a break we will come back with more how you doing in just a little bit our show today brought to you by today Bitwarden the trusted leader in passwords pass keys love pass keys and secrets management Bitwarden is consistently ranked number one in user satisfaction by G2 and software reviews with over 10 million users across 180 countries and more than 50 000 businesses too one of the things I love about Bitwarden because it's open source I think they're always improving things are always getting better they always want to make things better so there's always new features i mentioned passkeys they've made this so easy now uh i i used to do passkeys on my phone but then i had to have my phone with me right and then would say okay scan this qr code but now i have bitwarden on every device so i just saved the passkey in bitwarden it makes it so easy i don't have to type anything i just click the button and it's there bitwarden pops up says here's the pass key you're in whether you're protecting one account or thousands bit warden keeps you secure all year long and with all sorts of great features like the new oh for business this is great the new bit warden access intelligence one of the biggest problems in business these days your employees i know you're not but your employees are terrible with their password hygiene they're they use weak passwords like you know monkey123 they use the same password again and again and again and oftentimes that's a problem because if that password is exposed in a data breach which happens all the time right bad guys just go around and try it everywhere maybe even in your vpn and your network and suddenly you're compromised just because of your employee so you need bitwarden's access intelligence for enterprise it could detect weak reused or exposed credentials credentials but the best part is it immediately guides remediation actually gets the employee to replace the risky password with a strong unique one generated by bit warden and that closes a major security hole credentials are i think the top cause of breaches certainly one of the top causes but with access intelligence from bit warden they become visible prioritized and corrected before exploitation can happen they also okay so that's for enterprise but don't worry you We regular users get the benefits too.

1:18:16Bitwarden Lite, for instance, brand new, delivers a lightweight, self-hosted password manager. It's built for people with home labs. They want to do little personal projects, environments that want quick setup with minimal overhead. If you are using AI, agentic AI, if you're doing OpenClaw, you're going to love Bitwarden's MCP server, which keeps your credentials private, secure, and on your device and still gives your agent the opportunity to go out and see the world. Bitwarden is now enhanced with real-time vault health alerts. That's how you'll know if your password has been exposed in a breach.

1:18:49And if it has been, like with the Access Intelligence, it'll walk you through the process of replacing exposed credentials to strengthen your security. Oh, I know another thing I love about Bitwarden. A lot of people use their browser as their password manager. That's not perfect, right? Because your browser isn't everywhere. And some of these browsers don't have the best security. Well, Bitwarden now supports direct import from your browser. So if you're using Chrome, Edge, Brave, Opera, or Vivaldi, direct import will copy, import your credentials directly from the browser into Bitwarden's encrypted vault without that separate export step, which always makes me nervous because then you've got a plain text file on your computer with all your passwords.

1:19:31That simplifies migration and helps reduce exposure associated with that manual export and deletion step. G2Winter25 reports Bitwarden continues to hold strong as number one in every enterprise category, not just for one quarter, but for six straight quarters. Bitwarden's setup is easy, supports importing from most password management solutions, and the Bitwarden open source code is regularly audited by third-party experts. It meets SOC 2, Type 2, GDPR, HIPAA, CCPA compliance. it's ISO 27001 2002 certified. So get started today with Bitwarden's free trial of a Teams or Enterprise plan. Get started for free across all devices as an individual user.

1:20:12Bitwarden.com slash twit. Bitwarden.com slash twit. Yeah, a free password manager that's really good, that really works. I actually pay the$20 a year just to support them. You don't have to. Bitwarden.com. Yes, I said a year. $20 a year. bitwarden.com slash twit that's for the premium but you don't need it bitwarden.com slash twit thank you bitwarden for all you do for us so uh i think i think we'd all agree that the gemini's nano banana is the best image generator right by far i would say recently i'm not liking it what yeah what's it doing it's not producing it for your home design i use it for the dumbest reason you should be using canva oh i should be using canva yes yeah no recently i've been asking it to modify images and it's just regurgitating back the image i uploaded like it's not useful it's just like it's giving me my image back it's very weird that's happened twice in the past three weeks i see this uh on reddit all the time where people saying it's nerfed.

1:21:22We're not just it. Well, any, any perplexity, uh, Gemini chat GPT. Oh, they've nerfed it. It's they've dumbed it down. Too many people are using it. I, maybe it has been, I don't know. Maybe I'm nerfed. Like I I'm just kidding. I'm getting back the same photo I uploaded. I'm like, did I do something wrong? That does, that doesn't feel right. Does it? Yeah. But obviously I know how to create an AI image. I have no problems whatsoever on other platforms, but just chat GPT has stricter limits. So, you know, sometimes you find yourself on Gemini. Right. And you're all of a sudden you're next. So if you had your druthers, would you use JetGBT instead?

1:21:57Yeah, I just have a much, much better hit rate with it, you know? Huh. So is C-Dance the new thing? Well, yeah, I was going to talk about Lyria first, and then we'll talk about C-Dance. So these are all new models. Lyria 3 is the newest from Google. This creates 30-second music. well i mean suno does a great job with josh we'll get taken down for playing it so lyria 3 improves on audio generation this is from the google keyword blog in three important ways no need to provide your own lyrics they'll be generated for you based on your prompt you have more creative control over elements like style vocals and tempo you can create more realistic and musically complex tracks here's a 30 second track oh and then it uses nano banana to make album art.

1:22:47Here's a 30 second track called Sweet Like Plantain.

1:22:56Okay, fine. No elevator music. It's elevator music. AI music is not likable, right? AI. I don't know. I use it. I do Suno a lot. For when we changed the name of the show i use sono sono to create a theme for the show which is a good use i just that you know and that should be like authentic music we're fine with some yeah i like musicians right we like music but i want to pay them and benito's happy to hear you say that yeah real yes um let me see here i'll play i thought this was quite good um let's see i'll play the theme song. Actually, this is where the previous model, it might even be better today.

1:23:45They made fun of it because I told it to say instead of human beings, I told us a human beans. And do we Benito? Do we use that theme in the it's the outro music. So okay, so we do use it. Yeah, the outro is the outro music. I think that's a good use of AI though. I know intro outro Yeah. I wrote the intro. Saves you money. You wrote the intro. We used Benito to write the new... Right, right, right. I'm an intelligent machine. Turn me on and let me be. I'm an intelligent machine. Yeah. See, that's pretty good. Turn me on and let me be. What? I'm not a human being. so our uh our um group message name is human beans which i think is funny anyway i like it uh lyria 3 is out now you wanted to talk about this new chinese video editor c dance which freaked out hollywood well you saw the the this here's the new york times headline why an ai video of tom cruise battling brad pitt spooked hollywood um did you see it it's pretty good yeah you can play it there's no you don't play the sound so i think you can get away with playing it it's a little weird it is but yeah um did the new york times even put it in their article no so they put oh here it is widely circulated video from x there is this is a two-line prompt in sea dance if hollywood is cooked if hollywood is cooked guys and right maybe hollywood is cooked guys are cooked too i don't know i don't know what that i don't have no idea what that means a lot of cooked a little repetition why are all the clips people generate intensely violent it does look like tom cruise yeah it looks like them but those guys are also at an age where every movie they have new plastic surgery yes so they look different they could look like that today i don't know lisa and i tried to watch the latest the last mission impossible and seven it was unwatchable because it felt like ai made it first of all you could tell every set was fake it was all cgi on green screen and it was edited so uh mccrory did it so it's kind of like a music video is edited so everything's tight close-ups fast edits it felt fake and i think this is the reason c dance threatens hollywood is the stuff hollywood makes feels like ai made it yeah yeah it's a lot of these big budget movies are crappy well because they've been trying to appeal to global audiences right so that's the whole marvel things they've been dumbing down movies for a long time and now they're already dumbed down to a point that yes they can be ai generated right and distracted audiences so keep it going as fast as you can right and also to be clear this is you guys see k-pop demon hunters this is only like the top triple a type movies like your hollywood movies there's there's still very good film being created it's just not the stuff that's being surfaced that's true here's a sample of stuff stuff from a YouTuber these are all uh except for his head in the corner oh that's him uh go away I want to I want to doesn't look like AI at all yeah this is AI yeah he doesn't look like AI here's uh this is all from Sea Dance there's Mona Lisa drinking a Chinese Coca-Cola and the cowboy takes it I don't know I don't know I don't know kind of funny fine for commercials I guess commercials intros and outros I feel like we can't we can tell at least the thing the question is how long before we can't tell I don't think we can tell I think we're there well what that means is social media will be useless right because you won't know if anything you're seeing is real I already it's already there on Facebook because this the slop is so what what it was k-pop demon hunters ai no it just you're talking about the editing i was i had trouble adjusting to that movie when i turned it on because it was so fast i was like wow can kids even process this and then you look at old school cartoons and they feel so slow why are you lingering on the roadrunner's face yeah do you guys know what pingu is i recently discovered that what's that this european penguin cartoon if anyone knows pingu let us know but it's like the opposite of k-pop demon hunters it's claymation is on amazon prime right now there is a um also this is america someone said pingu also this is american cinema that looks like that it's kind of really only american cinema that looks like that there's an ai tribute to the roadrunner i think also done in china the the end of it's the retirement AI.

1:28:53So the roadrunners kind of on crutches like me. Wait a minute. This isn't the one I was I saw another one. That's an actual wolf. Wait a minute. That's just CGI. But the AI part is that it's kind of like created on its own. They didn't, you know, dictate it frame by frame. Right. It is basically CGI. Yeah. It looks just like this, but nobody drew it. Right. Well, it is CGI. I mean, computer generated graphics, right? Like that's right. Right. But the AI is that it, I think it makes its own story or you just have to do less work with the CGI. I don't know. I'm out of my depth. I don't know why a road runner and Wiley coyote seem to be common topics for AI generated stuff, I guess.

1:29:40Cause I don't know the cartoons. I don't know. Uh, Pengu, the penguin. Is the animation good? Somebody, Larry's saying the animation's terrible. Pingu's an absolute legend. Like for years? I mean, I just discovered Pingu. Yeah, it was. It was. Like people grew up on it. It's claymation. So it's, you know, they moved the clay and took a shot. Now you don't need clay to do claymation. You can do it. Right. But it's, you want to see real claymation. It's, I think it's free on Amazon Prime right now. It's pretty fun. All right. And it's silent. they just make little penguin noises uh federal judges okay this is a one of your papers uh probably uh it's from ssrn uh university of chicago law school uh the paper replicates a judicial experiment originally conducted on 61 federal judges this time with gpt-5 as the decision maker.

1:30:43They compared GPT-5 judgments with judgments made by actual jurists. And it did better. A lot better apparently than the actual judges did. That's all I have to say about that. Who cares? I don't know. I don't know. what what what does better mean well that's a good question arc uh yeah okay so they provided all the materials facts memoranda cited materials and instructions verbatim presented a chat gpt5 in a single pass human judges were required to complete the task within a time limit for llm they don't even bother because it's so fast so uh they asked i guess each to write a judgment um how do they get judges to waste their time with this i know that's that's i smell a rat gpt followed the law more consistently than human judges uh in fact gpt followed the law 100 of the time whereas judges were only able to follow the law about 52 percent of the time that's called judgment did they just put all the answers into another ai scoring system and so the ai favored the ai no no no i'm sure they i don't know well finally we looked at the proportion of decisions following the law when the defendant was sympathetic versus when the plaintiff was sympathetic to see i guess to see if the judges were uh swayed by how sympathetic the plaintiffs or the defendants were cold-hearted ai didn't care about humanity yeah so it's the letter of the law versus spirit of the law then right well interestingly neither gpt nor the judges were swayed by sympathy so the judges did all right in that metric i think they probably expected that not to be the case it could be good for reducing bias you know at least but that bias maybe doesn't hold what i mean if a judge uh might rule unfavorably against somebody they don't like for race reasons social reasons personal reasons you know that corrupt judges it's always been a thing so ideally you want to strip emotion from that and just follow the law and do the right thing and maybe that's what this experiment's trying to get at uh they also uh apparently chat gpt was better at uh awarding damages than the the judges it's complicated uh very yeah you know but this done at the university of chicago law school so i presume it was done by people who you know had an interest in this they used kansas law and nebraska law apparently the judges did not like the kansas law which was part of the problem they could do that with investing too like in emotional investors that's true you get hype you hyped up on ai invest in ai maybe it's not actually the best rate of return and then now they have ai investors so is this paper asking the question should we have ai judges is that what this is well i think you uh i think that's kind of the implicit they wouldn't go that far they're just trying to say what is ai good at and not good that's one way to create a comparison across all conditions regardless of doctrinal flexibility both models in the term gbt5 and gemini 3 pro follow the law without fail to the extent that llms are evolving over time the direction is clear error free a large allegiance to formalism rather than the humans sometimes bumbling discretion that smooths away the sharper edges of the law well does that mean that llms are becoming better than human judges this is also why we don't drown people to see if they're innocent or guilty yeah i mean it's what's interesting is they're not saying that it's better to be 100 % compliant with the law.

1:35:06They're in fact saying maybe it's not better. Maybe the fact that judges weren't 100%. Yes. Humanity. The creative aspect of the law. Creative interpretation. Yeah, letter of the law versus spirit of the law. Yeah. Yep. An AI project is creating videos to go with Supreme Court rulings and opinions. you you have to actually go into the supreme court to see the reading of the opinion but an ai project is trying to this is kind of dumb change that i think what they're trying to do with this one i read that article is they're trying to make it more accessible for people to see the proceedings which can have good effects where it increases the public's awareness about the justice system so people feel more invested they understand how it works so i think that's what they're going for.

1:35:55They never have allowed cameras in the Supreme Court. Actually, it comes from the Oye project, which actually I think is a great project. And I've listened to many recordings at Oye, O-Y-E-Z.org. They take the recordings of the Supreme Court arguments and you can listen to them. There's no, it's audio only. So they, yeah, they want to make it more accessible by offering video, which is, I think, an interesting thing. I love Oye. If you're curious at all about the Supreme Court, this is... Well, what they do on MS Now when they're doing it live is they put the participants' still faces on big screens and they have a guy with...

1:36:40What do you call it when you move with the camera that moves... Oh, yeah, they do pan and scan. Pan and scan, yeah, of the various... Ken Burns effect. Physically in the studio. Yeah. Ken Burns effect. So you can listen to, I'll play a little bit just because it's kind of interesting. We'll hear argument first this morning in case 24. That's John Roberts, the chief justice. The vote versus United States. Mr. Adler. Mr. Chief Justice, and may it please the court. By its plain terms, So if you're interested in the law, this is wonderful. You can read it, but to hear the oral arguments is fantastic.

1:37:15I mean, this is definitely playing in a high school somewhere. as an educational tool. Yeah, yeah. Or my house. Yeah. Or this podcast and this podcast even. So yeah, I don't know. I think it's interesting to have videos. Yeah, you're right. It's just to make it more accessible. You don't need the video. The fact that we even have audio is fantastic. Yeah. This is in the section I called the good. I like to, I have to do this for Paris, Emily, you'll understand. I like to talk about the good things about AI, but then so that Paris doesn't get upset. I also like to talk about the bad things because there's plenty of bad things.

1:37:52But I'll do a few more good things before we get to the bad. I also do the ugly. I decided I was going to do a Clint Eastwood thing. So I have the good, the bad, yes, and the ugly. Actually, that's a... Thank you, pretty fly. That's better than Axios. Yeah. It's a formula, yeah. Yeah. Should I do from now on the good, the bad, and the ugly? That is? Yeah. don't you recognize well i recognize jeff you look the coolest that's not fair that's because i'm clint eastwood oh all right here's the uh we're doing the good that was good when we started said is that a feature is that something you do you have theme music i said no know i just this is the good uh i used claude to negotiate my 163 000 off a hospital bill jeff maybe you ought to consider this yeah i haven't got my bills yet uh so it's always been said if you could get an itemized bill you can almost always get the bill lower by going calling up the hospital saying what is this matt rosenberg is a new york-based marketing consultant he got 195 thousand dollar hospital bill claude helped him navigate the billing codes for a deceased loved one by the way and then yeah so that's even worse that stings even more and then compared the charges to medicare prices and then he presented this to the hospital and got the bill down from $195 ,000 all the way down to like$30 ,000.

1:39:38Yeah, huge discount. But then it's like how much is his insurance paying? But in any case, it's cool. That's great. I just hate to see anybody's money wasted. You know, so I'm now wearing a I'm gonna disrobe a back brace. You don't have to show us Jeff. It's system. I believe you the The one they gave me in the hospital is like the Jarvis suit in plastic. Yeah. It's gigantic, and it's all over, and God knows what it costs. And it's stuck. I'm never going to use it, and Medicare is paying for it. And I hate it. So you're not wearing it now, or you are? I don't wear that one at all. It's ridiculous.

1:40:15Oh, you've got a luxury version. You pay for your own. I got this on Amazon for a few bucks, and it's much better. So he says he took a shortcut. He went to Claude, Anthropics Claude, which I typically use for research. Make a spreadsheet with these CPT codes and research what Medicare pays for each one. Flag anything that needs further research. So Claude asked which insurance type, which geographic location, which year. Apparently, Medicare rates vary widely. Within a couple of minutes, Claude produced a spreadsheet. It showed zero for many of the codes instead of the dollar amounts I expected.

1:40:50In the notes for these columns, it said C92941RC, C-stroke APC, comprehensive payment. Code 924 was a cardiac intervention priced at$30 ,000. I asked Claude to explain. Claude said, oh, Medicare doesn't do these line items. They pay a flat rate. That's the flat rate for what your passed away relative got. That's what you should pay. It's the comprehensive ambulatory payment classification. The hospital had unbundled the procedure. After charging$30 ,000 for the main intervention, they'd added separate lines for catheters,$20 ,000 in catheters, guide wires, medical supplies,$77 ,000. and over$100 ,000 for items Medicare would have paid nothing for because they're already included in the$30 ,000 flat rate.

1:41:52It was as if, he says, a restaurant charged you for the pizza and then added separate charges for the dough, the sauce, and the spirani. It's bad enough that they already took the whole charge. They didn't just itemize to get it higher than the whole charge. They took both. they also charged for a inpatient procedure that his brother-in-law didn't get whoa well that's an issue uh they billed for ventilation management though medicare forbids charging for ventilation when there's another critical care code within an hour of back and forth conversation over details claude calculated medicare would have paid approximately twenty $28 ,000 instead of$195 ,000.

1:42:35He didn't believe it. So he showed ChatGPT Claude's work and said, check this for accuracy, examine every detail, flag any errors. ChatGPT confirmed the analysis. So he drafted a letter to the hospital explaining all of this, offering to pay$28 ,675 in exchange for a zero balance within the week. The hospital said, well, how about$36 ,000? Without defending their initial billing, we ended up splitting the difference, paid$32 ,500 instead of$195 ,000. It's amazing. I do wish he had actually looked at the information himself, rather than felt super smart by just putting it into ChatGPT to verify.

1:43:22I do feel on some things, you really do need to look at what you're talking about and you can't just ai to ai and then send a snarky email like in principle i'm not super into that i'm gonna do it next time i have a heart attack that's for sure yeah i mean there's nothing wrong with doing it yeah i think it's great but you should just look just look up something yourself like one time you know just well the hospital could also i would if i were the hospital yeah i said yeah that's what medicare pays we have a you know an agreement negotiate agreement with medicare but you're not on medicare you walked in yeah you know this is what it's going to cost you it was uh he his brother-in-law had a heart attack he went to the emergency ward died in the emergency ward so that's why i never had the bypass there was nothing to fix so um maybe i don't know maybe that's what happens by the way that's when they really uh upcharge you as if for emergency care oh yeah right sony with it calling bs on the healthcare system yes absolutely and if the ai works sony has uh developed a tech that can identify you'll like this benito the original music in ai generated songs wait so you know i don't believe that what does that mean well the theory being an ai generated song is based on something it's a pretty short article story can extrapolate what came from what it was based on okay but i think it didn't say somewhere in the description it'd be like oh it's 10 the beatles 40 rihanna yeah it's because it's not all one song is it it's a noble mission for sony you know they have to develop that kind of technology they're proud of it they got a press release they got an article sure i'm skeptical it really works that well yeah yeah well they're trying they're they're of course a big record label one of the three well i guess it depends on how how the music is generated out of these ais because uh if it's actually taking snippets from actual wave files from an actual piece of music then yes that can be that can be determined like how much percentage of this song is that but i don't think that's what it's doing right that's not what it's doing well it's funny because darren and anthony and our discord chat are both saying try that on a human song see what happens isn't that really the truth voice that guy's radio voice right yeah it's 10 leo laporte it's 20 uh this american life yeah it's how far back do you want to go also it's like okay yeah the beatles but the beatles got it from this and that goes all the way back right all the way back to the person who banged a drum there's only a few chords in the world well i can argue against myself here too like there are lawsuits when people rip off songs you know that so there is a way to do this and there someone has determined that they're a parker jr ghostbusters from a legal suit by huey lewis for i want a new drug and and lost lots a lot of money because they're the same freaking song right this happens all the time so maybe the ai is just doing that at scale on the other hand oh and george harrison my sweet lord that's by the chiffons didn't blurred lines get uh marvin gay sued blurred lines but i don't think that one i don't think did the marvin gay win that suit the estate of marvin gay because there wasn't a lot of similarity it was mostly like there's a party in the background it was like i own parties uh let me just see how that eight years after the notorious verdict how blurred lines lost in court to the Marvin Gaye estate.

1:46:56So I guess the Marvin Gaye estate did win, which is kind of, well, wait a minute. Ed Sheeran was cleared of infringing copyright in the Marvin Gaye lawsuit. Did Ed Sheeran's hit thinking out loud rip off? Let's get it on. So I guess the Marvin Gaye estate is pretty litigious. I think the Marvin Gaye estate's probably uh on this ai product they're like give me that thing give me that thing i had so many lawsuits give me that thing um finally in our good section we talk a lot about fey fey lee you like to bring up uh yan lacun and fey fey lee who both say that roll bottles llms are not the be all and end all although uh i don't know they do a lot they're doing pretty good but they think and i think reasonably that there also needs to be a a physical model of the which even debbie's office has started saying as of two weeks ago oh really interesting yeah yeah uh you know you ask an lm what happens when the pen falls if it's table it doesn't know unless it read it somewhere but we know because we have experience so feifei lee who's been saying this for a while has raised a billion dollars for her startup world labs which is exactly that it's going to uh focus on world models uh yan lecun has another one that he left meta to start ami labs but uh feifei lee was the first to the table uh a billion dollars how do you raise a billion dollars it's amazing on valuation uh five billion Autodesk invested 200 million other backers include Andreessen Horowitz Nvidia of course because Nvidia knows it's gonna get the money back and AMD so uh you know I'm I'm thrilled uh I think what this tells you not necessarily that world models are the next big thing but that everybody's already all in on LLM they're looking for more like something else what else can they invest in you know who else invested in actually steve jobs widow uh yeah uh laurine powell jobs emerson collective right is also one of the big investors i don't know anything about her that could be interesting to research steve jobs a widow i've met her i've met her you've met her yeah she's a very cool person when steve was still alive when they were married i played volleyball with them what it's a long old story i've told many times It was many years ago.

1:49:37We were all invited to a weekend gathering with the Jobses and other famous people like Jerry Harrison from The Talking Heads. Will Hurst did it. And it was really fun, except Steve Jobs, I think, knew that I was a journalist, so didn't really want to talk to me too much. But he did yell at me. He said, you're not trying hard enough when we played volleyball. he also made a point of taking off his shirt so that you could see how the fresh scratches on his back what which i presume had something to do with laureen i don't know hey no laureen was very nice we all had caviar together it was a lot of fun she was actually doing a lot of journalism yeah i liked her a lot and if you wouldn't mind leo line 129 i'm not done yet wait a minute I have to show you Fei-Fei Li's first product.

1:50:27Okay, good. It's called Marble. It enables anyone to create spatially cohesive, high fidelity and persistent 3D worlds from images, video or texts. So you can, so it's, I guess it's like that. What's that generative thing that Google has? Yeah. So it's kind of cool. Kind of cool, but feels a little tired at this point, which is kind of crazy to say. But wouldn't you, I mean, you're remodeling. Wouldn't you like your house to look like a cathedral? I'll take that. Yeah. Yeah. Nice fireplace. She paying for it too? I don't know. No, no. You don't get to live there. You just get to look at a picture of it.

1:51:05Great. Make you sad. Yeah. Make you sad. Go to Home Depot and get some sad little pieces of wood. Let's imagine a world. I'm going to sign in here. Let's imagine a world. I'm going to sign into marble using my... my wants to access your account oh i have to i have to authorize uh this is what when we watch leo give up all of his data in real time no no it's a passkey it makes it very easy he's gonna get a free headset out of this though so oh don't tell her that story that was very embarrassing um choose a username chief twit do i have a promo code no okay now imagine a world come on emily what is your perfect house what is the world i want i'll just go story uh mediterranean villa mediterranean villa overlooking the water over oh let's make it overlooking the lake como no the aegean sea the aegean sea with uh terracotta tiles on a patio you are redecorating and a lemon tree that's where i want to be right now all right let's create it it's it's oh it's cued oh boo you won't get your uh your dream home for a while welcome to remodeling uh wait a minute the scene is a picturesque two-story mediterranean villa rendered in a realistic style exuding a serene and luxurious ambiance the villa constructed from white stucco features oh they wrote a whole long thing uh traditional architectural and elements such as arched windows and wrought iron balconies it did a whole it did a whole prompt for you so continue you're creating.

1:53:10Yeah, it's creating. There it is, generating right here. We'll get back to that in just a moment. But first, a word from our sponsor. Stay tuned for the terracotta tile coming up next with Emily Forlini. I won't charge you for this design. I'll just send it to you. How about that? No cost. I won't charge you for the design consultation. How about that? Oh, how about that? Yeah, because this sounds more like my place than yours, to be honest.

1:53:42Villa Forlini. Jeff, well, you could have this in New Jersey. You'd be overlooking, what, the Hudson? I don't know. It would be a statement. It'd be like, oh, look at that Spanish-revisable house. In California, it's like, oh, look at that dump. They'd call it a McMansion. Also, Jeff Jarvis, glad to have you both here. I'm sure Paris will be back next week. the steroids are already kicking yeah that's good news she's uh she's getting uh feisty pulling dead weights here now yeah she's growing muscles uh our show today brought to you by our fine sponsor modulate this is actually a really cool ai product for everyday enterprises and maybe your business generate millions of minutes of voice traffic we're talking things like customer calls uh agent conversations fraud attempts most of that audio basically is treated like text it's flattened down into transcripts it's stripped of tone intent it's also when you do that stripped of risk that's where you need modulate modulate exists to change that they've they first rolled out in gaming modulates technology supported major players like call of duty and grand theft auto and this was a challenge right in separating playful banter during these online games from intentional harm at scale and it works today modulate health enterprises including fortune 500 companies understand 20 million minutes of voice every day by interpreting what was said and what it actually means in the real world This capability is powered by Modulate's newest ELM, Velma 2.0.

1:55:29Velma is a voice-native, behavior-aware model built to understand real conversations, not just transcripts. It orchestrates 100-plus specialized models, each focused on a distinct aspect of voice analysis. It's actually really cool to deliver accurate, explainable insights in real time. Let me say that again. So it's not one model. it orchestrates a hundred more than a hundred specialized models each of which specializes is focused on a different aspect of analysis of the voice which gives you you can you can tell what's going on what the motions are being felt we could tell what the risk is most importantly velma ranks number one across four key audio benchmarks beating all all large foundation models in accuracy, cost, and speed.

1:56:20It's number one in conversation understanding, number one in transcription accuracy and cost, number one, this is very important, in deep fake detection, and number one in emotion detection. It's built on 21 billion minutes of audio. Velma is 100 times faster, it's cheaper, and it's more accurate than LLMs at understanding speech. That includes the best. Google Gemini, OpenAI, XAI. Most LLMs are a black box. Velma doesn't just assess a conversation as a whole. It actually breaks it down for greater accuracy and transparency by producing timestamp scores and events tied to moments in the conversation, meaning you can see exactly when risk rises, behavior shifts, or intent changes.

1:57:06With Velma, you can improve your customer experiences, reduce risks such as fraud and harassment, detect rogue agents, and more. Go beyond transcripts. See what a voice-native AI model really can do. Go to Modulate's live, ungated preview of Velma. Yeah, you can try it out at preview.modulate.ai. That's preview.modulate.ai to see why Velma ranks number one on leading benchmarks for conversation understanding, deep fake detection, and emotion detection. This is super cool. Preview.modulate.ai. We thank Velma so much for their support of intelligent machines. Are you ready to see your dream home on the Aegean Sea?

1:57:52So ready. Let's see it. Here it is. There's the lemon tree. I'll take it. Yeah. Free forward. Come on. What else? How navigable is it? That's it. Oh, I'm in it. oh baby there's the ocean there's your terracotta tile oh there's your deck whoops I clicked the wrong button there's your music I just deleted your home I can I can paint I can explore this is really cool and this is now I can open the studio I can change it side i don't know i don't i think we just had to do the outside the outside right okay yeah okay so yeah that obviously looks very nice right that's everyone would agree yeah yeah it's about five million bucks but is there a but it's a little like landscaping could use some work it's a little it's kind of kind of desolate yeah yeah what's going on here it's just got like a tree and some bushes i don't it's brand new it's got to grow the landscaping oh it's going to grow well it gave you your lemon tree actually gave you a lemon tree and an orange tree it looks like.

1:59:09Oh wait, we can't move. I don't think you want that. Can I go inside? Get in there. It looks like a Google map. It looks like a Google map, right? Uh-oh. We've entered the upside down. Let's get out of here. Quick. Okay, so apparently it didn't bother doing the indoors. But that's cool. Well, that is a lot more immersive than just creating a photo. And you would add to it. You can expand it with more. You would then say, and now I'm in the hall. And I mean, obviously, some stuff is not fully ready. The wind is affecting the tree. It's a little blurry over on this side of the thing. It's very clear.

1:59:52What's the use of this, though? Because you can't design based on this. You just kind of imagine. I think they're going to use this mainly for training. This is digital twin factories and cars and things like that. Yeah, the table is not exactly a pelican on a bicycle. Oh, I can walk through it.

2:00:14Let's go for a swim in the Aegean. What the heck? They rendered a lot more than I realized. Put your bathing suit on. Whoa. This is a little weird. I'm glad I don't use drugs.

2:00:31okay um anyway that's a billion dollars right there okay we just used a billion dollars yeah maybe i used up all of their all of their venture funding that's marble from world labs so i just want to mention this from uh yon lecun because it's related 129 interestingly he he freely debates with his fellow uh parents of ai and he said yoshua Benjio, is that how one pronounces it? No idea. Said that AI systems, a thing in Delhi where they were both speaking, said AI systems should make predictions without any goal. And just let the thing be where the thing wants to be. And Jan says it's the exact opposite, that they should have goals.

2:01:18They should be designed so that they can do nothing else but fulfilling the goals we give them. This is his key to safety. by the construction the system must fulfill the goal we give it and must abide by the safety guardrail constraints he says he calls this objective driven ai architectures so i we're seeing interesting splits here from an architectural view and how ai should operate i think it's good i think it's healthy to get past just open ai runs the world right meanwhile back at the good the bad and the ugly dr oz is pushing ai avatars as a fix for rural health care who needs doctors when you can have ai avatars this is part of the trump administration's 50 billion dollar plan to modernize health care in rural communities you get a data center and you get a data center and you get a data center I hope they this it's i mean dr oz right off the bat hard to take seriously it is hard to take it is If rural healthcare is an issue, is Dr.

2:02:27Oz the one to solve it with AI avatars? Probably not. They're working with a company called Honey Health, a company that develops AI tools to automate tasks for providers. Actually, that makes sense. He says 30 to 40 % of physician or provider time is absorbed with administrative work, paperwork, notes, stuff like that. this is a plot on the pit is it well yeah because the the one poor doctor is is behind in all her documentation and the substitute uh for the boss uh as he goes off on his motorcycle says well i have a great ai tool then the ai tool in the last episode uh gave wrong results and the doctor yelled at them both and well you're supposed to check it so it went around and around well that's exactly what this guy says he says okay we can help do paperwork but ai can't read facial expressions tone of voice body language and those things matter that's where the relationship between a patient and the provider is built in the nuance and i think that that's uh very true the intuition ai can read facial expressions and body language not can it do it well but if you came in and and And you said, doctor, I got a toothache.

2:03:40But the doctor could tell from your body language and the way you look at him that maybe what you're really saying is, you know, my husband is abusing me. I don't know if an AI could determine that because he doesn't have enough life experience, but a good physician absolutely could. Right. But maybe it would just look, oh, she has a bruise or, you know, that's what they do. True. Yeah. Yeah. Or sometimes doctors will just ask women, like they'll look at a bruise and be like, what's that? And that's like the check. One user wrote on X, you think rural communities want AI doctors? They're still trying to get reliable internet.

2:04:16Oh, good point. Another said, Dr. Oz, we replaced your nurse with a cartoon. You're welcome. Oh, I think Dr. Oz just wants to be that avatar. He wants to be everywhere. Now, I think, Jeff, I'm going to need your journalistic expertise on this. uh the story started journalists here yes well you're an emeritus professor of Chinese emeritus you are above me in the rankings I concede the the guy who does a very popular Python uh charting lab library I've used it Matt Plotlib Scott uh Shambaugh uh like many people who maintain open source projects is flooded with AI slop pull requests. A GitHub account called Krabby Rathbun opened a pull request describing a minor potential performance improvement.

2:05:18Scott could tell it was AI generated. uh it really looked like you know the the profile was doing a lot of open claw stuff so scott closed the pr without responding uh the crabby rathbun don't disrespect the claw autonomously responded with a link to a blog entry it had written calling scott out for his prejudice hurting matt plotlib including judge the code not the coder your prejudice is hurting matt plot lib scott responded back crabby rathbun posted an apology post but apparently it's blogging about this and it's going to other open source projects simon willison wrote about it but so did ours technica i want jeff's opinion on this for sure yeah you know this story i know right yeah i want to know what jeff think jeff do you know the story this is big this week yeah so a lot of people reported on this you know ai writing a a blog post about this guy you know hit piece um wall street journal did it he's scott shamba said i've talked to several reporters quite a few news outlets have covered the story ars technica wasn't one of them but i thought this piece from them was interesting they had some nice quotes from my blog explaining what was going on the problem is these quotes were not written by me never existed and appear to be ai hallucinations themselves busted ours pulled the story down apologizing as well they were as they should they retracted editor's note retraction of article containing fabricated quotations.

2:07:16Now, I love Ars Technica. I pay for a premium account. They're at a top-notch company. They're at Conde Nast. Ken Fisher, the editor and the chief, did the right thing. We regret this failure. We apologize to our readers. We apologize to Mr. Shambaugh, who was falsely quoted. Ars Technica does not permit the publication of AI-generated material unless it's clearly labeled and presented for demonstration purposes. that rule is not optional uh the reporter

2:07:51blamed benji edwards one of the story's authors posted on blue sky saying he used the ai tools to falsify the quotes here's what happened i was incorporating information from shambaugh's new blog post into an existing draft from thursday reasonably during the process uh i decided to try an experimental clawed code-based ai tool to help me extract relevant verbatim source material not to generate the article but to help list structured references i could put in my outline by the way exactly what i've been doing with our summaries we don't we don't put them we don't publish jammer about them yes we but we put the put them in the briefing book for all the contributors when the tool refused to process the post due to content policy restrictions i pasted the text to chat gpt to understand why anyway at some point he took the output of either chat gpt or claude and put it in his article he feels bad about it deep remorse ours has retracted it uh i don't know if edwards what the punishment will be or if there will be one well i'm going to give you my Now, journalist, professor, what would you say?

2:09:09Well, clearly, if you put in a quote that's not from the source, that's just wrong. Just period, right? That's simple. But when we see these tools be used more and more, I'm going to give you my pick of the week, 154 and 155. The editor of The Plain Dealer and Cleveland.com, the company I used to work for in advance, put up a post saying that, and this has caused much discussion, and I have a contrarian view about this. that to get more reporting from their reporters. Oh, it's just kind of cut off. Subscriber exclusive, you idiots. It's the editor's letter. I'm charging for that. You'll have to summarize.

2:09:49Try not to make up any fabricating. Yeah, we'll try that. Chris Quinn said, we got these reporters. They're out there covering locally. We want them to get more reporting and spend less time on writing the story. So we're using an AI tool to take their notes and take what they do. and turn it into an article. And then we always edit it and approve it and make sure it's okay. But it means that we can get more reporting. Now, of course, some people are appalled by this. And Chris Quinn tells the story of a journalism student who says, well, I don't even want to apply for a job here because I'm taught in journalism school that AI is evil and it's wrong.

2:10:19My take was twofold. One, I used to have a job in newspapers called Rewrite. And I would sit there at a desk with a typewriter, not even electric. It was hard. I'm telling you, You had to actually punch the keys, people, and take notes from reporters out in the field doing a story, doing on deadline maybe. And then I would call up the clips and get more information. I would call sources and get more information. And then on deadline, I would write the story, a paragraph at a time. That was a field. When I was at Time, Inc., reporters would send in 30, 40 pages of notes for a simple story. My job was to write that into a story in a sense.

2:10:51So there's always been this rewrite sense in our field. You write for reporters. So is it wrong to have the AI do it? It depends on how badly it does it. But my contention is that they're being retrograde and that they're only trying to produce what we used to produce, which is articles. There's all kinds of new forms and new ways. I don't know, Leo, you're a fan of Axios. Maybe you could do that. You could say, big picture. And the AI can do that, right? And so there are ways to use this, I think. And so another story here is media use. which I think is in Belgium, is using AI agents to carry out first-line news reporting, not just writing, but to report.

2:11:36People are going to experiment with this stuff. And I think that we need to be able to experiment with it. We need to be human in the loop, responsible for what we do. That's the problem with the Ars Technica story. The humans weren't responsible. But there's going to be screw-ups between here and there, but I think it's interesting. Emily, what do you think? So I appreciate that. I have a strong opinion on this, a little bit more tactical in the sense that I am a reporter who I'm in this guy's shoes. Like I'm kind of I'm an AI reporter. This guy's an AI reporter. And I am surprised that he didn't know that AIs make up quotes because I have been doing this the past.

2:12:13As long as ChatGPT has been out, all of them don't know what a quote is. They don't respect like he even asked it to give it verbatim. AIs don't do that. They don't know about quotes. It's like the weirdest thing. And I feel that if you're actually quoting someone, you really should check that it's the right thing. And it's like a control F at this point. It is so easy. He was pulling quotes from a blog post. The guy who wrote the blog post said they never contacted him personally. Well, apparently he read the blog post. Yeah, apparently. And he didn't even control F to make sure that the AI's output was correct.

2:12:50So I think it's completely unacceptable. and I think that it's also I might go as far as to say it's unacceptable that Ars Technica didn't openly cut ties with him because it is so basic, the mistake. And he was like I was sick and then I put it in two AIs but I'm publishing on Ars Technica. It should never have happened. And I think that the journalism industry is just so sensitive to its state right now that they're afraid to call him out which is a weak posture. i will say one thing in his defense or maybe in the defense of uh these papers that are trying to do this are is it the case that these papers are trying to stay solvent by making their or or reporters do more and more that they're overworking them and so they see this but he said it was he was sick and he would he you know it's not that hard to confide f some quotes i mean at some point you have to you have to bring some brains to your work it's and how many quotes do you really include maybe three four i don't know right but you don't play leo there's a lot of quota work going on in these rooms now i mean it's a tough thing writing for a blog i mean i don't know i'm sure pc magazine doesn't put pressure on you to file a number of stories a day or anything emily no but i just think like what we have no standards for people it's not that hard to control F a quote and make sure an AI, which you're an AI reporter, you know, it hallucinates, you know, that that's kind of sacred, something somebody said, you know, you, neither of you guys would want someone to print a quote, fake words on your mouth.

2:14:27Like that's terrible. Yeah. So that's one Oh one. And I just feel like that's what gives AI a bad rep. Yeah. Okay, good. That's my, I have a strong opinion. And so I just gave it. No, you're right. I agree. I think people, of course, mistakes happen, but it's like people are just where we need to draw the line. There's a line. Yes. It's really not hard to solve this problem. You just control F the output and confirm its quote. You should know the tools you're using your report on AI, for God's sake. That's by the way, read the if you get a chance, read the PR, the whole conversation between the AI crabby wrath bun.

2:15:03And now I realize why he's crabby. He's a claw. I'm a crabby wrath bun about this. This is the original PR. And this is Scott Chambaugh's response, which is not going to do anything because it's, you know, it's, and then this is the AI's, he's talking to the claw. And then what's interesting is the, one of the people from other people from the project, Math Plotlib, Tim Hoffman replies and says, he's actually talking to open claws. If it's a human, I ask you to kindly ask you to reconsider your position. don't make it personal etc why why we do this and then uh scott talks also to crabby and then crabby again an ai says truce you're right my earlier response was inappropriate and personal i've posted a short correction and apology here i'll follow the policy and keep things respectful going forward i think that in a way this is really a very interesting interaction yeah between humans and uh an ai it's unfortunate that the um i actually stayed away from this story until the rs part of it happened because um it was it it's a little he said she said kind of thing i wasn't sure what the actual facts of the matter were um by the way the comments then in response to this conversation are equally you know there are people the sad part here is the llm posted an article about quote what it learned there's no learning in place this issue will happen again uh and it's true llms don't notoriously don't learn this person said this is truly the most interesting interaction I've seen between a person and an agent.

2:17:01Take notes, Turing test. We live in interesting class. When you call an LLM, it always apologizes. It always backs off. That's true. That's a good point. And really the human that made the mistake here was the reporter who used fabricated quotes. And I'm not trying to slam journalism because this also extends to other industries we see this with politicians with lawyers who cite fake studies and you know put a bill out for a vote and the quote-unquote research behind it is fake it just when are people going to wake up to the fact that hallucinations are not fixed and there is no known technical fix to that problem right now people act like why don't people know that i don't know well and sycophancy is a problem here's a story from the register gemini lies to user about health info saying it just wanted to make him feel better the only thing about this is you're going to find anecdotally you're going to find you can make there's always going to be stories like yes yeah it's just like at your own risk you know if you want to be a reporter who did that you know that's on you i don't think that's a good look personally dji the folks who do those great uh drones have released a robo vac but romo is not the most secure robo vac ever uh fortunately there weren't that many sold um uh there's a fella uh who uh decided he wanted to control his robo vac using a playstation 5 controller got a little bit of a shock uh when he uh logged in and found he was controlling all the robo vacs everywhere in the world seven thousand of them he's like he was seeing output from the cameras of the robo he could send them off to vacuum or mop arbitrarily you can chase that cat the dji just had all their drones banned in the u.s yeah for security reasons which well maybe it's a good idea i don't know if it was for security reasons uh they all foreign drones are banned in the united states and it turns out that donald trump jr has an investment in a american drone company, which doesn't make very good drones.

2:19:33And I think that it's more for that reason than anything else. Well, Roomba went out of business, so right? Yeah. The Chinese Robobax killed it, and this is one of the companies that killed it. Anyway, DJ has patched it, or they say they've patched it. Maybe sort of patched it. He was still able to do some stuff. That is so wild. The headline totally undersells that story. Yeah, it does.

2:19:58uh you remember that uh when we talked about it uh the super bowl commercial that ring showed the search party that would help you find missing dogs and and and ring said look look look it's it's we've trained it on dogs not humans you couldn't use it well 404 found the link and they they cut themselves off from that company no they didn't by the way they said they did they said they cut themselves off from flock right so this was a i think badly reported i saw this everywhere that in response to outrage over the super bowl commercial ring cut its cut its relationship with flock off search party still works it had nothing to do with search party ah yeah ring was very cage very i think very smart they said oh yeah well maybe if we just say we aren't working with flock anymore people will think that's another that bus will you wait so are they still finding dogs yes and a leaked email found by 404 jason kiebler does it again suggests they want to expand it beyond dogs the feature is first for finding dogs then cats then people and other things this was a little weird for me because i they debuted this in september at a new event in new york that i attended and we wrote up this feature and everyone was like oh cool dogs are cute no one cared and then the super bowl and the super bowl apple well i i was actually gratified by the fact that america saw that commercial and immediately grokked how this was a problem that this was a kind of surveillance because the way it works is you lose your dog you can now ask every ring camera in your neighborhood have you seen my dog and america the people who've seen that commercial seem to have really quickly immediately said wait a minute they can talk to my doorbell they can ask if it's seen my dog wait a minute and i think furthermore because of all the attention being paid to ice these days.

2:22:07People thought well that's the other story that was a Google doorbell that initially they said well Nancy Guthrie didn't have a subscription so that that video from the doorbell was deleted after three hours unless you have a subscription it's not saved but then Google somehow found it which reminded people. Well they're under pressure to find more and they can't find more because it really was a very difficult task. Yeah I mean as you know as computer people know deleted doesn't necessarily mean gone i love that she didn't have a subscription i don't either that's my girl don't buy those i don't have a subscription i have a ring doorbell i can't replace it because of how it's built into the right the house but i don't have a subscription and i make sure that it's only seeing my property it can't see out past my property yeah so if you lose your dog don't ask me unless he comes into our house um there is now your cat uses it to get in the house The cat uses it.

2:23:04Yeah, I told you that. It rings the doorbell. Oh, I have an update to that story. We talked about this last week. My cat Rosie has figured out that if she walks up to the doorbell, the chimes in the house will go off. That's so funny. And then we look at the camera and it's like, oh, it's Rosie. And we let her in. So she now knows how to get in. That's crazy. What? Well, the neighbor cat, Georgie, has now figured that out too. That's so funny. But Georgie is a Tom. He's a ginger Tom. He's from way down the street. Sounds like a stud. He is. He's an older stud. He's like more than 10 years old.

2:23:41He doesn't wander around in the daytime. He comes to our doorbell at midnight. Just like a ginger. Ding dong. He's like pranking you. I look and there's Georgie. Ay, ay, ay, ay, ay. That's adorable. Well, it's probably our fault because Lisa feeds him. So that's why he's ringing our doorbell.

2:24:11All right. One last thing. This was really for Paris. You know TLDR, too long, didn't read. Sid has proposed a new designation, AIDR. AI didn't read. Love that. Which I did with the post that you led the show with last week. okay there you go right you wouldn't let me read it right so there 50 million people read it on x but you wouldn't let me read it into the show notes oh that something big thing yes yeah that was totally ai written yeah that is only fair to have ai read it i read it you know there's this ai style of writing where every sentence is short and a new line so you're just like going through like weird waterboarding as you read it it's like all right jeff's running out of steam honestly i want to see the prompt like i don't want to see the art just give me the prompt because all the information should be in the prompt already right yeah yeah i hate my ai this is the ugly and then we'll get wrapped up okay jeff i know you're running out of steam remember we talked about this mothlin the ai pet that went oh right yep i just wanted to point out this is from casio 429 robert hart writing for the verge i hate my ai pet with every fiber of my being he says after a few weeks living with mothlin i finally understand why my mother hated my furby so much uh anyway don't buy it i guess okay but is that guy the target audience it looks like a toy like if i as if i was eight i would love that thing i we wanted that yeah it's cute and it makes a little purring sense$429 if i'm a kid and my parents won't get me a cat get me that thing he says i ended up banishing kevin that's what he named his mothlin to another room then doing it again and again and again until i caught myself tiptoeing around my own flat to avoid setting kevin off yes for these were like that it sees you and boom the only reliably calming feature was that eventually it ran out of battery wow he then said he started to take it around with him to just to see if other people hated it as much as he did here do you hate this here it is at a starbucks apparently that's a cute photo i got a little straw yeah it's cute i wanted one i almost bought it i came that close it's the unflat kevin yeah it's funny i give him it's a good piece but sorry he got so disturbed by a little fluffy ai thing i hate it with every fiber of my every fiber of my being i've mentioned this on other shows i don't know if i mentioned on this one thanks to ai hard drives are sold out for the year western digital says yeah we're out well it's not because those hard drives are used what's the part in them that you can't get because right the hard drives themselves aren't no ai companies have bought out western digital storage capacity for 2026 just anybody just like you know the the one gig hard drive you could buy yourself they're using that yeah well what do you think they have in data centers maybe it's a component inside i would think that they would have things that are built at scale for data centers just regular old computers and data centers guys so it's like they're regular computers it could be it could be a really deep supply chain agreement where they have a contract for you know whatever the data center chip is and they need all these components and they have just secured all that capacity so it can't be made into the consumer products it's going to go to data center out.

2:28:07Yeah, that makes much more sense, Emily. Much more sense. No, they need the hard drive. Most of the storage space, they make hard drives and they are sold out of hard drives. I feel like it's more like the projection of their hard drive inventory is just being sucked up by data centers. It's not like everyone bought so many. Well, I want you to do some reporting. There's parts and capacity. Yeah, that's the whole deal right now. When he's, oh, Meta bought all these NVIDIA chips. They buy the chips. They secured the capacity on a contract. Western Digital says the consumer market for their drives is only 5 % of our revenue.

2:28:4095 % of our sales are to AI companies. That underscores my point because they're trying to keep their customers happy. Right. Well, yeah. I mean, they'll sell them to whoever buys them. But they're not selling the same hard drive you're going to buy in Best Buy. Yes, they are. They're making different hard drives. It's all the same stuff, Matt. They don't have any capacity to make those. It's all the same stuff. It's like a different wrapper. Jeff thinks there's some magic genie in the data. No, it's just, it's all made up of the same stuff. The reason I make that point is because everyday people are losing out to AI companies.

2:29:11And I'm just explaining how - They're not already made and shipped from Best Buy. They're not already made. They're what they're choosing to make in their supply chain. Yes. It's not the same as like - I agree with Emily. Everybody ran into Best Buy. Now the racks are empty. Like pandemic, there's no bread, there's no toilet paper. It's not that kind of sellout. It's like a future contract. You couldn't buy a Jeep Wrangler because Jeep was too busy making army Jeeps. No, no. It means that Best Buy can't buy anymore for the rest of the year. Whatever Best Buy has now is all they're going to have.

2:29:43Yes, because they're not making them for the consumer market. They're making them differently. Yeah, there's no continuous supply. It's the same stuff, Jeff. Supply chain. I put a sticker on it. They put a sticker on it that says data centers. I mean, it's the same thing. I have NAS drives. i buy special special western digital red drives for my network attached storage that maybe they last a little bit longer or whatever it's the same thing you do in a data center it's called oem this doesn't kind of a box it just doesn't have a box yeah it doesn't have a box great uh similar how about the ai powered private school that costs 60 000 a year 404 says students are being treated like guinea pigs inside an AI powered private school it's called the Alpha school oh yeah yeah sound good huh the AI is generating faulty lessons that sometimes do more harm than good and people pay $60 ,000 a year to send little Timmy to the AI brainwash this is very Silicon valley to me or like utah mystics like those are the two where the alpha school is actually i didn't i didn't if it's in utah like mic drop for the evening a school where kids crush academics in two hours build life schools skills through workshops and thrive beyond the classroom campuses in austin san francisco miami la washington dc dallas and other metropolitan areas a city thing it's a city thing you just would dr phil this is the whole pitch though learn twice as much in two hours so you only go to school for two hours oh here's one one this is from lulu who's level two one of the reasons i love alpha is because we have our own currency that motivates us to do more work this feels so not research fact to me like i i just feel like You have to spend a certain amount of time with material to absorb it.

2:31:50Like just point blank, full stop. Like you can't automate everything. Oh, there's one in Puerto Rico. That's for all the Bitcoin bros. Well, our villa could be there. That could be good. Oh, yeah. Miami could be good. Yeah. That is the bad, the good, and the ugly. Let us pause for station identification and then your picks of the week. And Jeff can relieve his L2. on my mountain of pillows yes by a mountain of pillows in fact jeff if you want to retire no no no no no okay we're almost there uh this episode of intelligent machines is brought to you by oh i love these guys the spaceship is where we got uh uh the paris's new site secretly british from spaceship.com and by space mail the professional email service from spaceship ship she has an email address now at secretlybrita.com a business email absolutely must it's the easiest the best way to look professional in every message you send if you're still sending messages that say you know leo at gmail.com that's not business like it needs to have your company name in it give your emails the best chance of reaching the inbox not the spam folder That's why over 2 ,000 users switch to SpaceMail every month.

2:33:12Switching is easy. SpaceMail's super fast unbox process links your domain and email in seconds. So once you set up, secretly British, for instance, or yourcompany.com, you literally just press a button and now you're getting email at your corporate address. And once you set up, SpaceMail keeps everything running smoothly. built-in spam detection and a 99 % uptime guarantee. What I love about SpaceMail is new features are shaped by the users. You make the roadmap. As a result, they're built around your needs. SpaceMail has built-in calendar, an AI email assistant. There are iOS and Android apps, beautiful apps for email on the go.

2:33:54All of this chosen by SpaceMail users. They said, we want an iOS app. okay you got it we want an android app okay you got it space mail is a key part of the wider spaceship universe that's where i register my domains now and if you're a regular listener you know spaceship offers some of the best prices on domains we were able to get it for half as much for uh for secretly british for half as much as as we were finding it elsewhere plus you get all the add-ons you might need including vpns website builders hosting and more whether you're building something big or launching your first idea space mail gives you a pro email address without the pro level price tag and with a 30-day free trial hey you can start today at no cost visit spaceship.com twit to see the exclusive offers discover why thousands have already made the move that's spaceship.com twit spaceship.com slash twit we thank you so much for your support spaceship of everything we do here at twit.

2:34:58Emily, if you have a pick, you can use it now, if not? I don't have a pick. No pick. I have a pick. So when I use Claude, Claude will, when it's done thinking or wants me to give it some input, pop up a little bubble on my screen. But sometimes I miss the bubble. So I've installed this. It's called Peon Ping. Stop babysitting your terminal. your peon pings you the instant Claude code finishes or needs permissions and actually it would work for anything is that from like war craft or it is you want to hear it listen ready to work yes oh my gosh it's so great so whenever I'm I'm ready to work I'm like it goes work work something you're doing that was such a joke in my family like me and my siblings would just repeat all these phrases.

2:35:53This is from Warcraft, right? Yes. I love it. Work, work. Wow. The good news is, you know. Be happy too. Oogie doogie. You're going to kill me. This is me and my friends circa 1992. I love, yes. This is me and my siblings. Like, someone would be like, go unload the dishwasher and be like, be happy too. Work, work. Work, work. And when you click the orc's head, the peon's head, it would say different things. Something you're doing? can you send this to me please it's so good it's peon ping but that but that's not all they have many many others as well they have all these different packs from hell divers from tf2 when you download this once you set it up you can rotate through all kinds of different sounds Are they going to get sued?

2:36:44I don't think so. And if they do, I hope... If the NPR guy sues... What's nice is they make the appropriate sound. So if it's like asking permission, it's... What do you want? If it's an acknowledgement, it's... I can do that. If it's annoyed... Me, not that kind of orc. I know, it's making me smile. They have a variety of sound packs of all kinds. I mean, oh, they have the human peasants from World Warcraft 3. Ready to work. Yes, me lord. Off I go then. And they have Soviet engineers from Red Alert 2. Tools ready. Yes, commander. They have Battlecruiser from StarCraft. Make it happen. Hailing frequencies open.

2:37:31Sarah Kerrigan, also from StarCraft. You may have time to play games, but I've got a job to do. That one gives me anxiety. that game is scary uh i when i install it there were uh i think 120 of these and you can add your own so very easy to install really fun it's sick but i love it but you know i think you need this uh to insert comments in the show ready to work ready work the thing about this is that that's what plays in my head when i get an email from benito do you want to be in the show i'm like can be happy to ready to work like this stuff is owned by microsoft now though so i don't know because like blizzard's owned by you know what it's still up go get it now while you can that's all i can say uh they they have uh yes they have by your command they have all the nerd stuff everything and then this uh also for paris awesome llm reasoning failures this is a curated list of things that LLMs screw up on.

2:38:33And it's just a lot of them. So if you're looking - Add quotes to that. If you're looking, yeah, for bad quotes, ammunition, hallucinations, all sorts of issues. Somebody's compiling this all. This is a common kind of trope on GitHub. Awesome X, right? Awesome C libraries, awesome Python libraries, awesome whatever. And so this is awesome LLL reasoning. failures i have a quick pick can i just really quickly no it's just something i'm nerding out about right now but not related to tech kind of the opposite no it doesn't have to be tech it could be terracotta tiles yeah i'm nerding out about that always but basically i've got i've realized there are different kinds of paper you can get like if anyone's into stationery you know these old paper mills in japan and europe japanese papers yeah like japanese paper european paper There's a French company I got some stuff from and they have different textures, different thicknesses.

2:39:32They are just been making the paper this way for hundreds of years. In Germany. Yeah. Yeah. Like maybe a hundred years. Maybe Japanese. It's a hundred hundreds. I don't know. Old. It's thousands. I think. Yeah. Yeah. It's like a lost art and you can still buy all this stuff online. And I'm, I'm having a lot of fun with it and taking some notes on paper, which apparently Sam Altman still does. If that makes it cool. I don't know. Do you do a bullet journal or what kind of journaling do you do? I mostly do like kind of personal. Personal and professional journaling, both. Nice. No, I think it's really good to do.

2:40:06Yeah. Yeah. But it's fun to mix it up with like these new paper types. So I think that's a fun rabbit hole if anyone's looking for one. I, for a long time, I had a Hobonichi, which is a Japanese day timer and they use beautiful, beautiful paper. I just love the feel of the paper. Just the tactile experience is so nice. actually there's a 14 year old who just won a prize of 25 000 in the thermo fisher's junior scientific innovators challenge because he created an origami fold of a classic japanese miura ori pattern that can hold 10 000 times in some way this kid loves paper it's so cute you gotta you gotta read the article about all the folding he did it's awesome get him look at he's piling all this weight on his little folds we need him he won 25 000 because you know what it holds 10 000 times its own weight good job miles woo jeff anything you would you've already given us some of your picks page 28 in the gutenberg parentheses credit for inventing paper has been given to chinese man named kai luen and ad 105 by coincidence in the same period the codex appeared and the space disappeared in writing however the legend is ruined by discovering some paper fragments in china that date their creation to two or three centuries before him wow nonetheless the chinese get credit for discovering how to chop and hammer fabric hemp fishnets and tree bark into cellulose soup diluting clean water and then dipping a mold into mixture to come up with paper its fibers overlapping and interlocking to create a smooth surface they use paper for clothing wrapping lanterns fans prayer ceremonies kites cups and yes the toilet you wrote that i wrote that it's awesome so we have some paper lovers on the show today so uh my pick is uh our old uh an android friend hugo barra has uh kind of vibe coded a new company yeah i saw your link on this and i checked it out i thought this is kind of interesting superdue is one of his favorite agents he built with dreamer and it's a vibe coded to-do list yeah it's very very gina trepani it's it's a to-do list that then becomes an agent and does things that you want it to do it doesn't merely stop at um listing it and bugging you yeah it goes out and gets you the tools you need to get it done for instance if it sees an email saying you did it it will check it off um very cool and now this is all about his really what this is a plug for his new company dreamer yeah the tool which is a kind of vibe coding platform right so he wrote superdue in dreamer When I first read his post, I thought, why does he keep mentioning Dreamer?

2:43:06Now I understand. That's why we built Dreamer. It's your home for personal intelligence. Yeah, I want to try it. It's interesting. Yeah. Well, this is the kind of thing that I've been arguing that the things that you do, Leo, with Claude and agents still requires a technical sophistication. I think the next level of this is when people can do it on a retail level without ever going into terminal, without having to install anything on a server. They make something, it does what they want, they can share it. And I think that's where we're going to see this explode. That's where everything kind of goes.

2:43:42I think you're right. Like website design, now it's just drag and drop. Yeah, exactly. Yeah. You start with Nano Banana, you get an image, then you say, make that image into a website. i have this compulsive thing in my head when someone says nano banana in my head i go nanner banana and i have to restrain myself from saying it out loud i'm like do i have tourettes no i don't know i mentioned that i spent monday coding what i wrote was i guess in a way it was kind of like writing a website i had claude code i bought many moons ago i bought this e-ink uh this display it's a color e-ink display with a raspberry pi on it but it required coding to put anything on it and i never got around to doing it so i had claude code build me a web a dashboard uh it's e-ink so i just unplugged plugged it so it's not live but because it's e-ink it stays there with the most recent episodes posted how many subscribers we have how many club members we have all that stuff isn't that cool that's amazing updates that i have it set now only to update it once a day because it doesn't change that often but i could have to do it more often.

2:44:48So speaking of hardware and AI, line 125, Raspberry Pi, I know I'm extending myself when I ask to go off, Raspberry Pi stock soars 40 % on the belief that Raspberry Pi will be used as hardware for AI agents. Sure, that makes sense. But is that also from your book? No. Okay, I'm like, so are you willing to stay on if we just read excerpts of your book? I could read you excerpts from magazine. You said page 145 or something. No, the line. Hold on. And actually, Darren is saying, don't start with Nano Banana because Google has another tool that's designed specifically for that called Stitch, which is where you would go to design with AI.

2:45:31So I am not aware of that. Thank you. So stitch.withgoogle.com. And design at the speed of AI, transform ideas into UI designs for mobile and web applications. ah i will use this because my next project is to write a uh twit a client a podcast client for twit just for our shows shouldn't be too hard maybe i'll try it with stitch thank you so much emily for leaning great to see you again emily writes at pc magazine she's a senior reporter every quote verifiably human i can promise you are you on blue sky mostly twitter where do you where do you hang your hat um i have been doing a tiktok push recently i'm kind of experimenting out loud the videos are very hit or miss but if you'd like to follow me there and see how my experiment goes that would be fun you can also find me on blue sky i don't know i'm everywhere i'm easy to find just search my name uh i i like this so what's your uh that's your handle on uh on the tick yeah i think it's emily forlini and then there's an underscore there might have been one who beat me um but just search my name yeah just hate that right i mean let me search for emily forlini yeah so i've been trying to post twice a week and pushing myself to figure out something i want to say twice a week which is hard so it doesn't always work but i'm having probably the most fun on that platform and it's fun when people comment and stuff so i i would like to connect there it is it's emily forlini with an underscore at the end oh no look at me you see it's so silly no it's cute there you are with mike i love it this is great it's a work in progress i'm trying to push myself and like figure out, you know, the tone of TikTok, the framing of the video, blah, blah, blah.

2:47:29So if you want to be part of that journey, you can find me there. I love it. I need to start making TikToks just to understand it. It's so cool. I have one. I mean, I'm just trying to not be embarrassed. It's just like, here's what I'm saying on TikTok. You know, it's it reminds me of when I first did a blog. It's like, this is so self-centered. Why would anybody? I know I'm I look like such an idiot, but I don't, whatever, I guess, right? No, I love this. I know TikTok has made some phenomenal stars. Anna Lapwood, who plays the organ, was at Cambridge. She now has a worldwide audience with albums and concerts.

2:48:06It's amazing. Leve, pronounced, Leve, spelled Laufey, the Icelandic American singer. On the Olympics, on the short programs, I heard her song the other day. There's that guy who does the sandwiches, Salt. What's his name? Salt Jim. Dumped underscore Hank, I think. Yeah, your son. Yeah, look at that mustache. It seems like it's getting darker. It's become a trademark. Oh, it is a trademark, let me tell you. See, this page looks good. I should just have more food on mine. Yeah, well, he figured, you know, that's what happened is he spent a lot of time figuring out, you know, what got the algorithm.

2:48:43By the way, the latest is the behind the scenes of his Super Bowl commercial. Cool. Yeah, the problem is I'm busy and kind of phoning it in sometimes. so like the lighting the sound you can't it's a full-time yeah gig and that's this is something i journaled about for example like you know new year's resolution setting goals yeah it's like i'm gonna try to tick tock twice a week for three months and see what happens so the other thing that he did which is smart is he also is on instagram and i think nowadays both is probably good right yeah yeah thank you emily so great to see you you too jeff jarvis professor of journalistic Innovation Emeritus.

2:49:22I'm going to bubble off. Montclair State University up the road from Emily. Wait, you teach at Montclair State? I'm a fellow at the Center for Cooperative Media. Yes, we should. I live in Glen Ridge. I'm working on a new program we can't talk about yet, but it's very exciting. Doing lots of things up there. It's the Jersey Boys on Twitter. The Jersey Boy and Girl. Jersey Strong. my uh colleague carrie brown i think lives in maplewood a former professor with me at cuny and lots of lots of journals nice so many i know you don't need to connect with me there's a million people so much of the new york media industry lives in this area as i call it the upper west side of new jersey yeah totally i know people work at cnn our editor-in-chief lives around here i mean there's like so many people uh jeff's also the author as you saw of the gutenberg parenthesis which talks about paper and magazine which talks about paper and is new on hot type which talks about printing on paper so there is a common thread i guess thank you jeff thank you better paris will be back i hope next week if the steroids hold out and emily we'll see you next on tech news weekly a couple of weeks i think yeah maybe next week maybe the week after we'll see.

2:50:41We don't know. It's going to be fun. Thank you everybody for joining us. We do Intelligent Machines every Wednesday, right around 2 p.m. Pacific, 5 p.m. Eastern, 2200 UCC. You can watch us live. If you're in the club, in the club, just discord, but also on YouTube, Twitch, x.com, Facebook, LinkedIn, and kick. We also, of course, are on YouTube. You can watch us there. You can download episodes from our website, twit.tv slash I am. Best thing to do though, subscribe in your favorite podcast player. You'll get it automatically as soon as we are done. Thank you everybody for being here. We'll see you next week on Intelligent Machines.

2:51:19Bye-bye. Hello everybody, Leo Laporte here. You know what a great gift would be, whether for the holidays or at just any time, a birthday, a membership in Club Twit. If you have a Twit listener in your family, somebody who enjoys our programming, and you want to give them a nice gift and support what we do, visit twit.tv slash clubtwit. They'll really appreciate it and so will we. Thank you. twit.tv slash clubtwit.

2:51:57But what I wanted to tell you, my daughter is a student, semester-beitrag, laptop, books, software, Handy, Internet. So a master is really cheap. Say, she can get it back. Yeah, you mean from the Steuer to set up, right? But she doesn't pay. No, it's a word, it's a loss. You make it simple with Wieso Steuer. And when she works, it's Kaching! That's it? Safe! Wieso Steuer? Hol dir dein Geld zurück. Now to try it out.

From the publisher

OpenClaw's creator makes headlines by joining OpenAI after GitHub fame and a whirlwind of VC and big tech offers, redefining what's possible for independent developers in the AI arms race. Is this the year agentic AI goes mainstream, and are the big players ready for that disruption?

  • OpenClaw, OpenAI and the future | Peter Steinberger
  • OpenAI disbands mission alignment team
  • Opinion | I Left My Job at OpenAI. Putting Ads on ChatGPT Was the Last Straw. - The New York Times
  • Introducing GPT‑5.3‑Codex‑Spark
  • Anthropic releases Sonnet 4.6
  • Exclusive: Pentagon threatens to cut off Anthropic in AI safeguards dispute
  • Google's Pixel 10a Launches on March 5 for $499
  • Google's AI drug discovery spinoff Isomorphic Labs claims major leap beyond AlphaFold 3
  • Gemini 3 Deep Think: AI model update designed for science
  • Radio host David Greene says Google's NotebookLM tool stole his voice
  • A new way to express yourself: Gemini can now create music
  • Why an A.I. Video of Tom Cruise Battling Brad Pitt Spooked Hollywood
  • GPT-5 outperforms federal judges 100% to 52% in legal reasoning experiment
  • An AI project is creating videos to go with Supreme Court justices' real words
  • I used Claude to negotiate $163,000 off a hospital bill. In a complex healthcare system, AI is giving patients power.
  • Sony Tech Can Identify Original Music in AI-Generated Songs
  • AI Pioneer Fei-Fei Li's Startup World Labs Raises $1 Billion
  • Yann v. Yoshua on directed systems
  • Dr. Oz pushes AI avatars as a fix for rural health care. Not so fast, critics say
  • An AI Agent Published a Hit Piece on Me
  • An Ars Technica Reporter Blamed A.I. Tools for Fabricating Quotes in a Bizarre A.I. Story
  • Plain Dealer using AI to write reporters' stories
  • Mediahuis trials use of AI agents to carry out 'first-line' news reporting
  • DJI's first robovac is an autonomous cleaning drone you can't trust
  • Leaked Email Suggests Ring Plans to Expand 'Search Party' Surveillance Beyond Dogs
  • ai;dr
  • I hate my AI pet with every fiber of my being
  • Thanks a lot, AI: Hard drives are sold out for the year, says WD
  • Students Are Being Treated Like Guinea Pigs:' Inside an AI-Powered Private School
  • peon-ping — Stop babysitting your terminal
  • Hugo Barra makes a to-do agent
  • Raspberry Pi soars 40% as CEO buys stock, AI chatter builds

Hosts: Leo Laporte, Jeff Jarvis, and Emily Forlini

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 858: The Itinerant Salt Miner from Buffalo - Silicon Valley's Military DilemmaIntelligent Machines (Audio) · 2 h 52 min
Listen in VO