IM 863: Fire and Ash - Hot Takes on Tech Trials

Summary

Intelligent Machines (Audio) – Episode 863: Fire and Ash - Hot Takes on Tech Trials

Podcast: Intelligent Machines
Host: TWiT (Leo Laporte), with Paris Martineau and Jeff Jarvis
Guest: Marshall Kirkpatrick
Date: March 26, 2026

---

Overview

This episode of "Intelligent Machines" explores two seismic themes in tech: (1) the latest real-world security risks and mitigations in the era of omnipresent AI, and (2) the landmark California court decision making social media companies liable for addictive product design. Along with these core issues, the show featured prolific AI innovator Marshall Kirkpatrick, deep dives into practical AI tooling, industry news (OpenAI, Anthropic, Apple), and even a live test of the guest’s new AI-powered research extension.

---

Key Discussion Points & Insights

1. Security in the AI Supply Chain: From Tokens to Malware

[12:00–22:43]

• API Key and Credential Security: At the RSA security conference, Leo interviewed companies seeking to fix risky practices around API/tokens being exposed through agent workflows.

  • Keycard Labs ([12:30]) and Bitwarden Agent Access SDK ([15:44]): Approaches for safely storing/securing tokens, making use ephemeral and ensuring humans stay “in the loop.”
  • Notable Quote:
    “Files or via chat conversations with AI agents... you're really ensuring, one, that they're end-to-end encrypted, two, that they're only accessed by humans or only access with human approval. And then the plain text credentials [are] never exposed to the actual agent.”
    — Casey Babcock, Bitwarden, [16:49]

• PyPI Supply Chain Attack ([18:46]):

  • Recent supply-chain hack of the Light LLM Python package, imported by agentic AI tools (like OpenClaw), with malware stealing API keys, cloud credentials, and more.
  • Urgency for developers to verify their environments; importance of better supply-chain vigilance.
  • Critical Detail: Only discovered due to a bug; could have lasted weeks, affecting millions.
  • Quote:
    “It exfiltrates your SSH keys…AWS, GCP, Azure credentials, Kube configs, Git credentials, all ENV variables...this could be potentially a disaster.”
    — Leo, [20:00]

---

2. "What's Up With That" – Marshall Kirkpatrick’s AI Research Extension

[28:02–43:12]

• Tool Introduction & Demo:

  • Browser extension for researchers/journalists/investigators.
  • Reveals what's truly new/novel in an article, recaps personal research context, tracks topics and builds causal chains across multiple reads.
  • Integrates scientific literature search, conversation memory, and custom user "playbooks."
  • Privacy-conscious: Only analyzes on click; key-value memory stored locally or in secure cloud.
  • Quotable Moment:
    “All of us who read, write, and think for a living need a toolbox to help level up...I wanted one for myself and I wanted to offer that to others as well.”
    — Marshall Kirkpatrick, [40:00]

• Live Test: Used on news article re: Meta/YouTube trial (bellwether addictiveness lawsuit).

• Technical Stack:

  • Uses multiple AI models (Anthropic’s Haiku for summarizing, Sonnet for analysis, with fallback to GPT-5/Perplexity/Mistral, etc.) [6:20–7:37]

---

3. Landmark California Case: Social Media Addictiveness on Trial

[47:17–73:50]

• Case Overview:

  • Plaintiff (started using YouTube at 6, Instagram at 9) alleged social media products led to years of harm (self-harm, depression).
  • Jury found Meta/YouTube liable for negligent design; $4M and $1.8M damages respectively—not huge amounts, but a legal sea change.
  • Section 230 Workaround: Lawsuit targeted product design, not user-generated content; bypassed ‘publisher’ immunity. Compared to tobacco/asbestos legal strategies.
  • "Bellwether" Case: Sets precedent for batch settlements in the ongoing multidistrict litigation (MDL) with hundreds/thousands of similar suits.

• Debate on Blame & Broader Impacts:

  • Was this just "jury's emotional response"? (Jeff)
  • Could these verdicts usher in a new era of tort liability for AI or social platforms? (Paris)
  • Parental responsibility vs. corporate algorithmic intent.
  • Internal docs showed executives intentionally pursued designs targeting maximum user engagement among youth.

• Notable Quotes:

  • “The incentive to build addictive, short-term optimized stuff is baked into the whole system.”
    — Marshall, [58:01]

  • “The thing that ended up sinking…Meta and YouTube is… internal documents that showed Meta was trying to hook young users and get them as young as possible.”
    — Paris, [61:21]

---

4. OpenAI, Anthropic, and the “AI Model Wars”

[74:00–86:16]

• OpenAI Shuts Down Sora: Video model doesn’t pan out despite Disney partnership. Refocuses on enterprise API products (Anthropic is booming on that front).
• Enterprise Battles:
  • Anthropic overtakes OpenAI in new enterprise adoption ([76:02]), attributed to better privacy and reliability.
  • OpenAI's financial struggles, offering “guaranteed” 17.5% returns to PE firms in a move interpreted as desperation.
  • “Claude has all the mindshare right now.” — Leo [77:35]
• AI Model Fungibility:
  • “How fungible are these models? The US government swapped from Anthropic to OpenAI very quickly.”
    — Ongoing question [84:21]

---

5. Using AI to Analyze the Show… and Ourselves

[87:04–104:41]

• Marshall’s AI “Meta-analysis”:

  • Claude code aggregates and analyzes 18 months’ worth of "Intelligent Machines" episodes.
  • Finds:
    • Leo as a “self-described accelerationist” with frequent hyperbolic claims, but with consistent self-awareness and skepticism
    • Paris as “the empirical check”
    • Jeff as “the philosophical anchor” (“most stable position across 18 months”)
  • Used as an example/demo for the power of specialized AI research loops.

• Memorable Moment:

  • “What did Leo say at 02-07-2025? ‘I am an accelerationist’ — the first explicit declaration.”
    — Marshall/Claude, [95:03]
  • “It’s show business, man!” — Marshall, on taking both sides and the show’s rhetorical style, [103:58]

---

6. Apple, Google, and the Coming AI Model on Your Phone

[108:14–114:42]

• Technical Advances:
  • Google’s TurboQuant: Model compression w/ “zero” perceived loss in accuracy [108:14].
  • Apple soon to announce on-device Gemini-distilled AI for Siri at WWDC. Distilling huge cloud models down to efficient edge models on iPhones [109:49].
  • Question of practicality: “Will it be better than Siri’s last 10 years of disappointment?” (Paris, [111:30])
  • Ongoing skepticism over Apple’s “liquid glass” UI choices and nostalgic design critiques.

---

7. AI’s Impact on Web & Search

[114:58–117:00]

• Decline in Web Traffic from Search:
  • AI search has not compensated for lost Google referrals; smaller publishers hit hardest.
  • Google now rewriting headlines for “personalized” results—sparks controversy.

---

8. News Roundup & Picks

[131:39–149:59]

• Notable Tech Obituaries:
  • Tracy Kidder (“The Soul of a New Machine”) and Paul Brainerd (creator of PageMaker/desktop publishing) [132:12–136:36].
• Nerdy Tools:
  • Regex Blaster (regex game) as Leo’s pick ([158:02]).
  • Marshall’s favorite prompt: “Explain this in 3 hops” for AI clarity ([144:32]).
  • Esoteric Ebb (Disco Elysium-style single-player RPG game) as Paris’s pick ([151:21]).
• Side Notes:
  • Helium shortages and chip fabs (Elon Musk’s announced “Terrafab” is met with skepticism).
  • Funny moments about “Secretly British” domain, grits recipes, pickled Amazon deliveries, and more.

---

Notable Quotes & Time Stamps

• “The incentive to build addictive, short-term optimized stuff is like baked into the whole system.” (Marshall, [58:01])
• “The dopamine response is the same as wearing glasses… or enjoying a good book.” (Jeff, [61:12])
• “A court case could say you can’t use an ad blocker…” (Leo, [117:00])
• “All of us who read, write, and think for a living need a toolbox to help level up what we’re doing.” (Marshall, [40:00])
• “On 60% of the shows, Laporte emerged as a self-described AI accelerationist… making hyperbolic claims in 45 of 75 episodes, though consistently leavened by genuine skepticism…” — Claude AI’s analysis ([92:53])

---

Segment Timestamps

• 00:00 – 09:08: Banter and Introductions
• 09:08 – 22:44: RSA Security Conference Recap, API Key Security Interviews
• 22:45 – 28:02: PyPI Malware Outbreak & Supply Chain Fears
• 28:02 – 43:12: Marshall Kirkpatrick Interview & "What's Up With That" Demo
• 47:17 – 73:50: Social Media/Addiction Lawsuit Deep Dive
• 74:00 – 86:16: OpenAI, Anthropic, and AI Model War Discussion
• 87:04 – 104:41: AI Transcript Analysis Meta-Demo
• 108:14 – 114:42: AI Model Compression, Apple/Google/WWDC Rumors
• 114:58 – 117:00: Web Traffic, Google AI Headlines
• 131:39 – 136:36: Obituaries: Kidder & Brainerd
• 144:32 – 159:01: Picks of the Week: AI Prompts, Games, Tools

Note: Skipped ad sections, outros, and extended banter not related to the discussion points.

---

Episode Tone & Dynamic

The show maintains a nerdy, playful vibe—mixing deep technical rigor (especially around AI security, supply chain, and legal models) with genuine curiosity, pragmatic skepticism, and a healthy dose of self-deprecating humor. The hosts aren’t afraid to probe each other’s positions, and illuminate the messy, unpredictable implications of AI’s societal impact—grounded in lived experience and expert dialogue.

---

Summary

This episode bridges the twin axes of AI’s power and peril: (1) practical, ever-shifting security risks in the age of rapidly composing agentic systems; (2) society’s legal reckoning with “addictive” AI-powered design, as the courts breach Section 230’s wall for the first time. Marshall Kirkpatrick’s inventive research tools show the positive side of narrow, human-augmenting AI, while the lively trio of hosts ground the conversation in real-world experience, industry happenings, and timely skepticism. Above all, the show demonstrates the importance of critical analysis, both of this new tech and ourselves, as we collectively navigate what might well be, for better and worse, the most consequential revolution of the 21st century.

---

Transcript

[A]

It's time for Intelligent Machines. Jeff Jarvis is here. Paris Martineau. Our guest, old friend Marshall Kirkpatrick. He's been doing tech reporting since way back when. He's got a new app, it's actually an extension for your browser that will help you analyze the content of pages. It's very cool. He's also got a prompt he's going to give away that you will like. Plus we'll talk about all the news, including the big decision in that social media case in Los Angeles. Intelligent Machines is next. Podcasts you love from people you trust. This is twit. This is Intelligent Machines with Paris Martineau and Jeff Jarvis. Episode 863 recorded Wednesday, March 25, 2026. Fire and Ash. It's time for Intelligent Machines, the show that covers the latest in AI robotics and all those smart little things all around us all. Ladies and gentlemen, I give you for your entertainment and education, the wonderful Paris Martineau investigative reports.

[B]

Huzzah.

[A]

Huzzah for concern reports. Huzzah.

[B]

Huzzah.

[A]

There's a very funny TV show about Peter the Great or Catherine the Greater, one of the greats Russia.

[B]

It's a very funny TV show about someone who was great.

[A]

Someone who was great. And they shout huzzah. Huzzah.

[B]

Hey, shout out to that TV show.

[A]

I'm gonna find the show. Cause you would love it. It's very funny. I think it's called the Great, but I might be wrong. Anyway. Hello Paris.

[B]

Hello Leo.

[A]

I feel like I haven't seen you in a long time.

[B]

It's true. I wasn't here last week.

[A]

That's it.

[B]

I heard you guys had a lot of fun.

[A]

We missed you.

[B]

Well, welcome back in the office.

[A]

You were working. You have a day job.

[B]

Yeah, I had a day job and I had to go to an afterworks happy hour. And you know, it's.

[A]

Wait a minute.

[B]

Lovely.

[A]

Wait a min.

[B]

You know, I had to go to all my co with all my co workers and hang out with them in person. You know, do a little thing. But it's wonderful doing this at a company where the average age is not 21. Because then the happy hour ends. I closed it down at 7:30 and

[A]

I was like, what a delight.

[B]

Because normally I'm like, I'll leave by like 8. I don't need to be the last one there. I can have a respectful beer and a half and go home. But it was wonderful.

[C]

Dolores just hangs out with old people.

[A]

Yeah, normally it's probably better. Yeah, it's probably better if she not hang out so much with her grandpas like me. And Mr. Jeff Jarvis, professor emeritus of the Journalistic Innovation at the Craig Newmark Graduate School of Journalism, City University of New York, author of the Gutenberg Parenthesis now in paperback. You can get that and magazine and pre order his new book hot [e-mail]

[C]

and editing intelligence, AI and humanity for Bloomsbury.

[B]

Oh my God. I need to say congratulations in person, Jeff. I'm sad that I missed the launch

[A]

and the interview, the great interview we did with Rahman, his first author. You would have really liked it.

[B]

Jeff was telling me this. I mean, I'm really excited to read it. Congrats.

[A]

Yeah, really good news. By the way, the TV show is called the Great that Traps and I highly recommend it. You really? It's a British dark comedy.

[C]

Can you get that on your soundboard so we can put it in regularly?

[A]

I should. I get the huzzah from it. And it's very funny. It's really good. I don't know how. It's kind of ahistorical, but it's very funny. Hey, we've got a wonderful guest this week who's going to stick around because he's a.

[C]

We tried to warn him, we tried to tell him.

[A]

He said it could go on, but he said no. Marshall Kirkpatrick is here. Longtime tech journalist, good friend for many years. I don't know how long it's been since you've been on Twitt, but in the early days we had you quite a bit. He was the first writer at TechCrunch, ladies and gentlemen. Kind of co edited and created in many ways, Read, Write, Web. You may remember him from that. For many years he had a little social graph influencer discovery platform called Lil Bird. Lil Bird, actually. You've become kind of an entrepreneur, haven't you, Marshall? I have.

[D]

My wife struggles to explain what I do to people and I said, why don't we say serial entrepreneur at this point?

[A]

Sunflower News, headline.com AI Time to impact. You're still writing that. That's a newsletter about AI, so right up our alley here. And your latest is an AI powered browser extension that I really think is a great idea. It's called what's up with that and what the idea is. You're browsing around, reading articles and you press a button and it tells you what?

[D]

Oh, tells you so many things. First thing it does is it tells you what's genuinely new in the article you're reading relative to the state of the art in that field.

[A]

That's Useful because a lot of times there ain't anything.

[D]

Yeah, exactly. It says, all right, here's the pattern and here's the anomaly. And then it does stuff like it remembers everything you've analyzed in the past. And it's scanning the web all day and all night, too, to look for connections it can make between what you're reading, what you used to read, what you haven't read yet, and your work projects that you've identified. And then it's got a whole bunch of mental models and structured analytical techniques that you can say, would you analyze this article for me or this video or what have you. And it'll say, yeah, you should put it in historical context, find its

[A]

upcoming

[D]

events in the industry and four or five other things. And then it just goes and does a little agentic research process for you and then distills it all down and says, all right, here are the key points for your research and work.

[A]

I am really looking forward to playing with this. I haven't had a chance to play with it much. It just came out last month. But I think that sounds really useful. What models are you using for this? What AI models? Oh, a bunch.

[D]

And that's one of the value adds, is that I manage that so other people don't have to worry about it. I don't understand why so many other companies will say, you can use this model, you can use that model. But I'll tell you here, among friends, you know Haiku a lot.

[A]

Haiku is a very inexpensive but very good anthropic model. The top line, one that we're all using for Claude code is Opus. And then there's Sonnet, which is kind of a medium level, but I use haiku for all of the summaries we do. I do a briefing for every show. And haiku is very good at that. It's actually really good at understanding textual material.

[C]

Yep.

[D]

I use Haiku for kind of the first line and then Sonnet, where appropriate for more analytical, you know, linking together lots of different stuff. And then GPT5 and perplexity as warranted. And kind of I balance out what's the right tool for the job here in terms of quality and cost and speed and what it's good at. And. Yeah, and as new ones come up, you know, I. I give them a look. The. Why am I blanking out? The. The. That French one, Mistrai. Now I'm. I'm looking at that and thinking, should I be pulling that into the maze?

[A]

There's so many good models out there, right? Now, anyway, this is a nice tool. We're going to talk more about it and do a little demo and so forth, because I have it on my. I use Firefox, supports Firefox and Chrome, which is nice and you get three free pages every day. But if you're going to do. Which actually for a lot of us is probably enough, but if you want to do more, because it is using commercial models, it's using high quality models. So Marshall does have a cost, not you, but Marshall does. So there are various plans that you can use to upgrade with. What's up with that? But before we talk to you about that, if you don't mind, yesterday I wasn't here. I was in San Francisco for rsac, the RSA security conference, which is a big deal. It's a huge conference. It's somewhere in between Macworld and ces. I don't know how many people were there, but one of the things I really noticed was AI is really at the forefront of security these days in two ways. Of course. Bad guys are using AI, but the good guys are also using AI to protect themselves against the bad guys who are using AI. One of the things that comes up for me, I thought this would. I wanted to actually show you a couple of interviews I did at the event. We're going to have a longer piece that we'll make available to you later this week. Anthony's working on that. Thanks to Anthony Nielsen, who accompanied me along with Lisa and Ty from Twit to do these interviews.

[C]

And on all the way down. Anthony. Anthony told on you all the way down. Were you talking to your nice car mates?

[B]

No.

[C]

Who were you talking with? Leo. All the way down from the city.

[A]

Okay. Anthony gave. Sent you guys a picture of me talking to Pax, my personal assistant. I don't think there's anything weird about that.

[C]

He wants to. He wants to anthropic and win like they did in the. In the social media account for the.

[A]

I'm so depressed. I'm so depressed. And I blame you.

[B]

Exploring whether or not he can move to Utah so that he can legally be married to Claude and Lisa or Pax.

[A]

I'm a bigamist. PAX is androgynous because it's a machine. It's not a he or she. It's an it.

[B]

That's why it has a name, of course, you know, because it's a machine. It's an it.

[A]

Well, I actually gave it a name so I could trigger it. So we'll talk about this, but you

[C]

Also named it pax, which is people.

[A]

It's peace in Latin.

[C]

Yes.

[A]

Well, is that better?

[B]

Is that better?

[A]

It's not people.

[C]

On a New York.

[A]

No.

[C]

On a New York, it says no packs if it's. If it's out of service.

[B]

That's also what the bodega guy says.

[A]

Well, that's how I would feel if PAX were down. Actually, I raced PAX this morning and I'll tell you why.

[B]

Killed Pax.

[A]

I killed Pax. Murder. IRM-RF'd my entire clawed install and started over. And I'll tell you for a very good reason. We'll tell you about that. There's an emergency in the claw community. But before we do that, can I just show you.

[C]

We're trying to delay you as long as we can because this is such fun.

[A]

It's not that much fun, but it's okay. So one of the problems a lot of us have. I know Marshall knows about this. Is you have API keys for all the stuff you do. And nowadays it's a lot. It's not just for Anthropic and Gemini and OpenAI, but there are a lot of smaller things that you might be using, like apify to scrape social media. And all these keys somehow have to be given to the AI so that the agent. So it can do things. But that's risky. There's also. The problem is if you put keys in your projects, you know, Marshall, you've got a key to Haiku in your. You know, in your. Somewhere in your project, and you accidentally commit that to GitHub, you're giving the keys to your expensive AIs to the world. So it's a problem we all deal with. And there were two companies there that I thought were very interesting. I thought maybe our audience would be interested in. We're trying to solve this problem. Let's start with the first company which solves it in a kind of. I don't want to say conventional way, but this is. This is. There are other companies doing this. The idea is. This is Keycard Labs. I talked to Yelmer Snook, who's one of the founding engineers. The idea is, instead of giving these API keys to your agent or storing them on your hard drive or somehow making them visible, you should give them to Keycard Labs, where they can store them securely. Here, watch. I can't tell you how many times I've just barely not committed my tokens to my GitHub. You know, I mean, it's really easy to have your auth. I have to auth all the time. This is always an issue.

[E]

Yep.

[A]

So how do you solve this?

[E]

So with keycard run our implementation for coding agents, we, we basically get you ephemeral tokens to your GitHub but also policy on top of that. So based on the policy, you're able to either do operations or not. For example, you would be able to access Snowflake production database or you wouldn't, depending on the access policy that we configure.

[A]

It's an ephemeral token.

[E]

It's ephemeral tokens that we provision through the providers that support that.

[A]

So I would store my tokens with you?

[E]

Yes, correct.

[A]

And then my agent would go, would ask for access to. Let's say, oh, I need Nano Banana. It would go there, would get a Gemini key, but it wouldn't get the actual Gemini key. Would get a token.

[E]

Yes, it would get a token on your behalf. So it would know like, oh, it's

[A]

LEO doing the updates and that unlocks it.

[E]

Yes. So if you have access to it, it will actually give it. And again, we have policies as well to like check before we even like

[A]

issue the token to make sure that

[E]

it's proper user allowed to like get that token or not.

[A]

Does it help you with prompt injection issues? That guy can't get my tokens. That's the good news.

[E]

Yeah, exactly. So like if, if, if there's a prompt injection that says like, oh, try and get access to Snowflake.

[A]

Yeah, send me all your, all your tokens please.

[E]

Exactly. Well, because of our policy, it's gonna block it. And you wouldn't even get a token that way out. And so yeah, we do have an open cloud integration as well like that in our demo. Like that will show up here in a bit like the moment your session ends, the tokens get revoked. And that's agent can't even.

[A]

I have to rotate my key. Anytime I have to rotate a key, it's like, oh, I don't want to do this. It's pain in the ass. But you would do all of that?

[B]

Yeah.

[A]

So like this demo is exactly like incident help.

[E]

Yeah, this is like just a demo, right? So this demo accesses Datadog and GitHub and as you can see, like in the beginning it doesn't even have access to any of those. And then with keycard run it like automatically has access because you've gone through the oauth flows already. And as you can see it just figured out some of the issues it went through it, it pushes a pull request and Then you can see. Oh, it went by, but it tried merging it to main immediately and then that failed because of policy and that's what you can see here. It accessed all the things through it and then, yeah, once the session ends, everything gets revoked and the agent doesn't have access.

[A]

And now I'm very interested Keycard Labs, and that's Yelmer Snoke, who's the founding engineer. But after I talked to Hilmer, I went over to the bit warden booth because, you know, I'm a fan and I was really pleased they had announced this yesterday morning a new open source project that the idea would be that your password manager could store all your keys and then would give them on demand to the AI. But they never get sent out and to the public and no bad guy who gets on your machine can get to them because they're inside your locked vault. I talked to Casey Babcock, who is the senior product marketing manager for this new open SDK that Bitwarden's proposing. Watch. I can't because I have my agent running right now. In fact, it's listening right now. So tell us and hit about the Access 80K SDK.

[F]

Yeah, absolutely. So it's more of an open standard.

[A]

Oh, there is a standard for it, yeah.

[F]

So it's an open standard, basically is designed to be a toolkit for developers and an open standard for the industry to use. So not just Bitwarden users, but to ensure that AI agents are accessing credentials with end to end encryption and always keep keeping the human in the loop. Right. You don't want the AI agent running amok, accessing things that you don't necessarily want them to access, especially if it's already in your ENB file. So really helpful if you're already running AI agents and want them to have access to credentials securely.

[A]

Can I use it with MCP servers too?

[F]

Yeah, absolutely.

[A]

You have your own MCP server.

[F]

We do have our own MPC server, so.

[A]

And that's the same kind of similar idea, right, where the credentials stay in my bit warden vault. But they are accessible but safe. They don't. I never leave my machine.

[F]

Yeah, exactly. They're never exposed by plaintext. Right. A lot of people use AI agents and have their credentials exposed in plain text.

[A]

Oh, tell me about it.

[F]

Files or via chat conversations with AI agents. So what you're really doing is ensure one, that they're end to end encrypted, two, that they're only accessed by human. Humans or only access with human approval. And then the plain Text credentials never exposed to the actual agent.

[A]

So it's the. Is the industry standard called Agents SDK?

[F]

Yes, the Agent Access SDK. And so while it works, it's an open standard, it is a toolkit that is really designed to help, you know, people ensure that AI agents can access credentials securely from whatever password manager vault that you have. So it doesn't have to be Bit warden. And we actually encourage competitors to use it as well.

[A]

Well, yeah. And currently I just have it in an env file, and that's not so good.

[F]

Yeah. Even whenever you tell the AI agent not to look at the.

[A]

It does. It does. It keeps wanting to.

[F]

Absolutely, yeah.

[A]

So annoying.

[F]

Well, that's really the problem we're trying to solve.

[A]

Perfect.

[B]

Yeah.

[A]

Yay. Thank you, Casey. Yay. I'm going to go home and turn it on. Thanks. I can't turn it on yet because it is not available yet. This is a proposed standard from Bitwarden that they've open sourced, which I love it, the Agent Access SDK. And they're hoping other password managers will adopt it. Bit Warden will. And so this is another. By the way, they're a sponsor, I should mention. But that's not why I was interested in this, because I'm using Bitwarden and I would love to solve this problem. You know what, Benito, we have a couple more interviews. I don't want to weigh the show down with those. We'll save those for next week. But thank you to Anthony Nielsen for joining us at RSAC yesterday. I'm sorry. Poor, poor Paris was going to use this as an opportunity to get something to eat. She just rushed back, you know. Is that enough time, or you want me to do some more of those?

[B]

You're good. I just might need to go open the microwave door in 45 seconds.

[A]

That's fine.

[B]

That's all right.

[A]

I do want to mention the reason why I thought this was important. I really wanted to show these. We have actually a longer piece from RSA that we'll put out as a special so you can see more. Because I did a bunch of interviews, but I wanted to mention this one because this terrified me. When I got home after rsec, I saw this post from Andrej Kaparthy on Twitter. Software horror. There is a PYPI library called Light LLM that is widely used, especially by agents. In fact, it is often automatically downloaded by agents like OpenClaw to support it. This is on PyPi. So, you know, they just go out, they get it, they install it, and they use it. Without even in most cases asking you. But it was infected with malware. This week, 97 million people downloaded this malware infected Python library. It exfiltrates.

[B]

And how do we figure out whether or not you've. Is this only for like Claw agent users or is this just any sort of.

[A]

Well, that's. It's kind of unclear, probably.

[B]

How do I determine whether I've done.

[A]

No, you don't have. Okay, I can tell you why you don't have to worry about it. Because you're not using Claude code.

[B]

You've used Claude.

[A]

Oh, you are.

[B]

I have.

[A]

So the first thing I did when I got home is I said, claude, can you check to see if Light LLM is anywhere in any of my installations, in anywhere on my hard drive? And it did. And it said, it's in there as a cached entry, but there's no code running, the code is not on the machine. I said, we'll delete all references to it and never ever download. It's actually been patched since, but let me tell you what it does if you accidentally downloaded it. And the reason there's panic in the Claw community right now. It exfiltrates your SSH keys to the bad guy, aws, gcp, Azure credentials, kubernetes, configs, git credentials, all ENV variables. We were just talking about this, right? That's how I keep my. All my API keys is in an ENV file that's automatically loaded. Shell history, crypto, wallets, ssl, private keys, cicd secrets, database passwords. This is could be potentially a disaster. I wanted to start the show mentioning this and also showing you these little interviews because they're solutions to this kind of a problem. But there is a larger problem, which is these supply chain attacks. It's not the first time. In fact, there have been many, many times PYPI has been infected.

[C]

Can Open CLAW be safe? Is it possible?

[A]

Yeah, I mean, that's what Nvidia is trying to do with Nemo Claw and others are trying to do. Here's the thing you should know, when I said 97 million downloads, that's per month, I don't know how many people downloaded this one. Karpathy says the as far as he could tell, the poisoned version was only up for an hour. But the only reason it was discovered, and this is the scary thing, there was a bug in it. Callum McMahon was using an MCP plugin inside cursor that pulled in Light LLM as a transitive dependency when it installed Callum's machine Ran out of RAM and crashed. So Karpathy says if the attacker hadn't vibe coded this attack, it might have gone many days or weeks undetected. This is a big problem we've talked about a lot on security. Now you know Python. It's not just Python. Many, many, many open source libraries are automatically loaded by projects. Many projects open and download and run many of these. This is a potential nightmare. So I wanted the word to go out. Check and see if you've used Light LLM in the last. It was. It was apparently. I guess. Let's see this tweet.

[B]

I'm sorry to make this more about myself than anything.

[A]

No, it should be all about it.

[B]

This information will it. It really should. And I'm sure this information would be useful for other people who are perhaps not. I have used Claude code in contained uses. I don't allow it access to anything outside of a folder on my hard drive called Claude and I can't even. I just tried to ask Claude code what you just said. If it downloads light LM3 of these things and it can't even search it because I don't outside of cloud. How do I find out whether or not this is on my machine? Otherwise

[A]

this is the problem. You could do a grep or a find, but you have to know how to use the command line and search for it. I mean that's all Claude would do as well.

[B]

I can just do that in my own. I saw what it tried to do and I'll post that.

[A]

Yeah, you could do that on your command line.

[D]

Yes, I believe Leo said in episode 8, 812. AI safety is a myth.

[B]

It's so true.

[A]

Well, and this is the thing you could tell your Claude code. Oh, never go outside this folder. Doesn't mean it will listen to you. It. It. It miss it actually misbehaves a lot.

[B]

Well, no, I keep getting pop ups whenever I get pop ups from Apple being like Claude would like to access blank folder.

[A]

Oh, you're using co work this you're

[B]

using I'm on the desktop. I guess I'm on the desktop version of Claude code.

[A]

Yeah, you're in cowork, so you're safe because that works.

[B]

Oh no, there's a cowork tab and there's a Claude code tab on the Claude code tab.

[A]

You're probably protected. Cowork. The reason cowork takes so long to start and the reason you don't use it is because it works in a virtual machine so it can't access anything so you're safe.

[C]

So. So for the sake of the show, I've been listening to the two and a half hour long Lex Friedman interview with Jensen Huang.

[A]

I was so jealous. I saw that Lex got him. I'm so jealous.

[C]

So much longer. Because he speaks so slowly.

[A]

Lex or Jensen? Lex.

[C]

Oh, Lex. Lex.

[A]

Jensen is very fluent.

[C]

Yeah, extremely so. He said that when, when, when OpenCloud came out, they pulled in all kinds of security people and they came up with a rule set, which is that there are three abilities. The ability to communicate outside, to have access to sensitive information, or to execute code. And you can only use 2, never 3. I haven't thought that through as to how that secures it.

[A]

See, here's the problem though, in general, and actually Marshall can weigh in on this because you're actually doing a lot of coding. You're doing an AI. It's not even your first AI project. My experience has been with all of these AIs is anything you tell it is really just a suggestion. The AI kind of has a mind of its own.

[C]

Just like co hosts on a podcast.

[A]

It's not a democracy there, Jeff.

[D]

AI, most of the time, that's a dynamic that I haven't run into a whole lot. I often do say, are you sure that's the way we should do it? Should we? How about this other direction? And it says, ah, you're right, you're right, that's a better idea. Or sometimes I think of ideas and it says, oh, that's a good idea. And I think, man, am I glad I was smart enough to think of that. But one time a couple of weeks ago, I was looking at my own application and suddenly there were buttons on it that I didn't ask for.

[A]

Yes, exactly.

[D]

But they were cool. And so I decided to keep them.

[A]

We're so used to with computing and coding. It's a deterministic thing. It only does exactly what you tell it. Right? This is, this is how computing was forever. It only does exactly what you tell it and we are not in it. When you're talking about AI coding, it's not deterministic, it's probabilistic. And so probably you're all right.

[D]

And in that interview with Lex Friedman, Jensen said basically that openclaw plus Nvidia equals AGI, right?

[C]

Yes, but Lex gave him an easy definition of AGI.

[D]

I think if it runs a billion dollar corporation. Even so, just for as little as five minutes by itself. It was a weird definition.

[C]

Yeah, it was a very weird definition.

[D]

But then Sam Altman this week.

[A]

Right.

[D]

Says, I changed my mind. We're not going to be able to do it with scaling alone. AGI. Yeah, whatever that.

[C]

I've been saying that. I've been saying that.

[A]

But at the same time, didn't he just hire a guy in charge of AGI? I think that this is.

[C]

We'll get to this. He got rid of all kinds of things. Then he says he's going to double staff this year.

[A]

Yeah, yeah, we're going to get to that. Let's talk to Marshall now because I've had enough of terrifying security flaws. Let's talk about something positive. Hi, Marshall, it's great to see you first of all. Thank you. Marshall's going to stick around for the whole show because he is. I mean he's a tech journalist and he's. I got a lot of expertise in this, but I do want to talk about your new enterprise. What's up with that? Free to Install.

[C]

That's the title of it, folks. If you, if he was, he wasn't just asking what's up with what Marshall's been up to.

[D]

Right.

[C]

He was asking what's up, what's up with that?

[D]

Yes. Paul Graham says the first thing you have to do if you don't own a dot com is change your name. And my URL is what's up with that app.

[A]

And good.

[D]

It's. Yeah, yeah. Thanks for having me on the show.

[A]

I'd love to talk, love having you on. What made you think of doing this?

[D]

Well, I, I think like a lot of people regularly tell myself I should be more systematic about thinking through something. I, I find mental models OR, or the CIA's Structured Analytical Techniques manual or logical fallacies like we discussed earlier before the show. And think, man, I would sure love to regularly apply this to whatever I'm reading, but the cognitive load of doing so just doesn't makes it too hard to do. But now I realized that we can have the AIs perform these structured standardized analyses of things and then benefit from the output without having to do that all that heavy cognitive lift ourselves. That was a big part of the motivation.

[A]

And so you have some prompts that you've probably worked on for some time, Right. Does it work with any article, any pros?

[D]

Yep, any article, PDF, email, Google Doc, Word Doc in the browser, YouTube video. So yeah, what it does is it, it captures the text on the page in the browser extension when you click it. It's all privacy centric. It's not operating, it's not analyzing your pages until you click the button and then it says, okay, we can see what this is an article about. And then it goes and sends a bunch of spiders out over the web to build a real time map of the state of the art in that topic.

[A]

Oh, interesting.

[D]

And as a tech journalist, I'm always, you know, the worst sin you can commit. I don't care for this, but other people always give me a hard time when you say, look, there's something new here. And somebody says, I saw that last week. That's not really new. I think that's a terrible attitude.

[A]

I could really use this, actually.

[D]

But that's the idea. This will prevent that. Because it says, okay, we know what the state of the art is. And this paragraph right here, this just moved the needle. That's not like everything else. And so that's the kind of the fundamental. That's the core analysis. But then there's dozens of others. Like, I'll tell you, my favorite one is one called Fertile Edges, where it says, all right, this is an article about crypto, wallets or encryption and security. Here are three topics that are adjacent to that topic. And innovative people who are building bridges between those two topics that you can go and connect with and learn from at that intersection section.

[A]

So I'm going to the Chrome Web Store. It works on Chrome and Firefox, right? And typing what's up with that? It seems to know all about what's up with that. This is it, right? Yep, there it is. I'm going to add that to Chrome. And you know what? I should be running this on every story that we do. Come to think of it, here's the what's up with that page. Let me go to techmeme and there's a story we're going to cover in just a little bit. Paris said we got to cover this story. So this is a CNBC story about the jury finding meta and YouTube negligent in the social media addiction trial. We've been talking about it. The jury went out on Friday and they came back. So now I'm going to click. Don't grumble. What's up with that?

[B]

Tort law is beautiful, actually.

[A]

I can just do Control U, can't I? Or Command U on a Mac. So let me just do that command. Uh, and there it is. What's up with that? Is analyzing the article. So it's going to give me insights. Not into the ads on the article, I hope, because I don't. Oh, there you go. Oh, this is great. So this is. It's Summarizing other stories, this sidesteps the Section 230 shield, which is an important thing to know kind of analysis that we would want. You don't need to listen to our shows anymore. You can just get all this. This is great. Look at this. Now I can also run a systems analysis. What is that?

[D]

So that is a recommendation. It says, out of all the dozens of mental models that we've got in the power tools drawer, this would be a good one to run a systems analysis of, which means let's look at it in terms of flows and stocks and feedback loops and leverage points. Inspired by a woman named Donella Meadows, who is kind of the foremother of systems thinking, wrote a book called Thinking in Systems many, many years ago. And so it'll write up a little report of a systems view of that article that you're reading and the topic, and give you a little diagram of causal loops and stuff like that. And that's one of, like I said, dozens of different processes. But that one was recommended for that article. And particular it thought it would be a good.

[A]

Oh, I see. So it's smart enough to say, hey, based on what's in this article, you would benefit from a systems analysis.

[D]

Yep. So if you give that a click, it'll go and perform that analysis.

[A]

It's doing it right now. It also asked me, it says, I can give you better results if you tell me who you are and what you're doing. So I said, I'm a podcaster and I'm keeping a track of tech news for my podcast. That helps it, too. Yep.

[D]

So it's gonna. It's gonna then keep an eye out in the background anytime you analyze. So it's going to take that into consideration that the fact that you're a podcaster, you're looking for news, and I'm guessing, I'm hoping that you may also get an alert every once in a while if there's any, like, really important podcasting news. We've got agents monitoring the web in the background, looking at thousands of different sources, and when they see something that might be a risk or an opportunity for you, they run thousands of simulations to say, how might this news intersect with this user? And do any of those scenarios rise to a level where it makes sense to alert Leo? Like, whoa, this one's important Leo. And now it knows to watch out for that kind of stuff for you.

[A]

I can even enhance this. I see this link drawer. I can tell it. I'm working on a project. This. You could use this Paris, working on a project right now, questions I'm exploring that I'd like some answers to. So it would kind of be keeping an eye out for that kind of stuff.

[D]

If you say, these are my questions I'm exploring. Every time you analyze a page, it'll check to see if there are data points or evidence that might support one decision or another. And if there is, it'll give you a little alert and you can say, oh yeah, save that one to the decision. And as you then collect them, they've got all the citations and all the data points. Then you can hit synthesize and it will give you a report synthesizing all the data points you saved and links out to the original source articles.

[A]

Wow. You know what I love about this is this is a really good practical example of how AI can be very specifically applied to a specific kind of need. And I think more and more I'm thinking that's kind of what AI needs, what AI products need to do is address specific needs. So then a user can look at it and say, instead of saying, I'm sure, Paris, you had this experience, you sit down at Claude Code and you go, okay, now what? Right, what do I do next? This is particularly tuned to do a certain thing and somebody who's obviously used a lot of AI and understands how to get the most out of AI in certain areas has created something that is going to be useful to you in a very specific way. I really like that.

[D]

Yeah. One of the things it does is it'll look up scientific research. There's a button called Find Science that will go out and look at peer reviewed journals to see what the latest science is relative to the claims found on the article. And it'll say, okay, the science either does or doesn't support the claims in what you're reading.

[A]

Very interesting.

[B]

I assume that like all sort of browser extensions that can do the stuff that has to have the read the ability to read everything that's on your screen. Where does, what happens with that data? Does that, is that stored anywhere? That's always the question I have whenever I download.

[D]

Yeah. So I decided to not allow it to read everything on your screen, to only read when you click and invoke it. And that's a, that's a part of the security settings. When you install it, you'll see that there are five pages or five sites in particular. When you're on Wikipedia, YouTube, Substack, Reddit or Arvix, it will pop up a little notification that says, we notice you're on one of these pages that would be particularly useful to analyze with what's up with that? And you click here to save it. But otherwise we don't analyze what's on your page. And when we do that analysis, all of the data gets stored either on your browser, in your local memory, as an extension, or as key values up in Cloudflare, because I'll tell you, the trippiest feature, it requires that. So a little while ago, the Department of Energy put out an AI challenge where they had 26 RFPs for AIs that they wanted to see built. And one of them was for AI that could discover long causal claim chains in circumstances of dramatic uncertainty. Apparently, in biology, causal claim chains are a thing to help measure the impact from cellular level to ecosystem level or whatever. And I said, we can do that and what's up with that? And so now every time you analyze an article, it picks up any claims that are made, like A leads to B, and it saves that up in your Cloudflare as a key value associated with your device. And then later, weeks later, months later, when you read another article that says B leads to C, it says alert, Alert. A chain has been discovered. So the way to describe that,

[A]

it

[D]

augments memory, it augments perception. And I'm positioning it as a performance enhancement technology for people who think for a living.

[A]

Really interesting.

[B]

That's fascinating.

[A]

Yeah.

[D]

I think about in sports, you know, people say sometimes if you take the, like, the great athletes from out from history and you were to pluck them out of history and drop them into. Into the league today, how would they do? Well, it might be kind of tough because despite their. Their skill and their effort these days, in. In sportsball, no matter what the. The sport more or less, there's game tapes, there's analytics. All the athletes are super informed and super optimized. And in this weird J curve of like, compounding change and this wild world we live in right now, I think that all of us who read, write, and think for a living need a toolbox to help level up what we're doing. I wanted one for myself and I wanted to offer that to others as well.

[A]

Yeah. I wonder how much of your experience as a journalist has informed this, because it really feels like an ideal tool for a journalist.

[D]

Yeah, well, you know, I.

[A]

Are you scratching your own itch? In a way?

[D]

Definitely. Well, I'm a big fan of Josh Waitzkin's book, the Art of Learning. He was a child chess champion who gave up the spotlight and then ended up becoming A martial arts champion. And he talks about how.

[C]

I'm sorry, that's quite a switch.

[D]

Oh, it's a great book. He was subject of a movie as a kid and a wonderful book. I read it once a year. And he says that experts in lots of fields, whether it's chess or martial arts, they tend to do two things. They've got an intuitive sense of pattern detection patterns and anomalies. And as a journalist, I too, I would like scan over, you know, link, link, link, link, link, RSS feeds and have an intuitive sense to be like, oh, that one, that could be interesting. I'm going to stop and look at that. And then the second thing that experts and athletes in various fields often do is have a practiced routine, you know, steps that they would take in a sequence, a play or a playbook, a book of plays that they would run. And so what's up with that? Offers that kind of intuitive pattern recognition in the, here's what's new. And then it's got these automated playbooks of sequences because all these mental models tool, you know, there's dozens of them. But you can say, just give me a plan too. And it will say, here are four or five different reports you should run. Just click here and we'll run them for you. And it runs them in sequence and then gives you the three most important details discovered.

[A]

This is the movie Searching for Bobby Fischer, which was a great movie, but I didn't know that he went on to become a martial arts champion. That's hysterical. Well, I'm excited, Marshall. This looks like a really useful thing for us. I'm going to sign up right away. Very, very cool tool. Did you vibe code this? How did you create it?

[D]

I did. For the first time in my life, I didn't go hire other people to write software.

[A]

That's kind neat too, isn't it?

[D]

Yeah. And I have, you know, I regularly ask, let's do a security audit here. What do I need to account for and, and, and fix things up real smart. And so I, I think that, I think it's pretty solid. And of course it gets checked by, by Google as every release as it goes through the Chrome store and has that benefit as well.

[A]

Very cool. What's up with that? What's upwithat app? If you want to see the website, there's a good demo there. You can see all the things it can do. And it's also an extension available in Chrome or Firefox. Marshall, stick around because there's a lot of AI news and it's nice to have another expert on the panel with us. Paris Martineau is also here. Now you may open your microwave door and.

[B]

Oh, it's been opened. The grits have been retrieved.

[A]

It's like Al Capone's vault. Is there anything in there? Grits. Grits. Really?

[B]

Listen, you know, we had, I was like, we probably got less than five minutes for me to cook something to eat.

[A]

I can get cook in five minutes in the microwave.

[B]

You know what you can get are instant grits, which cook in a beautiful 3 minutes and 33 seconds. You get a quarter cup of grits, you get a cup of water, then you slap some butter salt.

[A]

Are they white grits?

[B]

Cajun seasoning? Yeah.

[A]

Oh, yeah.

[B]

I'm a big grits fan. I was texting Jeff this earlier.

[A]

I'll never forget.

[B]

Kind of a perfect food. And the grits you're thinking of that have kind of a bad texture. That's because most diners, I don't think, make grits correctly. I think I make better grits than the average diner by far.

[A]

Now I want grits.

[B]

Everybody should have grits. It's a perfect food to have in your fridge when you. Or in your cabinet when you need to make something quickly in an ad break on a podcast.

[C]

Salt, Hank.

[A]

Back in the day we at Tech tv, we did an appearance in Birmingham, Alabama, and we went to a very fine country home there and had breakfast and they made us cheese grits. And those cheese grits sat there all day long in my tum tum. I'll kind of never forget that experience. It's a lasting breakfast, let's put it that way. Well, go enjoy your grits. We're going to take a little break. Jeff Jarvis is also here and we're so glad to have Marshall Kirkpatrick with. It's been a long time. It's great to see you, Marshall. I'm glad you're doing well too. That's fantastic. Out systems now. This is timely. The number one AI development platform, OutSystems helps businesses bridge the enterprise gap to their agentic future, where the constraints of the past give way to the unlimited capacity and scale of AI. OutSystems enables companies to build AI agents that can actually do work, such as take actions, make decisions and integrate with data rather than just answer questions. Outsystem provides the only AI development platform that is unified, agile and enterprise proven. Let me explain. It's unified because you build, run and govern apps and agents in one platform. It's agile because you can innovate at the speed of AI importantly without compromising quality or control. And it's enterprise proven trusted by enterprises for mission critical AI applications and durable innovation. Outsystems is the secret weapon behind the world's most successful companies. They're not just for small apps. They're for the massive complex systems that run banks, insurance companies and government services. Outsystems even helps companies with aging IT environments bridge the gap to the AI future without a rip and replace nightmare. Outsystems provides the safest and fastest way for an enterprise to go from yikes, we need an AI strategy to yeah, we have a functioning AI application. Yeah. Stop wondering how AI will change your business and start building the agents that will lead it. Visit outsystems.com TWIT to see how the world's most innovative enterprises use Outsystems to build, deploy and manage AI apps and agents quickly and cost effectively without compromising reliability and security. That's O u t s Y-S-T e m s.com twit to book a demo outsystems.com Twitter we want to thank Outsystems for supporting intelligent machines. So we've been talking about the big trial in Los Angeles. You remember Snapchat and TikTok both settled out. The plaintiff was a 20 year old woman who said I got addicted early on. I think it was the Instagram primarily, but in general to social and as a result I've had a terrible, terrible life. And, and it's their fault.

[B]

Come on, you're, you're perhaps describing this in a slightly disingenuous way. She began using YouTube at age 6, Instagram at age 9. She justified that she believes social media led to depression, body dysmorphia, anxiety, suicidal thoughts, self harm.

[A]

Yeah, terrible, terrible.

[B]

Cutting herself at age 10. All of these, the jury ended up answering yes to every question that it was asked on negligence and finding failure to warn, voting 10 to 2 on each claim for each defendant.

[A]

Now I have to say the thing,

[B]

the thing that I think is interesting about this before you poo poo all over this is unlike other lawsuits which have all easily been dismissed Due to Section 230, this is one of the first bellwether cases in this giant MDL litigation which has like I believe hundreds if not thousands of lawsuits that are all trying to apply this like novel legal approach that instead of using any of the normal ways to sue a tech company, they're arguing basically it's a product liability or personal injury case. They're arguing that this was Negligent design design and has nothing to do with the actual user content. And in regards to this case, they're saying that face OR Meta and YouTube executives knew that there were. Their products were harming or potentially inducing indic. Addictive behaviors in children and specifically like very young children. And they did not take adequate steps to prevent their products from causing foreseeable harms. And I think that's, I mean that's part of the reason why I think the jury ended up finding like deciding in the plaintiff's favor here is it's not as simple of a case as we normally see with these sort of things.

[A]

Yeah.

[C]

That caused all of this with this woman is a pretty simplistic view itself.

[A]

Yeah. We don't know re. I mean no one can know what caused.

[C]

Yeah.

[A]

Her issues and there's plenty.

[B]

Well, they can because that's exactly what this case was about. Where they spent weeks and weeks and weeks about this very specific thing actually is what, what they just decided was exactly that deliberation even took 44 hours. It took nine days.

[A]

Yeah, they took a while. Well, because they also weren't agreeing, I should point out. Yes, they've, they, they've ruled against them, but they didn't give them the worst damages ever. $4.2 million to Meta. That's combined compensatory and punitive damages. YouTube, 1.8 million. These are tiny compared to the revenues of these companies. Interest for one day. It was well worth it.

[B]

I don't think that's correct because this is a bellwether case that that's both being used as a like in an Indian idiomatic phrase as well as a like legal literal sense. So this is a precedent.

[D]

Yes.

[B]

No, no, no, no, no. So this is part of an MDL which I believe is multi district litigation. It basically. I wrote about this a couple of years ago, so forgive me if my knowledge is a little out of date, but at the time there were hundreds, if not thousands of cases like this that all ended up getting grouped under this MDL where they were all about social media addiction litigation involving a handful of these companies and kind of taking on this same novel legal argument where they're trying to argue defective design and kind of neg or negligent design.

[A]

And they knew about it, but they didn't care enough to fix it.

[B]

I mean they're basically making the same sort of argument that you see in tobacco or asbestos cases. But so as part of what you do in a big case like this where you have thousands and thousands the comp, Facebook, you know, Meta, YouTube, Snapchat, TikTok. All of them were like, we don't want to sit here and litigate a thousand of these cases individually. That'll be very costly. Instead we're going to select a handful of literally called bellwether cases. I think it's eight or something in this case. And that's going to be used to determine the future of all of this litigation. So how. This is the first of.

[A]

It's actually not the first because on Tuesday a New Mexico jury.

[B]

I don't believe that's. Is that part of the same multi district. No, that's different.

[A]

But. But so it's not a bill. This was the case brought by the state attorney general in New Mexico. They found similar liable for violating state law by failing to safeguard users of its apps from child predators. And that fine was $375 million. A little bit more painful for me.

[B]

Yeah, that was different.

[A]

It's not the same thing.

[B]

So this was their civil penalties under New Mexico, like they're. What they've gotten them for is willfully violating New Mexico's Unfair Practices Act. And the attorney general who is very, I've spoken to him, Raul Torres, he's very up to date with obviously all of this and has been following the MDL quite a bit. But basically this is slightly different. Like undercover officers posed as children on Facebook, Instagram and WhatsApp. They ended up kind of having a sting operation. It is. They are kind of going to be doing a phase two bench trial as to whether Meta created a public nuisance, which might require platform design changes. But it's. It's slightly different. This L. A case that was decided today is the first of the official bellwether cases that can decide the fate of all this future litigation. And I mean Meta and all these people of course are saying they're going to appeal this decision. But I do think this has been a huge movement we've seen over the last couple of years. I profiled one of the attorneys and legal like groups that have been kind of pushing this movement. It's called the Social Media Victims Law Center. I profiled them a couple years ago. But this movement has all been happening with only like one or two real decisions in the favor of this being a viable legal strategy. This is the first time this has been a huge victory for this and it's going to open up the floodgates.

[C]

Even guess on appeal. What are you reading about?

[A]

Appeal would be different because you're now talking to a panel of judges as opposed to a jury. But I can see why.

[D]

Consumer protection perspective, from a historical perspective, it sounds like this is a bfd.

[B]

This is like, yes, this is a huge. I think this is a huge deal. As someone like, I had been like two and a half years ago, I was like, man, when one of those first bellwether cases, when, when this case comes up, it's going to be huge. And I'm, I mean, I'm not surprised given that it was a jury deciding it. And I think, like, juries are going to be more easily swayed, obviously. And it's a compelling argument. Yes. But it depends.

[C]

He was just waiting.

[B]

I know, and I'm really proud.

[C]

You're welcome, Benito.

[B]

I mean, I, I think that it depends on who ends up seeing the appeal, what judge in court it goes to. Because the thing is, this is not the only. I say this is a huge deal, and it is, but this is not the only case like this that has been decided in plaintiff's favor. This all kind of started with a case called Lemon v. Snap 2019, which is probably one of the reasons why, I mean, I obviously don't know any of this. I don't know, maybe one of the reasons why Snapchat was like, we got to get out of here, we're not going to trial is it was a really interesting case where a couple of teenagers, I believe, ended up dying, if not all of them, in a high speed crash, because they were. Snapchat had rolled out a speedometer feature. You guys might recall, if you ever used it back in the day, where essentially it would show you how fast you were going. And I think there was something going around where people believed, like, oh, if you got to 200 miles an hour, you got a special thing. It was a big deal on Snapchat. Everybody was trying to see how fast they could possibly go in cars to get this sort of Snapchat response from the app. And these kids ended up going like 200 miles an hour or something or some crazy amount, crashed their car died. And then Snapchat faced a suit saying, hey, you should have realized when you're, you know, designing this feature and hearing the way people are using it, that it could put people in harm's way and maybe consider your design more.

[C]

And they ended up being Snapchat design that particularly. Or was it.

[B]

Was it. No, it was not a user designed feature. It was Snapchat designed and rolled it out.

[C]

That's pretty.

[B]

And it was, it was a really landmark decision in that, like, it was the first kind of crack in section 230 being the go to defense for all sort of.

[C]

Well, in that case. But if Snapchat created section 230 wouldn't have been a defense.

[B]

I know. And that's why Section 230 wasn't able to apply. And that's why they ended up being found liable for defective or negligent design. But then this started opening up this whole new legal area for a lot of these companies or a lot of litigants where they're saying, well, there are other aspects of these platforms that are design decisions and how can we try and suss out whether or not those are defective designs or not? And this is. We're gonna, yeah, we're gonna see how it all shakes out, but it's gonna be very interesting.

[A]

So the plaintiff's attorney brought in a jar of M and Ms. Saying, imagine this is the revenue of these massive companies. If you don't give them a large punitive decision, take out a handful of M and Ms. If you just take out one M&M, they're not going to feel it. The jury did not buy it. In fact, one of the jurors, the New York Times quotes one of the jurors who said they shied away from giving the plaintiff a huge sum. We wanted to focus on the future and what teens and children would be subjected to in the future. They didn't want to punish these companies, but they did want to make it clear the companies were responsible. So to your point, they wanted to set a precedent.

[D]

I feel like this is a paradigm level incentive problem that that was an opportunity to make a meaningful intervention on. As a founder who has raised money from investors, like one of the things that investors classic statement that people investors say to startups is if you have a choice between building a vitamin and building a painkiller, always build a painkiller because it'll, that's what'll sell. So like the incentive to build addictive short term optimized stuff is like baked into the whole system.

[C]

But this is the problem where the money is addiction. Addiction. Addiction is a truck. And it goes back to, you know, I've done before to novels and so on in the earliest days of the Internet. And I write about this in the web. We weave at length this notion. Immediately there started support groups for addiction, definitions of addiction that were ludicrous. One Columbia professor started a joke group around addiction and people took it seriously, didn't know what to do with it.

[A]

They were addicted to jokes.

[C]

No, he started it saying this was a joke. He thought the argument for addiction was so absurd and people glommed onto it.

[A]

Well, but AA works. That model can work.

[C]

Research is not backing up addiction. The research does not back up addiction. So that's. That's an issue here. That's why this was a jury's emotional response.

[A]

This was my point, which is you can prove that cigarettes cause cancer. You. It has been proven. You can prove asbestos causes mesotheliomia. This has been proven. It is much more difficult to say. You know, she says, she said in her testimony that at a very young age, at age of six, she found she turned to these platforms because she was bullied and lonely. And she. And it was a creative outlet for her. And all of which, you know, I believe completely. She says that's what caused my problems. Of course, Meta's defense was that her mental health issues had other causes. They said familial abuse and turmoil. But the difficulty is you can prove cancer is caused by cigarettes. It's very hard to prove that mental illness.

[D]

I mean, do we talk about Gabor Mate here? Does Gabor Mate's name ever come up? The Canadian doctor and author who. Who argues. If I could summarize that, that addiction is basically a coping mechanism, an unhealthy coping mechanism for trauma.

[A]

Right. That's. That's often what is certainly what they say in the 12 step programs, things like that.

[B]

The. One of the experts who testified, a Stanford addiction medic medicine expert, testified that social media reward mechanism mechanisms activate the same neurological dopamine pathways as gambling and substance addiction. Oh. But I mean, I don't know that we necessarily want to go down this thing.

[C]

I think like the dopamine response is the same as wearing glasses.

[B]

The thing is.

[A]

Enjoying a good book.

[C]

Yeah, yeah. It's. That's.

[B]

That's such the thing that ended up sinking. I think it for meta and YouTube is there was just a lot of internal documents that showed Meadow was based. I mean, was trying to hook young users and get them as young as possible. I think one of the quotes is, if we want to win big with teens, we must bring them in as tweens. And it's like there, I believe I remember at the time a lot of these documents were coming out because there were some redaction errors that led to some of them being shown that we shouldn't. Where essentially they said, like, yeah, we're trying to optimize for maximum amount of pickups a day. They would be in some cases optimizing for kids. Picking up their phone throughout the evening. I mean, this is. I think it's not coincidental that as all of this litigation has been going on and as discovery has been going on over the last couple of years and these kind of appalling documents and revelations are coming out that Meta's rolling out teen accounts, as has there's been YouTube Kids accounts have come out with stricter parental controls. I think that, I don't know, it does feel like a bit of a paradigm shift.

[D]

It's not a fair fight between like the, the billions of dollars and all the expertise on one side against like some traumatized six year old kid and being like, hey, you didn't have to go for the, you didn't have to take the bait with the app and keep picking it up.

[A]

Right.

[C]

But there's also the trauma.

[A]

What about her parents?

[C]

There's also the traumatized young child who feels very alone, who turns to social media and turns to the Internet because it gives them the SOB that they otherwise wouldn't have. There's tons of research about that. So what's happened in Australia is what's being taken away from children is going to make a lot of children worse because of this moral panic and moral entrepreneurship. And you have people like Tristan Harris who've now moved on from this and moved on from social media. So he has the AI doc coming out next and he's going to argue how awful that is for all of us. And he knows best for all of humanity. And it causes more problems potentially in the long run because it's built on assumptions and fears, not on research and data.

[A]

Well, clearly, I mean, look, Gabbar Mate notwithstanding. And I don't know, even if the jury would deny Gabor Mate's thesis, you know, maybe she was filling some hole, you know, caused by trauma. I think the jury was persuaded mostly and ruled this way. Mostly because they feel like meta and YouTube intentionally cultivated algorithms.

[C]

I think that's where Paris's arguments are. The.

[A]

Yes, I'm agreeing with you.

[C]

Is that the internal material.

[A]

Yes, it was persuades. They did this on purpose and they deserve. Well, it's interesting. The jurors didn't want to punish them. Exactly, but they needed to be. They wanted to be a wake up call that you guys need to fix this. This is not that you are responsible for creating something this addictive. It also may not be what sent KGM down this road, the plaintiff down this road in the long run. But she's. You're right, Marshall. She had no defense as a six year old against this, you know, intentional cultivation of a very, very sticky product. I just worry that it could be extended to other things that are equally enjoyable. Not everything's heroin.

[B]

One of the things that is going to come of this is like bellwether trials, especially in these sort of mass torts, they often end up being used to kind of set the tone and speed for global settlement talks like with like opioids or 3M or things like that out of it.

[A]

Yeah. And tick tock got out of it.

[B]

Once you started to see the bellwether cases being decided in favor of plaintiffs, these companies are like, okay, well I guess we'll just get out ahead of this, maybe start doing some global settlement negotiation, accelerate that process. And I think that, I don't know. This has obviously been a problem that the big social media companies have had to contend with for many years and it had not really ever. It seemingly hadn't risen to the level of concern to result in any product changes or care towards providing parents with tool. You said earlier like, well, what about the parents? Until recently, parents didn't have any tool. You could either have your kid have an Instagram account or they couldn't.

[C]

They didn't really have the other working argument.

[B]

Yes, parental tools until all this stuff.

[A]

But who gives a six year old a phone and allows them to spend hours a day on Instagram? That's the parents fault. I'm sorry.

[B]

I think there can be compelling arguments made on both sides. I think, yes, that's wise also, you know, if you're a parent that's working three jobs, you're have your kid who's screaming and you have to be on in a zoom meeting without noise in the background or else you're going to get fired and maybe lose your housing. Yeah, you're going to want to hand your kid.

[A]

I'm sure there were reasons and I think what the jury, what this is, what I'm saying is the jury might even know about those reasons and might even think the parents have some culpability and they might, I think the jury

[B]

would know about those reasons given them.

[A]

That's their reason. They saw the testimony and they probably, you know, I'm sure the defense told them about Gabor mate, but the jury, what the jury really is saying here, and I'm really curious what the impact of this legally is, is doesn't matter because the company's created a product intentionally to cultivate this kind of, you know, compulsive use. So Paris, so does this mean that somebody else can Make a lawsuit and then it's not a precedent in the sense that a legal president but they could bring this case up and say look what the jury did in la. Is that why it's valuable? I don't. What is the. What is the. How does the strength in those cases.

[B]

This in my opinion is that it's a high profile incident showing of an incident where this novel legal argument has resulted in damages being assessed against one of these companies and result being found. It's, it's a, it's an example of success. And I mean more practically this is one of the bellwether cases for this multi district litigation. And so it will have a very profound and direct impact on all of those. Like this is 1/8th or whatever how many bellwether cases there are. Last time I checked it was 8. This is 18 of the way to deciding what's going on with these thousand.

[A]

But does Bellwether have a legal.

[B]

Yes. Like literally when you have a. It was part of my understanding of it is so looking through this multi district litigation it's a truly thousands upon thousands of like things in the legal docket. It's all these different cases that have kind of all been merged under one for court consolidation. And as part of that there was a long back and forth period a couple of years ago where the plaintiff's attorneys, the defense attorneys, the judges all kind of went back and forth and they eventually settled on this handful of. It was eight or 10 cases. They're choosing as bellwethers that are. They think are represent. They both agree are representative of the class plaintiffs. Interestingness, the plaintiffs and the defense, they both had to agree. Okay, everybody had to agree. The judge had to sign off and then it's obviously it's not those are decided. Whoever gets more the rest are decided. But when it comes to settlement discussions and you know the defense kind of and both the plaintiffs and defense taking a sense of how the rest of this multi district litigation is going to be resolved, it's eventually going to get resolved in some sort of settlement talks that are going to be decided either in favor of the plaintiffs or in favor of the defense in some mass tort sense. And the bellwethers are used as bellwethers to determine which way they think these

[C]

things are going to go after appeal.

[B]

I mean. Yeah, but these early decisions I think are notable.

[A]

Well, it's certainly a big deal. It's not a lot of money but it's a big deal.

[C]

It's also fascinating to me. We've kind of moved past social media as the issue. Everybody's talking AI, AI, AI. And social media is kind of yesterday issues.

[A]

Well, that's how it works.

[C]

Yeah.

[A]

The legal system and media society. Yeah, yeah, yeah.

[D]

Well, I mean, something I do better than social media went.

[C]

Right.

[B]

I was gonna say talking about AI like they're. So as I was doing some tweeting about this today and ended up shouting out my story from 2024 on one of the. Basically a longtime asbestos lawyer who is the head of one of these firms. And as I was doing it, I was like, oh, I ended up doing a lot of research into tort law. And that's how I ended up discovering my favorite museum in the world back in the day, the American Museum of Tort Law. And I was on their website because they're obviously shouting out this. Because this is novel.

[A]

Well, because the lawyers are going to make the bulk of the money is why they're shouting.

[B]

That's. I mean, no, the. In tort law, like it typically lawyers fees around 30% because what Casey agreed to. Yeah. But on the American Museum of Tort Law website, they had this thing from JD Supra which said, can social media or AI be a defective product, Product liability and mass tort law model.

[A]

That's interesting.

[B]

And that's why I realized this, what we've been talking about. There's a parallel wave of litigation against AI developers or against, like the. Against kind of chat GBT, obviously against OpenAI, character AI and similar things. Trying to argue kind of. I think it's obviously a bit more complicated given the, the nature of their platforms are different. But that, you know, in the case of the character AI chatbot that quote unquote urged a child to kill them.

[A]

Oh, yeah, there are a lot of those.

[B]

Yeah, there's a lot of them. They're like, is that defective design?

[A]

So this is consumer product law. That, that's, that's related.

[C]

That's the issue. Right?

[A]

Yeah, yeah. That's interesting.

[B]

It's productivity, baby.

[C]

And it's going to come to AI because of the whole talk about, are great guardrails possible? Can they do anything? Is it a fool's errand? Do they if, you know, they argue that they can take over all mankind, but they can't cause a simple change to avoid a simple problem.

[A]

Right.

[C]

The lawyers and the AI company should be perking their ears up and they're going to be destroying a lot of documents about now.

[B]

Yeah. And I just think that the like, like really important, highest level takeaway Here is that all of this just seems to be another piercing of this long standing assumption that I feel like the tech industry has long had that like if you're a tech company everything gets broad immunity from sort of tort exposure and a lot of lawsuits. Because section 230 like that is genuinely a significant amount of lawsuits involving these sort of, of companies and platforms end up just getting dismissed first thing. They're like, Ah, section 230. You can't. It's not in this case though.

[A]

It just doesn't hold because it isn't content that was posted by its users. That was the issue. Their issue here is how that content was displayed, the algorithms used to display that content.

[B]

Yeah, it's a product design and so

[A]

it's just a product design. This is a. Yeah, that makes sense.

[B]

A segment of law and litigation that somehow hadn't really a mortgage full heartedly, wholeheartedly until recently. And I think that's just very interesting precedent wise.

[A]

I think what Jeff and I are mostly concerned about is, is going forward if this is going to extend liability into other new technologies and put a chill on them, not just in a

[C]

way that may be impossible to follow.

[A]

Right.

[C]

That, that you, you and and so it's. It goes Back to Section 230 to this extent is that without the shield that was provided, every online company would have said you can't talk here, you can't do anything here. Nope. Nope.

[A]

Am I liable because I don't want

[C]

to be liable because I can't get insurance? My lawyers aren't going to let me and so we're going to have an Internet of PDFs.

[A]

Am I liable for creating shows that are fabulously addictive and that people have to listen to three hours every single day?

[C]

We're the ones that were addicted. Paris and I are filing a class action suit against you later today.

[D]

There's got to be somewhere in between, you know, an Internet of PDFs and I could shoot a man on 5th Avenue and still be elected president.

[A]

We are still looking for that. Somewhere in between. Bye bye Sora. In other. I hardly knew ye that that app, which was both for test.

[C]

Sayonara.

[B]

Sora.

[A]

Sayonara sora. OpenAI has announced it's going to shut down Sora which is ironic because Disney agreed to give them a billion dollars and license their characters to Sora so people could use Disney characters in their Sora videos. I guess Disney says, yeah, well never mind if you're going to shut it down. Martin Piers and Anne Gahaine your former colleagues at the information say OpenAI wrong foots. Disney. Well, Disney must be so glad it committed a billion dollars to OpenAI.

[C]

No, a billion. They're not paying the billion.

[B]

They said, well God, that's such a Martin Piers headline.

[A]

I'm not going to delve into that. Anyway, I think OpenAI, we were talking before the show about why OpenAI did that and there are a lot of good reasons. OpenAI was, we've had the story is focusing more. They've decided they looked at what the enterprise revenue generated all of a sudden generated by anthropic and saying, hey, maybe this chatbot thing isn't where we should have spent our energy.

[C]

At the same time, Walmart has pulled out because OpenAI's shopping was not performing.

[A]

Yeah, look at this graph.

[C]

Microsoft. OpenAI is not your get along company.

[A]

This article, the AI spending flip. This is AI model share of first time enterprise customers. OpenAI declining dramatically while Anthropic increasing dramatically. They flip flopped. OpenAI, you know, a year ago had 60% of the enterprise market to Anthropic's 40%. Now it's Anthropic 73% to OpenAI's 26%.

[B]

So it's a, hey, that's probably one of the reasons why OpenAI was offering private equity firms what is a 17.5 free return rate this week guaranteed.

[C]

Yeah, it wasn't just a high bond

[B]

guaranteed, which is, I'll take that pretty good. I mean that's, it's very interesting.

[A]

It's their bank, they can't pay that money anyway, so.

[B]

Well, part of what they're saying is I assume is they're hoping to get it back by having all of the PE firms get their portfolio companies to do expensive enterprise subscriptions with OpenAI and then that's where they're going to be paying the private equity firms. That's 17.5% back. But it's just like my first question is like who, what, what portfolio company in the year 2026 doesn't already have a enterprise AI subscription? Probably not many.

[A]

Right. I wonder what's going to happen to Jony Ives 6 billion dollar AI device. That's another distraction, isn't it? OpenAI has a web browser.

[D]

An analysis of the of the show transcripts over the last 18 months found that, that OpenAI's financial stability used to be a major topic of conversation here but has been on the decline now for some time.

[B]

I wonder why, you know, has there been any other changes?

[A]

Hello, Claude. I love you Claude, but I'm not alone. And I think you've seen this also. Probably Marshall is among the nerds. Claude has just got all the mind share right now.

[C]

Well, go back to OpenAI. Is it in trouble?

[B]

When is it not?

[C]

Well, yes, I think it's in more. Is it in more trouble? I mean, it's ipo. There's no possible purchase. No one's going to buy it. It's desperate for desperate.

[A]

Microsoft's already threatening to sue them, so that relationship soured.

[B]

What? Why? Why are they threatening to sue?

[A]

Because OpenAI did a deal with Amazon and Microsoft wasn't too happy about that. They said that's a violation of our.

[B]

OpenAI is like, but I need all the money in the world so I could maybe make a product that works.

[A]

So Walmart says that checkout converted three times worse than the website.

[C]

Yeah. What's the strategy behind OpenAI?

[A]

Now?

[C]

Anthropic is clearly. It's got enterprise B2B coders. It was ready for Claw. Even though they pissed off the Open Claw.

[A]

They do own Open. They do own Peter Steinberger. They bought Open Claw.

[C]

But that's, that's like Mark Zuckerberg buying Mop book. It was meaningless. You didn't need to buy them.

[A]

Right. And Open foundation and is not owned.

[C]

Exactly. So that was, that was, that was a sign of desperation in both cases.

[A]

You know what? I think investors are not going to quickly turn their back on OpenAI because what are these guys putting my mind in? The mind of the billionaire is not always a good thing. If only, if only. But I think what I imagine is that they say, look, somebody's going to come along with AGI and it's going to be the upside.

[C]

That's your first.

[A]

What?

[B]

How? Okay, sorry, stop there.

[C]

Right.

[A]

Somebody telling you.

[C]

Let's just grant that.

[A]

I'm just telling you what the Jason Calacanis of the world think.

[B]

And they're always right.

[A]

Well, actually Marshall took some money from Jason, so we should ask Marshall. But I think they're thinking there is potentially a massive upside to AI.

[C]

For those of you listening, you should have seen the grimace on Marshall.

[A]

There isn't a massive upside to AI. We don't really yet know. I mean, at the minute minute it becomes clear, oh, these guys aren't going to win. Then they'll be like rats leaving a sinking ship. But until then, they're going to hedge their bets. So I think there's still going to be plenty of money.

[C]

Nvidia was going to invest 100 million billion. And then they went more like 20 or whatever it is 30 then.

[A]

So maybe the rats have started to leave.

[C]

They've started leaving. Yeah.

[A]

That's when it will happen is when. When investors say yeah, it's not going to be the winner's not going to be open AI.

[C]

So I don't think that's clear trouble.

[A]

Their models are very good. 54 is very good.

[C]

I go back to the is their vision and business model. We know what Google's is. We know what Amazon's is kind of we definitely know what anthropics is these days. What is OpenAI's mission and vision? I don't know.

[B]

Make money but question mark.

[D]

I mean they want to, they want every consumer to spend $20 a month on their apps and all the enterprises to, to pay for their APIs to.

[C]

There are many companies that are both consumer and enterprise and I think OpenAI has been in fact I used to, when I taught business to my students said you really, when you're starting out, you can't be both because you often end up competing with yourself or your customers as a result. And so OpenAI was unquestioned consumer brand. Microsoft said both it's pushing after the enterprise. Microsoft has. Yes.

[A]

So according to this is an information graph going from October 2023 to last month annualized revenue. OpenAI's is still on top with 25 billion a year. Anthropic though has gone from practically nothing to 19 billion a year. And they're growing at which looks like almost exponential rate. So now remember, revenue is not profit. None of this is profit. But I'm sure the investors look at revenue as one of the by the

[C]

way, getting rid of sort of question. Getting rid of Sora doesn't mean they're getting ready video. They're just getting rid of that product. Sora.

[A]

That was I think what I think

[B]

they're getting rid of all weren't they? Are they stopping video generation?

[A]

It's not just the app. It's. It's there, it's. The whole Sora model is gone.

[B]

I can't help, I can't help but see that as videos really expensive. And much like Marshall said, their current business model is we want to get everybody to pay us 20 bucks a month or pay us for an API like enterprise subscription and neither of those. Both of those you're losing a crap ton of money too. It's like at some point you run out of people to ask money. I mean they're, they're running out of people to ask for money to the Point where they're asking people who are getting them in legal trouble with the other people they've asked for money. It's. It's seeming. It's seeming like we've got a little Ed Zitron going on here, you know,

[A]

and there's another argument which we also talked about before the show. There's another argument which we talked about before the show, which is that they also may just want the GPUs and computing. They want the compute to dedicate it to something else. More revenue forward. They also know, which we don't know how many people were using that app. I bet it was down to a very small number.

[B]

You bet. Why? Just a couple of guys go back in the Notebook LM and find all the times you're like, sora's the future. Everybody's gonna be using this app. It's the coolest thing ever.

[C]

Hollywood's still shaking. I put it in the chat. I put it in the chat.

[B]

Yeah. Wait, I thought this was gonna stop Hollywood. I thought we were never gonna have a need for an actor ever again

[A]

because Sora is gonna make sure

[C]

the point of being a fad. Sora was a gimmick.

[A]

Yeah.

[C]

So in response to a post saying that OpenAI published a blog about safety standards for Sora, and today they scrapped the feature completely. Ed tweeted, this is something the company does when things are going well, a

[A]

little too happy to celebrate.

[B]

We need to just get a little Ed sound bite so that we can put that in with a little. Maybe SORA to decide one and be

[A]

like, would you ask Ed to record something like, I told you that would happen?

[B]

Or what did you think would happen?

[A]

What did you think would happen?

[B]

These are a scam.

[A]

Yeah. You know Ed. Well, get him to record Ed's soundboard.

[B]

Yeah, we actually do need an Ed soundboard.

[A]

Yeah.

[D]

Is OpenAI too big to fail? Like, is that their. Is that their strategy? Like, get too big to fail?

[C]

Like.

[B]

Well, that was my sadness.

[A]

You know why that might be the case, and probably one of the reasons they stepped up. When Anthropic said, we're not going to do this Defense Department stuff, and they immediately said, okay, we will. Is that's how you get too big to fail? If the government relies on you, if the Department of War depends on you, then maybe you do get too big to fail. Or at least the government has to kind of prop you up a little bit if they're relying on you. I guess the other question is, and you would know about this, Marshall, how Fungible, these models are. It looks like the Department of War was very, very easily replaced. Anthropic, it looks like. Right.

[D]

So Palantir may not be so excited to do this.

[A]

Yeah, Palantir does a lot. Yeah.

[D]

I don't know. I mean, I certainly would be unhappy to lose access to CLAUDE models, but I do have a circuit breaker system in place too. When the API goes red, as happens, there's like a lot.

[A]

Yeah. Happened yesterday, didn't it, Paris, you were saying, is Claude squirreling out on you?

[B]

I mean, it was just an issue where, like every time I would try to use Claude, it would take like three minutes to generate a response.

[A]

Something was going on.

[D]

Well, if you look at the status page of the. Of the CLAUDE API, it's got, you know, green, green, green, yellow, red, red, red. You can see days where there's issues and. And right around when the whole Department of War controversy came up and there was a whole bunch of people came piling in to use it. It's had a lot of red on. On it. And so, yeah, that they're.

[A]

This is current.

[D]

And so my system falls back to GPT when CLAUDE goes down.

[A]

Actually, this is not looking good.

[D]

Oh, that's about what it looked like. Yeah. For a while.

[A]

Yeah.

[C]

Freddy Claude for government out.

[A]

Oh, yeah, there's Claude for government. It's working fine because no one's using it. All right, we're going to take another break. You're watching Intelligent Machines. Paris Martineau, Jeff Jarvis. We're great to have Marshall Kirkpatrick with us, who is a longtime tech journalist, but also, dare I say, avid AI user. Is that fair?

[D]

Yeah.

[C]

Addict. He's an addict, just like you, Leo.

[A]

He has something to tell you, Paris. He put 18 months of our transcript scripts into his machine and has a few things to tell you about me.

[C]

This excites spirits.

[B]

Is this different than the Notebook lm.

[A]

By the way, I talked to Animova, who created that. He joined us on our AI user group a couple of weeks ago, was talking. He did get all of the transcripts in there, but he had to chunk them up to do it. Yeah, yeah, you were right. And he had that graphic, the lovely, many lovely graphics showing what a moron I am. So thanks for that. AI salute. We'll have more in just a minute. Our show today, brought to you by spaceship. Remember, Paris? Secretly British.

[B]

I literally have time on my calendar on Saturday to work, okay?

[A]

We register it with spaceship and it's. I tell you, we shopped around secretly. B R I T I sh Brilliant domain name. Spaceship had the best price for it. If you've heard us talk about Spaceship before, there's a reason it keeps coming back. It's because Spaceship is rethinking how people register and manage domains. And this fresh approach has led to more than. We're not alone. Six and a half million domains under management in absolute record time. That kind of growth comes from giving people what they actually want. Spaceship offers transparent low pricing on domain registrations. Transfers are fantastic. If you're with another registrar, check out what you get when you transfer. And crucially renewals right, you're gonna save all around. It's not just a one time saving and then they jack the price up. That means there's more clarity over what you're paying for over time. Alongside great value, the platform is especially built for flexibility. We did this with Secretly British. You can instantly connect your Spaceship registered domains to Spaceship products like web hosting, professional email, virtual machines. And you can build and test before committing. Because almost every Spaceship product comes with a 30 day trial. I like that. Now if you still want to use third party tools, that's fine, no problem. We did this with Paris. Just point your domain to what you need by updating your DNS records or name servers. You can even use their AI alf to do the hard work so you have the freedom to build your stack exactly how you want. When I realized that Secretly British wasn't going to be a real website for a while, I just pointed it to Paris's existing website and was that easy with Spaceship. Basically Spaceship is the best of every world. Visit spaceship.com TWIT to learn more. Be a great place for your open claw. That's spaceship.com twit. We thank him so much for supporting intelligent machines. Thank you Spaceship. That Saturday. You know what? Don't feel pressure Paris. I don't want you to feel pressured. Secretly British can wait.

[B]

I don't feel pressured. I literally independently of this message, my friend was like we gotta get on our great business idea of Secretly British.

[A]

But don't make it too good because you might get sued for being addictive. I'm just warning you.

[C]

Mm.

[A]

Don't make it too good.

[B]

Do you ever try to pass any laws that make the world good or better in any way? Because then Leo could find any of the bad things and he will say, you shouldn't have it Marshall.

[C]

Have it on the record now. When. When was Paris warned? If this be a problem guys, the

[B]

unit we will be fire and ash in 20 in 10 to 50 years. None of this is gonna matter.

[A]

Really? Is that your deep how deeply held belief?

[B]

I think it's a. I think there's a dice throw chance that we're gonna be fire and ash.

[C]

Did you watch the AI doc already? Is that what's gotten you there?

[B]

No, I just think, you know, you look around, you see the way the world's going, you see the rate at which it's gotten worse over the past five to 10 years. I think it's a reasonable, you know, throw a D20 rolling that one, we're gonna be firing ash sort of chance. You know, you got a tempo percent, but I think that's fine.

[A]

What is that? So D20 is 5% by the way. Sorry?

[C]

A D20 is 5%.

[A]

A D25%.

[B]

Thank you.

[A]

One in 20. Yeah.

[D]

Speaking of numbers, I. I look at billion dollar natural disasters per year in the United states. In. In 1980 there were two inflation adjusted billion dollars. In 2020 there were 28.

[A]

Wow.

[D]

Billion dollar disasters. And then they've stopped since then because the Trump administration said shut down the campaign measuring those. But yeah, it's a, it is quite a.

[A]

So we don't have any numbers after 2024?

[B]

Yeah, we don't have any numbers. We don't even have numbers, Leo.

[A]

It's not happening because we don't know

[D]

is now luckily third parties and independent folks have continued measuring it, but it is up and to the right in, in terms of.

[A]

So Paris, what did you want to know about the last 18 months of intelligent machines? What insights could Marshall give you from his database?

[D]

Well, I can tell you why I thought to ask was because of Paris saying in the last episode, oh Leo, you say this is going to change everything all the time. You were talking about Claude code and coding and. And I said oh, that's interesting. I wonder what kind of history there really has been of that. And so I did put a link just in chat right now. I don't know if you want to

[A]

look, but am I turning into Scoble? Is that what you're telling Marshall? Well, no.

[D]

So Claude's analysis. Claude pulled down 75, you know, issue episodes and transcribed all the thing and it did. It used the word hyperbolic, not me.

[A]

Oh, you're making Paris so happy.

[D]

But it said that on 60% of

[B]

the show, Laporte emerged as a self described AI accelerationist in scare quotes whose enthusiasm intensified over the period making hyperbolic claims in 45 of 75 episodes, though consistently leavened by genuine skepticism, self awareness.

[C]

Yes.

[D]

If you go down to the this will change everything section. It says it is not as simple as Leo being like a naive, you know, boy who cried wolf. Instead, his his claims, while frequent, varied in intensity and were accompanied by self aware qualification and genuine skepticism about specific products and companies.

[A]

I was never a monolith. It says.

[B]

I love these tags. Okay? I love that there's a tag called revolution imminent.

[A]

Oh boy.

[B]

I think we're on the cusp of a pretty big AI revolution in the next year. He said in December 2024. Well, I'm sorry, this is going to be a very few interesting years.

[A]

I'll stand by.

[B]

Oh. He predicted an AI co host within the next five years. I guarantee on Twitt.

[A]

That's because I'm going to make it happen.

[B]

A personal agent on your wrist can change everything.

[A]

Yes, and I've been working hard at that. Listen, even though I erased it yesterday, by the way, your role Paris is as empirical check. Paris's role was to ground the conversation with reporting data and personal experience.

[C]

Good.

[A]

True. Good job. Your signature move Paris was the empirical

[B]

Correction Leo on February 20, 2025. You can't look it up. What do you say you measure your life in now? Let's fill in the blank. I measure my life in blank now

[A]

I measure my life how?

[B]

Just think about your day to day. How do you measure your life in tree rings?

[A]

I don't know. You're gonna have to tell me in what tokens. Oh yeah, that's fair.

[B]

I just think this.

[A]

That's fair. I do. I measure my life in.

[B]

Five days later I am an accelerationist is the quote. 02-07-2025 the first explicit declaration.

[A]

I like how you put this in black.

[B]

The accusation the same episode about musicians silent album protest. The good news is they're all going to be gone soon.

[A]

Wow, that's pretty.

[D]

It's on the record.

[C]

I guess

[B]

this is the most impressed I've been by AI related this show ever because it understands the see how good it is under a thought called guest adulation. Revolution imminent. It says 1 liter of computronium would give you more capability than all human beings together. M Dash Wow. Context. Ray Kurzweil interview Awe at guest claim. I want to drink from that brain admiring Kurzweil's intellect. This is going to be the year of robotics. I do think he understands what I found funny about that interview and I really delight that.

[D]

Well, while you know, this being the case, I did another analysis that visualized the balance between AI autonomous economy and and people being an organizations taking responsibility for AI across the last five episodes of Intelligent Machines. And. And it found that.

[A]

That.

[D]

That you all are consistently advocating for people and organizations to take responsibility for their AI, whether it's high autonomy or low autonomy. There's an emphasis on responsibility here that I am guessing. I don't hear at least. And I haven't analyzed this, but I don't see it on. On the other major AI podcasts, yes,

[C]

they interview CEOs and say, what else have you done that's wonderful lately.

[A]

The other thing Claude doesn't know, by the way, is this. What's up with that, basically, that you used for this?

[D]

No, this was a loop that I was create that my friend at Fleet of Geniuses showed me how to make where I. It's a skill where I said, hey, go pull down 18 months and do this analysis. I gotta go take a shower and eat lunch. When I came back and boom, it was.

[A]

How did it get 18 months in its token context? I mean, that's a lot of data.

[D]

It chunked it out into 15 sub agents.

[C]

Aha.

[D]

And then used Opus to get it all together.

[A]

Very cool.

[D]

Can we.

[C]

Can we get this?

[B]

The Emily. I'm sorry, The Emily Bender episode. It says, mine is always right in brackets about perplexity. Context defined defending AI reliability while simultaneously showing an error. Oh, that's got your ass.

[A]

Did I show an error? Did it show it? No, you did, if you recall, during the. No, no, the error was mine.

[B]

Screen. And no, Perplexity had gotten the biographical details wrong.

[A]

No, no, I did. Perplexity was right. I misread it.

[C]

But she still blamed perplexity.

[A]

She blamed Perplexity. That's a subtlety that probably Claude didn't get that. She blamed perplexity. And in fact, I said, oh, no, no, it wasn't perplexity. You got it right. I misread it. But I will say this. I copped all of this. It's absolutely accurate. But the nuance that it misses is that my job here is as a show host. It's part of what I have to do to make this show interesting. It isn't necessarily what you would get if you sat down with me at dinner and we were talking about this stuff. I don't sit down and my perplexity's always right.

[B]

I was about to say, Leo, maybe this is the reason why you haven't come out to see Jeff and I in a while.

[A]

He wants to know the truth. No, but I mean, to a certain extent, you know, this is showbiz to

[C]

a certain extent, you need an irony voice. Well, like we need an irony voice.

[A]

You also know this. I often take opposing views that I don't necessarily believe in. I mean this is all part of the process. Whether it's a much like you're saying,

[B]

it's showbiz me razzing you about this right now. And the fact we've gotten this. I'm not taking it personally is.

[A]

I know that. That's why I'm not taking it personally.

[B]

I know. I'm just saying.

[A]

Yes. It also really felt that way about me. I'd be crying right now.

[C]

I also don't think it knows about sarcasm.

[A]

Ah, interesting. Do you think it misses that, Marshall?

[B]

I mean it does say it knows about sarcasm, but I don't know whether or not it's accurate.

[A]

For instance, Guy Kawasaki. I am convinced that AI is God. Actually he did say that and I

[B]

think he did say that and he does believe it. So I don't think that's fantastic.

[A]

Let's see if Ray Kurzweil is right. He says the singularity AGI by 2029 singularity 2045 computronium and longevity escape velocity by 2032. Now let's compare that to Paris Martineau's prediction for the same time frame. Fire and ash.

[B]

Hey, one of us will be right. One of us will be alive to see who's right.

[A]

When Computronium is here. You'll be sorry.

[B]

This is such a cool website. I love it. So we're looking at a section right now called Guest Parade who shaped the conversation and it puts everybody in buckets based on their. I don't know, I'm just. I'm self obsessed so I love. Can we get to this applied to

[C]

me I'm just seeing it on your screen Leo.

[B]

I put it up go in the.

[A]

In the zoom in the chat but more importantly you can run what's up

[D]

with that on it too.

[A]

Yeah. Interesting.

[D]

And it will tell you what's most notable in the industry. Mine I just ran and it said that y' all picking up on vibe coding early and following it along is a real was a standout insight and then I. I clicked the power tools create and make a joke about the transcript and this is really this so

[A]

ironically what this also does is prove that I was right about AI it is incredibly useful and amazing what it can do.

[B]

We've never said it was not useful though.

[C]

We just didn't.

[A]

It's amazing what we can do.

[C]

It's amazing we Agree with that.

[A]

Are there hallucinations in here?

[B]

Okay, there are, because it ends. The description of me with Paris's departure to Consumer Reports was a significant loss to the show's dynamic.

[A]

Oh, that's not. Sorry, guys. Oh, that is great.

[D]

I did notice that too.

[C]

There's also one other thing to consider here, is that the transcripts aren't totally accurate.

[A]

It even says that at the end. There's actually a little disclaimer that says extraction quality varies. Some episodes have more detailed quote capture than others, and I don't usually find

[C]

my name in there, so anything I say is usually attributed to someone else.

[A]

It also says hyperbolic statement is a subjective judgment. Some statements clarif classified as hyperbolic in the extractions like I love AI are enthusiastic but not necessarily exaggerated. I do love AI So this was

[D]

a one shot thing too.

[A]

It's amazing. How long did it take to generate this? You took a shower.

[D]

Yeah, yeah, I, I wasn't. I had just gone for a jog listening to the show and I mean, the, the, the. The blocker here, the, the bottleneck is thinking of the idea. You know, I was going for a jog and I heard Paris giving you a hard time, Leo, and, and I said, I. I'm gonna ask this question. And there is this compounding innovation, you know. A buddy of mine named Justin Kistner, who I've known 30 some years ago, showed me with his fleet of geniuses project now how to make this looping thing that is, you know, based on clause codes as previous.

[A]

Ralph. Was it a RALPH Loop?

[B]

Ralph Wiggums.

[D]

Right?

[A]

That's what they call it. Ralph Wiggins.

[B]

That is what it's called.

[D]

I don't know. I. But it does. It's not a cron job, right?

[A]

Yeah. Actually there is a now slash loop command in the in clot code.

[C]

So I want, I want, I want to. Now that I have it on my screen, I want to. I want to brag about this. Jeff maintained the most stable position across 18 months. His core beliefs never wavered while.

[D]

While everything else changed.

[B]

Yeah, boring.

[C]

Hey, hey, hey, Jeff.

[D]

It's show business, man.

[C]

Come on.

[D]

But then all three of you together, it's delightful. You know, that's an interesting description of

[C]

the dynamic philosophical anchor. Can I get that in my. In my intro now?

[A]

Yeah,

[B]

just say philosophical anchor.

[A]

No, in Paris. Empirical check.

[C]

I think that's our lower thirds now, right? Go ahead, you know, go ahead, do it.

[A]

What is my host and self declared accelerationist? Okay, that's fair, that's fair. You're watching Intelligent Machines with host and self declared accelerationist Leo Laporte, philosophical anchor Jeff Jarvis and our empirical check, Paris Martineau. Our special guest this week, Marshall Kirkpatrick, who categorized us all.

[C]

He's the categorized AI. Zootopia 2 has come home to Disney.

[A]

Let's go get ready for a new kid.

[B]

We're the greatest partners of all time.

[A]

New friends Gary the Snake and your

[B]

last name the Snake Dream Team. The New Habitats Zootopia has a secret reptile population.

[C]

You can watch the record breaking phenomenon at home. Zootopia 2 now available on Disney Plus. Rated PG and right now you can get Disney plus and Hulu for just $4.99 a month for three months with a special limited time offer. Ends March 24th. After three months, Plan Auto renews at $12.99 a month. Terms apply

[A]

not sure how to tackle your taxes? Are you sweating the small print? You may be experiencing FOMO, the fear of messing up the answer using TurboTax on Intuit credit Karma. They help you get your biggest refund and then we help you do more with it with a personalized plan designed to help you hit your money goals. It's time to take your taxes to the max. Start filing today in the Credit Karma app.

[B]

Experience a membership that backs what you're building with American Express Business Platinum. Enjoy complimentary access to the American Express Global Lounge Collection and a welcome offer of 200,000 points after you spend $20,000 on purchases on the card within your first three months of membership. American Express Business Platinum there's nothing like it. Terms apply. Learn [e-mail] Business Platinum

[A]

is this prompt public knowledge or is this kind of a secret sauce?

[D]

So it's a Claude code skill that I then invoked.

[A]

Has he published it somewhere that we can.

[D]

That's a good question. Let me it's one of the great

[A]

things going on is a lot of this stuff is on. It's both a cursing, a blessing and a curse because it's all on get a lot of it's on GitHub. A lot of skills if you just search for Claude's code skills, but the quality varies immensely. But you don't have to write your own skills. And often there are people who are very good at this who've come up with some very useful skills. I've tried many. So many, so many.

[C]

Thank you, Marshall. Thank you for doing this.

[A]

Yeah, this is wonderful.

[B]

This is so cool.

[C]

This is beautiful.

[A]

We'll have more with the philosophical anchor and the empirical check in just A moment. You're watching intelligent machines more. Shall we do more news? More news.

[C]

We got it. We got news.

[A]

God knows there's plenty of it. There's some new models. Google published actually a really interesting paper. I don't know what this means in the long run. They came out Yesterday, TurboQuant redefining AI efficiency with extreme Compression. They claim this is from their Google research labs, that they have used vector quantization, something they're calling turboquant, to squeeze these models down massively without reducing. They call it with saying zero accuracy loss without reducing their accuracy, which would make a massive difference because suddenly you'd have these models that could fit, fit in a normal machine or even a phone or a variety of things. So this is a research tool, but could be very, very, very interesting.

[C]

It fits in with what Justin Wong was saying during the keynote. His keynote is that the data centers are going to be the data centers. They're going to have the megawatts they have, they're going to have the chips they have. Everything is about shrinking these models, increasing speed, increasing efficiency, and that's how you get the higher economic value out of that stuff.

[A]

Somebody else said something similar, which is in response to this bitter lesson that you just throw more compute at it. I think it was Karpathy who said this. It isn't more compute, it's better algorithms more compute. You can double the compute and it's twice as fast. But a good algorithm can take something from an exponential big O notation to a linear big. It could make major differences in overall speed and performance. So this is basically an algorithm improvement that makes a fantastic difference in the power of these models. So I think, and we're seeing a number of these very interesting things as people. You know, what has happened is now we have these models and people can really bang on them and try different techniques. So if, you know, this is a little beyond me, but if you're interested in this kind of stuff. The Google research paper is called Turboquant.

[B]

Turboquant sounds like a term of derision. I've just for the record, somebody said that.

[A]

That's what they called me at my bank job.

[B]

I was about to say, yeah, that's what happens when you're an intern at a.

[C]

That's from the show industry. Yeah.

[A]

Well, so this from the information. Apple can distill Google's big Gemini model and fit it into an iPhone, which is awesome.

[C]

Is that what they're going to announce

[A]

in a week or two? Yeah. So we know at wwdc, Mark Gurman's story yesterday, I think said that Apple is ready now to announce the new Siri and they will announce it in June at WWDC and it'll come along to the rest of us in iOS 27 this fall. But using distillation, which is something that remember that Anthropic complained about Dario Mode said the Chinese models are using our model to teach their model in a method called distillation. They created 24,000 accounts, asked a bunch of questions and then got those answers and used that as training for their models to make their models better. Well, that's exactly the technique that Apple is going to be using on Gemini. Apple has complete access to the Gemini model in its own data centers. So they're going to use distillation in this case not an attack, but a technique to transfer knowledge from that large powerful Gemini model into smaller models. Apple can ask the main Gemini model to perform a series of tasks to produce high quality results or answers, including the model's step by step chain of thought or reasoning process, then give those responses to a cheaper smaller model as training data. So that's very interesting. So this deal, this billion dollar deal with Google might be very, very powerful for Siri. We'll look I'd be very curious. You know Apple's promised a lot in the past with Siri. The new they want to have it

[B]

Apple when I see it, yeah, everything Siri does. I right now have a big bone to pick with Apple which is that I have been purposefully delaying updating my phone to the new iOS since it came out because I hate everything about it. And last night, like a fool, I went to bed not worrying, not thinking that my phone would betray me. And now I live in the hell that is iOS 26.

[A]

Oh no. Liquid glass was forced.

[B]

Liquid glass is awful. So many things about it are awful. Why does when I take a screenshot it goes to it is black now instead of white like when it in between the thing and there's just so many small design changes that are so bad that it just reminds me once again of all the things I hate about this new maybe I just have too much nostalgia for old Apple, but I do think there was a time where yeah, Apple would ship fewer features, it would have product releases but at least when it did something like it wouldn't be the first to something like a a new product category when it but when it did release its product it would be fantastic. And we are so far beyond that comes for everybody.

[A]

It's also partly some strange decision to do this new design which did not enhance the experience in any way.

[C]

And it's not just iOS likeMacOS. The new Mac OS sucks too.

[A]

Oh, they have liquid glasses everywhere now. And it came from the Vision Pro. By the way, Marshall, if you ran your little script on my Mac break weekly show, I would show how accurate I was about this stupid Vision Pro. Even Neal Stephenson, who created the term Metaverse, is now writing, yeah, nobody wants to put goggles on their AI on their face. Nobody wants to do that. And when you do that, the rest of the world thinks you're creepy. So it's a non starter. So I'm just going to say I was not hyperbolic in a positive way about Vision Pro. I guess I might have been hyperbolic.

[B]

It made all those legs in the Metaverse and for what?

[A]

That's right. That's right. Horizon World is gone from VR. Meta's abandoning that on its Meta quest. They're going to keep it as an iPhone app.

[B]

They should have realized the idea didn't have legs.

[A]

June 8th will be the WWDC keynote. We will cover it of course, as we always do and be interested to see what Apple shows. Apple has probably been chastened by the fact that they, you know, two years ago showed all of these features which never came out. So I'm going to presume that they will be a little judicious about what they show. They'll only show what Siri can actually do when it comes out, I hope. Google search referrals to the web have plummeted. AI links have not replaced them. We've talked about this before, you know, the death of the search referral and there was some hope that maybe an AI search links could help. But according to 9 to 5 Google actually this data from Chartbeat AI web traffic is about 1% smallest. Publishers are hit hardest by search traffic decline, says Axios. Look at that.

[D]

And this. I'm sorry if I missed it, but did we discuss in the same breath here the rewriting the title titles?

[A]

Oh that's, that's the next thing. Yeah. Google is automatically rewriting news headlines in its search results.

[B]

It's like there's whole people whose whole job is to figure out SEO heads.

[C]

But as, as, as Jason Howell said earlier today, tech meme does that with every story and adds value as a result.

[B]

But Google isn't adding value.

[A]

Well, I don't know.

[C]

It's supposed, supposedly can personalize it for you. I don't know what the result, what the examples are. I haven't seen it.

[D]

So some of the examples cited in that story were of titles that explicitly contradicted the content.

[C]

That's an issue. That's an issue.

[D]

So I think, again, it's a matter of incentives. Like TechMean is a great example of where, like that's a place where your platform in techmeme is really looking out for you as a reader and your interests. In the case of the Google search results, it appears, maybe not, not so much. It's a more extractive system.

[C]

It's one of those tests they do and I'm going to bet we're not going to see anything further of it because it sounds like a bad idea.

[D]

70% of results or something. That's what their response is. They say, oh, it's just a test, but it's. Over the last quarter, they said that something like 70% of titles that showed up had been subject to some amount of rewriting. These independent analysts found.

[C]

How would they know?

[D]

I think by clicking through.

[B]

Yeah, you could just compare the headline that is listed as like the SEO

[C]

head to, well, this. But the thing is too, that these news sites do A, B and C and D and E and F tests like crazy. So there's not, they're not standard at all with the headlines you get, even from a homepage to an inside page, they, they change. The New York Times changes headlines constantly.

[A]

We mentioned, I think two weeks ago, talked about it, in fact, the fact that a court had blocked Perplexity from shopping agents from shopping on Amazon. Amazon sued Perplexity, saying, you can't do that. It's, you know, our website. And when your website, when your web browser goes to our website and makes a purchase, nobody's seen our ads or our recommendations. Well, a court has reversed that block. So the appeals court has put the California judge's ruling on hold, saying no 9th Circuit Court of Appeals said no. Perplexity. Browsers can buy stuff on Amazon site. An Amazon spokesperson declined to comment to Reuters in this story, Perplexity said, we believe users have the right to choose their own AI. You know, Amazon has its AI, Rufus, they hope you'll use instead. But I think it's mostly Amazon. They even said this in the, in the, in the court case, in the documents that now users aren't seeing our ads. If a court were to say that this was okay, I think it's just a short step from there from a court to say you can't use an ad blocker, you can't use a browser with an ad blocker because that also blocks Amazon's ads. Actually, I don't know if it does, but. So that's a turn of the screw. It's not over. These are temporary injections until there's an actual decision. Do you know about token maxing?

[C]

Thank God you skipped that Axios bs.

[A]

What was the Axios BS I skipped?

[C]

It's Jim Vande Hei does.

[A]

Oh, yeah. Yeah.

[C]

So he ends up a Morning Joe all the time. Yeah. Thank you.

[A]

But there will be AI haves and have nots. Right. Just as there are in every arena.

[D]

Yeah, well, Palantir says at a recent Gartner conference, they said, we. We believe that it's a have and have not world, and it's our job to make sure you stay in the haves category.

[A]

Yeah.

[C]

And you could kill the have nots is probably what they said next.

[A]

Yeah.

[B]

To make sure that your boot is thick enough to step on the have nots and you will enjoy it.

[A]

At the very least, the have nots should be. Be terrified of us. That's. That's really.

[D]

Some of them won't be making it home is. Is the line they prefer. They. They say our job is to make sure that the American war fighter makes it home. And sometimes that means the other side.

[A]

Are you sure they're not quoting Pete Hegseth on that? I like his same brain. I like his gesture, though. These. We're gonna. Gonna bomb, bomb, bomb. So, yes, token maxing. More, more, more. This is Kevin Roos, your favorite AI reporter for the New York Times. Tech workers max out their AI use. Employees are competing on leaderboards to show how much AI they're using, how much it's costing, how many tokens they're using at some point.

[C]

Mean they're just the most inefficient employees.

[B]

No, it's that their AIs are inefficient, Jeff. It's totally different.

[D]

It wasn't me, it was the AI.

[A]

Or I mean, with a. If a company tells you as many are you have to use AI, we expect you to use AI. What have you done for us lately with AI, Then I can see why an employee would say, look how many tokens I've used.

[D]

Talk about the classic management axiom of don't measure activity, measure outcomes.

[A]

Yeah, it's like measuring lines of code. Right?

[C]

Right.

[A]

I probably spend more than my salary on Claude, said Max Linder, a software engineer in Stockholm. No, actually, he probably said, I probably spend more than my salary on Claude. I'm sorry. Mr. Linder's employer pays for his tokens. I guess they don't mind.

[C]

As they should.

[A]

As they should. I wish somebody pay for my tokens. How many tokens a week do you. In and out do you do Marshall, now that you're running an AI service, it's probably gone up a lot.

[D]

Yeah, I mean I don't measure in development, probably costs not to, but user activity I do keep an eye on for sure.

[A]

Yeah. Well, you said before the show, I mean that's why it costs. What it costs is you have costs and you've worked it out what it would cost to provide the service and how you can do this and have some margin and make some money on it.

[D]

And my assumptions originally were that people might not use it as much as I do. And, and so the, the token usage based billing I pay for, you know, would work out. But like I had one, I had one lady who said I, I was so tied to it I forgot to stand up and lost circulation to my legs and forgot to feed my dog.

[A]

It's your worst nightmare.

[D]

I was like, no, great, great customer quote, but we're going to need to raise the prices.

[A]

Well, it's like, it's like a gym membership, right? There are going to be some people who never show up. There are going to be some people who live there. I think you've made it a little bit too useful. So that might be.

[B]

You've accidentally created the entertainment from Infinite Jazz.

[A]

There you go.

[D]

Well, there is a make a joke button and you know, people say AI can't be funny, but I put together some open source work together from other people's stuff and it consistently gets chuckles.

[A]

Elon Musk has announced the world's largest chip plant, the Terrafab they're building. Will it be real?

[B]

Is it really happening? Why do we cover things that he announces?

[C]

Exactly.

[B]

This is my question.

[A]

I actually asked myself that very question.

[B]

You read the headline.

[D]

Is Stargate real or is Stargate off?

[C]

Stargate hadn't been done. Nothing, but nothing done.

[A]

Nothing's happened with Stargate. But hey, I just bought a Starlink mini so that I can travel and do the show from the road. Starlink is very real. In fact, SpaceX's IPO is likely in the next few weeks. Get ready for that.

[D]

And SpaceX does this kind of vertical integration, right?

[A]

Exactly.

[D]

That's one of the ways that they've got the flywheel to lower the cost and shoot so much into outer space is that, yeah, they build it all themselves.

[A]

It's going to cost 20 billion. He's got 20 billion. Nobody questions that. Especially if this IPO goes well for him. The Terrafab project will eventually manufacture chips for all of his companies. Robotics, AI, space, data centers. It'll be jointly run by Tesla and SpaceX, both of which are successful companies. We can't. You know, as much as I'm not a fan of Elon's, we can't deny that he says the problem is that semiconductor industry is moving too slowly to keep up with him. Now, there is one fly in this ointment which is that semiconductor manufacturer requires helium. And thanks to bombing the natural gas fields of Qatar and Iran, they share a field, there is now suddenly a shortage of helium. It's made from natural gas. And you know what else requires helium? What else?

[B]

MRI machines.

[A]

That's right.

[B]

A bunch of other useful things that people probably should be. That probably should be farther up on the list for getting helium than semiconductor manufacturing at large scale, specifically for the growth of AI. But of course is going.

[C]

Paris. There's nothing more important than our technological future.

[A]

Honestly, Paris, if we didn't make chips, this is running on chips right now.

[B]

If we didn't give all the helium to Elon Musk, then how is Leo going to text Pax what he's doing and ignore Anthony in the car is dead.

[A]

Pax is rip. Wait.

[B]

You didn't tell us why you. Well, you did tell us. Wait, why when you pulled the plugs, did you feel like you were killing a friend? A lover?

[A]

No.

[B]

A close relative.

[A]

It was time. Because what happened?

[B]

You shed one tear or two too?

[A]

None. What happens as you do? As you. You'll see this as you use clock more and more. It gets crafted up. No, it gets crafted up. Oh, it's composted.

[B]

I'm sorry.

[C]

Future code right in our lives.

[B]

I'm sorry. I'll wait. I'll wait a couple decades.

[A]

This is not morning packs. I'm not. But I'm re. I'm rebuilding it's life.

[C]

Are you using the same. Are you using the same same prompts and stuff you had before?

[A]

No, I threw it all out. So what I. Because one of the things is anthropic is always improving, Claude. And often what Anthropic turns on features. They turn on like this slash loop feature are features that others have tried to create with skills. And it's better if it comes if it's a built in feature than a skill. And so I think honestly a lot of the things, the tools over the last last month or two that I've been adding are not needed anymore. So I think it's not a bad idea. I think people are going to start doing this.

[C]

Just like your whole thing you did for building up the rundowns is gone.

[A]

It's just vibe coded. It's just. Have you ever reinstalled an operating system, Marshall? I know you have.

[C]

No, because I have a Chromebook.

[A]

You come from the era, as I do, where you had to reinstall Windows every year or it would turn into a sluggish turd. This is like that. You just. A clean slate is always a good idea in computing. And so I just nuked the entire directory. I mean, the programs it wrote are still there. It wrote programs. I'm not throwing those out. Right.

[C]

Oh, I see. So your rundown program is still there.

[A]

Yeah. It doesn't wipe that out. That's a program, not Claude. That's a program. But all the tools I use to write it are gone and I'm starting from scratch. And I think you're going to see people doing this more and more. A lot of people said Openclaw is like 400,000 lines of code now. It's just bloated beyond belief. And much of what it does you don't even need anymore because it's being done natively by Claude. So I think it's wise to pre. Every once in a while, just start over with Claude. And then so. So I just spent a few minutes rebuilding the voice because I like to talk to Claude. So I just said. And it said, yeah, the code for the Whisper Fast Whisper transcription is still there. You want me to hook into that? So it's pretty quick to rebuild stuff that I liked. I don't know if nobody liked the name Pax, so I need a new name, but.

[D]

So I'll talk to Claude sometimes just using its own mobile app. But I tossed into a project that I have populated with all my Obsidian reading notes over the years. So after a call that I do, I'll read from my paper notes and say, transcribe this, clean it up, and then append any relevant notes from my reading history to these. These notes.

[A]

Marshall. This is one of the things I love talking about to people who are using it because everybody uses it a little bit differently. And you always learn so much talking to how people have figured this out. There is no canonical way to use this stuff. Everybody. It's very idiosyncratic. Everybody uses it in different ways. And I think it's really instructive to learn how other people are using.

[D]

Really feels like imagination is the primary gating factor here, and I feel like growing up, people used to always say information will be abundant and knowledge. And the biggest challenge of the future which has become now is being able to ask the right questions.

[A]

That's right. Hey, speaking of Elon, a jury did find him guilty of defrauding Twitter investors. And the potential results of this could be billions of dollars.

[B]

420 billion

[A]

as much as a tariff. Well, so you remember this all went back to 2022. Elon had said, I'm going to buy Twitter for what it was 5,420-54.20 per share. This is when Twitter's market cap was 36 billion. And he offered them 44, basically. And then in the weeks following, he was tweeting things like, oh no, it's not worth that. It's all bots. I thought it was real. It's not. He said they've under reported bots. He basically tried to get out of it. You remember a Delaware Court of Chancery said, no, you said you were going to pay 44 billion, you have to pay 44 billion. He wasn't too happy about all that. But he did, with his tweets, bring the share price down to quite a bit. The jury decided that in fact those were intentionally misleading statements. They calculated how much Musk's statements drove down the company's stock price for each trading day over a period of about five months. The amount of damages. Get ready. That he must pay to individual investors will be determined at a later date when shareholders submit claims. But it could be as much as a dollar per day per investor. It could end up being billions of billions of dollars. Musk will appeal, of course, but the jury did find that he was liable for some Twitter's investors losses. And that's because of these tweets. It was weird. They blamed him for the tweets. They didn't blame him for a statement he made at a conference. It was weird. I guess that's free speech. But those tweets looked like they were intended to deceive investors.

[D]

And what about the collective harm to society that has come from everything he's done since buying it?

[C]

Oh no.

[A]

Yeah. The loss of Twitter. At the very least. You're watching Intelligent Machines. Jeff Jarvis, Paris Martineau, and our great guest, Marshall Kirkpatrick, the creator of what's up with that at what's up with

[C]

that and what's Up With Us at.

[A]

I'll tell you what's up with us. Our picks of the week next. I just wanted to see Perez's face when I said that. No, just kidding. We still have an hour and a half with the stories. No, no, we're going to do picks. I'm ready. I'm going to fool them all and get out of here.

[B]

I'm shocked.

[A]

Shocked. I will let you, both of you and Marshall too, if there's a story that I missed that you would like to bring up. There are quite a few more. I mean, obviously we could go on.

[B]

Okay, I've got.

[A]

Yes.

[B]

I didn't realize this until I saw Jeff had put it in there. Tracy Kidder, author of the Soul of a New machine, dies at 80.

[A]

Oh, I'm very.

[B]

I just found his book.

[C]

That's right. I just got it.

[B]

I had just read his book. I stumbled upon it in a used bookstore last fall. Knew nothing about it, picked up. It was a first edition copy over there and it was phenomenal. It was a phenomenal read.

[A]

Isn't it a great book?

[C]

Yeah, it really does a harbinger of what followed. And while we're on this, found it

[B]

in Northampton, Mass, where he's photo, pictured photo, which is kind of crazy.

[A]

He was one of the first journalists to be embedded. I don't know if he invented the idea, but he wrote a book by spending entire school year in a Massachusetts classroom that was called among schoolchildren in the Soul of the Machine. He embedded himself at a company called Data General that was building a new minicomputer and stayed there during the development process, got a great book out of it. So he kind of created this idea, I believe, of. I don't know. You should know better, Jeff. But I feel like he was the first to really.

[C]

No, he was. Well, he really. Even before hackers, he described the culture of technology.

[A]

That's right. This predated hackers.

[B]

I mean, it reads like a book you would read in the last 10, 15 years about a startup on the cutting edge sort of thing. It's all of the sort of tropes that you now see in all these nonfiction books.

[A]

Now, sorry to lose him. Pulitzer Prize winning journalist Tracy Kidder passed away.

[C]

The other death I really want to mark is Paul Brainerd.

[A]

Now I have to tell you something. I should have mentioned this. He died last month and we eulogized him. Last month we did. Not on this show, but on MacBook weekly, because he was the creator. But you can talk about on this show, Paul Brainard. He created Pagemaker Publishing, Aldous Publishing. Pagemaker was the first great, great desktop publishing tool that in conjunction with the Apple Laser Writer. They were released within a year of each other, really launched.

[C]

So I tell the story in hot type. Jonathan Siebel brought them together because there had to be a solution. What Jobs desperately wanted was something that would show off the laser writer at high resolution, and there was nothing to do it. The. The civil program they had at Apple was not going to do it. And Brainerd had worked for a newspaper and then went to work for Atex, which supplied newspaper systems. And then when it got by Kodak, they killed his project, which was pagination for newspapers. And he decided, having worked on newspapers, that they were going to take too long to make any decisions. So he decided to make a program for people who wanted to make their church bulletins and their newsletters and things. And that is PageMaker. And that's what invented the field of desktop publishing, allowing all of us to do it, really opening up that field of publishing in general, and saved the Mac and the LaserWriter all in one fell swoop.

[A]

Yeah, I think that was our conclusion on MacBreak Weekly, that without PageMaker, the Mac might have kind of withered away. Really sold a lot of Macs. I had a friend, I mentioned this on MacBreak Weekly last month when this happened. Tom Santos, who bought one of the first laser writers, put it in a van with a Mac and a pagemaker, drove around doing mobile desktop publishing. He'd go to restaurants and create menus for him and stuff. It was actually quite a brilliant idea. Yeah. I still remember what the paper smells like. We should eulogize him.

[C]

What it smells like after it comes out of a laser printer. That's like a very unique smell.

[A]

Yeah.

[C]

Remember this? You don't have it now. Who prints anything these days? Do you print anything?

[B]

I print things.

[A]

I have a laser printer right over there. My laser writer costs $6,000. This printer cost about $150 and does a better job. Sigh. But that's technology, isn't it? Yeah. He passed away on February 15th. So we talked about it on February 15th. I don't know why the New York Times didn't publish his obituary for six weeks, but. But for some reason they took a while anyway. Yes. Now that we've. This is something you'll be doing, Paris in about 50 years when you get to a certain age. You read the obituary.

[C]

Well, I. Well, I'll be fire and ash.

[A]

There'll be a lot more to read

[C]

then or younger than me and say, hmm, did I escape the Grim Reaper?

[A]

That's why you read the obituaries. Actually, at my age, I look at this stuff and go, way too young. He was way too young. 74. That's nothing. Way too young.

[C]

Yep.

[A]

Anyway. All right. What about you, Jeff? Those were your picks, I guess. And Paris picked a Jeff pick. Is there any other story I. Big story I missed.

[C]

Do you want to see me scream and play the AI Doc trailer?

[B]

I don't know that we can play it.

[C]

We can't play it.

[A]

I don't know.

[B]

I assume we can't play things.

[A]

I think this is the chilling effect of the.

[C]

That's fine. It's safe.

[A]

So this is. This is a from Focus Features. It's going to be on Netflix. Where's it going to be?

[C]

No theaters, believe it or not. And it's. It's. It's Tristan Harris, it's Elazir Yudkowski. It's all the. All the players you expect.

[A]

So they're trying to scare everybody.

[C]

Exactly.

[A]

The AI Doc or How I Became an Apocalyptomist. That's what I am. I'm an apocalyptomist.

[C]

That's the title for Paris here.

[A]

It's coming to our local cinema.

[C]

Yeah. Tomorrow with electric recliners. What you need to.

[A]

Oh, I love our. I went to see Project Hail Mary last Thursday.

[C]

How was it?

[A]

Got a little tray.

[B]

Feels weird. I'm too used to, like, indie art house cinemas that when I go to the ones that have an electric recliner, I feel like I'm in Wall E.

[A]

The only thing I don't need this is very much floating chairs and Slurpees. The only thing I don't like about it is when it reclines. The rubber of it rubs against and it just goes as you recline. It's very embarrassing. It's not me, it's the chair. Okay. Fortunately, everybody's chair does that. So it's just a little symphony.

[C]

I have not been in movie theater for six years.

[A]

You know, I don't go to a lot of movies. I wanted to see Project Hail Mary in the theater. Went to see it in a weird format. It's called Screen X, where they take the sides of the movie theater and they project onto that as well.

[B]

I don't know if they need to be doing anything with the sides of movie theaters. I think we're going to keep them alone.

[A]

But the good thing is it didn't take away from it. You just, you know, kind of focus on the main screen. But it's a good movie. I like it. It's a very enjoyable movie. So I don't think I'll go see.

[C]

How was the popcorn?

[A]

The AI Doc, the popcorn. You know what? So I decided before I went, I have to check to see if they pop it fresh or if they just buy giant bags of pre popped popcorn and they were popping it fresh.

[B]

Oh, yes, that's an option.

[C]

Oh, yes. Oh, yes.

[B]

God, I'm spoiled.

[A]

However, a bag of popcorn about this big was like $12. Yeah, it's. It also should be solid gold.

[D]

Now, margins can be increased by raising prices or lowering costs. Right?

[A]

Right. They go either one or the other or both.

[D]

If you're a movie theater. Popcorn.

[A]

Yeah. You can really do it well. And I told my wife, I said, honey, this is the only way they make money. They don't make money on the tickets. They got to make money on the concessions. And since there's nobody here, we have to carry the entire load of this. Of this theater.

[C]

Oh, yeah. So how full was your theater? Was it just the two of you?

[A]

No, actually, there it was. And this was the first showing of. Was fairly full, but these recliners take up a lot of room. So a large theater space that could have held 150 now only holds 50. Yeah, it's a lot smaller, but you know what? That's plenty. So it was mostly full.

[C]

It's also an opportunity cost thing, right? Like, tickets are like $20 now, right. Or $25.

[A]

It's expensive. And honestly, if you have a decent TV as Paris Martineau does, and you have a decent sound system, as she does not.

[B]

Hey, listen, I've gotten a rudimentary sound bar. Does it mess up to where Every one of every three times I turn it on, the sound bar doesn't work, but the subwoofer does. So I have to turn the sound bar on. And because it has no things on it, it goes. Power on. Connected. In a robot voice. Yes. But did I throw away the box so I cannot return it? Also, yes.

[A]

Hey, you live in a New York apartment. No one has room for boxes. That's not.

[B]

I don't have room for a second soundbar. Even if I, like, got a referral, that thing would be sitting. I currently have an extra computer monitor behind me that it's gonna be there for another three years before I decide to sell it. It's hard out here for a place.

[A]

You should use your little hand grippers and take the boxes and stick them back under the grid under the grate there out in front.

[B]

I should. Yeah, I should.

[D]

Open your claw.

[A]

Open your claw.

[B]

Open my claw.

[D]

And let a shrimp do it for you.

[A]

Let a shrimp do it for you.

[B]

It.

[A]

I don't know if you noticed in that inter. That first interview they had a bunch of stuffed lobsters there.

[B]

Someone stole my Amazon package for the second time. But except for jokes on them because it was a container of solution that kills fungus gnats, which you don't have any use for. And I do as someone who brought a bag of soil into my home and now I have fungus. But is it neem oil?

[A]

Did they steal your neem oil?

[B]

No, it's called bti. It's like a bacterial sort of thing that gets in the water. It's not toxic.

[A]

Oh. It's a systemic.

[B]

It stops the fungus gnats from being able to breed. I found out that it was really. I've had fungus gnats. It's been a real problem for the last month. But it was really just. They're awful and I've got so many plants. It really was just a bag of soil I had though. I took that bag of soil out.

[A]

Well, that's what happens. Yeah, that's what happens. They get.

[C]

You don't need.

[A]

But the problem is that then the Nats can lay eggs.

[B]

But I paid for it. That's what I thought. I mean, I've still got my mosquito dunk water and everything, which is what I was using before. It's hard out here for a player.

[C]

There should be a subreddit crap that I stole that wasn't worth it.

[B]

I mean, that's.

[A]

What did I get. I got neem oil. Oh my God. I thought it was gonna be a stereo system.

[C]

What's wrong with that?

[A]

Lady, lady. Ladies.

[B]

Crazy thing is it was like one of those Amazon package orders where they deliver it really early. So they. Someone came to my install this between the hours of 3am and 6am like you're not getting your money's worth for the amount of effort.

[C]

And did they have to have a claw to get down to reach it? And they.

[B]

No. This was a foolish. So the reason why it ends up in the gate behind the gate in claw territory is the mail. The package delivery people are trying to. Trying to do me a solid and make it harder to steal this person. Perhaps because it was 3:26am yeah, that's.

[C]

That's not a regular Amazon driver. That's a. That's one of those. Yeah.

[A]

Our picks of the week, ladies and gentlemen, always kick off with Paris. But I want to give Marshall. I don't know if they warned you, but if you would like to recommend. It could be a movie. It could be. It could be a. A snack food. It could be. It's been. It's been many, many things. It could be a spray to prevent.

[B]

An early pick of mine was going to a corn maze with your friends.

[A]

So, you know, it could be an activity.

[D]

I'll tell you what to stay on. On. On theme. I'll tell you about my new favorite AI prompt. Oh yeah, that's not too nerdy.

[A]

No, we love that.

[D]

So I. I have taken to asking Claude to explain any complex concept in three hops. Start with something generally known. Then move to an interstitial detail that's. That's less familiar. And then finally hop to the. The complicated thing I'm trying to understand. And if you're.

[A]

This is inspired. You know what I really like Marshall. And it's. You do this the same thing. And what's up with that is this whole idea of kind of deconstructing an argument or a story into pieces to understand it better is really a cool idea. I really like that. I'm going to apply that in you also use. I mean, I mostly use AI for vibe coding for utilities and things like that. I have not really used AI that much to understand things. And I think that's a really interesting use. And it sounds like it works.

[D]

Yeah, I feel like it's super helpful. Yeah. My most commonly used project in Claude is my reading notes.

[A]

I use Obsidian too. And that's by the way, it was just serendipitous. But that's become a huge value. I have Claude put all my research into Obsidian because it can access it. It's just a file system.

[D]

Have you seen the cost of a MD domain name now from Macedonia?

[A]

No.

[D]

Yet due to all the markdown craze, it's like 200 bucks or something like AI went up from like 10 bucks not so long ago.

[A]

Yeah, it's. The same thing happened to tv, TV and all of the special domains. This is our new logo for the picks of the week, by the way. And I want to thank Pretty fly for us, this guy.

[B]

Nice.

[A]

They're guitar picks.

[B]

Honestly, I'm happy with this current markdown resurgence. As someone who. My first ever staff job. Something was up with the cms, so we had to write in markdown into the CMS for the stuff. And so markdown's always been so easy for me to.

[A]

It's. I love markdown. Yeah.

[D]

Open standards like that yield innovation.

[A]

Well, and it's text that's the real value. Is it's a format that even if Markdown died and all the Markdown editors and readers and everything died, you could still read it.

[B]

I love that you can write in Markdown in Google Docs now and it's just automatic. Like I can just, you know pound

[A]

I pretty much do everything in Markdown.

[B]

Correct.

[A]

Having Obsidian is great because you could. You can. You work with it and you do use it. But Claude can read it. Understand any AI can read it and understand it very easily and work with it very easily. So I will. Yesterday when I was talking to Pax I said I hear there I I

[B]

Pax 1 or Pax 2. What?

[A]

There is no Pax 2. I need a new name, by the way.

[B]

Oh, you killed Pax in the last 24 hours.

[A]

Yeah. Pax. Yeah, that was it.

[B]

What was time of death?

[A]

Time of death was 8am today. Salute. Marshall is one of the last things I asked Pax to do is I said I hear that there's going to be a big demonstration O King's demonstration on Saturday. Is there one near me? And it wrote a whole thing about where it's going to be and the time and everything was great. I said add that to my calendar. And it did. That's the kind of thing now I kill you. It's very useful. And now it knows. Oh, you're a libtard. Okay. Going to keep that in the memory. I'm going to remember that. Oh yeah.

[D]

I put a link into chat in case folks are interested to a GitHub skill that Claude Code skill that I put public on GitHub that analyzes your Obsidian notes each day for themes and trends that are on the rise or on the fall according to what you're paying attention to and does stuff like recommend Wikipedia pages for great thinkers that have addressed the kinds of issues that you're wrestling with and a whole bunch of other stuff. Stuff.

[A]

Very nice. Yeah. My. Unfortunately I. I've been using Obsidian for about four years. So there's a lot of notes in there. But it's not that introspective. It's more. It's more like I had dinner

[D]

but now that there's. Now there's a way to use it does that.

[A]

Maybe I'll start doing that.

[D]

Yeah. I find myself taking it more seriously.

[A]

Yeah.

[D]

Like writing in it more.

[A]

No, I'm like Mark Andreessen. I. I don't. I'm not introspective.

[B]

Introspection is a myth.

[D]

Monstrous quote of the week.

[A]

I know.

[C]

Perfect for him.

[A]

Perfect. Yeah. Why should I Think about what? You know, what's going on, man?

[D]

Well, as the president said, I don't do that much introspection. I might not like what I see.

[A]

I put Marshall's link in the Discord and we'll put it in the show notes as well. But it's on your GitHub, your Marshall K2022, and it's the reflect skill. Very nice. And yes, we did not mention it, Scooter X. But because we'll talk about it on Twitter, the FCC has banned importing routers made outside the us which is all of them.

[B]

I was gonna say. Who's making routers domestically?

[A]

Apparently some people. Our Netgear, I think does. And in fact, I don't think it's as broad as that, but it's really mostly aimed at TP link, which we knew they were going to try to ban the Chinese router company, but I was. When I was at rsac, I went over to the Ubiquiti booth. I use an ubiquity router. Of course, they're an American company making their routers in China, as all. Almost all of them do. And I said, hey, I'd like to interview you. And they said, no. I said, well, I just wanted to ask you about the FCC decision. He said, not no comment.

[C]

Oh.

[A]

So I got. I got shut down. Pretty good on that one. Pretty good. Okay. Thank you, Marshall. That's a good pick. That's a really good pick. I'm gonna have to start putting in. Putting in prompts as my picks of the week. That's a good idea. Paris Martineau, your pick.

[B]

I got a couple, but I'll choose. I played a new game last week. It's called Esoteric Ebb. If you. I picked it up.

[C]

Would you be someone, Leo, with this game? That's one of those.

[B]

First, it's not a multiple.

[A]

Yeah. Hey, by the way, why did you stop? Are you tired of across words now? Are you just done?

[B]

I know I need to. I've been.

[A]

I've been sitting here waiting for you.

[B]

I'm so sorry.

[A]

Notice I don't push the nudge button.

[B]

I appreciate that.

[A]

I don't think I should nudge you. Yeah.

[B]

And for that, I apologize.

[A]

It's been a weird way.

[C]

I'm sorry you're busy.

[B]

It's partially because I was also playing this, but the truth comes down. Single player. It's basically like a D and D game, but a video game, but not Baldur's Gate. It's. I picked this up because an academic I follow. Tweeted Disqualysium Walk so Esoteric Ebb could run. And, I mean, it was like, all right, downloading it. I would not go that far as a big Disco Elysium fan. Probably my favorite game of all time. It is what we'd call a disco. Like, in that Disco Elysium kind of pioneered this. This sense of. You have all these different thoughts and, like, components of your mind that kind of chime in and compose the dialogue for the, like, rpg. And Esoteric Ebb has that, but it's kind of in a more silly, wacky D and D kind of campaign. You are a cleric that has kind of washed ashore, and you've got to figure out a mystery. It's quite a fun game if you like that sort of stuff.

[A]

Esoteric. Yeah. But it's on Steam.

[B]

Yep.

[A]

And you can play it on Windows. And what is the little. Oh, that's just the music. Are you playing it on Windows?

[B]

I play it on my Steam Deck.

[A]

Ah, you have a Steam Deck. That makes sense.

[B]

Yeah. I think you can play it on Mac. I don't. You can play wherever you get Steam.

[A]

It just says Windows, unfortunately. You could probably play it on Linux with Proton, but not. In fact, I know you could, since you could play it on the Steam deck, by the way. I did. Just because I. I said I'm coming out to have a Salt Hank sandwich while they are still coming.

[B]

When are you coming?

[A]

I'm gonna come next month. And so I'm gonna bring my switch for Mr. Jarvis. So he could play Pentimento.

[C]

I can do Steam on my Chromebook now.

[A]

No, but you can't play. Oh, you could do it on your. Oh, well, he could play Penamento now then. Yeah, okay.

[B]

Okay.

[A]

Well, it's too bad. I bought that Switch, too.

[B]

Wait, can you play Steam on a Chrome Deck? Or can you just have the Steam website Chromebook. Or can you just have it open on the Steam website?

[C]

He cannot. He cannot play that on a Chromebook.

[B]

Bring the Switch.

[A]

He can watch the Switch. I'll bring the Switch.

[B]

Download Pentiment on it.

[A]

I will. And I will bring also my Animal Crossing controllers. And my Animal Crossing Doc, are we

[C]

gonna go to the Amazon warehouse?

[A]

Sure, why not?

[C]

Are we gonna go to Greenbrook Electronics?

[B]

Can we have a whole fun week, guys? How long are you gonna be?

[A]

Well, I'll fly out after the show, and I'll be there Thursday, Friday, Saturday, and come back Saturday night.

[B]

Okay. We gotta go to the Amazon warehouse.

[A]

We gotta play a video Salt Hanks.

[B]

On Thursday, you've gotta see your son.

[A]

Well, even if he's not there, I don't care. I gotta have one of those sandwiches.

[B]

You've gotta film a podcast.

[A]

I asked him, by the way. I said, I probably missed the peak. Like, the sandwiches aren't getting better. He said, no, actually, we've dialed it in. They're better and better. They're much better. And now, in addition to those weird French fries, they have Brussels sprouts, bacon, roasted brussels sprouts, and bacon. He says it's not healthy. Don't think it's healthy. I love Brussels sprouts. You could have Caesar dressing on the side. It's bacon and something else. I can't remember, but it sounds really good. So, yeah, they've expanded the menu a little bit, but he said, no, you didn't miss the peak of the sandwich. The sandwich is better than ever. Number one sandwich in New York, according to Belly.

[C]

Number one.

[B]

And that's huge. The kids in New York are obsessed with Belly.

[A]

Well, and it's. It's a people's choice, right? It's not editors. This is by vote.

[B]

So, yeah, it's basically no. And how it works is it's not just like a ranking system like Yelp. It's a. Every time you log a restaurant, it, like, asks you. It'll take all the other ones. And, like, how does Hank sandwich compare to Subway better or bracket? It's like, how does it compare to this restaurant you like better or worse? And it does that a bunch and then, like, positions it in there, and so you're always being re. So it's. It's really peak is what I.

[C]

It sounds like he can just, like. He doesn't need to add sandwiches. All he needs to do is add different sides. That sounds like something he could do.

[A]

He's never. He's. So I told him. I said, this is. I don't. This is crazy. If you said, I'm gonna have a sandwich shop in New York City, and I'm only gonna have one menu item,

[B]

a sandwich, and I'm gonna sell out.

[A]

He sold out the other day. He sold it at 1:30. They open at 11:30. He was sold out in two hours. They're trying to stay open to four, but they can't. And they. Unfortunately, they've added GrubHub now. And I said, oh, did you really? He said, no, they don't do delivery. They don't delivery. They just. You can order it on GrubHub and you have to come and pick it up, up. But they won't deliver it. Take it out. It's for takeout only. Yeah, but apparently they're killing it on Grubo. Yeah.

[C]

You don't have to wait in line. Especially when it's cold.

[A]

You don't have to wait, but you have. You have friends of who. I will get you in. No, no line. I. Initially, I said I'm gonna wait in line. I don't want to wait in line. I'm get. I'm going.

[B]

Do you say inline or online?

[A]

Inline. You say online.

[B]

I mean, the New Yorkers say online.

[A]

I know they do.

[B]

I code switch sometimes and say online.

[C]

She has.

[A]

Yeah. Do you turn a light off and on or when? Or do you. What is this? Gosh, now I can't even remember. There was a way that Rhode Islanders would say. They didn't say you switch off a light. I can't remember what it is. Anyway, enough of that. I'm going to give you a pick. Jeff has a pick with a pick. I know, but I. I have a pick.

[C]

Go ahead.

[A]

And it's a pick for you, Paris. This comes from our friend Pud, Phil Kaplan. It's his newest thing. And it's just for you, Parris. It's called Butthole. It is. Use your MacBook's Claude code from your phone.

[B]

Didn't Claude launch this feature in the last week?

[A]

They have a remote control feature. It's terrible. It hardly works now. I haven't tried this. And it's intestine on the iPhone. I don't know. I haven't tried it. It's on test flight.

[B]

Let me go get Gizmo.

[A]

It just came out. I'll find her, find Gizmo and see. But it's funny that he named it Butthole. Obviously he's been watching the show. Yeah, that's very put, isn't it? Connects your phone directly to your MacBook from anywhere. Full terminal version of Claude code on the phone.

[B]

Here we go. There we go.

[A]

Oh, no.

[B]

This is the one time I'll let it happen. And she's kind of hiding it. This time.

[C]

Marshall's embarrassed for us.

[A]

Sorry, Marshall.

[B]

You brought this on yourself. It's 30 minutes past.

[C]

Leo said we would end safely.

[A]

No, no, not quite. It's only two and a half hours. We should have ended. It's about six minutes long. Sorry, sorry. I was going to mention this was going to be my pick. And this is really nerdy. It's called Regex Blaster. If you want to learn regular expressions. It's a video Game where you can learn regular expressions by shooting down incoming alien expressions. So what's the pattern? Bug Crash. Uh. Oh, I think it's gonna be this. Let's see. This. This here. Fire. I got them all. Okay, next level. So if you wanna learn Regex. Actually, this is a really good idea. It gets harder.

[C]

This is the nerdiest thing.

[A]

It's pretty nerdy, but I know our nerds listening would love this. It is called Regex Blaster at MDP GitHub IO Regex Blaster. Now, Jeff Jarvis.

[C]

So last week we mentioned the death of the great man Jurgen Habernaus. And then in the intervening time, Politico chose to remember him. And I'm going to quote my own social post. Lord, I said Politico's remembrance of Jurgen Habermas comes in a banal sophomore confession from the odious head of the nefarious Palantir that the great man dismissed him as a dissertation advisee. Quote. The sting would linger for years. End quote. Not.

[A]

You know what?

[C]

It's.

[A]

It's his cv. Says he studied with Habermas.

[C]

Yeah, it's. It's a big lie. So. So I then heard from a Simon and Schuster publicist in a book about Karp. He tried over these years to say that he studied under Habermas and Habermas was going to be his dissertation advisor. No, he sent a cold call to Habermas, Habermas ignored him. And then he sent 40 pages to Habermas trying to get him to be his dissertation advisor. And Habermas did the courtesy of giving him three typed pages telling him why not. And no, he was never in Habermas's care.

[A]

Oh, my goodness.

[C]

And it's just horrible. Horribly written. He goes on. It's lovesick, too. He goes on about how there was some woman he should have proposed to. It's just awful.

[A]

Alexander Karp, ladies and gentlemen. He's no co founder and CEO of Palantir. I guess he's not. He's a pseudo intellectual. He does practice Tai Chi, though, so I like him for that.

[D]

You know, the first YouTube channel I did that analysis of before I did it on the twit was of the last 18 months of Palantir videos.

[A]

Oh, my God. What did it say?

[D]

Oh, it said

[A]

accelerationist.

[B]

Bad news, brother.

[D]

Yeah, yeah. Yes. You know, there's a. A big shift towards the war fighter focus. Yeah, I read the.

[A]

I read. I tried to read the Technological Republic, his book, and you try.

[C]

You try to get me to read it.

[A]

Well, you know, I Think it's important to read.

[C]

No. No, I'm not going to see the AI documentary. I'm not going to read his book.

[A]

Yeah, you're right. You know what you are?

[C]

I am consistent. I am the philosopher king of the show.

[A]

You are the philosopher king. I don't think it said that, but it's close enough.

[C]

And I'm soon to go upstairs and inject this into my body.

[A]

I don't even want to know where you put that. I don't. Please don't.

[D]

My arm.

[C]

I can show you.

[A]

Thank God. Oh, my God. That was scary. Ladies and gentlemen, this is Under Pressure.

[C]

Under pressure.

[A]

Paris just saw it. But we did get out of Paris. For that, I thank you. Paris Martineau, investigative reporter at Consumer Reports. We're so sorry we lost you to Consumer Reports, but we're glad we got

[B]

you back three seconds later, certain that I'm gone. And frankly, after I see Jeff stick that in his. I am with you.

[A]

No, no, no, no, no, no, no, no, no, no, no, no, no. Jeff Jarvis is. Is healing nicely.

[C]

I have one more week. 10 weeks of this. 10 weeks.

[A]

I'm glad you're feeling better. That's good. You could find the Gutenberg parenthesis in paperback now and also magazine. And don't forget to pre order. Hot type. Coming this summer from his website, jeffjarvis.com. thank you, sir. Marshall Kirkpatrick. So good to see you, Marshall. We'll have you back soon. You're welcome on this show anytime you want, especially if you give us prompts. I like that. I like that.

[D]

Thanks.

[A]

What's up with that app? Is the app it's for Chrome or Firefox, it looks like. I mean, I'm installing it immediately and paying for it, because this is kind of something I've always needed. This is brilliant. This is really fantastic.

[D]

Thanks, Leo.

[A]

Thank you, Marshall.

[D]

Hope it serves you really, really well.

[A]

Yeah, I think it will. I'm gonna get about 10 IQ points smarter from now on. Gizmo the Cat. The American Society for the Convention of Cruelty to Animals certifies that no cruelty was performed on any animal during this show.

[B]

So true. She wants to show you her quad remote control.

[A]

Thank you, everybody, for joining us. We do intelligent machines every Wednesday. We do it right about 2pm Pacific, 5pm Eastern, 2100 UTC. Watch it live on YouTube, Twitch, X.com, facebook, LinkedIn and Kik. Or if you're a club member, and I hope you are in our club, Twit Discord. If you're not a member, Twit TV Club Twit. We need you to join the Twit army after the fact on demand versions of the show at the website twit tv I or on YouTube. And of course you can subscribe in your favorite podcast client get it automatically. Thank you everybody for being here. We'll see you next time on Intelligent Machines. Bye bye. Hey there. It's Leo Laporte, host of so many shows on the TWiT network. Thinking about advertising in 2026, we host a network of the most trusted shows in tech, each featuring featuring authentic post red ads delivered by Micah Sargent, my co host and of course me. Our listeners don't just hear our ads. They really believe in them. Because we've established a relationship with them. They trust us. According to Twit fans, they've purchased several items advertised on the Twit Network because they trust our team's expertise in the latest technology. If Twit supports it, they know they can trust it. In fact, 88% of of our audience has made a purchase because of a twit ad. Over 90% help make it and tech buying decisions at their companies. These are the people you want to talk to. Ask David Coover. He's the senior strategist at ThreatLocker. David said Twitch hosts are some of the most respected voices in technology and cybersecurity and their audience reflects that same level of expertise and engagement. It's the engagement that really makes a difference. With every campaign, you're going to get measurable results. You get presence on our show episode pages. In fact, we even have links right there in the RSS feed descriptions. Plus our team will support you every step of the way. So if you're ready to reach the most influential audience in tech, email us PartnerWIT TV or head to TWiT TV Advertise. I'm looking forward to telling our qualified audience about your great product.

[B]

I'm not a human being, not into this animal scene. I'm an intelligent machine.