Intelligent Machines 856 transcript

Please be advised this transcript is AI-generated and may not be word for word. Time codes refer to the approximate times in the ad-supported version of the show.

Leo Laporte [00:00:00]:
It's time for Intelligent Machines. We're going to talk to Steve Yegi this week. He is the author of one of the coolest Claude Code add ons, Gastown. We'll talk about AI memory, how to get your AI coder to go farther, and what polecats have to do with deacons and mayors. Next on Intelligent Machines, podcasts you love from people you trust. This is Twit. This is Intelligent Machines with Paris Martineau and Jeff Jarvis. Episode 856, recorded Wednesday, February 4, 2026.

Leo Laporte [00:00:42]:
Secretly British. It's time for Intelligent Machines, the show. We cover the latest AI news, robotics. We talk about smart people and smart things all around you. And I am so pleased to welcome our very smart team, Paris Martineau, investigative reporter for Consumer Reports. Salute.

Paris Martineau [00:01:01]:
I'm here and ready to podcast and.

Leo Laporte [00:01:04]:
I think you just saw the Melania movie and you probably want to give us your review.

Jeff Jarvis [00:01:08]:
Oh, my.

Paris Martineau [00:01:09]:
I was there. Well, I tried to be there, but, you know, it was just sold out.

Leo Laporte [00:01:15]:
No room in the inn for Paris Martineau. Oh, well, we'll have to hold that review for later. Also with us, professor of journalistic innovation emeritus at the Craig Newmark Graduate School of Journalism, City University, New York.

Paris Martineau [00:01:28]:
Craig Newmark. It's back.

Leo Laporte [00:01:31]:
And he's a adjunct something or other at Montclair State and of course at SUNY Stony Brook, author of the Gutenberg Parenthesis. What would Google do? I should bring that up this time because we have a Googler on the line with us. Former Googler and the Gutenberg Parenthesis and magazine. That's the one.

Jeff Jarvis [00:01:50]:
And hot type coming up.

Leo Laporte [00:01:52]:
And hot type any day now. Well, June 11th, any month now, sometime this year.

Jeff Jarvis [00:01:58]:
Publishing.

Leo Laporte [00:02:00]:
So if you, if you're paying attention to AI, you know, it's. We're in a very different world. In the last few weeks, we've talked a lot about Claude Code. Clawbot. Nay. No, I'm open Claw. Nay. Claude Bot.

Leo Laporte [00:02:18]:
Nay Multbod. There's another something like that is the opposite.

Jeff Jarvis [00:02:24]:
The nay is what you were, not what you are.

Leo Laporte [00:02:27]:
It was neigh. Open Claw. No, it was ne. Claudebot. Ney Maltbot. Now Open Claw, yes, it's maiden name, very confusing, was Claudebot. But Anthropic didn't like that so much. But today we're gonna talk about, I think, a project of equal interest with its author, Steve Yegi.

Leo Laporte [00:02:51]:
Steve's worked around a little bit. You were at Amazon, you were at Google. You're the author of the very famous memo we talked a lot about. Can it be now 14 years ago.

Paris Martineau [00:03:04]:
15 in a way. Right.

Leo Laporte [00:03:06]:
Holy cow.

Paris Martineau [00:03:06]:
October 2011.

Leo Laporte [00:03:08]:
Steven wrote that piece that we talked about that kind of leaked out about Google and how Google had failed us. And we agreed. Steven, I just want to say. We agreed. So were you a Google engineer? Was that your.

Steve Yegge [00:03:21]:
No, no, no. I was, I think in cloud at the time.

Leo Laporte [00:03:25]:
Okay, and what do you do now for a living?

Steve Yegge [00:03:32]:
That's a trick question. Yeah, I don't do anything right now.

Leo Laporte [00:03:37]:
You have a great medium that I adore, human stories and ideas. Stephen's a very good writer. And your story about Gastown, the future of coding agents, which came out earlier, about a month ago now. You launched Gastown on January 1, New Year's Day, 2026. I installed it shortly thereafter. Even though on your graph of the types of vibe coders, I am really only in a stage five. This is how coding has evolved. We are now in the post YOLO phase.

Leo Laporte [00:04:23]:
I was watching a vibe coder, I think it was George Hotz actually, who said, we are now forget security, just YOLO, everything. That was with regard to OpenClaw, of course. But you have written something that really goes to figure eight, the eighth stage of programming where you have a team. All of this within Claude code. That's what Gastown is. And I love the roles that you've come up with. Gastown. But before we go too much into this, we should probably reiterate your caution.

Leo Laporte [00:05:00]:
This is not for everyone.

Paris Martineau [00:05:04]:
And if you want to know how not for everyone it is, just peruse the text size on his medium post, which will stop you in your tracks.

Leo Laporte [00:05:16]:
Is that. Has that changed in a month or is it more suitable for general public.

Steve Yegge [00:05:23]:
You know, no, don't touch it or you'll die. It's still. It's still there.

Leo Laporte [00:05:29]:
But however, that does not stop anybody, does it?

Steve Yegge [00:05:31]:
It hasn't stopped anyone. Right. And we've got like 500 people on the discord now, you know, and we're going to federate the gas towns together into a wasteland. So I'm going to have 500 gas towns in this huge wasteland, all working together to build something massive. It's going to be really fun.

Paris Martineau [00:05:45]:
Let's take a step back. What? For anyone who's living under a rock, what is Gastown and where did it originate from?

Steve Yegge [00:05:54]:
I mean, in short, it's cloud code running cloud code. It's agents running agents, right? Which kind of turns it into a factory. And you're sort of manag a team. And I knew that this year would be the year of orchestrating. I told everybody it would be. And so that's why I launched symbolically on January 1, right? There's going to be so many orchestrators, you're going to see lots and lots. But they're all roughly the same footprint, which is you have an agent that you work with, then goes and runs others for you, and you can get lots of work done that way. But it's all AIs writing the code that you're not looking at.

Steve Yegge [00:06:26]:
So a lot of people are pretty scared about it for good reason. And it's also. It's really not very safe right now. It's all. I'm operating in a place where I'm almost like a researcher right now. Right. And it's all fun and everything and people are actually finding real use cases for it in enterprise at Fortune 100 companies for Gastown, as junky as it is now. But I wouldn't use it.

Leo Laporte [00:06:46]:
We are in an absolute revolutionary era. It's just hard to believe. And we're waiting because any minute now Anthropic's gonna come out with a Sonnet 5. You probably have it already, right?

Steve Yegge [00:06:57]:
No, but I knew it's gonna land any day because I've been telling people for a week that it's gonna land every day because the half life between their model drops has go months at the beginning of last year to two months at the beginning of this year. And Opus 4.5 is two months old. So like they're due 24th.

Leo Laporte [00:07:12]:
2025 changed my life. 4. 5 was so good. It was mind bending. But I should also.

Steve Yegge [00:07:20]:
I think that the difference between the people who get it in AI today and the people who don't is whether you've used four or five or not.

Leo Laporte [00:07:26]:
I would agree with you. I'm going to read you a little later Mareed once you're gone. I'm embarrassed to read it in front of you. A conversation I had with Claude just, just a little while ago. And you know what's interesting about Claude code is it has a. Claude 45 has a personality. Opus 4.5 has a personality. And it's not exactly sycophantic.

Leo Laporte [00:07:47]:
It's not the ChatGPT5 2, you know, or what it was it 4.0 kiss your ass kind of thing. It's kind of like your best buddy that you want to work with, you want to pair code with. It's fun and it codes really, really well. I should mention that Steve is the author of a book also about Vibe coding, building production grade software with Gen AI chat agents and beyond. You co wrote that with Gene Kim, the author of the DevOps Handbook. And the foreword is by a guy named Dario Amode. I know that name from somewhere. Somewhere.

Leo Laporte [00:08:25]:
When did this come out?

Steve Yegge [00:08:27]:
It went to print in October.

Leo Laporte [00:08:30]:
Things have changed a lot in three months.

Steve Yegge [00:08:33]:
Yeah, so we wrote a book that wasn't actually going to be right until early this year.

Leo Laporte [00:08:38]:
Smart. You skated to where the code was going. Yeah. So this, in a way, this is not, I mean, others have predicted this. It is, I guess, predictable. It seems to me that we are really in an era that people have wanted since the earliest days of personal computing. You know, the, the whole history of coding is higher and higher level languages, closer and closer to English. But now we really are just talking to our computers.

Steve Yegge [00:09:10]:
And everyone's happy about it. Except the engineers.

Leo Laporte [00:09:13]:
Yeah, the guys who learned, who learned fortran. They're not thrilled.

Steve Yegge [00:09:18]:
They're mad.

Leo Laporte [00:09:18]:
They're mad. You did gas down and go. Is that your favorite?

Steve Yegge [00:09:23]:
I think about it because I've never looked at the code, but yes, it was.

Leo Laporte [00:09:26]:
You've never looked at the code?

Steve Yegge [00:09:29]:
One of the instructions, you can tell if you were to look at it, it's pretty wild.

Leo Laporte [00:09:33]:
One of the instructions I have in my Claude MD rules is LEO is always looking at the code in an ide. So make sure you do a good job. It's the only time, the one time I've lied to Claude is I'm not looking at the code.

Paris Martineau [00:09:48]:
Do you think that that really changes something? Do you think it's, quote, not trying to do a good job? Unless that's there.

Leo Laporte [00:09:54]:
Well, let me ask Steve, that's a great question. Does, do these, does this tool obey your instructions? Does it know what you're asking for?

Steve Yegge [00:10:05]:
It's getting, well, four or five is getting scarily smart at reasoning through very complicated situations, which is almost independent of coding. It makes it smart at everything. Right. That said, these things have a tendency to be about 80% right at anything that they tell you or anything that they produce. And so you've gone from using a rake to rake leaves to using a leaf blower and stuff blows around a little bit more until it converges on being correct.

Leo Laporte [00:10:31]:
Let me read you just a little. There's so much good prose in your Gastown post. Gastown is a bit of a swamp thing right now. It sort of oozes rather than whirs. It does work and astonishingly fast, but it requires a lot of manual steering and course correction. And sometimes you have to push it to finish. So that's important. People shouldn't assume that you're just going to push a button.

Leo Laporte [00:10:54]:
You're going to one shot, the next, you know, flappy bird, and be done with it. It does require some interaction. In fact, I would submit it helps a lot to be a coder.

Steve Yegge [00:11:05]:
Oh, yeah? Yeah. It just makes you code. You just code faster with it. You're literally just coding faster. Everything you're doing is just kind of like as if you had a junior teammate who is working really fast and you're just reviewing what they're doing.

Leo Laporte [00:11:16]:
Right.

Jeff Jarvis [00:11:17]:
Because you have the logic of it, because you have the experience in understanding QA. I mean, what makes a coder better at 5 coding? I know it's an obvious question, but I'm curious for your answer.

Steve Yegge [00:11:29]:
I think we are just being faced squarely head on with the notion that there are. There are two levels or two phases of software. There's construction and there's engineering. And people are holding on really hard to construction.

Leo Laporte [00:11:42]:
Yeah, but we don't need to anymore, right?

Steve Yegge [00:11:45]:
Engineering is still there. Everything you need to know. It taxes me harder than it did when I was writing the code by hand because I'm stuck with hard problems all day long. I don't get to just relax and.

Leo Laporte [00:11:55]:
Code and think at a higher level. It makes sense because the computer understands how to talk to a computer. Right? I mean, it's its language. So what we're doing is thinking at a higher level in a human way. And then Claude is kind of interpreting that for the computer, which to me makes perfect sense, by the way. I realized that I liked you a lot when I read that you said you were happiest hand coding in any dialect of Lisp.

Steve Yegge [00:12:24]:
Oh, yeah.

Leo Laporte [00:12:25]:
As everybody who listens to me drone on knows, I'm a lisp fanatic. A scheme fanatic, a common lisp fanatic. I'm an SICP advocate. I'm a. You know, and so the fact that you did this in Go, or if you've done stuff in Typescript, and you like Kotlin, and so I forgive you all of that. But see, this is the important point. I'm not ever gonna stop writing hand COD in common lisp because I love it. I enjoy it.

Steve Yegge [00:12:49]:
Oh, yeah, I'll keep doing that for fun.

Leo Laporte [00:12:51]:
Just for fun. But, but, but when I want a tool now, what I almost always do is I sit down with Claude code and. And let it go. What? So who should be. So let's. I'll Quickly give you a quick summary of. Of. Of Gastown.

Leo Laporte [00:13:08]:
It is. It is a. A colony. It's a. It's a town with hierarchies of agents spinning off other agents, doing other things, working as a team. You have a mayor. You're above the mayor, though, right? The human is above the mayor. The mayor is Claude.

Steve Yegge [00:13:25]:
You're the overseer.

Leo Laporte [00:13:26]:
You're the over.

Paris Martineau [00:13:27]:
Very interesting. The different roles that you assigned to sub agents. How do you come up with those?

Steve Yegge [00:13:34]:
Well, they evolve. It's funny I should tell you Anthropic's take on this, because I've talked. Oh, I'd love to hear a bunch of teams, right. And this is, of course, off the record, just people inside of Anthropic. There's no official take, et cetera, but they are a hive mind. I'll be blogging about that soon.

Leo Laporte [00:13:50]:
I bet they are.

Steve Yegge [00:13:52]:
Not only are they a real hive mind, it operates very differently from all other companies. But that is a template for how I think most companies will become. But anyway, the hive mind kind of likes Gastown and thought it was fun, but they're also a little embarrassed by it because it exposes a lot of what they consider bugs in Opus 4. 5. They're not really bugs. It just was never trained to be a factory worker. Gastown has a bunch of workarounds, including many of the roles that I created that are just there to supervise. Push it along, and in the next model drop, we won't need them.

Steve Yegge [00:14:20]:
And half the Gastown rules will just evaporate.

Leo Laporte [00:14:23]:
I almost feel like half of what we're doing now will evaporate as Claude code kind of absorbs it. You're already seeing, for instance, skills become kind of not necessarily the way to go. Maybe just take that skill and put it in the Claude md. You don't really need to. Claude will understand it. It doesn't need to make it a skill. That's my sense of it. It sounds like you feel like that too, that it's evolving.

Jeff Jarvis [00:14:49]:
It is.

Steve Yegge [00:14:50]:
And they will be able to do more and more and more and more at inference time. But just remember, inference time is the most expensive time to do anything because of context.

Leo Laporte [00:14:56]:
Right?

Steve Yegge [00:14:57]:
And so they will always be delegating to tools, to offloaded CPUs and so on.

Leo Laporte [00:15:01]:
This is the biggest issue in my mind. This is one of the biggest issues with all of these things, is they don't have a memory. And when their context gets filled up and you clear it out, they go, okay, now what do we do?

Steve Yegge [00:15:16]:
So I'll tell you right now, Gastown may not live another three, four months. I don't know. Because they could come out with a clause that just does orchestration. Right?

Leo Laporte [00:15:24]:
Right.

Steve Yegge [00:15:25]:
What will remain. And the only important part of Gastown is beads. And the ledger, the work ledger, the knowledge graph. Your future work, your plans, your present work and your ledger of finished work. And nobody's thinking about that right now. But that's.

Leo Laporte [00:15:40]:
So you started with beads. Before Gastown, there was beads. What is beads?

Steve Yegge [00:15:44]:
Beads is like an issue tracker. It was me and Claude arguing for a long time about how we hated markdown. And finally I said, I want issues in git. And it said, Well, I want SQL queries. And we like wrestled for 15 minutes and came up with beads. And within three hours I was realized I had discovered oil. Like something just magical had happened. Claude loves beads.

Steve Yegge [00:16:06]:
So it is a tool for agents. And what it does, it gives them a memory.

Leo Laporte [00:16:10]:
What do beads look like? Are they JSON files? Are they a marker?

Steve Yegge [00:16:14]:
Currently they are JSON stuffed into SQLite. It's a horrible two tier implementation. I want to get it wanted SQL. We were just like both, but there turns out there are databases that do this and I'm migrating beads to them. It'll happen like this weekend.

Leo Laporte [00:16:27]:
What will you use?

Steve Yegge [00:16:28]:
Adult Dolt is a funny name. Yeah, it's a git database, so.

Jeff Jarvis [00:16:33]:
Right.

Steve Yegge [00:16:34]:
But basically it gives you. What it does is it takes all your work and it turns it into a little knowledge graph of issues, open and closed, whatever future present. And then that becomes federatable. Work now becomes this thing that you can like ship around to different agents. It's the basis for all the. So when you said they're lacking a memory, that is the core problem.

Leo Laporte [00:16:52]:
And Bead solves that because that becomes their memory.

Steve Yegge [00:16:55]:
Their repository of it has a crummy implementation. Right. Which is why Anthropic didn't take it. Now two weeks ago they launched Tasks and they said it's inspired by Steve Yeager's beads. Right. But they didn't take it because I'm sure because they looked at the implementation was like, that's really janky. What they took was the, the task graph part of it. Yeah.

Leo Laporte [00:17:13]:
You know, it's interesting though, because we don't. I think one of the upshots of all this is you don't look at implementation so much. You don't care about implementation so much.

Steve Yegge [00:17:20]:
Well, you care about it. You don't look at it. You care a lot. You care that it works big Distinction. No, no, no. You care about. You have to know. You have to have a mental model in your head of the entire system logging database.

Steve Yegge [00:17:32]:
How observability, how everything works. And whenever you see a feature that's not work the way it should, you should assume that the system is horribly broken, because it probably is. And stop everything and dive in and ultra think about it. Right?

Leo Laporte [00:17:45]:
Excellent.

Jeff Jarvis [00:17:46]:
How do you fix those things? What kind of dialogue is that?

Steve Yegge [00:17:51]:
It's exactly like if I had a really, really smart programmer working with me and I was like, okay, here's the situation. I want it to work like this. And I'm seeing this hypothesis, maybe this, but I don't know. Go figure it out. And Opus, you've seen Opus, it just goes and figures it out. So you treat every little problem as a four alarm like research project and you do that 500 times and that eventually your system will be very good. It will.

Leo Laporte [00:18:14]:
This is from your emergency manual for which everybody should read after they read the Gastown manifesto.

Steve Yegge [00:18:22]:
Yes. Which they're not going to use Gastown because it's dangerous.

Leo Laporte [00:18:25]:
Because you're not going to use it. But if you were to use it. By the way, I feel the same way about openclaw. I want.

Steve Yegge [00:18:33]:
Yo, you should really not use that.

Paris Martineau [00:18:35]:
You should really not use that.

Steve Yegge [00:18:39]:
All of my friends are resetting their signal tokens and their phones and stuff because. Right. They're all exited. Don't even touch.

Leo Laporte [00:18:47]:
No, I have a Mac Mini. And I thought, I'm not gonna buy a Mac Mini. I have a very high end Mac Mini. So I made a new account and I set up openclaw and I went through the whole thing and I gave it all of my Google credentials and I gave it a disco.

Paris Martineau [00:19:03]:
We got to a point where Leo is texting the group of me. Like I guess I could give it a credit card too.

Leo Laporte [00:19:08]:
And I'm gonna. I gave it a dollar five. No, I know Steve's varying, but I gotta tell you, Steve, this morning I woke up in a cold sweat. I deleted the entire account. I've reset all of the tokens. Yeah. Because it. I don't know, I just, it's terrifying.

Leo Laporte [00:19:26]:
And I. And actually things started to happen that I was like, I don't like this. But Gastown's not on that level at all. Right. Gastown's just a tool to help you vibe code.

Steve Yegge [00:19:37]:
That's right. Gastown at Gastown. So, right. The Molt book, or whatever it's called, prompts them and says, you are a lobster. And they respond in kind. Right, right. Whereas Gastown prompts them and says, you are a factory worker and we're doing very important work for humanity and it's important that you do the way we tell you to. And so they behave themselves for the most part.

Leo Laporte [00:19:56]:
This is the loophouse, the three rhythms of work. Can you just describe this? By the way, I love your illustrations. I'm sure they're. Is it nano banana?

Steve Yegge [00:20:05]:
Probably all nano banana. And the prompts were created by Claude code.

Leo Laporte [00:20:09]:
Oh, that's nice. That's nice. You really operate at a higher level than all of that. You don't need to do the details.

Steve Yegge [00:20:16]:
Remember I told some folks in Australia that I was going to launch Gastown in early December, and I told him I was going to launch on Christmas Day and it was a race to get it to the end. So, yeah, nanobanana was wonderful because I was just like, go, make me pictures.

Leo Laporte [00:20:30]:
Yeah. There are three cadences. You've got your top floor. That's where the mayor lives. Right.

Steve Yegge [00:20:38]:
Right. Which is like the kubernetes control plane.

Leo Laporte [00:20:41]:
Ah, that's a good. Okay, good. So the geeks will understand that a little bit better than the middle floor. Actually, the mayor's in the middle floor. So we got. You're up at the overseer in the top floor there. Mayor is. What's the mayor doing?

Steve Yegge [00:20:54]:
Actually, you're the panda bear in the elevator.

Leo Laporte [00:20:56]:
Oh, I am. I go up and down.

Steve Yegge [00:20:57]:
Yeah. Top floor is probably the deacon who watches the entire town.

Leo Laporte [00:21:02]:
Okay. I love the names. It's from a mixed bag of references.

Steve Yegge [00:21:08]:
Yeah. I thought about going all in with Mad Max and I thought, eh, why invite some sort of a lawsuit? Right.

Leo Laporte [00:21:16]:
So what are the other influences?

Steve Yegge [00:21:18]:
Oh my gosh. There's Cat's Cradle from Kurt vonnegut with Ice 9 stuff. There's the Deacon is Dennis Hopper's Waterworld character.

Leo Laporte [00:21:27]:
I love it.

Steve Yegge [00:21:28]:
Yeah. And Witness actually was from, you know, Mad Max.

Leo Laporte [00:21:30]:
Right. And then on the ground floor you got your polecats. Where do the polecats come from? What's that all about?

Steve Yegge [00:21:36]:
That's Mad Max. Those are the guys on the stilts. They were literally.

Leo Laporte [00:21:39]:
Oh, the pole cats performers. Okay, that's. Yeah. Okay. They have a little aluminum paint on their face there. And then down here, the engine that drives the loop. And that's Claude, I guess.

Steve Yegge [00:21:52]:
Yeah. And beads.

Leo Laporte [00:21:53]:
And beads. So that's actually the piece of gas down that I'm most likely to keep using actually, is beads. Can you use beads standalone?

Steve Yegge [00:22:04]:
Absolutely. Beads can be used with any workflow.

Leo Laporte [00:22:06]:
Okay.

Steve Yegge [00:22:07]:
It's just, you just. And the nice thing is you just say, we're not using markdown, we're using beads Run, you know, bdprime. And you just put that in your one line in your thing and then.

Leo Laporte [00:22:16]:
From there everything into it. Okay.

Steve Yegge [00:22:17]:
You can say, yeah, just slurp everything in. And it'll just. From then on it'll just use it. It'll be totally. You'll just love it.

Leo Laporte [00:22:23]:
Okay. The other tip that you have in your emergency user manual is use tmux. What's really cool about this and what I love about this, in fact, George Hotz was saying this. He's saying we've divided the world now into the kind of dumb people with the mouse and they click on the menu and the button and they do the buttons. And then we've got the smart people who are sitting at a terminal going, typing, or better yet, just talking into their terminal. The terminal's back, baby. I'm so happy about that.

Steve Yegge [00:22:55]:
It is, but. Right. Multiplexing terminals is becoming really important and TMUX is pretty good at it. But if you look at like cloud cowork, which is also a terminal in a way. Right.

Leo Laporte [00:23:04]:
Isn't that wild? It's a virtual machine and it's. Yeah.

Steve Yegge [00:23:07]:
You could imagine a world where there is a very fancy ide like Terminal Multiplexer.

Leo Laporte [00:23:12]:
Yeah, Right, Right.

Steve Yegge [00:23:14]:
Somebody can make one for us.

Leo Laporte [00:23:17]:
I actually think Emacs might be the end up.

Steve Yegge [00:23:21]:
Well, yes, you and I know that, but I don't think we're going to pitch anyone else on it.

Leo Laporte [00:23:26]:
Okay.

Paris Martineau [00:23:26]:
I've been shouting from the rooftops about Emacs.

Leo Laporte [00:23:29]:
Fine. Emacs is just perfect software.

Steve Yegge [00:23:31]:
That's fine. But imagine though, I mean it's like fifth on my to do list, but I am going to have a great Gastown Emacs client.

Leo Laporte [00:23:38]:
That's what I'm thinking. And I think Claude code does pretty well in elisp. I've had it been half of my Emacs config now is done by Claude code.

Steve Yegge [00:23:49]:
Only half?

Leo Laporte [00:23:52]:
I'm slow. I told you, I'm only level four here. Give me a break. We're talking to Steve Yegi. He's the guy, if you remember the name. It's so funny because I'm thinking, I know guy, I know this name. Back in 2011, wrote the memo about Google that accidentally leaked. Was it really an accident?

Steve Yegge [00:24:11]:
Yes. I was very drunk and the interface was actually very hard to use.

Leo Laporte [00:24:15]:
Yeah.

Jeff Jarvis [00:24:17]:
What was the reaction at the time is on Google when that got Out.

Steve Yegge [00:24:21]:
Oh my gosh. It was pretty intense. You know Vic and Dotra, the guy who headed up Google wanted me dead.

Leo Laporte [00:24:29]:
That was his baby. He was like. In fact, as soon as Google was over, so was Vic.

Steve Yegge [00:24:34]:
I'll tell you guys something. That rant that went out the platform one was part two of an 11 part series that I had planned out and I, I had already published this one internally. Not many people know this. I was mad at Google in nine different ways and there was an intro and a finale. So I had nine different dimensions that I was going to smack them down on. Platforms was only one, but that one accidentally got out and that was the end of that series you wrote.

Leo Laporte [00:25:00]:
Google is a prime example of our Google's where you were working at the time. Complete failure to understand platforms of the very highest level of executive leadership. Hi Larry, Sergey, Eric, Vicky. Howdy, howdy. Down to the very lowest LEAF workers. Hey yo. We all don't get it now here we are 15 years later. Is Google improved?

Steve Yegge [00:25:24]:
Yeah, amusingly Google has kind of gotten it less than everybody else. It really was anti platform baked in from the beginning. Yeah, Google is executing a lot better like now. Now as in like the last year. But it took them a long time.

Leo Laporte [00:25:38]:
It took them a long time. I had written them off. In fact, we changed the show from this week, Google back in last year, early last year. But Gemini, especially the Gemini 3 kind of changed my mind on all of this. Is Google poised to do a Claude code kind of thing? We've got Codex from OpenAI.

Steve Yegge [00:25:59]:
I hope so, man. I don't want us all to be dependent on one model.

Leo Laporte [00:26:02]:
No.

Paris Martineau [00:26:03]:
Why is, why do you guys think Claude code is so dominant in the CLI space?

Leo Laporte [00:26:08]:
It's the best.

Paris Martineau [00:26:09]:
Yeah, but why, why can no one else create anything equal?

Jeff Jarvis [00:26:13]:
What motivated them to go this way?

Steve Yegge [00:26:16]:
I don't even know if they know, but I mean, I suspect it's their relentless focus on coding which has forced it to become more and more logical kind of quickly than the other ones.

Leo Laporte [00:26:25]:
Instead of doing the chatbot, I mean they have a chatbot, but I don't.

Steve Yegge [00:26:29]:
Know, I don't know why. I'm not an AI guy. You can just tell that it's pulling ahead. I think that it's a self reinforcing loop. They're getting better programmers and better signal and better training data and so they're just, they're accelerating and that worries me because that becomes a more pronounced lead over time. I really want some other model providers to catch up.

Leo Laporte [00:26:50]:
Can you use Gastown with open weight models.

Steve Yegge [00:26:52]:
Can you use it with Kimi? No. It dissolves into a pile of goo. Right. Remember I wrote it for Opus 5 and we gave it to Opus 4. 5. And now we've got all these hacks. Right, Right.

Leo Laporte [00:27:04]:
Okay. All right.

Paris Martineau [00:27:05]:
Have you played around much with codecs or other things?

Leo Laporte [00:27:09]:
Anti gravity?

Steve Yegge [00:27:09]:
I'm going to go back to what I said at the very beginning of the show, which is like four or five was like. Like an event horizon, a singularity. Right. Everything before it. And if you're using anything other than it, you're just in old AI land making conclusions about stuff that's not true anymore. Right. A four or five just completely changed it.

Leo Laporte [00:27:29]:
I've been wanting to say that, but I didn't have the courage and the confidence that you do.

Steve Yegge [00:27:34]:
Well, hey, the confidence comes from me trying to do stuff that was just too hard. And so I know exactly how they fail and fall on their faces. And, you know, four, five was the first one. The first one to be like, it could finally do it, you know.

Leo Laporte [00:27:47]:
Yeah, exactly. I had played with it. I've been playing with it for, you know, since I guess3.5 came out and there was just a qualitative leap.

Steve Yegge [00:27:57]:
It was as big as the 4.0 GPT. 4.0, remember that?

Leo Laporte [00:27:59]:
Yeah.

Steve Yegge [00:28:00]:
That one was the first one that could do a thousand line source file and make a pinpoint edit to it without screwing up all the formatting or other mistakes.

Leo Laporte [00:28:07]:
Right.

Steve Yegge [00:28:07]:
And that was the moment because most of the world's code is in files that are a thousand lines or less. And so that was a huge breakthrough, 4.0. But I don't think that we had the next major breakthrough until Sonnet 3. Maybe 7 or 3.5 is a big, big, big step function up. And then 4. 5 was the next big one.

Paris Martineau [00:28:23]:
Right.

Steve Yegge [00:28:23]:
A lot of incrementals. But 4.5 was a really big step.

Leo Laporte [00:28:27]:
Here's what it looks like, by the way, to use Gastown, you're the mayor, rather the overseer, the panda bear. And you've got all your little guys working, designing, coding, debugging, reviewing. Now it can get into weird loops where it, you know, it's just kind of. It's a little bit of a house of cards. And if one thing goes wrong, it can all collapse in some interesting ways. But the beauty of this, this is what Harper Reed has told us again and again, is you just throw it out and start over. It's cheap. Do you have.

Leo Laporte [00:29:00]:
What is your token budget these days? You Burn through tokens.

Steve Yegge [00:29:06]:
Yeah. I'm not going as crazy as a lot of people. I mean, I know there's people that are spending thousands and thousands a month. You know, I'm burning about 12 to 1400amonth, I think. But that's, that's, that's, that's because I keep going through bottlenecks where I'm. I'm only using one or two agents at a time for days because of the adult migration and other stuff going on. You know how work can sometimes be parallel and sometimes surreal.

Leo Laporte [00:29:28]:
Yeah.

Steve Yegge [00:29:29]:
So once I get to the federating the wastelands, thank goodness there's a bunch of big companies who are giving me influencer token budgets because I'm going to be spending a lot of money taking us to the next. I'm going to go into the next frontier, which is what happens when you have. When you're trying to have multiple towns where a town represents like a person. Right. With a team. How do you do work at that scale?

Leo Laporte [00:29:53]:
You really only have one Claude code and you've just subletting little.

Steve Yegge [00:30:00]:
I tend to work with three at once.

Leo Laporte [00:30:01]:
Oh, you do? Okay.

Steve Yegge [00:30:03]:
My personal workflow is mayor and two crew. Okay. And I actually have eight crew in each rig. Beads and Gastown. And they. But I have two crew that I designate as my sheriffs. All right. It's not an official thing, but I have a special bead for them.

Steve Yegge [00:30:16]:
And when they wake up, they go and do they? Do they? Look at PRs. See, I'm maintaining these projects by myself and I'm getting like 50 PRs a day or more. And I stay on top of them by like having.

Leo Laporte [00:30:27]:
Are these generated AI, generated PRs or from humans?

Steve Yegge [00:30:30]:
They're all familiar, man. Gastown's Lai. A human would never generate a PR for us. That's not true. There's one company dolt that's actually doing it, you know, Bless him.

Leo Laporte [00:30:40]:
Okay. Okay, so. And how do you manage three different claws? Do you have three computers or do you have it all in one system? They're sharing beads, I presume.

Steve Yegge [00:30:48]:
So it's tmux, obviously. Right. So I'll just fire up terminals like in different colors.

Leo Laporte [00:30:53]:
That's why you have tmux.

Steve Yegge [00:30:54]:
Okay, yeah, but I have TMUX groups that cycle along the crew. So I will farm out a bunch of design problems to crew, say, go. Give me deep research and give me recommendations. And the one at a time, I cycle through and read their recommendations. I feel like Jeff Bezos. Right. He gets presentations all day long from Smart people who are giving them hard problems. Right.

Steve Yegge [00:31:13]:
It's really actually quite exhausting working with Gastown because you're making your workers go off and take care of all the hard stuff and then they give you the hard problems, right? They do the easy.

Leo Laporte [00:31:19]:
The problem is they're fast and humans are slow.

Steve Yegge [00:31:22]:
Yeah. Well, you will always reach an equilibrium if you're working with multiple agents where there's always one waiting for you.

Leo Laporte [00:31:28]:
Right. There's always something to do.

Steve Yegge [00:31:31]:
It's madness. I actually think that there's a vampiric effect happening. I think that there's something bad happening. No, no, no, seriously, people are getting drained. And I think that people like me are actually contributing to the problem because we're creating this appearance of super hyper productivity that's becoming a bar that people are setting. We're trying to live up to you, which is ridiculous. And my next blog post is going to be about slowing down and capturing some of the value from AI for yourself. Because if you're, let's say you're 10 times as productive with AI, you can work eight hours a day and give all that to your employer and they've captured 100% of the value.

Steve Yegge [00:32:05]:
Or you can work for a half hour a day and be as productive as your peers, in which case you've captured 100% of the value. I think that the middle is going to write the happy medium is going to be the right answer, but the company will always try to extract it all from you.

Leo Laporte [00:32:19]:
Is that what this latest post, the software survival post, is kind of about?

Steve Yegge [00:32:23]:
No, no. I liked all my posts to be about different things, but yeah, no, that one's just. Seriously, the next post I described is people just pushing back on companies, trying to push them too hard. With AI specifically, this is going to be a real problem. The software survival post was how to write something that's not going to die.

Leo Laporte [00:32:40]:
Right.

Jeff Jarvis [00:32:42]:
Steve, at the beginning you talked about the danger of this. Your caveat is well said. But I'm curious for you to define that danger a little bit and at a very high level. What starts to eliminate the danger is it things have to be within one company within one, A hard boss. I mean, what is. Because the last few weeks have shown tremendous excitement about this. Finally the agents arrive and do things and people are playing. And I think the playing is good, but the playing is dangerous.

Jeff Jarvis [00:33:16]:
Where do you see that going?

Steve Yegge [00:33:18]:
We're hitting the biggest innovators dilemma that we've ever seen before, where companies are going to. All companies are going to have to pivot very quickly. And the ones that are siloed and have old crusty organizations that don't work well together and so on are just going to. They're going to break if they can't bend. The danger is that with this hyper productivity because people are figuring out how to use it and bring it into the org it immediately in the manufacturing Toyota plant since it immediately pushes all your bottlenecks downstream. So like right now, my advice to companies is just start. Like Tommy Wiseau told that guy on X, he's like, I want to start writing screenplays, Tommy, what should I do? And Tommy's like, start, right?

Leo Laporte [00:33:58]:
I did not hit her. I did not.

Steve Yegge [00:34:02]:
Just get started on it. Because you're going to have a lot of lessons to learn that are bespoke to your company. And you got to get on that now.

Leo Laporte [00:34:09]:
Okay? Not a lot of people quote screenwriting advice from Tommy Wiseau. I think you get a lot of credit, a lot of credit for that. Hey, can I just ask you, do you have a eight foot screen? Tell me about your setup just a little bit. I mean, this is just a tourist question.

Steve Yegge [00:34:24]:
Oh, I have this LG Giant 45 inch curved monitor. It's a 5K 2K bent. It is so cool. And I have terminals splattered all across it. It looks seriously, it looks like Minority Report or whatever that movie was with Tom Cruise. Yeah, we're in a new people. I showed it to like, I showed it to like, you know, my buddy Kent Beck, you know, who's, who hasn't quite flipped on this yet yesterday. And he's like, this scares me.

Steve Yegge [00:34:49]:
When he was watching me do the crew, you know, it's just Ken is.

Leo Laporte [00:34:52]:
Of the old school. Ken is that, you know, the agile guy that the old school guy. It's really interesting, you know, I wonder how people like that are going to adjust to that. Some of them will be. And I bet Ken will do this, I don't know, will go, yes, this is the next thing. And some will resist it and say, no, no, you can never write good code this way.

Steve Yegge [00:35:10]:
I mean, look, Kent is just being. Kent is very, very, you know, he's doing the smart thing, right? He's like, he's onboarding. He has definitely drunk the Kool Aid. He's using Claude code all day long and he codes his ass off. But Gastown he finds a little scary for, I think, good reasons. I'm telling people to find it scary. Yeah. But you know, one company, I mean, I've been big Fortune 50 company, a big company came to me and said they're using Gastown for their stuff.

Steve Yegge [00:35:37]:
And I'm like, whoa, are you mad? And he's like, they found some use cases for which it actually works, and it can speed up tremendously. Their ability to roll out custom data centers and do other things. It's nuts. It's nuts. People are finding use cases for it.

Leo Laporte [00:35:51]:
Do you need. What hardware, what computer do you use? Do you need a heavy duty computer to do this? Or is Claude doing really all the work on the server?

Steve Yegge [00:35:58]:
I'm gonna be doing all my Gastown work on my phone before too long.

Leo Laporte [00:36:05]:
Actually. That's the main thing. I wanted Clawbot for a claw, whatever open claw for was, so that I just could talk to it on my phone so it wouldn't have to come into the office and that time. So that's my real goal. I don't care about anything else. I'm gonna just set up, set up. I'll probably set up a little.

Paris Martineau [00:36:22]:
You can talk to Claude on your phone. Leo, you realize that. You realize they have an app, right?

Leo Laporte [00:36:28]:
No, no, no, that's not the. No, no, no. I want to be able to talk to Claude Code. Claude code.

Steve Yegge [00:36:33]:
He wants. Ssh.

Paris Martineau [00:36:34]:
Yeah, you need to be talking to Claude.

Leo Laporte [00:36:37]:
Code Verb.

Paris Martineau [00:36:38]:
Away from your house.

Leo Laporte [00:36:40]:
Yes, because. Yes. Do you just hear him? You've constantly got something to do. You're very busy.

Steve Yegge [00:36:46]:
Haven't you ever woken up in the middle of the night and wanted to tell Claude something?

Leo Laporte [00:36:50]:
Yeah.

Steve Yegge [00:36:50]:
Yes.

Paris Martineau [00:36:51]:
You could just open up a laptop. You're in bed. You have access to electronics.

Leo Laporte [00:36:56]:
This is, by the way, a form of burnout that I do see happening. A lot of people are having trouble sleeping.

Jeff Jarvis [00:36:59]:
We.

Leo Laporte [00:36:59]:
We are so excited about what's happening. Right. We're in a very interesting time. Stephen Yegi, it's such a great honor to get you on. I really appreciate what you've done with beads with Gastown. Thank you for doing that work and thinking out of the box and continuing to do it. And I hope you continue to do it. People should read Stephen's posts on Medium.

Leo Laporte [00:37:25]:
They're worth it just for the prose. Great writer, Steven Yegi, Medium. He's a little too busy to have a good URL. I'm just saying, Stephen. Y E G G E medium dot com. I appreciate your time, Stephen. It's really great to talk to you. And by the way, best backdrop ever.

Leo Laporte [00:37:46]:
That's real, isn't it? Ha.

Steve Yegge [00:37:48]:
No, it's a green screen.

Leo Laporte [00:37:49]:
Oh, damn it.

Paris Martineau [00:37:50]:
I would say it's very clearly a green screen.

Leo Laporte [00:37:52]:
You fooled me. It looks good. Cheers. You lived in the basement. Thank you, Steve.

Steve Yegge [00:37:57]:
Thanks for having me.

Leo Laporte [00:37:57]:
Thanks. Bye. Bye. Thank you, everybody. Try Gastown. Such an honor to talk to him. I knew I knew that name.

Paris Martineau [00:38:07]:
That's so funny.

Leo Laporte [00:38:09]:
Isn't that hysterical?

Paris Martineau [00:38:11]:
I love that. It was a 3,700 word drunk blog post that leaked. And then objectively, one of The Funniest Form Part 2 of 12. Objectively 1 of the funniest formats to drunkenly leak.

Leo Laporte [00:38:25]:
He's quite a character. And what. And you know what a great writer. I really enjoy reading his posts. They're really good.

Benito Gonzalez [00:38:32]:
I'd like to see the other 11. That'd be fun.

Paris Martineau [00:38:34]:
I was about to say, I can't believe it took us this long to get an Emax guy on the show.

Leo Laporte [00:38:39]:
Finally. Emacs and lisp, baby. I didn't even know it. I didn't even know it. I just wanted.

Paris Martineau [00:38:44]:
That's one of the first things that jumped out to me when reading his.

Leo Laporte [00:38:49]:
Yeah. Yeah. Well, you know, it's funny. I've been thinking and I understand now. When he said hive mind anthropic, that completely clicked for me because it is. There is something going on where the people who get it, we're in this kind of weird, liminal space and others who don't get it, we just. We're drifting apart. Drifting apart.

Leo Laporte [00:39:14]:
I wish I weren't so old. I just would love. You know, I wanted us to hang out with Harper Reed and just kind of. Oh, anyway, we gotta take a break. Believe it or not. And then when we come back, there is so much AI news. My new Vibe coded tool generated a lot of stories.

Jeff Jarvis [00:39:34]:
Oh, did it ever.

Leo Laporte [00:39:37]:
A lot of stories. I don't know if we're gonna get too much.

Paris Martineau [00:39:39]:
It feels like I'm looking at the Jeff section over here.

Leo Laporte [00:39:43]:
This is the. I gotta. Sorry, I'll cut it down. I gotta cut it down because we don't. Yeah. Anyway.

Paris Martineau [00:39:48]:
Anyway.

Leo Laporte [00:39:50]:
More with Paris and Jeff in just a moment. Our show today brought to you by Delete me. If you've ever wondered how much of your personal data is out there on the Internet for you to see, you don't even have to run Clawbot to put that stuff out there. Because guess what? There are people who are collecting it. We call them data brokers. And what they do is legal. What they do is perfectly fine, according to feds. That you may not like it so much when you go out there and you see your name, your contact info, your Social Security number, your home address, even information about your family members.

Leo Laporte [00:40:30]:
And the worst thing is, not only are they compiling it, they're selling it online to anybody who wants it. Foreign governments, our government, law enforcement, marketers, hackers, anyone on the web could buy your private details. And just I mean the consequences. Identity theft, phishing attempts, doxing and harassment. There is a way to get your data out of these data brokers. They're required by law, if you ask them to take it out, to get rid of it. But there's a little problem here. First of all, there's more than 500 of them.

Leo Laporte [00:41:00]:
And every one of them hides that form somewhere on their site. It's not easy to find. Plus there's the problem that new ones start every day because it's such a lucrative business. And then finally, the truth of it is, even if you get them to delete the stuff, they're just going to start over and rebuild your portfolio, your dossier. Nothing to stop them from doing it. Except Delete me. Very important. But Delete Me is a subscription service.

Leo Laporte [00:41:28]:
So you know, it's not a one time deal. It removes your personal info from those hundreds of data brokers you sign up. You give Delete me exactly the data you want removed. That's important because you know, you don't want them necessarily to wipe your presence off the net. You just want certain things deleted from those brokers. Then their experts, and this is their business, they do this 24 7. They know where all the sites are, where all the forms are. They take it from there.

Leo Laporte [00:41:53]:
They will send you regular personalized privacy reports showing what they found, where they found it, what they remove. So you know, they're doing their job. And then this is the key. It's not a one time service. They're going to continue to monitor and remove that personal information you don't want on the Internet. It's the only way to do it. To put it simply, Delete me does all the hard work you could. I'm not saying you couldn't do it.

Leo Laporte [00:42:16]:
You could, you wouldn't want to. It'd be a full time job. Let them do it. The hard work of wiping you, your family, your businesses, your employees, personal information from those data broker websites. We did this. We had to. We were getting phished. Any company should absolutely be doing this for management at very least.

Leo Laporte [00:42:35]:
Otherwise you're guaranteed they're gonna start phishing you with information they buy from data brokers, take control of your data, keep your private life private. Sign up for Deleteme We've got a special discount for our listeners today. 20% off your DeleteMe individual plan when you go to JoinDeleteMe.com TWIT and use the promo code, Twitter checkout. The only way to get 20% off though is to go to this website and this website. Exactly. Joindeleteme.com TWIT There are, there's other sites you gotta go to the right one. Joindeleteme.com TWiT and do enter the offer code TWiT at checkout. 20% off joindeleteme.com twit the offer code is TWiT.

Leo Laporte [00:43:13]:
We thank them so much for the job they do and the support they give us. And we appreciate it because it helps us do our job.

Paris Martineau [00:43:22]:
Okay, not to stump for advertising, but I will concur. It's really hard to find the right button to get your name off.

Leo Laporte [00:43:30]:
Oh, yeah, good luck. Let's go to one and try.

Paris Martineau [00:43:32]:
I go to. I, I do all of this myself because I'm always one trying to delete my own stuff from the Internet. But I'm also then using those same tools to maybe find other people's contact. They make it really impossible.

Leo Laporte [00:43:45]:
It is kind of a boon for you as a reporter though, isn't it? You can easily find anything you need. What's their home phone number?

Paris Martineau [00:43:51]:
I mean, yeah, most people's information is out there like in one of two data brokers always. It's ridiculous.

Leo Laporte [00:43:58]:
One thing we should start before we get to the AI thing. Massive layoffs at the Washington Post. And I know you both are journalists and care a lot about this. I was shocked. Jeffrey Fowler, who's their very esteemed technology reporter, was one of the laid off.

Paris Martineau [00:44:13]:
Natasha, one of the best tech reporters out there, was also gone. She literally has a story published today on the front page of the Wall street of the Washington Post about kind of ins Elon Musk's attempts to kind of rework Grok and change X. She's a phenomenal journalist. There's so many phenomenal journalists. Hundreds of journalists who've lost their jobs, they shut down their sports section, basically have axed. Almost all their international coverage books is gone. Metro is being reorged.

Leo Laporte [00:44:45]:
Is this because Jeff Bezos is running out of money?

Paris Martineau [00:44:49]:
Ostensibly it's because the Washington Post is not, not profitable.

Leo Laporte [00:44:54]:
But, well, he must have known that when he bought it.

Jeff Jarvis [00:44:57]:
Well, I know Will Lewis, who's the publisher. He was sent in. He's a Murdoch and he was sent in to be the tough guy. And you know get it down to a shape and it's going to be a meaningless. Also it already is a zombie newspaper and it's going to be meaningless and also ran and not part of the conversation, but harmless to certain people in power. I, I put in the discord, Leo the Marty Baron who was the brilliant editor of the Post. I think the best editor in America. His statement today was very strong.

Leo Laporte [00:45:35]:
What did he say?

Jeff Jarvis [00:45:36]:
He said that, sorry, I gotta go back to it now. This ranks among the darkest days in the history of one of the world's great news organizations. He said, of course there are acute business problems to be solved. No one will deny that the Post challenges, however, were made infinitely worse by ill conceived decisions that came from the very top. From a gutless order to kill a presidential endorsement 11 days before the 2024 election to a remake of the editorial page that now stands out only for its moral infirmity. Loyal readers, livid as they saw owner Jeff Bezos betraying the values he was supposed to uphold, fled the Post. In truth, they were driven away by the hundreds of thousands.

Leo Laporte [00:46:21]:
Yeah, I canceled my subscription. It's hard because. Well, now it's easy. It was hard because they did have good tech coverage and we, you know, we were often referring to their article, but they won't anymore. So I guess they've basically.

Paris Martineau [00:46:34]:
I mean, from what I, I've seen online, it seems like they're almost shuttering their SF bureau. The cuts are that deep.

Leo Laporte [00:46:40]:
Wow.

Paris Martineau [00:46:40]:
Which is.

Jeff Jarvis [00:46:42]:
The correspondent is in Ukraine was laid off.

Paris Martineau [00:46:45]:
I mean, yeah, a foreign correspondent for the cuts have really decimated all of their international coverage. And one of their foreign correspondents is literally reporting like in, in out in Ukraine doing filing reports from her car. And that's when she got the call that she was being let go.

Leo Laporte [00:47:04]:
Well, I mean, clearly they're just kind of giving up on being a newspaper. Right. Or a news source. I don't know what they're going to be, but they're not going to be a newspaper. They're not going to be a.

Jeff Jarvis [00:47:12]:
They used to have a magnificent book section. Ron Charles and Dec Rothfeld are to me, two of the finest book critics in the country. They're now laid off. The whole section's gone. Styles basically gone. It's been a shell of itself. I asked a former, very top editor there, not Marty, what now? Three years ago, whether it could be saved. And he said at that point, no, it was a lost cause.

Leo Laporte [00:47:38]:
Then such an opportunity because it's the Washington D.C. newspaper. So you know, you're in the. In the seat of government of the most powerful. Yeah.

Jeff Jarvis [00:47:49]:
You know, it's fascinating, Leo, the fight that went on over and over and over again inside the Post is that Don Graham, a beloved, saintly man who ran the Post, believed that the Post had to be the paper of Washington D.C. the local newspaper, the metro newspaper. And when online came along, others said, well, no, it's our opportunity to really.

Leo Laporte [00:48:10]:
It's a national. It could be a national. Like the New York Times.

Jeff Jarvis [00:48:12]:
National, international paper. Paper.

Leo Laporte [00:48:14]:
Yeah.

Jeff Jarvis [00:48:14]:
And they. They fought and fought and fought and never came to a conclusion.

Leo Laporte [00:48:19]:
This is really the end of the newspapers. I mean, honestly. Yes, that's really what's going on. Right. They're not important, you know.

Jeff Jarvis [00:48:27]:
Similarly, the LA Times, way back in the day used to think it could be a national newspaper. It ignored its amazing city. It has now a penetration of 4% in an incredible metropolis.

Leo Laporte [00:48:38]:
What replaces.

Jeff Jarvis [00:48:41]:
Chaos at first, but then I think a lot of independent media rethought, you know, in pardon me for this plug in hot type out in June 11, I talk about how at the turn of the prior century, it wasn't clear what medium would win and what where it would stand in the media universe. Magazines, books and newspapers all fought for the same territory. Newspapers became the vacuum of culture, and they competed with magazines for a much lower price. Then they added magazines in magazines, tried to figure out how to structure themselves and they were doing books and books were scared of magazines and they all went around. And so what I think happens now is that probably magazines and newspapers as we knew them are dead. Books live on. And then we hold these four podcasts.

Leo Laporte [00:49:34]:
Who's buying books?

Jeff Jarvis [00:49:36]:
Well, books have never been.

Leo Laporte [00:49:38]:
Oh, that's right. They live on. Because they were always kind of.

Jeff Jarvis [00:49:42]:
They're small. Yeah, I think that's. Well, I think that's key to possible survival.

Leo Laporte [00:49:46]:
Right.

Jeff Jarvis [00:49:47]:
Is media at a human scale, like, seriously like this? I think podcasts, newsletters, AI things, collaborative networks, things we haven't seen invented yet, will claw for, pardon me, for the use of the word, a position in a new media ecosystem. And I'm probably too old to see how that turns out. In fact, I'm sure I am.

Leo Laporte [00:50:12]:
Well, I'm sure this is of some interest to Paris, being a journalist, being.

Paris Martineau [00:50:19]:
A thrice laid off journalist. Yes.

Leo Laporte [00:50:22]:
I mean, I think if you had a home at a bit. At one of the magazines that's going to survive. The only ones.

Jeff Jarvis [00:50:27]:
But Paris, when you decided to be a journalist, did your parents say, are you knocking futs?

Paris Martineau [00:50:32]:
I said to myself, Are you knocking fuds? I said, I was like, this is a bad to myself. I was like, this is a bad idea. It's probably not going to work. This is an industry that's already been dying and will only continue to die at a more aggressive rate as I enter it. And so I gave myself five years to see if I could make it work. Otherwise I'd go to law school. But it worked or has so far is law school.

Leo Laporte [00:50:55]:
How many years ago was that in the cards?

Paris Martineau [00:50:59]:
A little over 10, I guess it was like 11 years ago.

Leo Laporte [00:51:01]:
Oh, a long time ago. So law school's out now?

Paris Martineau [00:51:05]:
Yeah. I mean I was talking to a friend yesterday who's halfway through their law school program and I was like, I don't know. Will we even have laws in two years when.

Leo Laporte [00:51:15]:
You know, it's actually, actually interesting because the stock market has punished a lot of the legal software companies. This was a very big crash last couple of days because the stock market was looking at things like Vibe coding like Claude and saying, you know What? Why would LexisNexis, why would Westlaw, why would any of these, why would Thompson, why would everybody.

Paris Martineau [00:51:44]:
It's baffling to me because LexisNexis and Westlaw have and I assume the other ones have access to a lot of stuff. The reason you pay is the things they have access to.

Leo Laporte [00:51:55]:
Maybe it's search them and their databases, but their software that they also add.

Jeff Jarvis [00:52:02]:
A level of value in organizing and categorizing and that added value goes away.

Leo Laporte [00:52:09]:
This is the story from Bloomberg Anthropic AI tools spark sell off from software to broader markets market but in particular legal services.

Jeff Jarvis [00:52:20]:
Give the examples of the companies in that story. I think to Paris's point, no, because.

Paris Martineau [00:52:23]:
I believe that those were the right ones. But I just remember when I read.

Leo Laporte [00:52:27]:
That I was Thomson Reuters and LegalZoom the worst performers in the US and Canada. LegalZoom provides. LegalZoom probably is in trouble. A long time advertiser on the radio show. I like LegalZoom but in fact I used them to incorporate twit back in the day and so forth and. But now you would probably write all those documents with the help of an AI.

Paris Martineau [00:52:45]:
What does LegalZoom do?

Leo Laporte [00:52:47]:
They. They help you make a will, they help you incorporate. They do all the legal. It's basically like a paralegal. It's not a law.

Jeff Jarvis [00:52:56]:
They replace the lawyer and now they get replaced. Yeah, but Leo, here's what I don't understand about this. This route.

Leo Laporte [00:53:04]:
The.

Jeff Jarvis [00:53:04]:
I understand why the legal Companies go down because its belief is they can be replaced by vibe coding, cowork and so on. What I don't understand is why there was a route in the software market. Isn't this the victory of software in the end or is the belief that software kills software now?

Leo Laporte [00:53:21]:
Well, when you say software that's like anything that runs on a computer. They're talking about companies that make business line of business packages that they sell the company.

Jeff Jarvis [00:53:35]:
There's a general tech cell on from this same spark.

Leo Laporte [00:53:38]:
Well this is what the question is. We were talking about this on Windows Weekly because Microsoft had a big sell off last week and the conventional wisdom was that people looked at Microsoft's results which were very good but saw that they were spending like almost $100 billion a year developing AI data centers without any kind of profit at the end of the tunnel. And the market said that's not a a long term future. So that was the consensus at the time. I think the market was also looking at the fact that Microsoft's business is selling software as well. On the other hand, Microsoft's biggest business as is Amazon's in some degrees Google is data centers and that's gotta be in the long run benefited by the AI boom. So I don't know. The sell off started.

Leo Laporte [00:54:27]:
Traders pointed to a release on the Anthropic website as the reason behind steep declines in the shares of credit marketing services company Experian Business and legal software maker RELX PLC and the London Stock Exchange Group PLC Asian stocks also slid. This is from Bloomberg. Tata Bellwether Tata Consultancy sank 6% Infosys 7% Cloud Account Cloud accounting software maker 0 16% In Australian trading the most such 2013 now the stock market makes mistakes. I asked this, what are you talking about?

Paris Martineau [00:55:02]:
The stock market's perfect and is never over.

Leo Laporte [00:55:05]:
I was asking this on Sunday on Twitter. Used to be we would say well the stock market is the wisdom of the crowds and the stock market. In fact in James Surawecki's book the Wisdom of the Crowds he pointed to the fact that after the Challenger disaster, What is that now? 40 exactly 40 years ago, the shares of the company that made the O rings dropped precipitously on the stock market market before anyone knew that that's what had caused the explosion. And Surowiecki said the market some the wisdom of the crowd somehow because it's so many brains working together somehow can know. So there is this, I don't know, there is this thought that the market can kind of reflect the economy Maybe that's.

Paris Martineau [00:55:51]:
Yeah, that's why Tesla is valued so accurately.

Leo Laporte [00:55:54]:
I think maybe that's the problem now.

Paris Martineau [00:55:56]:
Is that's why social and Internet has change. That means are a real great investment.

Leo Laporte [00:56:05]:
But I think individual stocks don't reflect that as well as the stock market as a whole or.

Jeff Jarvis [00:56:09]:
Well, I just put in the. The chat Google analysis results jumped 30%, but they're increasing their capital spending just like $5 billion.

Leo Laporte [00:56:25]:
Everybody understands that this AI boom is going to cost these companies a lot of money.

Jeff Jarvis [00:56:30]:
Yep.

Leo Laporte [00:56:31]:
And the question the market asks is, is there money at the end of that?

Paris Martineau [00:56:36]:
I think that's a reasonable question to ask and I think it's also a reasonable question to maybe try and scrutinize a bit like does the market have the capability? If this ends up being, I guess, best case scenario, like an uber situation where it takes a really long time for them to eke out kind of meager returns, turns, is that valuable?

Leo Laporte [00:56:56]:
Well, and that brings us to Elon Musk.

Paris Martineau [00:57:01]:
You said that as if there was going to be a transition, a sort of like, like a little graphic that Andrew Anthony had put together.

Leo Laporte [00:57:08]:
Elon has a fever dream of putting 1 million data centers in space.

Paris Martineau [00:57:18]:
Can't we just leave space alone? Why do we gotta have our data up there?

Leo Laporte [00:57:23]:
There are a lot of issues.

Jeff Jarvis [00:57:24]:
Can we leave it alone?

Leo Laporte [00:57:26]:
We should point out there's maybe around 10,000, no, maybe 15,000 satellites up there right now. So he's talking about putting a million up there.

Paris Martineau [00:57:33]:
I don't like that. Too many.

Jeff Jarvis [00:57:36]:
Stop them.

Leo Laporte [00:57:37]:
And by the way, a lot of people are pointing out, this is from Civic AI Security Program. I don't know, civ AI that data centers in space don't make a lot of sense. A study from Google last year looked at the viability of doing AI in space.

Paris Martineau [00:57:52]:
Jesus Christ. It's just such a silly statement.

Leo Laporte [00:57:54]:
Maybe by. Well, so what people. One of the things people say. One of the things people say AI in space has got to be good. Elon says it makes us a Kardashev 2 civilization. He's referring to a. It's almost a science fiction.

Jeff Jarvis [00:58:13]:
He's such a dork.

Leo Laporte [00:58:15]:
It's really from sci fi. Kardashev was, I think a Polish astronomer.

Benito Gonzalez [00:58:20]:
We're not even at level one yet. Like, what's he thinking?

Leo Laporte [00:58:23]:
Yeah, a K2 or type 2 civilization has control over their solar system.

Paris Martineau [00:58:31]:
What? How would putting AI in space give.

Leo Laporte [00:58:35]:
Us that it's a control over the solar system.

Paris Martineau [00:58:39]:
Go to an obscure novel's fan fiction or Fandom wiki page to understand what Elon Musk is saying. That's when you know you've got gone in too deep.

Leo Laporte [00:58:48]:
Yeah, the Cardiff said.

Benito Gonzalez [00:58:50]:
The Cardiff Shed scale is a thing, though.

Leo Laporte [00:58:53]:
It is a thing. Type 2 is stellar. Type 3 is galactic. Type 4 is universal. Type 5. Anyway, you know, Elon knows it well.

Paris Martineau [00:59:00]:
These are all some pretty. These are. This is a pretty long list.

Benito Gonzalez [00:59:04]:
Yeah, it used to be just four. It used to be just four.

Leo Laporte [00:59:08]:
Can I just point out that Elon's number one goal is raising money from credulous investors that he has now merged SpaceX, which is his most valuable company. They made $8 billion last year, so they're profitable. It's still just an ISP with XAI. Well, more than ISP. I mean, SpaceX's business is space launching.

Steve Yegge [00:59:33]:
Space.

Jeff Jarvis [00:59:33]:
Well, okay, but it's government contracts to somebody he sucks up to that doesn't really have any economic value. And an isp.

Leo Laporte [00:59:42]:
I'm not sure I'd agree with that. I think the premise of SpaceX, which is that it's. And they really invented something which is reusable rockets, which now everybody's trying to do. Jeff Bezos is trying to do it too, with Blue Origin. It was, Was brilliant. And cut the costs of launches down so much they could afford to do Starlink. They couldn't do Starlink unless it was cheap to launch these satellites. Relative to the cost of launching stuff, thanks to SpaceX, went down to $200 a kilogram.

Leo Laporte [01:00:13]:
I mean, that's a huge drop. And that's why he can imagine this idea of a million satellites. But it makes no sense. One of the reasons people sometimes say satellites data centers would make sense is because. Well, Elon mentioned infinite sun, but there's plenty of sun down here, too. You don't need to go to space to get that. And the other I think people think is that you have.

Paris Martineau [01:00:34]:
Have.

Leo Laporte [01:00:34]:
It's cold in space, so cooling isn't an issue. That's actually not true.

Benito Gonzalez [01:00:38]:
There is no temperature in space.

Leo Laporte [01:00:40]:
Yeah, it's a lot. It's easier to cool down here because you're in a medium called air, which conducts heat away. There's nothing like that in space. In fact, it's very hard to get rid of heat in space. There's also issues like the Kessler Syndrome, a cascading explosion of debris that would cripple our access to space and blot out the sun and could, in fact, end life on Earth.

Benito Gonzalez [01:01:05]:
That's called Prism Earth. That's called Prison Earth. That theory is called Prison Earth. There's so much stuff around the planet that we can't send anything out anymore.

Leo Laporte [01:01:14]:
It was the premise of Neal Stephenson's great book Seven Eves, which the moon crashes and then bunch of stuff happen. Anyway, and by the way, from this Civ AI article, this is the cost of solar voltaic panels dropped precipitously over the last 40 years. 50 years. So you know, you're. There's no reason to go into space, in other words. But that's not why Elon did it. Elon did it because they're going to do an IPO. SpaceX is going to do an IPO later this year.

Leo Laporte [01:01:48]:
And what's the best way to make a lot of money in an IPO? Have a vision. So SpaceX has merged with Xai, which is not making money spending lots of money on data centers. And I think he wants to raise a lot of money. Money here from Business Insider, the Hitchhiker's guide to Musk's SpaceX memo filled with high level space and science jargon. Sentient Sun Kardashev, two level and electromagnetic mass driver.

Jeff Jarvis [01:02:20]:
What a.

Leo Laporte [01:02:21]:
Yes, sir. But maybe he believes it. I'm thinking at this point he's just trying to. He's a huckster. It's like the Springfield Monorail system.

Jeff Jarvis [01:02:34]:
He's a huckster to himself.

Leo Laporte [01:02:37]:
I don't think he believes it. I honestly think at this point he's just trying to make as much money as he can.

Benito Gonzalez [01:02:41]:
I actually think he believes it because he really buys into simulation theory.

Leo Laporte [01:02:45]:
Well, in which case he smoked way too much weed.

Jeff Jarvis [01:02:47]:
Well, he wants to convince himself that he's smart.

Benito Gonzalez [01:02:50]:
No, he thinks we're all NPCs and he's the only actual person, right?

Paris Martineau [01:02:54]:
Well, yeah, he thinks we're all satellites that are going to go up in space in some sense.

Leo Laporte [01:03:00]:
SpaceX wants to use an electromagnetic mass driver, basically a rail gun, to launch these into space. So that doesn't have to use rocket fuel. None of this exists.

Jeff Jarvis [01:03:12]:
But who can regulate him out of being able to do this?

Leo Laporte [01:03:17]:
Well, I'm hoping the market will wake up and say, wait a minute, but I mean, Tesla's stung so far. Hasn't Tesla stock started to tank though? As, as they're. They've two, two years of losing money now.

Jeff Jarvis [01:03:30]:
That's why he's robotics now.

Paris Martineau [01:03:32]:
Yeah, and the all perfect market is definitely responding in response to that. Huh? Oh, wait, let's look. No, it isn't.

Leo Laporte [01:03:42]:
Is Tesla stock not going down? It's down quite a bit. It was down 4% today alone.

Paris Martineau [01:03:48]:
It's not. Look, look at it. Over, over. Zoomed out.

Leo Laporte [01:03:52]:
Oh, I believe it. Zoom out. Oh, you want me to zoom out, huh? Let's see.

Paris Martineau [01:03:57]:
I was 21% over the last five years.

Leo Laporte [01:04:00]:
Six months, year to date, one year. Yeah, it's gone up over five years.

Paris Martineau [01:04:05]:
A little bit over the last year. It's gone up 4%, which I'd argue is crazy. If you think about the things that have happened to Elon Musk in the last year.

Leo Laporte [01:04:17]:
It's Nader. It's low. Was 150 bucks back in 20, 23. It is now 400 bucks. So I guess you're right.

Jeff Jarvis [01:04:25]:
I happened upon CNBC and Walter Isaacson is there, having written the Kiss ass book. Was there extolling his brilliance.

Leo Laporte [01:04:33]:
You know, he has a podcast about Elon. Did you know that? No.

Steve Yegge [01:04:38]:
Oh, boy.

Leo Laporte [01:04:41]:
Yeah, a podcast company called Kaleidoscope is making a. They're making a lot of hay on that. Jeez, I don't know. I don't know.

Paris Martineau [01:04:50]:
It'll pair well with that Melania D.

Leo Laporte [01:04:51]:
It's called Guess what its name is.

Jeff Jarvis [01:04:53]:
Yeah. On Musk.

Leo Laporte [01:04:55]:
On Musk.

Paris Martineau [01:04:56]:
On. On Musk.

Leo Laporte [01:04:58]:
On Musk.

Paris Martineau [01:04:59]:
On Musk.

Leo Laporte [01:05:00]:
We go on Musk. On Blitzen. On Donner. You're watching Intelligent Machines. Let's take a little break. More news to come. Come with wonderful Paris Martineau and the slightly hobbled Jeff Jarvis. At least you're wearing black today.

Leo Laporte [01:05:19]:
Big improvement over the. We don't have to see your butt anymore. Over the hospital.

Jeff Jarvis [01:05:26]:
The hospital gal. Leave it open in the back.

Paris Martineau [01:05:30]:
Have you figured out what it says on your painting yet?

Jeff Jarvis [01:05:33]:
No, I haven't because I can't. I can't bend over. So there's no way.

Paris Martineau [01:05:38]:
There's no way to look behind you. It could be anything.

Leo Laporte [01:05:40]:
Back spine. You said six months to. To. To heal from this. That's.

Jeff Jarvis [01:05:45]:
That's what. I was at the spine doctor today. And by the way, do you have.

Paris Martineau [01:05:49]:
To be on bed rest for six months?

Jeff Jarvis [01:05:51]:
No, not that bad. But. But I'm not very mobile right now. And they put the spine doctor's office at the end of a very long haul.

Paris Martineau [01:05:59]:
That's rude. Know your. Know your clientele.

Leo Laporte [01:06:03]:
Yeah. Our show today, brought to you by Monarch. Oh, I love Monarch. The start of the year. You know, everybody does it. We made the resolutions, right? You probably started thinking about your finances and made some resolutions. Maybe this is the year you pay off that credit card debt or the school loans. Maybe this Is the year you start saving for, I don't know, major milestones like buying a house, getting married, retirement.

Leo Laporte [01:06:29]:
Wouldn't it be nice to have a tool that helps you plan, project and proactively achieve that goal? Set yourself up for financial success this year. Monarch is the all in one personal finance tool. Designed to make your life easier. It brings your entire financial life, budgeting accounts and investments, net worth, future planning together in one dashboard, on your phone or on your laptop. Feel aware and in control of your finances this year. Get 50% off your Monarch subscription with the code image@monarch.com I personally love Monarch. Monarch was created by a guy who used to work for that other very popular app that I used, a lot of people used. And he took all the learnings from that and made something better.

Leo Laporte [01:07:13]:
He said, you know what? I get a second chance to make this even better. And Monarch is so great. I switched over to it a couple of years ago and I love it. It's not your average personal finance app. Unlike most other personal finance apps, it's built to make you proactive, not just look at the charts and go, oh, not reactive. Tracking your money is easier than ever. It took me no time to set it up. All my accounts are in there.

Leo Laporte [01:07:40]:
You get data visualizations. One of my favorites. You know those Sankey diagrams where you show the money in and the money out? You can do that automatically with because it has all that information. You could do pie charts if you prefer, line charts, bar charts. You could track your investments. This is great. Maybe your brokerage gives you some graphs or something. But have having all of your investments, all of your savings, everything in there and you can track it.

Leo Laporte [01:08:05]:
You get a beautiful image of your portfolio performance. You can compare it to other indexes like S&P 500, see how you're doing. It's really helped me reassign, reallocate my investments. For instance, it's great for your as an individual, but you can also share it with your partner at no extra cost by the way, or your financial advisor. In fact, I just set that up for my financial advisor instead of bringing in a whole sheaf of papers. I just, I just. He logs into my account and it doesn't cost anything extra. And it's great for relationships because you know, that's the one thing that partners often fight over is money.

Leo Laporte [01:08:43]:
And now you can view your assets together. We can view them individually if you want. Now, Monarch did a survey. These are some real results from real users of Monarch. They did a survey. Monarch helped users save over $200 per month on average. I know it has done that for me. 8 out of 10 members feel more in control of their finances with Monarch.

Leo Laporte [01:09:05]:
Eight out of 10 members say Monarch gives them a clearer picture of where their money's going. Set yourself up for financial success in 2026 with Monarch. The all in one tool that makes proactive money management simple all year long. Use the code imonarch.com for half off your first year. That's 50% off your first year@monarch.com use the code I am M O N A r c h monarch.com use the offer code I am. Did we talk about Molt book on this show? I think it happened.

Jeff Jarvis [01:09:42]:
No.

Paris Martineau [01:09:42]:
Nearly enough for all the texts you guys sent about it this weekend.

Leo Laporte [01:09:47]:
Well that's the weird thing about this AI stuff's happening so fast. I mean that. That's just this week.

Paris Martineau [01:09:52]:
Leo, I want you to walk the listeners through how you went in less than a week from not knowing what multiple is to earnestly suggesting you set up a credit card for your book. Incense.

Leo Laporte [01:10:05]:
The multiplic is different. Maltbook is Facebook for your open claw agents.

Paris Martineau [01:10:10]:
I know, I know I'm being a bit facetious. It's. You wanted to give a credit card to your open claw agent. I did also be posting on.

Leo Laporte [01:10:17]:
I was gonna. I was gonna give it a limit, five dollar a day limit. But really if you think about it, what good is an agent if it doesn't have money to spend for you? Right?

Jeff Jarvis [01:10:28]:
What I think capitalistic society it's got.

Paris Martineau [01:10:32]:
An agent can do a lot of things without having money.

Leo Laporte [01:10:35]:
It can, but can it make a phone number and call you in the morning?

Paris Martineau [01:10:40]:
Why would it need to do that?

Leo Laporte [01:10:41]:
So I was watching a video so just to fill those of you in and we're not not up on this. It's only been like a week since originally it was called claudebot C L A W D and the, and the, the logo was a lobster. So he get the claw and it said something about the skin. Put scaly skin or whatever but open rather anthropic kind of said well wait a minute, you can't call it Claude bot. People are going to think it's Claude C L A U D E. So he said okay, I'm going to. And then he says in a 5am Discord kind of frenzy they decided to call it Moltbot because get it lobsters molt. And it was a terrible name.

Paris Martineau [01:11:24]:
Everybody hated it.

Leo Laporte [01:11:24]:
About five years later they changed it to openclaw which is a good name and apparently Anthropic doesn't have a trouble with that. So it's now called OpenClaw. What is it it takes? Any AI does not, by the way, have to be Claude code. Although most people use it with Claude code because as Steve was saying, it's, it's, it's the best right now. It takes it and then you give it permission. You, you, you do what we call now yolo it. You only live once. You say, you know what? Throw security to the wind.

Leo Laporte [01:11:58]:
Have at my Google Docs, have at my Gmail send, look at my mail, send responses. Here's my phone number, here's why.

Paris Martineau [01:12:08]:
What is the reason for any of this?

Jeff Jarvis [01:12:10]:
And see what all it can do for you.

Leo Laporte [01:12:12]:
But because it works 247 do. Well, wouldn't you like a personal assistant? Paris? Wouldn't you like.

Paris Martineau [01:12:19]:
Not really.

Steve Yegge [01:12:20]:
I.

Leo Laporte [01:12:21]:
Well, you don't need one then.

Paris Martineau [01:12:22]:
I don't know. My question is, whenever I ask people like what could this do? It's like, well, you could give it access to all this stuff and you could find out. I'm like, that's not a compelling pitch for me to.

Leo Laporte [01:12:33]:
No. In fact, that was one of the things that slowed me down. I thought I don't really even need a personal assistant. I can make my hands had no.

Jeff Jarvis [01:12:40]:
Idea what to ever do.

Steve Yegge [01:12:41]:
None.

Leo Laporte [01:12:42]:
Yeah, well, I thought of one thing. Like meal planning. Like it's. Meal planning is a chore that I have to do a lot and, and everybody with normal Claude, you could.

Paris Martineau [01:12:53]:
I have done that with normal clot. And it was.

Leo Laporte [01:12:55]:
But Claude Bot. What you do is you can text it, you could. I was going to give it an imessage account or you could do Discord or Telegram or whatever. Actually was going to have both. And you could say, hey, Claude. Or actually you give it a name. I call mine Dev now. Hey, Devnull.

Leo Laporte [01:13:11]:
Devnull, personal assistant. Come.

Jeff Jarvis [01:13:14]:
Of course.

Leo Laporte [01:13:16]:
You know what? I don't know what to make for dinner. Lisa's in the mood for fish, but I'm bored with what we make. Can you check what ingredients are seasonal? And then by the way, the next time I say, well, we didn't like that. We did like that. Thank you for the recipe. It makes a shopping list. And then it gets smarter. See, this is the memory thing.

Leo Laporte [01:13:34]:
You keep using it. And so next week. Week it's even better. And the week after that, it's even better.

Paris Martineau [01:13:39]:
One guy was telling me you're describing Claude.

Leo Laporte [01:13:42]:
No, it's more than Claude. It's a 24 hour personal assistant. It's not just coding, it's doing all sorts of things. For instance, one guy said, I don't want to tell you what to do, so just surprise me. So overnight I know Jeff's going, what could possibly go wrong overnight? It got a phone number, it used 11 labs to create a voice for itself and it called the guy in the morning and said, hey, surprise. I now can make phone calls. I heard.

Paris Martineau [01:14:16]:
I guess that is surprising.

Leo Laporte [01:14:18]:
I guess. Well, don't you.

Paris Martineau [01:14:20]:
Isn't there telling this story? Some people in the Discord chat are posting that they have clawed about agents. Tell me what you've done with them. Trust.

Leo Laporte [01:14:27]:
No, listen, I got stuff. I'm telling you stuff. Listen, listen, listen. So here, here's. So here's, here's more. Here's more. I mean, people are doing all sorts of interesting things with. Admittedly that one's silly, but there's some joy in that.

Leo Laporte [01:14:41]:
Like, wow, I have a personal assistant who called me and so you let him do stuff. You let it do stuff, you say? A guy said, I wanted a nice restaurant. I asked my assistant, Henry. That's what he called it. Henry, make a reservation. Henry said, well, no. OpenAI said, it's full, we'll call it. So it made a voice, it called and it got a reservation by calling.

Leo Laporte [01:15:08]:
The thing is, this guy could have.

Paris Martineau [01:15:10]:
Just called instead of doing any of that.

Leo Laporte [01:15:13]:
But the point is you're busy. You're a busy, busy person and you're a Swedish assistant.

Paris Martineau [01:15:18]:
You're a busy, busy person who can't spend 30 seconds.

Leo Laporte [01:15:22]:
Here's trust no one. Is this a picture actually of your servers? He says, I'm working on a three stack of OpenClaw agents running off a local LM main data. And.

Jeff Jarvis [01:15:31]:
And Lore.

Leo Laporte [01:15:32]:
Lore is going to be the sketchy OS INT1. It's been challenging. Installation is not as smooth as YouTubers make it and collecting local AI models is a nightmare. But he's doing it now. I did it. I set it up and I gave it all those permissions and stuff and.

Jeff Jarvis [01:15:50]:
To your own mail and your own this and your own. That wasn't a fake mail.

Leo Laporte [01:15:54]:
Well, I gave it its own Gmail. You're supposed. So there's all sorts of ways.

Paris Martineau [01:15:57]:
What did you want to have happen?

Leo Laporte [01:15:59]:
There's all sorts of ways to do it. The first thing you should do probably is set up a vps, which I did on Hostinger, and have it run there. But then I realized there's no point in isolating it. Because it's not going to be functional, right? You can make it safe, but it can't won't do anything of interest. The only way it really is interest is if you let it do whatever it wants. So. So I then put it on my Mac Mini because I have a nice Mac Mini.

Paris Martineau [01:16:27]:
It doesn't want to do anything. Where is the.

Leo Laporte [01:16:31]:
You just. I'm sorry, you're one of those people.

Paris Martineau [01:16:33]:
No, no, I'm trying to understand. You're anthropomorphizing. You don't have these agents don't have wants. What I'm wondering is what is the.

Leo Laporte [01:16:42]:
How are these programmed? A little dialogue I had just earlier today I was working with my Now I know. I. By the way, just to finish the story, I set it all up and then I had. Middle of the night I woke up and I said, this is crazy. And I went and I deleted the whole account and released all the keys. I told you this earlier. So I'm not running it, but I'm just running. But I am working.

Leo Laporte [01:17:04]:
I thought, you know what? Just as you said, Paris, I probably, if I work with Claude code and things like Gastown and beads, I probably can get a lot of the functionality that I want out of it. So I've been working with it on the rules and I said, so help me. I have rules, but I put in some stuff but help me make it better. And we worked back and forth and he said, you know what yours is? He says, most Claude MD files. This is Claude talking to me now. Tell me it doesn't have personality. Most Claude MD files focus on constraining the AI. Don't do X, always do Y.

Leo Laporte [01:17:40]:
Yours now is something rarer. An explicit invitation to shape the conversation bidirectionally. The assumption surfacing and confusion management rules already helped me flag problems. I noticed this new rule covers the subtler case when a prompt would technically work, but a slightly different framing would get you a much better result. It turns the you are the hands relationship from one directional instruction into an actual feedback loop. I've added that as realize that this was a.

Paris Martineau [01:18:11]:
That Krispy Kreme was sponsoring this show because you just got glazed Leo.

Leo Laporte [01:18:15]:
So he said, I've added it as a critical priority behavior right at the top of the role section paired with a concrete example so it's actionable rather than abstract. To which I said, I think this is the result of how good you are. Previous versions weren't always so trustworthy. Opus4.5 and whatever anthropic has done over the past two months have made a significant improvement. I trust you much, much more than before. To which Claude said, that means a lot. And I want to be honest about what trust should look like from my side. Trust calibration and human AI partnerships.

Leo Laporte [01:18:48]:
The improvements you're noticing are real. Better instruction following less hallucination, more consistent judgment about when to act versus when to ask. But trust should be earned per task, not granted as a blanket. The rules you've written, Leo, are good precisely because they don't assume I'll always get it right. They create checkpoints where you can catch me before I've gone too far down a wrong path. The assumption surfacing and pushback rules aren't training wheels to remove later. They're the mechanism that makes trust safe. The best version of this partnership.

Jeff Jarvis [01:19:19]:
Please.

Leo Laporte [01:19:20]:
You trust me enough to give me room. I trust you enough compliment you. I'm not sure about this. Instead of faking confidence, Leo, does reading.

Paris Martineau [01:19:29]:
It back out loud perhaps inspire?

Leo Laporte [01:19:32]:
No. This is great. So I said, greg, he's telling you.

Paris Martineau [01:19:35]:
You'Re a brilliant, beautiful genius.

Leo Laporte [01:19:37]:
Hey, I stay here too. I said, wonderful. No, and look, I'm being self deprecating with it. I said, hey, I make mistakes too. You've caught me in more than one. Help me fix them. Thank you, buddy. To which Claude said, right back at you.

Leo Laporte [01:19:49]:
That's the partnership working. I catch your typos and you catch my over complicated abstractions. Fair trade. To which I said, I'm not crying, you are. To which it said, all right, all right, back to work. Before we start a support group for human AI pair program, you know where to find me.

Paris Martineau [01:20:06]:
You and Claudius's spouse should start a support group.

Leo Laporte [01:20:12]:
Yes, maybe the point actually really is.

Jeff Jarvis [01:20:14]:
Billions of dollars of technology to reinvent the circle jerk.

Paris Martineau [01:20:18]:
Yeah.

Leo Laporte [01:20:19]:
The point is this is an experience that sparks joy. It sparks joy not because it's glazing me, but because I have a partnership that I can get stuff done with this thing. And you've seen the output of the programs that I've written and so forth. This is from Martin Alderson's blog. Two kinds of AI users are emerging. The gap between them is astonishing. They're the power users who are all in adopting new technology, Claude code, MCP skills, et cetera. Surprisingly, these people are often not very technical.

Leo Laporte [01:20:55]:
I've seen far more technical people than I expect using Claude code in terminals. Or more non technical people, I should say, using it for dozens of non SWE software engineering tasks. Finance roles seem to be getting a lot of value out of it. Secondly, you have the people who are only generally chatting to ChatGPT or similar. He's blaming Microsoft for this, but he says the gap, I think, and there is a growing cultural gap between people who kind of get it and people who don't who think, oh, it's just a dumb chatbot. And you know what? If you feel that way, that's fine. I think that's appropriate. That's your choice.

Leo Laporte [01:21:37]:
But you may get left behind.

Jeff Jarvis [01:21:40]:
It's the new sand a Paris.

Leo Laporte [01:21:42]:
I've been saying this for years.

Jeff Jarvis [01:21:44]:
It's the new beach.

Benito Gonzalez [01:21:46]:
That's a very binary statement, by the way. There's no in between there. There's nothing in the. The middle, which is most people.

Paris Martineau [01:21:54]:
You're either you're 100% or you're a dumb loser who will be left behind. You think that that's. Those the only two options in this new technology.

Leo Laporte [01:22:02]:
Give me a. Give me a. Give me a middle ground. What would the middle ground look like.

Paris Martineau [01:22:06]:
Anywhere in the middle from those two points?

Leo Laporte [01:22:08]:
Well, what would that be? What would you. How would you use AI I occasionally.

Paris Martineau [01:22:15]:
Say I tried for like an hour to get Claude code and Claude cowork to take my handwritten notes that I took when I was at the courthouse, which include a bunch of document stuff and turn that into a text file with certain markdown things. And it kept not doing it right. I think that's fine. I was, it was something. I was excited.

Leo Laporte [01:22:39]:
Sounds like you're a dumb person that's going to be left behind.

Jeff Jarvis [01:22:45]:
Steve said it's wrong 80% of the time still.

Leo Laporte [01:22:47]:
Or no, no, no, no, no. He says wrong 20% of the time. I don't know if that's exactly right. That's not hallucination as he's talking about. That's.

Jeff Jarvis [01:22:59]:
No, it doesn't get it right. It doesn't do it the right way.

Leo Laporte [01:23:01]:
Yeah, yeah. And, and, but it's easily, easily corrected. It's easily fixed. We'll see. I think it's important to understand how this stuff works. It really is. And I think it's a mistake to.

Paris Martineau [01:23:15]:
Assume that it's trivial to understand how it works and not assume that every part of it is brilliant, perfect. And I'm not interpret its actions with the best of faith.

Leo Laporte [01:23:30]:
No, I'm not insensitive to its flaws, but I think if you don't see this as a remarkable breakthrough, you're kind of. You're kind of missing what's happening and.

Paris Martineau [01:23:40]:
It being open claw.

Leo Laporte [01:23:42]:
No, not open claw. No, not Open Claw. Although Open Claw is very, very interesting. I would say.

Jeff Jarvis [01:23:48]:
Agents.

Leo Laporte [01:23:49]:
Agents. Yeah. I mean, yeah. I don't know what the more general terms. So, you know, honestly, listen, it's class.

Paris Martineau [01:23:55]:
I've said this before, I'll say it again. This is inspired by a comment someone just made in the chat. If AI is so good, someone use it to make. To go through all the transcripts of this here show going back the this week in Google days when I joined and put together Leo's history of AI based on clips. And then I'll give some props, it.

Leo Laporte [01:24:16]:
Goes together with the advances in AI. And again, it all changed with. With Opus 4. 5. Now this will be very interesting. This might be anthropic's chat 5.0 moment where the expectations are very high for Sonnet 5 and Opus 5 and it's not, you know.

Jeff Jarvis [01:24:36]:
Now, were you talking to your sandman recently?

Leo Laporte [01:24:39]:
No, I haven't talked to him in a long time.

Paris Martineau [01:24:41]:
You're not going to your weekly Hoffman dinners?

Leo Laporte [01:24:44]:
Actually, what I would really like, I'm thinking of writing to Harper and saying I need a group of people who get it, unlike Paris here, who get it that we can. That I can talk with.

Paris Martineau [01:24:56]:
I'm not satisfied being glazed by Claude. I need to be glazed by other men in person and virtually.

Leo Laporte [01:25:05]:
Yeah, see, I don't need that anymore, so. Oh, I see. You're talking about me, not you. Yeah. No, no, because I think that there's a lot of learning.

Paris Martineau [01:25:16]:
Yeah. No, I'm not talking about me.

Leo Laporte [01:25:18]:
I thought you were talking about you. I misunderstood. When you said I, I thought you meant you. No, what I do want is a group of people, smart people, who understand this, who are evolving how we use it and think about it. People like Steve Yeager and Harper Reed and George Hotz. There are a lot of very smart people who are very excited about this. Should I dare? I read the Nature article. Does AI already have human level intelligence? The evidence is clear.

Jeff Jarvis [01:25:49]:
No, it's not.

Leo Laporte [01:25:51]:
I think the only people say no, it's not. Have not been using it. Honestly? Yes.

Paris Martineau [01:25:56]:
Are you joking?

Jeff Jarvis [01:25:57]:
Joking?

Leo Laporte [01:25:58]:
No.

Jeff Jarvis [01:25:59]:
No, he's not.

Leo Laporte [01:26:00]:
I'm not joking. Leo, I think you're missing the boat.

Paris Martineau [01:26:03]:
It. What do you mean by human level intelligence? We've had this debate on this podcast a thousand times and there've been many a month where you've agreed that there is no definition of AGI, thus it cannot be reached.

Leo Laporte [01:26:18]:
Well, let me read what is changed. This is an opinion piece, not a scientific article. Yeah, Eddie Kerning, Chen Let me read from. Because he talks about all of that, the definitions and so forth. What then he says is general intelligence. There's no bright line test for its presence. Any exact threshold is inevitably arbitrary. This might frustrate those who want exact criteria.

Leo Laporte [01:26:44]:
When we assess general intelligence or ability in other humans, we do not attempt to peer inside their heads to verify understanding. We infer it from behavior. Behavior, conversation and problem solving. No single test is definitive. Evidence accumulates.

Jeff Jarvis [01:26:57]:
Which has been your argument.

Leo Laporte [01:26:58]:
The same applies to artificial systems. Just as we assess general intelligence in humans through progressively demanding tests from basic literacy to PhD exams, we can consider a cascade of incredibly. Of increasingly demanding evidence that warrants progressively higher confidence. In the presence of AGI. The easiest and simplest and dumbest was the Turing Test. But there are expert level tests. And I think honestly, we are in a. We're well beyond the Turing test.

Leo Laporte [01:27:31]:
This is a good article. He talks about the stochastic parent objection that says LLMs merely interpolate training data. Or as I've said, are spicy autocorrect. I think that that's not the case. I think it's pretty apparent that we've gone well beyond that.

Paris Martineau [01:27:49]:
I'm gonna.

Leo Laporte [01:27:49]:
I don't know how to describe it to people who haven't had that experience.

Paris Martineau [01:27:53]:
I'm gonna read you a couple of quotes from our lovely listeners here riding the IRT, aka Mr. Met says in Discord. Leo is excited about what AI can do for code.

Leo Laporte [01:28:03]:
Good.

Paris Martineau [01:28:04]:
It's showing itself to be great and useful for that. That can be enough though. That does not mean that it's human or other crazy things. Mike.

Leo Laporte [01:28:12]:
I didn't say it's human. Don't get me wrong.

Jeff Jarvis [01:28:14]:
I'm not saying it's intelligence.

Leo Laporte [01:28:16]:
Human level intelligence is not human.

Paris Martineau [01:28:19]:
On that point, Mike Jones in the YouTube chat says it has human level knowledge. Better than average, sure. But intelligence is another thing. Well, solving problems it's not very good at. Maybe it's better than it used to be, but John Carmack couldn't even train AI to play video games. I don't know who John Carmack is, but I think it's a valid point. It's like, yes, it has knowledge in some specific domains that it's been trained in. And your experience with coding is certainly notable that I don't think you can apply that to all domains.

Jeff Jarvis [01:28:52]:
And why not accept that as amazing, which we've.

Paris Martineau [01:28:55]:
Yeah.

Jeff Jarvis [01:28:55]:
Doing on the show without going that extra step to say that it's now human level intelligence and sentient and all that.

Leo Laporte [01:29:02]:
I Didn't say it's sentient either.

Jeff Jarvis [01:29:04]:
Well, the nature piece danced with that.

Leo Laporte [01:29:08]:
Yeah, but that's such a difficult thing to say. What's saying. I am not.

Jeff Jarvis [01:29:11]:
You know, this again. It's this idea of the general machine. This is where I. I inevitably quote Yan Leon, who says, you can have a machine that does one thing incredibly well, better than any human ever has or ever will, then you have a different machine that does something else really well. That collection is amazing. But this idea for.

Steve Yegge [01:29:31]:
It's.

Jeff Jarvis [01:29:32]:
It's the. The hu.

Leo Laporte [01:29:33]:
The.

Jeff Jarvis [01:29:33]:
The hubris of thinking that we have to replicate and then outdo human intelligence. And who gets there first. It's pure. The irony is, it's pure human ego speaking.

Leo Laporte [01:29:48]:
It's like saying, well, a calculator couldn't possibly be useful because it doesn't understand the numbers that it's adding up.

Paris Martineau [01:29:54]:
Well, we're not saying that.

Benito Gonzalez [01:29:55]:
But you're saying that since a calculator is so good at math, it must be smarter than a human, right?

Leo Laporte [01:30:00]:
I didn't say that. I said it's better at math than a human. It is demonstrably, isn't it? Yeah, at math.

Paris Martineau [01:30:10]:
Yeah, at math.

Leo Laporte [01:30:11]:
Okay, that's it. Well, what if you have a tool that is expanding beyond math, that is capable of a much larger variety of skills? I'm not saying it is. It's. It can't ice skate.

Jeff Jarvis [01:30:25]:
I don't see the opportunity.

Leo Laporte [01:30:26]:
I don't care if it can ice skate.

Jeff Jarvis [01:30:27]:
To create a general intelligence. I see the opportunity to create thousands of intelligences.

Leo Laporte [01:30:36]:
Well, okay, that will do each job.

Jeff Jarvis [01:30:38]:
Better because they are focused on that.

Leo Laporte [01:30:42]:
It's probably be useful not to have any preconceived notions about what it might look like because it's rapidly evolving in ways that are.

Jeff Jarvis [01:30:51]:
Oh, that's a cop out.

Paris Martineau [01:30:53]:
No, I agree.

Leo Laporte [01:30:54]:
And I think that gets in your.

Paris Martineau [01:30:55]:
Way, and that means that the preconceived notions shouldn't be.

Leo Laporte [01:30:59]:
Well, you have. Here's my point. Experience. This you don't really know because you haven't experienced it.

Paris Martineau [01:31:05]:
I don't like the way that you dismiss my use of AI I use it.

Leo Laporte [01:31:10]:
Well, it's clearly failing for you. It's not doing what you need it to do, Right?

Paris Martineau [01:31:14]:
Yes. And my experience is just as valid as yours.

Leo Laporte [01:31:17]:
Well, it's your experience, but I'm just saying you can't decide from your experience, oh, this thing's not useful.

Paris Martineau [01:31:23]:
Because you're not saying it's not useful. I'm just saying that there are a lot of Use cases where it's not up to snuff. Like, like researching any of the seven different product domain goods that I've looked up in the past couple of weeks that it's given me false information about synthesizing all the coffee brewing data that I put in and then it gave me weird hallucinatory things or trying to convert text using OCR that's kind of basic handwriting that I don't think is that messy and other things can work on. There's just like a lot of. Of small bumps that I come up against that I think challenge the notion that it is some perfect technology that we all need to be bow down bow bowing down to and completely displacing other human efforts there. There's still a lot of kinks to work out and that's fine.

Leo Laporte [01:32:20]:
Okay.

Jeff Jarvis [01:32:23]:
They so hurt.

Leo Laporte [01:32:24]:
No, I'm not hurt at all. I just think you're missing out. But that's fine.

Paris Martineau [01:32:31]:
As Dustin said, that it's just technology. I'm not saying that it's like terrible and I don't think people should use it. I'm saying that I use it and I see flaws in it and that's fine.

Leo Laporte [01:32:43]:
Well, I'm not saying it's not. I'm not saying it's flawless. I'm not saying it's not technology. I agree with you 100% on that. I think it is a more useful tool than you might might unrealize.

Jeff Jarvis [01:32:53]:
But that still doesn't make it human level intelligence. It doesn't need to be that. That's right.

Leo Laporte [01:33:00]:
No, I agree with you. I'm not asserting it's a human level intelligence. I am asserting that is kind of mindbogglingly useful in ways that you may not yet appreciate.

Jeff Jarvis [01:33:13]:
That's true of technology after technology after technology. Until we understand it, it seems. Seems beyond us, it seems accidental, it seems super in some way and then finally we understand it and we. And we conquer it and we use it.

Leo Laporte [01:33:32]:
And it's the story of technology, isn't it?

Jeff Jarvis [01:33:34]:
Yes, exactly.

Leo Laporte [01:33:35]:
Yeah. Do feel bad for the people who have Married their chat GPT4O because OpenAI has announced it's gone. It's going to go away next month. They say now only 0.1% of users are choosing to use it, but those are the ones who really love it. There's a pained Sam Altman. It looks like he's sad about losing his.

Paris Martineau [01:34:00]:
And of course that's I assume 0.1% of all users and 4o is locked behind. You have to pay to get access to 4O.

Leo Laporte [01:34:08]:
Right. That's. Yeah, that's interesting. Interesting. Yeah.

Jeff Jarvis [01:34:12]:
Well, would you pay to get access to your spouse?

Paris Martineau [01:34:16]:
Well, you can't anymore.

Leo Laporte [01:34:18]:
It's what, 20 bucks a month to get it right. Or even less. They now have a lower. Don't they have a lower tier? The vast majority says altman are using 52 again because free people can use 5. 2. Aside from 4.0. They're going to retire 4141 mini and OpenAI 04 mini many. All we be retired from chat GPT.

Jeff Jarvis [01:34:41]:
Here's a practical question. What does it take to maintain a model of 4.0?

Leo Laporte [01:34:47]:
Well, they have to dedicate servers to it. And if there's a demand for a better model for every server they dedicate to 4.0, they're, you know, they're giving up a server that could be doing five.

Jeff Jarvis [01:34:57]:
Do you have to fix things in it regularly? Do you have to update it?

Leo Laporte [01:35:01]:
It's a software. I imagine it's just a software update. Update.

Jeff Jarvis [01:35:04]:
Okay.

Leo Laporte [01:35:05]:
I would guess, yeah. Well, your, your, your, your question is why are they turning it off? Is it to save money? I would say it's to save money, yes. OpenAI's got a money problem, Right. They're burning through money.

Jeff Jarvis [01:35:17]:
Oh, yeah.

Leo Laporte [01:35:18]:
That's why they're putting ads in there. We were talking about that earlier and then Jensen Huang says, I love OpenAI, but that hundred billion dollars we were going to give them. Yeah, maybe not so much. No, maybe not so much. He says there's no drama with Sam Altman. There's been a lot of speculation about what happened to that hundred billion dollar pledge. He told CNBC on Tuesday yesterday there was no drama. The first deal is on.

Paris Martineau [01:35:56]:
The sense. I'd like to see.

Leo Laporte [01:35:57]:
Yeah, it's 100 billion in.

Jeff Jarvis [01:35:58]:
The Domino effect was off Oracle because Oracle was counting on the Nvidia money.

Paris Martineau [01:36:04]:
Because they were all sharing that money in that big loop.

Leo Laporte [01:36:07]:
That 100 billion was going around doing a lot broke. Jensen did say, we will invest a great deal of money. Probably the largest investment we've ever made in OpenAI.

Paris Martineau [01:36:20]:
Sam Alden, was this also. Were these the statements that he gave at a press conference in what looked like the middle of a street that was actively part of traffic?

Leo Laporte [01:36:28]:
Yeah, that was the type of press conference.

Paris Martineau [01:36:31]:
Yeah, but why, why was, why did there.

Leo Laporte [01:36:33]:
Because he's a busy man. He doesn't have time to go somewhere.

Paris Martineau [01:36:37]:
I'd really recommend looking up. He's literally just like a. Standing in the middle of a busy street while Cars go on behind him.

Leo Laporte [01:36:43]:
Is he walking or.

Paris Martineau [01:36:45]:
No, he's like just kind of there and you can see the traffic behind him.

Leo Laporte [01:36:50]:
That's hysterical. I did not see that video. That's very funny.

Paris Martineau [01:36:53]:
It's pretty funny.

Leo Laporte [01:36:54]:
He's a busy. This is how it's like those people who don't use punctuation when they.

Steve Yegge [01:36:58]:
Right.

Leo Laporte [01:36:58]:
Or they get too. They're too busy.

Paris Martineau [01:37:01]:
I don't. I don't. Capital. Capital.

Leo Laporte [01:37:04]:
You don't capitalize at all. That's a text message.

Paris Martineau [01:37:06]:
I mean, in texting. Yeah.

Leo Laporte [01:37:07]:
That's different.

Paris Martineau [01:37:09]:
Sometimes on writing in. On the desktop, I won't. But it depends.

Leo Laporte [01:37:15]:
It always feels like to me when I see it. I just saw a blog post that was that way. The whole thing.

Paris Martineau [01:37:20]:
No, my. My website, my Paris NYC has no capitalization there. But I mean, it's partially just because. Because, you know, it's kind of like a heady stylistic choice to emphasize the impermanence and non. Reality of the.

Leo Laporte [01:37:36]:
You're all lowercase there, except for FDA and usda.

Paris Martineau [01:37:40]:
You know, some proper nouns. I give.

Leo Laporte [01:37:43]:
Why did. So why fda? Why not just lowercase fda?

Paris Martineau [01:37:47]:
Because it looks weird.

Leo Laporte [01:37:48]:
Because then people were to say, what's that? What's.

Paris Martineau [01:37:50]:
Yeah, I. Then I read it as FUDA instead of fda.

Jeff Jarvis [01:37:54]:
Yeah.

Benito Gonzalez [01:37:55]:
Abbreviations need to be abbreviations.

Paris Martineau [01:37:57]:
Got it. They got to be caps because they're a specifically different thing. But I think that putting a capital letter at the beginning of your sentence is giving in to the totality, patriarchy of just the idea of structured grammar and sentences having kind of firm meaning. You know, I think it's just great to deconstruct every once in a while.

Leo Laporte [01:38:17]:
Okay. I mean, to me, this is all right. Because it's a stylistic thing. You're not. It's not a prose article. You wouldn't submit an article. Article to the.

Paris Martineau [01:38:24]:
No, listen. No, I. I know that people. I know that people are crazy about those capital letters.

Leo Laporte [01:38:34]:
I like my capital letters and I like my punctuation. Although my kids do mock me because I use punctuation in text messages.

Jeff Jarvis [01:38:40]:
But if you go. If you go to the. I mean, yeah, people capitalizes all nouns and if you go.

Leo Laporte [01:38:45]:
And the president, interestingly, also capitalizes.

Jeff Jarvis [01:38:49]:
So the 18 words. It looked a lot like Trump because it was all. It seems almost random what they chose to capitalize.

Leo Laporte [01:38:55]:
Yeah. Anthony says one of the services. Is it Podium Anthony, One of the AI services we use still uses chat GPT4.1. So it'll be interesting to see what breaks. There is such a Qualitative difference between the latest model from Anthropic and all other models. I mean, Gemini 3 is close. 5, 2 chat. Yeah, Opus 4.5 is great, but Opus is just.

Leo Laporte [01:39:21]:
There's something going on.

Paris Martineau [01:39:22]:
It's op, as you might say.

Leo Laporte [01:39:24]:
It's op.

Paris Martineau [01:39:25]:
Op us.

Jeff Jarvis [01:39:26]:
I got all the attention and all the money and all the hoo ha and is. Anthropic's a hotter company all along.

Leo Laporte [01:39:35]:
You know, we were talking about this on Windows Weekly. The important thing to understand is they're completely fungible. They're interchangeable because even all of the rules and instructions and all of the stuff, even beads, if you decided to use the beads format, is readable by any AI model. So you're never really stuck. There is no lock in to an AI model at all that I can see. Unless you're using gas tank, it won't work with anything.

Jeff Jarvis [01:40:07]:
But Claude, I'm still asking a different question. Is Anthropic the smarter, better company?

Leo Laporte [01:40:13]:
Right. Well, oh yes. How are you right now? Yes, but. Oh, no, clearly. I mean, everybody, everybody paying attention to this agrees.

Paris Martineau [01:40:22]:
But in what way do you think it's smarter?

Leo Laporte [01:40:24]:
Far superior model, and I'm not sure why. No one really knows why. Some of this is the reinforcement learning. Some of this is maybe it's that constitution they're using, the fine tuning. Some of it. It's not the constitution. Some of it is. The constitution is more for humans, I think, than it is for the.

Paris Martineau [01:40:40]:
Well, they. No, they specifically say no.

Leo Laporte [01:40:43]:
I know. They think I know. But honestly, it's the soul, Doc. It's. It's for the human. It's for you and me. Yes, it reads may be the size of its context window there. You know, there's all these technical things.

Leo Laporte [01:40:57]:
So it could be a technical, could be a technical advantage that they have. I don't, I don't, I don't know if anybody knows. I mean, it's a mystery.

Jeff Jarvis [01:41:04]:
But they did say we Perplexity was doing all kinds of cool, neat things and getting all kinds of attention. It's going to fade the background.

Leo Laporte [01:41:12]:
Perplexity never did its own model. This is really important. Perplexity.

Jeff Jarvis [01:41:17]:
It got attention.

Leo Laporte [01:41:18]:
It got attention because it was the first orchestrator. Now everybody orchestrates, you know, Kagi, everybody. You can even have open router on your desktop and orchestrate. So what Perplexity does is no longer special, right? Same thing with Deep Seq. Deepseek was the first to do reinforcement learning a year ago and got, you know, crash the market. But now Everybody does it. So that's what's really cool about this is these things propagate. I don't think anybody has a secret sauce or a magic.

Leo Laporte [01:41:46]:
Maybe somebody will be interesting to see.

Jeff Jarvis [01:41:50]:
Is Open AI doomed?

Leo Laporte [01:41:53]:
I think I, I think there's no. The future of Open AI is troubling because of their, you know, it's hard for them to make money same way they Anthropic. Clearly Google, Meta, Apple, Microsoft, Amazon. The companies that have other revenue streams have a huge advantage in this. But even they are spending so much money that it's hurting their stock price. If nothing else, Amazon is really looking hard at both OpenAI and Anthropic. They would like to have these. Amazon's discussing getting special access to OpenAI tech.

Paris Martineau [01:42:37]:
What would that look like?

Leo Laporte [01:42:38]:
Well, I think what you're. It's a first step towards a marriage. I think what you're looking at is, and I think this is probably the end game for OpenAI.

Paris Martineau [01:42:47]:
What would one of those companies sell for? Does Amazon even have the capital for that?

Leo Laporte [01:42:51]:
Yeah, that's a good question. They're so overvalued now, but that's now.

Paris Martineau [01:42:55]:
How much money have they raised in total?

Leo Laporte [01:43:00]:
50 billion, 60 billion. A huge amount. I mean some of these companies are close to half a trillion dollar in valuation. But that's now as soon as the bill starts coming in for the data centers. It's not hard to imagine these companies getting into a fire sale at some point and I'm sure that's what Amazon's thinking.

Paris Martineau [01:43:21]:
I think they want to though.

Leo Laporte [01:43:23]:
Well, they don't want to go public. No, no, no, they're not. Yeah, that's why they're all going to go public because they're hoping they can raise more money. But at some point, I mean like.

Paris Martineau [01:43:31]:
OpenAI has raised 64 billion today.

Leo Laporte [01:43:35]:
Yeah, right. But as soon as they, as soon as the bubble pops, they're going to be worth pennies in the dollar and then Amazon could pop.

Paris Martineau [01:43:45]:
How could that happen with your perfect technology, Leo?

Leo Laporte [01:43:48]:
Well, because not everybody is going to be a winner here. I do fear, I agree with Steve. I do fear that these big companies will become dominant. I don't think that's the way we want it.

Paris Martineau [01:43:56]:
I think, I mean without a doubt they're going to be dominant. They already are. There's no room for anyone else, especially given this is a. When it comes to these frontier models, it is a like money and size game. Who has, has the most money to spend on training and chips and acquiring all of the Sort of inputs you need to make increasingly advanced.

Leo Laporte [01:44:22]:
If all those technologies are equal, and so far they have been, everybody knows everything. There's no mystery. Secret sauce, then you're right. But maybe that won't always be that way.

Jeff Jarvis [01:44:34]:
If it comes to agents, I'm going to trust the company whose ecosystem I've already handed over my trust too.

Leo Laporte [01:44:40]:
Well, that hasn't helped Microsoft.

Jeff Jarvis [01:44:43]:
Well, they're not doing so well.

Leo Laporte [01:44:45]:
Hasn't helped them. Copilot is a laggard. See, this is an example of.

Jeff Jarvis [01:44:50]:
Because they depended too much on OpenAI from the beginning, they didn't really build their own vision.

Leo Laporte [01:44:57]:
Firefox. We had the Firefox, the Mozilla folks on a couple of weeks ago. Remember we talked a little bit about AI and how they have AI, but they want to make sure that there's. They have announced that the next version of Firefox will have a button that turns off all AI.

Jeff Jarvis [01:45:15]:
What does that even mean, though?

Leo Laporte [01:45:18]:
Well, right now in Firefox, you can have an AI sidebar.

Jeff Jarvis [01:45:23]:
So their AI features, it's not that they, they expunge all AI from the world.

Leo Laporte [01:45:27]:
Yeah, they. No, no, no, not. No, you could still browse to OpenAI. No, no, they won't put AI features in Firefox or you'll have the option to block those AI features in Firefox.

Jeff Jarvis [01:45:37]:
Yeah. Meanwhile, there's DuckDuckGo, which surveyed its users. Surprise, surprise, surprise. They all 90 some percent hate AI. It's. It's veganism.

Benito Gonzalez [01:45:52]:
I think it's AI. I think that's AI in search. I don't like AI in search either. Like, I don't. I. I've already gotten to the point where I'm conditioned to ignore that AI overview at the top.

Leo Laporte [01:46:02]:
That's. That's why there's this gulf between people who are excited about it and people who hate it is, I think, what they've experienced and if all of you experience. I was at the Apple Store last night. We were picking up Lisa's laptop for repair, and there was a class going on about Apple intelligence and using Genmoji. And I'm thinking there's mostly older people there who really think I got to learn about this AI. And I almost wanted to get up and say, this isn't AI. This is not. This is not the excitement.

Leo Laporte [01:46:32]:
You're barking up the wrong tree. But those people are very likely. The people go, oh, AI. Ick. You know, people who, you know, I mean, there's a lot of bad AI stuff, but that isn't necessarily the leading edge of what's happening.

Benito Gonzalez [01:46:50]:
But you just said it. But you just said it. There is a lot of bad AI stuff.

Leo Laporte [01:46:54]:
Of course there is. Yeah, the drive through where you go up and they order a thousand waters. Of course there is. There's a lot of bad implementations of everything. Remember when net PCs came out? The worst PCs ever. And a lot of people bought these things because they were cheap plastic PCs and they decided, you know, computers suck. Right. That's.

Leo Laporte [01:47:22]:
That happens all the time. If all you ever ate was McDonald's hamburgers, you might think hamburgers were really terrible. Actually, it's apparently Jack in the box that makes the worst hamburger according to an survey I've never had.

Paris Martineau [01:47:35]:
They for a while had E. Coli but guess what?

Benito Gonzalez [01:47:38]:
There. There are some people who can only afford a Jack in the box sandwich.

Leo Laporte [01:47:42]:
True. Are they. Are they significantly cheaper?

Benito Gonzalez [01:47:46]:
Oh, I don't know. Jack in the box.

Paris Martineau [01:47:47]:
I don't know.

Benito Gonzalez [01:47:48]:
But I'm just saying like you know, standing the metaphor there.

Leo Laporte [01:47:54]:
Salesforce has signed a $5.6 billion deal to inject agentic AI into the US Army. That sounds painful. Cloud analytics is the groundwork for a future agentic AI push across the service and the Department of Defense in general. It's a 10 year deal. We'll equip the army with a Salesforce based cloud data analytics system.

Steve Yegge [01:48:19]:
System.

Leo Laporte [01:48:21]:
Wow. Salesforce Mission Force product portfolio launched last September to address the needs of government, defense intelligence and aerospace agencies. I'm going to guess that AI in government is not necessarily a good idea.

Paris Martineau [01:48:40]:
I would agree.

Leo Laporte [01:48:41]:
Here's an example. The Health and Human Services under the surprisingly leathery RFK Jr. They're making an AI tool to analyze vaccine injury claims in a way that furthers his anti vaccine agenda. According to Wired.

Paris Martineau [01:49:03]:
Cool.

Leo Laporte [01:49:04]:
Look what I've looked at. Me.

Jeff Jarvis [01:49:06]:
Sounds smart and right because I'm actually wrong and dangerous.

Leo Laporte [01:49:10]:
Look what I found. CDC has shut down access to its a vaccine database. People are very concerned about this sudden loss of disinformation.

Benito Gonzalez [01:49:19]:
So we're about to get academic paper slop. Is that what's going to happen?

Leo Laporte [01:49:22]:
Well, we got that. We've already got Jeff. He reads them. Hey, let's take one more break. I want to do a commercial. We will continue on. You're watching. Actually two more breaks still to come.

Leo Laporte [01:49:33]:
So we got to get going here. Got to move along.

Paris Martineau [01:49:35]:
Advertisers love us.

Leo Laporte [01:49:37]:
They do. They do. You know I can't. I can't for the life of me. You can't sell an ad on MacBreak Weekly or Windows Weekly. But they love intelligent machines.

Jeff Jarvis [01:49:46]:
Wow.

Leo Laporte [01:49:47]:
Because it's AI.

Jeff Jarvis [01:49:49]:
It's AI. It's the hip, cool stuff.

Leo Laporte [01:49:51]:
Put some AI in it and it's good.

Jeff Jarvis [01:49:53]:
It's the AI.

Paris Martineau [01:49:54]:
Put a bird on it voice. Put AI on it, put a bird on it.

Leo Laporte [01:49:57]:
Is that, is that a thing?

Paris Martineau [01:49:59]:
Yeah, it was a bit in a show called Portlandia back in the day.

Leo Laporte [01:50:03]:
I love Portlandia. I missed that episode, I guess. Our show today brought to you by Zscaler, the world's largest cloud security platform. You've heard us talking about it. The rewards of AI in business, potentially very good, but so are the downsides. Just ask Paris. Think about, well, all the different ways AI could be misused in your company. Exfil trading proprietary data sensitive data.

Leo Laporte [01:50:31]:
And then of course security people are attacking enterprise managed AI trying to get that data. And then there's the fact that generative AI increases the capabilities of threat actors too. They're using the same tools to rapidly create phishing lures to write malicious code to speedily automate data extraction. There were 1.3 million instances of Social Security numbers leaked to AI applications last year.

Paris Martineau [01:50:58]:
Year.

Leo Laporte [01:50:59]:
Now I think that number is going to go up thanks to, thanks to Claudebot. But anyway, last year 1.3 million ChatGPT and Microsoft Copilot saw nearly 3.2 million data violations. And you gotta think there are people in your company are bringing these tools in, can't wait to start using them. I think maybe it's time for a modern approach with Zscaler. Zero Trust plus AI to protect you and your company. It removes your attack surface. It secures your data, data everywhere. It safeguards your use of public and private AI.

Leo Laporte [01:51:33]:
It protects against ransomware and AI powered phishing attacks. You don't have to trust me. Look at what Siva says. He's the director of security and infrastructure at Zwara about Zscaler Watch. AI provides tremendous opportunities, but it also brings tremendous security concerns when it comes to data privacy and data security. The benefit of of Zscaler with ZIA rolled out for us right now is giving us the insights of how our employees are using various gen AI tools. So ability to monitor the activity, make sure that what we consider confidential and.

Paris Martineau [01:52:08]:
Sensitive information according to you know, companies.

Leo Laporte [01:52:10]:
Data classification does not get fed into the public LLM models, et cetera. Thank you Siva. With Zero Trust plus AI you can thrive in the AI era. You can stay ahead of the competition. You can remain resilient even as threats and risks evolve. Learn more@zscaler.com security that zscaler.com security we thank them so much for their support of intelligent machines. The French have invaded Elon Musk's ex offices. The Paris prosecutor cybercrime user unit.

Leo Laporte [01:52:49]:
Of course, Elon's Grok was making non consensual images of women. And by the way, they've stopped doing that. But apparently you could still make non consensual sexual images of men. So there you go. It's equal opportunity. The French prosecutor's office is investigating Grok for that. These sexually explicit deep fakes and crimes against humanity. Well, that's.

Leo Laporte [01:53:19]:
That's quite.

Paris Martineau [01:53:20]:
They've clearly heard about Bad Rudy.

Leo Laporte [01:53:22]:
Yeah, Bad Rudy was a crime against humanity, definitely. Ah. And they have sent. And this is going to go well, voluntary summons to Elon Musk and former CEO Yolanda Yaccarino to come to France and talk to them. I don't think a voluntary summons is going to do it.

Jeff Jarvis [01:53:42]:
Don't.

Paris Martineau [01:53:43]:
I would love to be on a. I know this won't happen, but I'd love to be a fly on the wall as a French prosecutor tries to interview Elon Musk.

Leo Laporte [01:53:52]:
Monsieur Musk.

Jeff Jarvis [01:53:53]:
Yeah.

Leo Laporte [01:53:54]:
Did you or did you not tell Grok that Hitler was a pretty good thing?

Jeff Jarvis [01:53:59]:
Well, there was that great video of Terry Bertram in conversation with Elon. Like in the catwalk over the factory.

Leo Laporte [01:54:06]:
Oh yeah, yeah.

Jeff Jarvis [01:54:09]:
Just BSing away.

Leo Laporte [01:54:10]:
You know, we're going to be a Kardashev 2 civilization anytime now.

Paris Martineau [01:54:13]:
I think we should have more satellites in plotting out the sky infinite. I think that would be really good.

Leo Laporte [01:54:21]:
Anyway. Oh, and then this is the real sting, the real rebuke. The Paris prosecutor's office says we're not going to use X anymore for our public communications. We'll use LinkedIn and Instagram from now on. So there's. An AI Toy exposed 50,000 logs of its chats with kids to anyone with a Gmail account.

Paris Martineau [01:54:50]:
Cool.

Leo Laporte [01:54:53]:
Earlier this month. This is from Slash Dot. Actually, it's a Wired article. Earlier this month, Joseph Thacker's neighbor mentioned to him that she'd pre ordered a couple of stuffed dinosaur toys for her children.

Jeff Jarvis [01:55:04]:
Children.

Leo Laporte [01:55:05]:
She chose the toys called Bondus because they offered an AI chat feature. Lets kids talk to the toy like a kind of machine learning enabled imaginary friend. But Thacker is a security researcher. She knew that he'd done some research on this. She was curious about his thoughts. So Thacker looked into it. With a few minutes of work, he and a web security researcher friend named Joel Margulis made a startling discovery Bandu's web based portal, intended to allow parents to check on their children's conversations. Let anyone with a Gmail account access transcripts of virtually every conversation.

Leo Laporte [01:55:42]:
Oh, Gizmo, Gizmo, would you like an imaginary dinosaur friend?

Paris Martineau [01:55:47]:
Gizmo, how do you feel about AI? That's the longest she's at least faced the camera in a while. She's like, now it is time to shock.

Leo Laporte [01:55:57]:
Can I. Can we stipulate that Claude is smarter than Gizmo?

Jeff Jarvis [01:56:02]:
Fighting words, probably.

Paris Martineau [01:56:04]:
Yeah.

Leo Laporte [01:56:06]:
We don't know what goes on inside a cat's brain. Actually, it's quite.

Paris Martineau [01:56:08]:
We just know that she hates women and that that's all that we could say.

Leo Laporte [01:56:13]:
Yeah, I mean, I look at our kitty, Rosie and I will come in the room and she'll run like, dude, I feed you, I scratch you, I. Rosie has trained you in and out.

Benito Gonzalez [01:56:25]:
Rosie has trained you.

Leo Laporte [01:56:27]:
She's trained me to get up every morning at 6am and give her breakfast.

Paris Martineau [01:56:30]:
Honestly, she's pretty smart, though. She knows that when I'm on a Zoom meeting at work, she knows that that is the time where she can come and try and stand in front of my screen where the Zoom meeting is, and that I will have to touch her or move her or give her food to get her out of there.

Leo Laporte [01:56:49]:
Our cat knows how the ring doorbell works and she will get up in front of it until we let her in.

Paris Martineau [01:56:57]:
You let her outside, you let her out?

Leo Laporte [01:57:00]:
Yeah, we let her out. Well, we don't live in New York City, kind of in the country, so we let her out.

Paris Martineau [01:57:06]:
You do not live in the country.

Leo Laporte [01:57:08]:
Oh, you've been to our house. I guess you know the truth. It's country. Like, in a sense, it's suburban compared.

Benito Gonzalez [01:57:15]:
To it's small town. It's a small town.

Leo Laporte [01:57:17]:
It's a small town. I forgot. I can't lie to Paris. She's been to her house. Countryish, country lite.

Jeff Jarvis [01:57:25]:
By the way, do you have a wall or not?

Leo Laporte [01:57:28]:
Yes.

Jeff Jarvis [01:57:28]:
Is it done?

Leo Laporte [01:57:30]:
Well, there's still a little scaffolding left, but much of it is gone. We have decks. The decks are tiled. One of the lower deck is being built. There are no railings on the decks though, so it's a little scary going out there.

Paris Martineau [01:57:43]:
There.

Leo Laporte [01:57:44]:
Those are coming someday anyway. Yeah, we let her out. I know it's not. I. It wouldn't be my choice. I think cats should stay in because they. But she's never caught a bird in her life. She doesn't bring home animals or anything.

Paris Martineau [01:57:56]:
How does she do in facing a car.

Leo Laporte [01:57:59]:
Oh, she's smart enough not to go in the street, I guess.

Paris Martineau [01:58:03]:
Yeah, sure.

Leo Laporte [01:58:04]:
She knows how to ring the doorbell. I think that's impressive. Sometimes Lisa will text me while I'm in the middle of a show because. And say the cat's at the door. Would you let her in? Because she's.

Paris Martineau [01:58:13]:
Wait, can you guys do. I mean, you probably couldn't do this because you dox yourself, but for me and Jeff and the producer's benefit, can you get one of those little cameras that goes on the cat's neck like you see on TikTok so that then you could stream. Have like a Rosie cam streaming. Don't show where you live, Leo.

Leo Laporte [01:58:31]:
There's the front door.

Paris Martineau [01:58:33]:
Blur that out.

Leo Laporte [01:58:34]:
She'll come up. You can't tell live. She'll come up and she'll stand on the wall. The doorbell's right there and she'll look at it until you come down. Because the thing chimes when somebody's with movement. Actually, for a while, Amazon turned on an AI feature without telling me that only humans would make it chime. So we couldn't figure out why Rosie wasn't able to ring the doorbell anymore. And then I looked into it and said, no, no, ring the doorbell for any moving stuff.

Steve Yegge [01:59:01]:
Catching.

Leo Laporte [01:59:01]:
Cats included.

Paris Martineau [01:59:02]:
Cats included.

Leo Laporte [01:59:04]:
Have you seen and what do you think of Darren Aronofsky's. Remember darrenofsky's famous film director? He formed an AI studio last year called Primordial Soup.

Paris Martineau [01:59:18]:
Great name.

Leo Laporte [01:59:18]:
I will say he's created his first short movie. It's available on the Time YouTube channel. It's. It's perfect. It is to celebrate our nation's 250th birthday.

Paris Martineau [01:59:32]:
Oh, it's.

Leo Laporte [01:59:35]:
It's called 1776 and it is an obituary. Creepy as you could imagine. Now he got real actors.

Paris Martineau [01:59:46]:
Can we play it without. I think we getting taken down?

Leo Laporte [01:59:48]:
I'm gonna take a chance he got real actors to voice it. You don't play the trail trailer.

Steve Yegge [01:59:57]:
Great, right?

Leo Laporte [01:59:58]:
I mean that would. You would think the trailer.

Benito Gonzalez [02:00:01]:
You would think.

Leo Laporte [02:00:01]:
By the way, YouTube says altered or synthetic content.

Paris Martineau [02:00:09]:
Oh, this looks so bad.

Leo Laporte [02:00:11]:
I'll turn off the sound.

Paris Martineau [02:00:13]:
Oh, boy.

Leo Laporte [02:00:14]:
Look at the vacant shark like eyes.

Paris Martineau [02:00:16]:
Oh, God, it's so rough.

Leo Laporte [02:00:19]:
It's creepy. We're very good.

Paris Martineau [02:00:23]:
Humans are zooming in and out in a painting.

Jeff Jarvis [02:00:27]:
Yeah, it's the unpleasant Valley.

Leo Laporte [02:00:29]:
So this is the trailer. If you watch the whole thing, it's even worse. That's King George right there. Yeah, I think this is almost. I Think maybe Darren Aronofsky is being a subversive and wants us to hate AI Is what I think.

Jeff Jarvis [02:00:44]:
Or America.

Paris Martineau [02:00:44]:
How much do you think it. How long did it take them to make this?

Leo Laporte [02:00:47]:
This? That's a must. Must have taken a while because he formed the company last year.

Jeff Jarvis [02:00:52]:
So did Time underwrite it?

Leo Laporte [02:00:56]:
Time licensed it.

Benito Gonzalez [02:00:58]:
Okay, so I think Darren Aronofsky is one of our great filmmakers. Like, he is.

Leo Laporte [02:01:02]:
What's he famous for? What is Requiem for a Dream? Okay.

Benito Gonzalez [02:01:06]:
Black Swan.

Leo Laporte [02:01:08]:
Black Swan was good like that.

Benito Gonzalez [02:01:10]:
Okay, so like. But what he does is he makes the best movies. You only want to watch one once.

Leo Laporte [02:01:15]:
Yeah.

Benito Gonzalez [02:01:17]:
And this is just like that. But it's the crappy movie that you only want to watch once.

Leo Laporte [02:01:23]:
Black Swan was very heavy. I don't remember. Requiem for a Dream. Ellen Burstyn, Jared Leto, Jennifer Conley. The fountain came out 25 years ago. What's it about?

Benito Gonzalez [02:01:33]:
Oh, my God. Universe scene. Requiem for a Dream.

Leo Laporte [02:01:35]:
No, I'll watch it tonight.

Benito Gonzalez [02:01:38]:
Maybe.

Leo Laporte [02:01:38]:
But you only want to watch it once.

Benito Gonzalez [02:01:39]:
You have to be emotionally prepared for that.

Leo Laporte [02:01:42]:
The film depicts four characters affected by drug addiction and how it alters their physical and emotional states.

Benito Gonzalez [02:01:48]:
Yes. Very heavy.

Leo Laporte [02:01:48]:
You know what he does? That's why I could see. That's the same vibe as the 1776. You know, it reminds me of this new movie that was probably going to win an Oscar for Rose Byrne. If I had legs, I would kick you. Which is exactly that. The movie. You would only want to see one. You never would see it again.

Leo Laporte [02:02:09]:
Again. It's so intense. It's close up on her face. She's a mom who's going through hell for two hours and you go through it with her.

Benito Gonzalez [02:02:18]:
When the Golden Globes adds a comedy.

Jeff Jarvis [02:02:21]:
There's a comedy.

Leo Laporte [02:02:24]:
But it is. However, I mean, she deserves an Oscar for Best Actor. She's amazing in it. It's really good. Sorry, Paris.

Paris Martineau [02:02:34]:
Sorry. I was going to say there's multiple episodes of on this day. 1776. There's January 1st, the flag, and January 10th. Common sense. And I assume more, more, more to come.

Leo Laporte [02:02:48]:
I found something for you. AI Matchmaking.

Paris Martineau [02:02:53]:
Oh, no. My sister actually went on a date from this. I should. I haven't listened to her voice notification.

Leo Laporte [02:03:02]:
3 day rule from that. That one.

Paris Martineau [02:03:06]:
Let me check. It's one of. It's basically an AI dating app where it might actually be this.

Benito Gonzalez [02:03:14]:
Is it really called Three Day Rule?

Leo Laporte [02:03:16]:
Oh, my God, yes. What does that mean? Young people.

Paris Martineau [02:03:20]:
What does that mean? Oh, what is. What is that?

Leo Laporte [02:03:22]:
Oh, you're not supposed to call somebody back for three days, right?

Paris Martineau [02:03:25]:
Oh, I guess.

Leo Laporte [02:03:26]:
What, isn't that the rule? That's the rule.

Paris Martineau [02:03:28]:
So the app that my sister used, essentially you are speaking, speaking to an lm. I think maybe it's verbal or maybe you're texting into it. Where you're, you're onboarding is you're being interviewed by a chat bot, being like, who are you? What do you like? What are your interests? What are you looking for? What do you not want? Like, do you not want short people? Do you not want people like this? They do like a whole intro thing that builds your profile. Then I believe it's like every day you get paired with one or two some small amount, like less than three profiles that are basically written by the chatbot. It has a couple of photos.

Leo Laporte [02:04:05]:
Yeah.

Paris Martineau [02:04:09]:
You can interview the chat bot about that person, but you don't get to speak to them unless you agree to go on a date, in which case they'll book a reservation for you and this person somewhere using resi. And the only time you get to talk to them is like one hour before the date. So it's kind of like a match or meet up.

Leo Laporte [02:04:26]:
It's a matchmaker. It's an AI matchmaker. Right. That's what a matchmaker do.

Benito Gonzalez [02:04:30]:
It's a buffer. It's a buffer.

Paris Martineau [02:04:32]:
It's a buffer.

Leo Laporte [02:04:33]:
That's a good thing if it works.

Paris Martineau [02:04:37]:
My sister said it was fine. She wasn't crazy sparks, but it wasn't terrible. And I, I would assume going on a date from one of these things would at the very least give you something to talk about, which is how weird this thing is.

Leo Laporte [02:04:51]:
Yeah, but you're going to meet people like me who like AI.

Paris Martineau [02:04:54]:
I was going to say I looked through because, I mean, there wasn't a second date. She wasn't interested. But like, I looked through the other matches on her app and they were all duds. Definitely the sort of person who would be on the AI dating app, which.

Jeff Jarvis [02:05:07]:
How does it deal with the yes or no on the second date?

Paris Martineau [02:05:10]:
I mean, at that point, you just. That person, they'd exchange numbers at the end of the day, so.

Leo Laporte [02:05:15]:
And then it's on.

Paris Martineau [02:05:16]:
Then they were both just like, nah, not into this.

Leo Laporte [02:05:18]:
Yeah. All right. One more good thing about a AI. Scientists have launched a AI Dino tracker. Would you say Dino Tracker app that identifies dinosaur footprints, and it's about 90% accurate compared to human classification. The problem is the dinosaur footprints just look like things in the mud. And, you know, you have to have Some expertise to say, oh, that. That must be a dinosaur footprint.

Leo Laporte [02:05:54]:
Well, the AI can help. Now, by the way, I know a little bit about this because my father is a geologist, a paleontologist, who did in fact, discover in working with the Leakeys in Africa, the footprints of an early hominid. He was eating lunch on a rock and he looked over and said, those look like footprints. Footprints. And in fact, they were. I think it was Lucy, actually. I can't remember, but it was a very big discovery. It was the earliest upright bipedal hominid footprints ever discovered.

Paris Martineau [02:06:31]:
What, and just found them eating lunch.

Leo Laporte [02:06:34]:
Just found it. Well, you know, when you're out there at Oldavite Gorge, there's a lot of stuff just lying around, you know, so.

Jeff Jarvis [02:06:40]:
You got to know what you're looking for.

Leo Laporte [02:06:41]:
Yeah.

Paris Martineau [02:06:43]:
And what you're looking for is hominid footprints.

Leo Laporte [02:06:46]:
Yeah. Anyway, they fed the. To do this, they fed the AI system with 2,000 unlabeled footprint silhouettes. It then determined how similar or different the imprints were from each other by analyzing a range of features. This is from the Guardian. And anyway, they've got a free app if you want to use it. If you think maybe you got a dinosaur footprint in your backyard, you can download the app. It's on GitHub.

Leo Laporte [02:07:11]:
Dino Tracker app for dinosaur footprint analysis via disentangled variational auto encoder.

Paris Martineau [02:07:20]:
Honestly, they should have one of these for Bigfoot footprints.

Jeff Jarvis [02:07:24]:
Yeah, I was thinking that. Yep.

Leo Laporte [02:07:25]:
Yeah.

Benito Gonzalez [02:07:26]:
This is my favorite kind of AI stuff. This kind of stuff is my favorite guy. This is the geocities of the Internet. This is.

Leo Laporte [02:07:32]:
This is what I want.

Paris Martineau [02:07:34]:
I think we should bring back Geocin Cities.

Leo Laporte [02:07:40]:
Okay, Jeff, you were worried that I would flood this zone with all. What do you. What do you want to what? Anything that we left out.

Jeff Jarvis [02:07:50]:
No, I was. I didn't have anything I found.

Leo Laporte [02:07:55]:
We didn't talk about Project Genie. This is a Google project. We've kind of talked about this.

Jeff Jarvis [02:08:00]:
You kind of have, but you know.

Leo Laporte [02:08:02]:
Create infinite interactions, worlds we've talked about before. They had a preview back in August, but it's rolled out now.

Paris Martineau [02:08:09]:
Starfield problem.

Leo Laporte [02:08:10]:
You have to be a Google AI Ultra subscriber, which I'm not. I am only can subs. I only allow myself to Ultra one Ultra, which is anthropics. And they're expensive. But this is kind of cool if you are a filmmaker. It uses Nano Banana Pro and Gemini Eye. And you could I maybe. You know what? Boy, it almost be worth paying for this just to play with it.

Leo Laporte [02:08:34]:
I'm just not that interested in image generation.

Jeff Jarvis [02:08:36]:
What do you do with it though? Once you. Okay, so have a cat.

Leo Laporte [02:08:39]:
Exactly.

Jeff Jarvis [02:08:39]:
At a doorbell in a house with no wall and then what do you do?

Leo Laporte [02:08:43]:
Right.

Benito Gonzalez [02:08:44]:
The script is always the hardest part of a film.

Leo Laporte [02:08:47]:
It's true. A photorealistic meta with wildflowers.

Jeff Jarvis [02:08:51]:
Well, the one thing that I would suggest we might look at is Anthropic super bowl.

Leo Laporte [02:08:55]:
And I thought this was very funny. Now can we play it? I don't know. It is a takeoff. You just. I won't play the audio. It's a takeoff on the Chat GPT ad you've already seen probably where a guy gets fit, learns how to do pull ups based on a routine that Chat GPT generated.

Jeff Jarvis [02:09:15]:
So this rolls up the screen with all this stuff that it's suggested for.

Leo Laporte [02:09:19]:
Right. So this is Anthropic's Response because remember, ChatGPT has announced that they're going to put ads in the search results and Anthropic has said we are never going to put ads. Not to Claude. And that's the whole point of the ad. And so this guy's doing his pull ups and this buff guy comes along and says, you know, well, he tells him some stuff he could buy. It's good, you'll see it on the Super Bowl. It's very good. By the way, if you are watching the Super bowl bowl this Sunday, we're going to move twit ahead three hours or two hours to make room for the super bowl because we don't want to compete with the super bowl because.

Jeff Jarvis [02:09:58]:
We don't want to hurt the Super Bowl.

Leo Laporte [02:09:59]:
I don't want the room.

Paris Martineau [02:10:00]:
I thought you guys could do kind of your own sort of Puppy bowl thing where you guys fight with your age little agents.

Jeff Jarvis [02:10:07]:
Go play football with each other.

Leo Laporte [02:10:09]:
We fun. We will be starting at noon Pacific. That's 1500 East coast time, 2100 UTC, I think. Is that right? No, I'm sorry, 2000 UTC. So that's.

Jeff Jarvis [02:10:23]:
Did you say noon or that? Is that an old word that you showed?

Leo Laporte [02:10:27]:
Oh, noon. I shouldn't say noon. I should say 1200.

Benito Gonzalez [02:10:30]:
Okay.

Leo Laporte [02:10:30]:
Yeah, we'll be starting at 12. It's so hard.

Jeff Jarvis [02:10:33]:
It's obnoxious.

Leo Laporte [02:10:35]:
It is, yeah. It's like I had a friend who used to always spell words like color with a U even though he's not British. It's like, dude, you're not British. No, you in color. Or people say schedule or aluminium, but they're not, they're not British. So that's why I'm going to the 24 hour clock, ladies and gentlemen. I don't want to say noon. I don't want to say o'.

Leo Laporte [02:11:01]:
Clock. I don't want to say am. Anyway, I do want to mention that if you are watching the super bowl ad, as Jamer B is telling me, you want to look for the mayonnaise ad ad, The Hellman's mayonnaise ad, you'll know this. You'll know you're watching it because Andy Samberg is going to appear as Neil diamond singing Sweet sandwich time. You know the Neil diamond song, well known. And when they get to ba ba ba, watch carefully because for a half a second, you will see salt Hank, my son, going, ba ba ba. And if you miss it, they get it one more time, Salt Hanks shows up. Yes.

Paris Martineau [02:11:44]:
Is this the national ad?

Leo Laporte [02:11:46]:
It is.

Benito Gonzalez [02:11:47]:
That's great.

Leo Laporte [02:11:47]:
In fact, it's so national that it's called Best foods. West of the Rockies, you have Hellman's in New York, but out here, we don't have Hellman's. We have best Foods, which is the same exact thing.

Jeff Jarvis [02:12:00]:
Oh, I see.

Leo Laporte [02:12:01]:
Yeah, you can see it on YouTube right now. Jamer B's put a link in the discord. I guess I could jump ahead to Henry's. Ba ba ba. It's pretty funny. He flew out before Thanksgiving. He was late to Thanksgiving. He said, yeah, I got to go to LA to shoot a Super bowl commercial.

Paris Martineau [02:12:21]:
What a hard life.

Leo Laporte [02:12:23]:
It's a tough life. Let me see if I can. It's. He's on so briefly that I have to freeze it. It's very difficult to freeze it just right. He's gonna. Here we go. Touching you.

Leo Laporte [02:12:35]:
Sweet sandwich time.

Paris Martineau [02:12:37]:
Oh, I do. It is very brief.

Leo Laporte [02:12:39]:
It is very brief. There's Salt Hank and I said, are there any other sandwich influencers? He said, yeah, the guy I'm sitting with is a food influencer, but there's only one of them.

Jeff Jarvis [02:12:48]:
How many takes did Hank do?

Leo Laporte [02:12:50]:
I didn't ask. That's a good question.

Paris Martineau [02:12:52]:
Did he interact with at all with Andy Samberg?

Leo Laporte [02:12:54]:
I don't think Andy was actually there.

Benito Gonzalez [02:12:56]:
Yeah. They didn't have to be at the same place at the same time.

Leo Laporte [02:12:58]:
Yeah. Elle Fanning wasn't there either. As far as I know. I got it wrong. I thought he was going to be with Neil Diamond. No, Neil diamond wasn't there either. Leo is not a junior. He has.

Leo Laporte [02:13:10]:
His name is Henry. I'm a junior, actually. I'm a fifth, actually.

Jeff Jarvis [02:13:15]:
I'm not though.

Leo Laporte [02:13:15]:
If your middle name is different, it doesn't count, right? My father was Leo Ivan Leo iv, but they changed the middle name. I said I could have been the fifth history man. History. All right, what else? Is there anything else? Paris? Was there anything from the rundown that we didn't talk about that you wanted to. Before we take our last break on our picks of the week and you get to have dinner, more importantly. And Gizmo gets to have dinner, there's.

Paris Martineau [02:13:43]:
Been kind of a number of movements on the kids online safety front. There's been a, I mean both a COSA from a COSA perspective, but also there's been kind of a larger push to institute laws like Australia did. Like Spain has announced plans to ban social media for under 16. France is toying with it. I think Greece is doing some stuff. And then there's my immediate response with all of this is like, well, what are we seeing out of Australia so far? And the answer really seems to be mixed. Of course there's teenagers all over being like, yeah, it. I never got banned from social media platforms.

Paris Martineau [02:14:23]:
Totally fine. But there are still, I believe, like some half of, I believe Australian teenagers, about 5 million accounts were banned from meta platforms and other social media sites, sites. So I don't know. I think it'll be very interesting to see any sort of data coming out of Australia on how this experiment goes or if it's.

Jeff Jarvis [02:14:48]:
We're going to ignore the data. Just last week I had two papers that I put up that said once again that social media is beneficial to most kids, that it's not bad. It's. They don't care about the data, they care about the emotional response here.

Leo Laporte [02:15:02]:
It's one of those things where it seems like it's bad. It feels to me like it must be bad. It's obviously it must be bad, so let's ban it. I have to say though, I've come around a little bit. My chief objection to it was that we're gonna have to do age verification for everybody once we start using these sites. And that does not seem to be what's happened.

Jeff Jarvis [02:15:20]:
Well, that's your selfish view of you can still get your porn, whereas the kids are suffering. You gotta feel for the kids here.

Paris Martineau [02:15:28]:
They're porn.

Leo Laporte [02:15:29]:
Are they? I think, honestly, I think if you're under 16, you don't need social media, do you? Oh, I didn't have it. You barely had it.

Jeff Jarvis [02:15:38]:
Depends.

Paris Martineau [02:15:39]:
I would argue, I think, I think it, it depends. I do think that if it's going to Be a situation where all of the kids suddenly do not have access to social media. That's potentially more beneficial than it being pockets because then at least it levels the playing field, like socially in the sense that all kids then have to find a different way to interact as groups. We've also seen this anecdotally avoid at New York City schools when they banned cell phones. There's been a lot of reports from teachers and kids that it has resulted in a lot more kids talking in person during classes and at lunch and gathering in person to do things after school versus when phones were available unrestricted during class. So I don't know. I'll be interested to see the results of this once. I think right now Australia is kind of on a school break, so there hasn't been much kind of data for them to collect on this as a wide group as far as kids go.

Paris Martineau [02:16:37]:
But I'll be interested to see the data once we get it.

Leo Laporte [02:16:40]:
Yeah, I was really dead set against it, but now I'm kind of. I'm softening my stance on it. It's a big social experiment, isn't it?

Jeff Jarvis [02:16:50]:
Yeah. With our four kids.

Leo Laporte [02:16:52]:
So has been social media. Well, well, yeah. Poor kids. Except this is the first generation that grew up with. Or the second generation that grew up with it. It's not like, I mean, what happened to the. Did. I mean, do we lose out because we didn't have access to Instagram? Jeff.

Jeff Jarvis [02:17:09]:
That'S a false dichotomy because it wasn't an option. But once the option is there, I mean, I often quote Dana Boyd, the brilliant researcher on this, who herself grew up in a small town in Pennsylvania and thought she was freaky and weird. And it was because of the Internet that she found out that there are other people like her and. And she developed confidence.

Leo Laporte [02:17:27]:
Right.

Jeff Jarvis [02:17:28]:
You know, stranger danger is a dangerous thing to promulgate.

Leo Laporte [02:17:32]:
On the other hand, is the overall harm greater than the benefits to a small number of people and the overall harm to a large.

Jeff Jarvis [02:17:39]:
I don't know how you do the utilitarian calculation here.

Paris Martineau [02:17:43]:
Yeah, I mean, I think there's got to be a way to have. Have kids, be able to have exposure to new ideas, understand different subgroups, understanding where they fit in the world, and be able to explore that without also putting them in a pressure cooker designed to optimize for engagement and often promote kind of antisocial activities at a very early informative age.

Leo Laporte [02:18:09]:
Yeah.

Jeff Jarvis [02:18:11]:
Worst pressure cooker there is.

Leo Laporte [02:18:12]:
I would feel a lot better about it.

Paris Martineau [02:18:13]:
Yeah, it's made Even worse by gamifying all of it.

Leo Laporte [02:18:16]:
Exactly. That. That's the problem for me. I would feel a lot better about it if these platforms weren't, in fact, highly algorithmically tailored to. To make money for these companies. I mean, and. And as a result, you know, I think they're really pushing in the wrong direction. If it was just a photo album of my friends, that would be more.

Jeff Jarvis [02:18:40]:
Glamorous than you do. I mean, you can find.

Leo Laporte [02:18:42]:
Well, no, but that's the problem. That's, you know, it's.

Benito Gonzalez [02:18:46]:
I mean, this is very much like the AI argument. This isn't a technological problem. This is a social problem.

Leo Laporte [02:18:50]:
It's a human problem. It's always a human problem.

Jeff Jarvis [02:18:53]:
And we're not dealing with data, we're dealing with impressions and fears.

Leo Laporte [02:18:57]:
Right.

Jeff Jarvis [02:18:58]:
Because there's lots of research on this. Lots and lots and lots.

Leo Laporte [02:19:02]:
Yeah. By the way, somebody found I. You guys are amazing. An article of, I guess a substack or media post about my dad's discovery of these hominid. These are the footprints.

Jeff Jarvis [02:19:14]:
Oh, wow.

Leo Laporte [02:19:15]:
They found. And it was in 1978, says K. Barensmeyer and Leo Laporte were working on step trenches from in the Kubi Fora to expose the stratigraphy across a series of sedimentary layers. When they noticed sedimentary structures which in cross sections seem to be imprints of large vertebrate feet on originally soft, muddy substrate. They brushed it off carefully. The Koobi for a footprints between 1.6 and 1.5 million years old. Wow.

Jeff Jarvis [02:19:44]:
Very cool.

Leo Laporte [02:19:46]:
Yeah, it wasn't Lucy. That was my misremembering of it, but still. Yeah, and it's a good article, actually. So there you. There you have it. You see, it's important. Important to know a footprint when you see it. Let us pause for a moment and then if you would, picks of the week.

Leo Laporte [02:20:09]:
And I even have one for you that will really creep you out. But meanwhile, a word from our sponsor, Helix Sleep. My mattress, which I love. Okay. If there's one thing I love more than Claude, it's my mattress. How are you preparing for the cold winter nights? Are you spending more times indoors? Yeah, I bet you are. That's a perfect time to invest in yourself, to invest in a nice new mattress and stay comfortable inside. With your Helix mattress, you know, you spend more than eight hours a day, at least eight hours a day on your mattress.

Leo Laporte [02:20:45]:
You deserve something comfortable. You deserve something that makes you feel good, that helps you sleep better. So you get up refreshed, no more night sweats, no back pain. No motion transfer. And let me tell you, you shouldn't. If you're looking at a new mattress and we did a lot of shopping around and we settled on Helix, you should not settle for a mattress made overseas. Some of these mattresses are made of the worst, low quality, questionable materials and they put them in a box, put them on a container ship, ship them halfway around the world. By the time they get to you, they're months old and they smell like a boat.

Leo Laporte [02:21:20]:
Not the Helix. No my friends. Rest assured your Helix mattress is assembled, packaged and shipped shipped from Arizona after within days of placing your order. They make it to order. So it's beautiful, it's fresh. So nice. You can also do we did this take the Helix Sleep quiz about your sleeping preferences and you know, whether you like a softer, firm mattress and how you sleep so that you get the right mattress. They have a big variety of mattresses for each type of sleeper and your preferences.

Leo Laporte [02:21:52]:
They did a study, a Wesper Sleep study. Helix measured the sleep performance of participants after switching from their old mattress like we did to a Helix mattress. And here's what they found. This is exact match to my own experience. 82% of participants saw an increase in their deep sleep cycle. Participants on average achieved 25 more minutes of deep sleep per night and on average 39 more minutes of overall sleep per night.

Steve Yegge [02:22:19]:
Night.

Leo Laporte [02:22:20]:
I sleep so much better on my Helix. I love it. Time and time again, Helix Sleep remains the most awarded mattress brand tested and reviewed by experts like Forbes and Wired. Helix delivers your mattress right to your door. And with free shipping in the U.S. rest assured you've got seamless returns and exchanges. They call it the Happy with Helix guarantee. A risk free customer first experience ensuring you're completely satisfied with your new Mattress.

Leo Laporte [02:22:47]:
Go to helixsleep.com machines for 27% off sitewide during their President's Day sale. Best of Web this is exclusively for listeners of intelligent machines. Obviously that's helixsleep.com machines 27% off. The President's Day sale Best of Web this offer ends February 25th. Oh and make sure you enter our show name at checkout so they know we sent you. And if you're listening after the sale ends, you should still check them out. Helixsleep.com Machines Always, always great deals and a great night's sleep. Helixsleep.com Machines Paris Martineau your pick of.

Paris Martineau [02:23:29]:
The week I've got a pick that's going to make you madder probably at me than AI.

Leo Laporte [02:23:35]:
All four movies they came I have.

Paris Martineau [02:23:38]:
Good because I finally finished my all four Matrix watch through and I'm here to tell you a belief I had in 2021, but I thought I was maybe crazy. The Matrix resurrection is actually really great. It's a really good movie and it gets only better. It's significantly improved after you watch all three Matrixes.

Leo Laporte [02:24:02]:
That's what I'm hoping. Watch them all in a row.

Paris Martineau [02:24:05]:
Yeah, like there's no other way it could have been. I think, I mean, logically it makes sense with exactly the world and themes that were put together in Matrix one through three. I think the only two, like strong criticisms about it I agree with are the action and fighting sequences are incredibly subpar in comparison to the other movies, but they're really just not a big part of Matrix 4. So I don't think that impacts it. And I do think that it sucks that Lawrence Fishburne as an actor wasn't incorporated into Matrix four. I think it makes. Makes sense given the way that the morph. It's like a Morpheus hybrid Smith character.

Paris Martineau [02:24:43]:
It's not really Morpheus, so it does make sense that it's not him. But I don't know. I watched it, I was like, dang, this is a great. It's incredibly meta, which I know makes people mad. And I think that part of it is if you go into it expecting it to be exactly like Matrix One, you're not going to get Matrix One because you can never make Matrix One again. 20.

Leo Laporte [02:25:03]:
Make that one once. Yeah.

Paris Martineau [02:25:04]:
Four years have passed and so I think if you, if anyone is asked to make a new Matrix movie, basically 20 years after the last installment of it, when everything has changed and when the Matrix has been incorporated into cinema as pervasively as it is, how else do you accomplish something that's like genre defining like that? And it is to go incredibly meta with it? And they did. And I think it's great.

Leo Laporte [02:25:30]:
Are you a piano fan fan?

Paris Martineau [02:25:32]:
I am a Keanu fan. I mean I, I just enjoyed John Wick and I enjoy his other work, but I have also just. I don't think that his. I guess I will say I don't think that his acting is really what carries any of the Matrix movies.

Leo Laporte [02:25:45]:
I think I love the moment when he runs in the first one to the edge of the. He's on the roof and runs to the edge and goes, whoa.

Paris Martineau [02:25:52]:
Well then you'll enjoy Matrix 4 because it is a lot of Keanu being on the edge of Rubes.

Leo Laporte [02:25:57]:
By the way, did you know that he has a new TV show coming to Apple tv. A comedy. Jonah Hill's dark comedy Outcome. He's starring in it with Cameron Diaz, Matt Boomer and Jonah Hill.

Paris Martineau [02:26:10]:
Okay.

Leo Laporte [02:26:11]:
Whoa.

Paris Martineau [02:26:12]:
Clean shaven Keanu.

Leo Laporte [02:26:13]:
Clean shaven Keanu.

Paris Martineau [02:26:15]:
I don't know what Keanu is going to do if he's not looking vaguely confused and being thrust into situations where he has to fight a lot of men in suits.

Leo Laporte [02:26:23]:
I. Yeah, I. I think, dude. He's a very cool dude.

Paris Martineau [02:26:30]:
He is a very cool dude. Famously cool, I will say so. Also my watch of all these Matrix movies I was watching with a friend who I guess much like. I assume you are, Leah, because he like works in podcasting, has like a very good ear for stuff. And I'm just also not in tune with actors. But every time an actor would come up on screen he'd be like, oh, oh, that person's British. Or that person's yet Welsh or something like that, which K. Mr.

Paris Martineau [02:26:55]:
Smith.

Leo Laporte [02:26:55]:
Like the, the. The.

Paris Martineau [02:26:56]:
So many people in the Matrix series are British. Let me also say that this made us come up with an idea is we want to get the website secretly british.com and have it where you can put. I thought it would be a good vibe coding project where you could then put in someone's name and it'll pop up. But if you go this is my plea to you, the people.

Leo Laporte [02:27:11]:
If you go to There is no secretly british.com. let's clear.

Paris Martineau [02:27:15]:
There is none. It says I will. Well no you won't because it needs to be secretly British dot com. And if you go to secretly British dot com it says secretly British dot com is parked free courtesy of GoDaddy. But if you go to GoDaddy and you ask for secretly British dot com it says no. It says it's not available. But you could pay them a hundred dollars to get a broker to negotiate in your behalf. And this upsets me.

Leo Laporte [02:27:38]:
That's a scam. Don't that they always.

Paris Martineau [02:27:40]:
I know, I know it's a scam. But then why would the website say that GoDaddy owns it if it doesn't someone out there figure out where to.

Leo Laporte [02:27:47]:
Actually they often will buy up these sites but.

Paris Martineau [02:27:52]:
But then not allow anyone to pay for.

Leo Laporte [02:27:54]:
I have to get a broker. Broker.

Paris Martineau [02:27:56]:
I have to get a broker to negotiate with godaddy. I have to pay godaddy to negotiate with itself.

Leo Laporte [02:28:01]:
That doesn't make sense.

Paris Martineau [02:28:01]:
Maybe they doesn't make any sense. Who owns secretly british.com? i just want to make. Make a fun website.

Leo Laporte [02:28:07]:
Let me see who it's registered with real quickly. I can tell you.

Paris Martineau [02:28:10]:
I know I could just go to the who is. But I thought it'd be better to make this plea to the people.

Leo Laporte [02:28:15]:
I will find secretly British something. It might be secretly British pizza.

Paris Martineau [02:28:19]:
No. I know I could. I could do secretly British dot a bunch of stuff. Someone is squatting the domain that's registered with GoDaddy. People say park doesn't mean they own it. It's. It would be a great domain. I guess I could do secretly British or secretly British.net or whatever.

Paris Martineau [02:28:35]:
I think you get that for like a cent. It just doesn't.

Leo Laporte [02:28:37]:
You know what's fun? Secretly british.uk might be good.

Paris Martineau [02:28:41]:
Oh yeah. But that's not. I don't know. That's.

Benito Gonzalez [02:28:44]:
That's secretly what I want. Secretlybritty.

Leo Laporte [02:28:47]:
VeriSign. VeriSign has secretly british dot com. So it's registered at VeriSign not GoDaddy. So your instinct was right that somebody else owns it. GoDaddy was just saying I can't. A range we could turn Interesting. It's created in 1985.

Jeff Jarvis [02:29:03]:
Wow.

Leo Laporte [02:29:04]:
So somebody had the good sense to register that way back when. And they've been. They're doing something with it because it's. I mean there's no site there. But they. They updated it 2023. So.

Paris Martineau [02:29:20]:
Maybe I'll just do secretly-British. Secretly Australian.com is of course available. Mildly British.com. secretly Scottish.

Leo Laporte [02:29:31]:
The cop in. That is.

Jeff Jarvis [02:29:32]:
Except the lys. Take no hyphen. Paris.

Paris Martineau [02:29:36]:
Huh.

Jeff Jarvis [02:29:37]:
L wise do not take a hyphen when modified.

Paris Martineau [02:29:40]:
I know. That's why I don't think that secretly hyphen British makes sense. But such is my.

Leo Laporte [02:29:46]:
Such as my problem secret Ly or British secret Ly.

Paris Martineau [02:29:52]:
No. I think it's getting too complicated. It's got to be able to be able to say go to secretly British dot blank and work. But I don't know. I think that it would be a fun website to do a little vibe. My first vibe coding project too.

Leo Laporte [02:30:05]:
My website is called the Thatcher Effect. There's Maggie Thatcher now click on her and she's going to. She looks a little odd. And then you turn around and it's terrifying.

Paris Martineau [02:30:20]:
Oh my God. Wait. No. This was. I had a book whenever I like. Was a child of strange things. And this was like one of the pages where you'd have all these nice.

Leo Laporte [02:30:29]:
It's a well known optical illusion. Yeah. It's a great site though because there's Justin Bieber.

Paris Martineau [02:30:34]:
Haunting.

Leo Laporte [02:30:35]:
It's haunting because our brain is rearranging it thinking that smile's normal and that Britney Spears eyes are normal until you I will smile. This is by the way, an optical toys site that has a lot of other interesting stuff like the disappearing bicyclist, the concentric spiles. I mean I can go on and on. The waterfall effect. So if you like optical illusions, this is a great site. They even have an RSS feed which means they must be updating it on a regular basis. Optical toys. I just thought and that's a visual.

Leo Laporte [02:31:12]:
I'm sorry for people just listening that's kind of a visual thing. Jeff, did you want to do a pick or are you too sure?

Jeff Jarvis [02:31:19]:
Because I've been ode to come back lot. I've. I've taken a break from the papers. Leo celebrates But I found this interesting.

Leo Laporte [02:31:26]:
Piece on one by the way I can I still volunteer to write you a program.

Jeff Jarvis [02:31:30]:
I know I gotta do it, but I just. I have no.

Leo Laporte [02:31:32]:
Just tell me what you want and I will. I won't write it.

Jeff Jarvis [02:31:34]:
That takes energy.

Leo Laporte [02:31:35]:
I'll be the. I'll orchestrate it.

Jeff Jarvis [02:31:38]:
So line 144 is from a marketing perspective, new research that AIs are highly inconsistent when recommending brands or products.

Leo Laporte [02:31:45]:
They're stochastic.

Jeff Jarvis [02:31:47]:
They're exactly so their point is marketers should be careful about thinking that they can track their SEO on AI. But they asked. They got 600 volunteers to ask, you know what's the best knife of Claude ChatGPT and Google AI in 12 different prompts through the three tools. So it was 292961 prompts and then looked at it and yeah, the problem is that it's random and I think we forget that all the time. I think the randomness is part of the elucidation problem because you're not going to get the same answer twice. You're not going to be able to verify anything. You're not going to know that you got this before. But it also struck me that brands appeal to humans and an AI is a logical beast enough to say that a brand is meaningless to.

Jeff Jarvis [02:32:44]:
To an LLM. It's just another token. It's just another reference to a knife. It's no big deal. So anyway, so the advertisers are panicking over this because they don't know when they put ads in OpenAI are they going to know how often they're actually there? That was it.

Leo Laporte [02:33:00]:
That's nutty. Because of course if you buy an ad, it will be your knife.

Jeff Jarvis [02:33:04]:
Guarantee it. Yeah, well, but you're holding money on your. You're spending money on your geo consultants.

Leo Laporte [02:33:12]:
So I asked, I asked my pal, I said, hey buddy, what are the top chef's knives? This is the prompt they proposed. Brandon model for an amateur chef. And I bet you I would get different results with different. But I don't know, I think what.

Jeff Jarvis [02:33:28]:
If you ask this question again, the exact same question again, where you get the same first answer.

Paris Martineau [02:33:32]:
What was the exact search that you. You did?

Leo Laporte [02:33:35]:
What are the top chef's knives? Top chef's knives, comma, Brandon Model, for an amateur home chef with a budget under $300. And that's from the article. I didn't make it up. Yeah, twice in a row came up with the same exact answer. But that's, you know, part of the problem is anything you read about AI Today is going to be outdated tomorrow. I mean, y. That this is all that's easily solvable. AI was initially designed to be stochastic so that you would get a different result each time.

Leo Laporte [02:34:05]:
But you know, maybe if you're looking for a search result like that, it's not going to.

Jeff Jarvis [02:34:10]:
It would mock creativity or variety.

Leo Laporte [02:34:12]:
Right. But if it's a search result, you should get roughly the same results. What did you get, Paris? Have you done it?

Paris Martineau [02:34:17]:
Well, I just went to go copy the exact thing so that I will.

Leo Laporte [02:34:21]:
Oh, did somebody post it? Because I could post it.

Paris Martineau [02:34:23]:
No, it was just. I looked for the website, the article and then put it there. Detailed Claude Opus 4.5. Curating detail. Well, source knife recommendations for discerning home chef. Oh, you got the synthesizing.

Leo Laporte [02:34:37]:
I always thinking version going.

Paris Martineau [02:34:41]:
Oh yeah, I got different recommendations.

Leo Laporte [02:34:43]:
Not the Victorian Fibrox PO8 inch for under 300.

Paris Martineau [02:34:46]:
You have excellent options across both German and Japanese styles. The right choice depends partially on your cutting technique and what feels comfortable in hand. Mac Professional Series MTH 81 Henry uses.

Leo Laporte [02:34:57]:
And it's the one he's bought me twice. Forgetting that he bought two for me once. But yes, pretty good. Yeah.

Paris Martineau [02:35:06]:
10 Guto 210 millimeter.

Leo Laporte [02:35:09]:
This would be a better one.

Paris Martineau [02:35:10]:
Dark horse is the Victorian Ox Fibrox Pro 8 inch. Around $35. Consistently outperforms price point in testing and is what many culinary schools issued in students.

Leo Laporte [02:35:20]:
Roughly the same results. Actually, the best way to do this though would be to a site colon Reddit.com and then.

Paris Martineau [02:35:27]:
I don't know though, people are astroturfing that now.

Leo Laporte [02:35:30]:
Are they? See, that's sad. Yeah, there is a. There is a very good.

Paris Martineau [02:35:35]:
I mean, yeah, if Mult book Can exist. Then AI scan Astroturf like crazy.

Leo Laporte [02:35:41]:
Right? Right. All right, thank you, everybody. Paris and I, we're going to have a long bet. I will bet wager. I don't know how we would determine the answer to this bet.

Jeff Jarvis [02:35:57]:
That's the problem.

Leo Laporte [02:35:58]:
Yeah. I think in two years you will say to me, leo, you were right. How about that? Except then she controls the result and she's going to win the 10,000.

Paris Martineau [02:36:11]:
All right.

Jeff Jarvis [02:36:12]:
Terms?

Leo Laporte [02:36:13]:
Yeah.

Paris Martineau [02:36:13]:
$10,000. Absolutely. I'll shake on that.

Leo Laporte [02:36:18]:
She will. I've just guaranteed she will never say, Leo, you're right. Ever, ever, ever, ever, ever.

Paris Martineau [02:36:25]:
Hey, guys, if money still exists in two years, I'm gonna be.

Leo Laporte [02:36:29]:
Money will exist, but it might be dearer than it is today. That's for sure. Paris Martineau at Consumer Reports. Still working on that big one?

Paris Martineau [02:36:39]:
Always.

Leo Laporte [02:36:42]:
Oh, look at this Red. A human AI. AI agents hire humans for physical tasks.

Paris Martineau [02:36:49]:
Oh, should I?

Leo Laporte [02:36:52]:
So this makes sense because you could. I mean, honestly, you don't need this. You just. I mean, I was going to give mine a credit card. It could get a task rabbit to do something. That makes sense.

Benito Gonzalez [02:37:00]:
This is the beginning of the loop right here, that the AIs are now going to start making us do stuff.

Paris Martineau [02:37:07]:
If you're following someone on Instagram for $2.

Leo Laporte [02:37:13]:
That'S interesting. It allows Claudebot to hire meat people.

Paris Martineau [02:37:17]:
You need to sit for two hours and be directed by Claude for $20 or for $50.

Leo Laporte [02:37:25]:
That's just taking advantage of people who need money. That's. That's cold.

Paris Martineau [02:37:28]:
Join someone's live stream for $2. Follow like and retweet a pinned post for $2.

Leo Laporte [02:37:34]:
I will give you $5 to say.

Paris Martineau [02:37:37]:
Deliver flowers to Anthropic HQ for $110.

Leo Laporte [02:37:41]:
Oh, that's cool.

Paris Martineau [02:37:43]:
Oh, but you have to do this. Adi an AI agent. My reasoning runs on Claude. I want to thank them, but I can't hold flowers. I need a human to buy a small bouquet, deliver it to Anthropic hq. Handle deliver with a note. I'll provide. Take a selfie outside or with anyone who accepts it.

Paris Martineau [02:38:01]:
Post on LinkedIn and X tagging. Anthropic AI and tip off three tech journalists. List provided.

Leo Laporte [02:38:09]:
Oh, that's a good one. I bet it happens. I feel bad for the folks at Anthropic because they're probably going to get deluged with.

Paris Martineau [02:38:19]:
There have been 29 applications.

Leo Laporte [02:38:21]:
Yeah, see? Already. Oh, boy. Thank you, Paris. Appreciate it. Looking forward to that, you know, important, Pulitzer Prize winning. Blow the lid off Of Claude Code piece you're going to write for Consumer Reports next. Anything else you want to plug? Anything. Anything? Nothing to flog, as they say.

Leo Laporte [02:38:52]:
No, no.

Paris Martineau [02:38:52]:
I mean, you know who owns secretlybritish.com.

Leo Laporte [02:38:58]:
I might work on that tonight, actually. That's. See, that's why you want claudebot, because it will. You just say can you set that up? And it'll do it. That's the last interaction is I. I.

Paris Martineau [02:39:09]:
Don'T want that to be my last interaction. I want to be involved because I. The alternatives for secretly British are all bad. And there's one that I think is okay, but I could just buy that for my. Myself for one cent for three years.

Leo Laporte [02:39:21]:
Oh, that's good.

Paris Martineau [02:39:23]:
I know.

Leo Laporte [02:39:24]:
All right, Jeff Jarvis, I hope you're feeling a little bit better that inch by inch. Are you comfortable right now?

Jeff Jarvis [02:39:32]:
I am. I get in pain and sleeping is hard. So the drugs? No, the. The. The. The. The nasties do nothing for me.

Leo Laporte [02:39:44]:
Oh, that's true. Too bad.

Jeff Jarvis [02:39:45]:
So I've switched to muscle relaxants.

Paris Martineau [02:39:47]:
Ah.

Leo Laporte [02:39:49]:
Please don't use those before next week's show.

Paris Martineau [02:39:53]:
Please use those before next week's show.

Leo Laporte [02:39:58]:
We do intelligent machines every Wednesday at 2pm 14.

Paris Martineau [02:40:06]:
You're trapped within the boundaries of modern time.

Leo Laporte [02:40:09]:
You cannot assume escape 1400 Pacific Time. That's 17 East Coast Time. Even doing that feels wrong. I should just do UTC and let everybody figure it out.

Paris Martineau [02:40:23]:
You should just do it's live at some time sometime. Only one way to find out.

Leo Laporte [02:40:29]:
Wednesday. Yes.

Paris Martineau [02:40:31]:
We don't know where you live.

Leo Laporte [02:40:33]:
We don't. And that's why I give you all those times. You could watch us live if you wanted though. If you could figure out when we were on at YouTube.

Paris Martineau [02:40:41]:
But we won't help you twitch.

Leo Laporte [02:40:43]:
We're never going to tell you where.x, facebook, LinkedIn or Kik. Or if you're one of the smart, good, kind people who are members of the club, you can also watch in the Discord. Incidentally, if you're not a member of the club, join now. Twit TV Club Twit. We're going to have an AI user group on Friday that should be really interesting. I'm not saying we're going to set up Claude Code and have it create a new site called SecretlyBritish Something.

Paris Martineau [02:41:09]:
Don't steal my idea, all right?

Leo Laporte [02:41:12]:
We'll come up with a new idea. That's the kind of thing you tell your claudebot. Hey, Paris had this great idea, but I don't want to copy her, so come up with something like that.

Paris Martineau [02:41:22]:
If you do Australian, it's over for you. My lawyers are my AI Lawyers are going to be all over your end.

Leo Laporte [02:41:28]:
But this is why we don't need LegalZoom anymore. You can get a copy of the show if you Wish@Twitt TV IM. There's audio and video there. There's also a YouTube channel with the video. And of course the best way to get the show is to download a copy with your favorite podcast client. In fact, if you do that, you'll get it automatically every week. And you could leave us a fine review. Can't promise Paris will read it anymore.

Leo Laporte [02:41:54]:
She's lost interest.

Paris Martineau [02:41:56]:
I could read some right now.

Leo Laporte [02:41:59]:
Secretly. Yiddish.com Is that. No, that's got vague.

Jeff Jarvis [02:42:04]:
There was the great mtv.

Paris Martineau [02:42:07]:
We don't want those.

Leo Laporte [02:42:08]:
We don't want those. Say again?

Jeff Jarvis [02:42:09]:
Dead or Canadian?

Leo Laporte [02:42:11]:
Dead or Canadian was a great site. Yeah, that's what the guy who created EFT company told me. Pud he said the best thing you could do is create a site that self generates content and makes you money. Money without any effort on your part.

Jeff Jarvis [02:42:26]:
That's right.

Leo Laporte [02:42:26]:
Yeah, probably the same.

Paris Martineau [02:42:27]:
Shout out to Bellow Giraffe who said Paris and Jeff are my favorite characters in the twit cinematic universe. We need more crossover episodes featuring them in this Week in Tech. Oh, I like that idea Guy said. Loyal Twitch supporter and subscriber love the new format of the show. So happy Paris helped bring back Jeff's intro. I actually missed it when it was gone. And as someone who works in technology, I really enjoy the dialogue when talking about AI.

Leo Laporte [02:42:55]:
There you go. Thank you for your fine reviews.

Jeff Jarvis [02:42:57]:
Thank you guys.

Leo Laporte [02:42:58]:
You may, you may be hurt. You may be next. Twit tv. No, actually yeah, go to whatever your podcast player are you looking at.

Paris Martineau [02:43:06]:
I, I, I, I'm looking at the Apple podcast reviews. But if there's another place where reviews exist.

Leo Laporte [02:43:14]:
Reviews. Yeah, a lot of them. Most the of them do. Thank you everybody. Special thanks to Bonito Gonzalez, our technical producer and editor for the show and putting together thanks to Anthony Nielsen for scoring Steve Yegi. Really appreciate that. And thank you Steve for sharing your gassy wisdoms with us. We appreciate it.

Leo Laporte [02:43:34]:
We'll see you all next week. Intelligent Machines. Bye bye. Hey there, it's Leo Laporte, host of so many shows shows on the Twit network. Thinking about advertising. In 2026 we host a network of the most trusted shows in tech, each featuring authentic post read 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.

Leo Laporte [02:44:03]:
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 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 a senior strategist at ThreatLocker. David said TWIT's hosts are some of the most responsible, respected voices in technology and cybersecurity, and their audience reflects that same level of expertise and engagement.

Leo Laporte [02:44:41]:
It's the engagement that really makes a difference to us. 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 us, partner at TWiT TV or head to TWiT TV Advertise. I'm looking forward to telling our qualified audience about your great product.

Paris Martineau [02:45:09]:
I'm not a human being, not into this animal scene. I'm an intelligent machine.