Intelligent Machines 884 transcript

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

Leo Laporte [00:00:00]:
It's time for Intelligent Machines. Jeff's here, Paris is here. Our guest this week, Dan O'Dowd, is the founder of the Dawn Project. You may have seen their ads on the Super Bowl a couple of years ago. He says he's making computers safe for humanity. He says Elon's not. That's next on Intelligent Machines. Podcasts you love.

Leo Laporte [00:00:21]:
From people you trust. This is TWiT. This is Intelligent Machines with Jeff Jarvis and Paris Martineau. Episode 884, recorded Wednesday, August 19th, 2026. Cyber Gym. It's time for Intelligent Machines, the show where we cover the latest in AI, robotics, and the smart things all around us. Today, not self-driving cars, but we'll talk about that in just a moment. First, let me say hi to our hosts.

Leo Laporte [00:00:51]:
Parris Martineau is here from Consumer Reports. Hello, Parris.

Paris Martineau [00:00:55]:
Hello, Leo. Bonjour.

Leo Laporte [00:00:57]:
Ça va bien?

Paris Martineau [00:00:59]:
Bien.

Leo Laporte [00:01:00]:
Okay. She speaks like a Parisienne. Also here, he speaks like a Brooklyner. He is, of course, the Emeritus Journalism Professor of Innovation in Journalism at the City University of New York at the— okay, I'm gonna say it— Craig Newmark Graduate School of Journalism.

Paris Martineau [00:01:19]:
You have to hand it to him on the eve of his publishing.

Leo Laporte [00:01:24]:
I know. Any day now, hot types start to arrive.

Paris Martineau [00:01:26]:
Tomorrow!

Jeff Jarvis [00:01:27]:
Tomorrow.

Leo Laporte [00:01:27]:
This is kind of a hot one. The magnificent machine that gave birth to mass media.

Jeff Jarvis [00:01:33]:
Aren't you wonderful? Thank you, Paris.

Leo Laporte [00:01:35]:
And we are a mass medium. Our guest today is very, very interesting. You may know the name Dan O'Dowd from the Dawn Project. Dan, it's great to have you. Welcome.

Jeff Jarvis [00:01:47]:
Good to be here.

Leo Laporte [00:01:48]:
When he got out of school at Caltech, I'm going to go back in time, Dan, a little bit. You helped the Mattel game device of the time, which I had. I think a lot of us had the Mattel Electronic Football Game.

Dan O'Dowd [00:02:09]:
Right.

Leo Laporte [00:02:09]:
The must-have Christmas toy of 1977. National Semiconductor in '78 designed the architecture of the NS32000, a 32-bit microprocessor that was used in NASA's Global Surveyor. You may remember Surveyor mapping Mars. In fact, that's what Dan's been doing a lot of, is writing highly reliable software. The Integrity Real-Time Operating System, which is the operating system for the B-1B Lancer nuclear bomber. You don't want to— you don't want a nuclear bomber to have crashes or software glitches if you've ridden in an A380, which many of us have. You were riding on Dan's software, the B-2, the B-52, the F-16, F-22, F-35, NASA's Orion. Is this right, Dan? You worked on an unhackable laptop for the FBI?

Dan O'Dowd [00:03:09]:
Yep. Every FBI agent had one. They could hook up to the internet securely and communicate to FBI headquarters. through a Starbucks link, through Starbucks Wi-Fi.

Paris Martineau [00:03:23]:
The most secure internet of all.

Leo Laporte [00:03:27]:
What is the secret? What is— I mean, that's— first of all, to say anything's unhackable is to invite hackers. Was any of this ever cracked?

Dan O'Dowd [00:03:37]:
No, I mean, I invite them all the time. You know, go ahead. You know, you say you can hack my software. It's right out there. It's in tons and tons of places. It's not everywhere it needs to be. That's one of our problems, actually, that we're using it in all these different places. It's gone through all the reviews.

Dan O'Dowd [00:03:58]:
It's gone through NSA's reviews. We got from the NSA the highest certification, security certification anyone's ever received. No one's even close because that's what we do. We build weapons of mass destruction, or at least we build the software underneath them, the operating system software. And it just, it can't ever fail and it can't be hacked. It just would be catastrophic for that to be true. And we've been doing this for more than 30 years for all the most critical military equipment. That's what they use.

Dan O'Dowd [00:04:30]:
They don't use Linux. They don't use other commercial software that just wasn't developed to the standards. We spend thousands of dollars per line of code. We test everything so many times over and over again. I've looked at the same piece of code 100 times, along with the other people. And that's simply what we focus on, the most critical applications out there. And that's part of why we're here with the Dawn Project, which is the critical infrastructure, particularly the power grid, is the number one problem. But there's some other— Oh my God.

Dan O'Dowd [00:05:08]:
Some runners up there. The power grid, we cannot live without it. People don't understand that if we lose the power grid, there will be mass starvation in the United States of America in weeks. There will be no one, you can't, without electricity, you don't have any money. All your money's gone. You can't call the bank, you can't send anything, you can't send a check. There's no way to pay anything. No one's gonna deliver anything to you if you don't have money, 'cause they've got something and you give them nothing.

Dan O'Dowd [00:05:39]:
There will be no trucks, there'll be no trains from Iowa to California because no one will be paying them.

Paris Martineau [00:05:45]:
There'll be no food.

Dan O'Dowd [00:05:46]:
Yeah. It'll be gone in very short order. The fuel, you know, you won't have fuel to drive your car. You could drive 100 miles or 200 miles, same thing for your electric car. But if the whole country is down, there's nowhere to go. There's nowhere to escape. The FBI director Recent FBI director, not our current one, but he said this is the defining threat of our generation. The NSA director said we're going to get hit.

Dan O'Dowd [00:06:14]:
They're already prepositioned. The hackers are prepositioned in our networks, just waiting for a signal to bring down the networks.

Paris Martineau [00:06:25]:
So, why has our power grid stayed so insecure, despite seemingly all of the people in a position to change this being aware of this problem for quite some time?

Dan O'Dowd [00:06:36]:
They don't want to change it. It's a huge investment. It's a trillion-dollar investment that they've got in there, multi-trillion dollars. They would have to redo all the software that they're using right now, and that is going to cost a lot of money. I'm not saying it's not, but we have to. We have to do it. They refuse. I've tried.

Dan O'Dowd [00:06:54]:
I've met with them. I've talked to them. They come back and say, well, we're meeting all the government regulations, so... We're good. We checked all the checkboxes and that's good.

Leo Laporte [00:07:04]:
Same thing telecoms have been saying after the Salt Typhoon hack. The Chinese are still in our telecom system, but they're unwilling to fix it.

Dan O'Dowd [00:07:12]:
Well, they're unwilling, they don't know what to do. Yeah.

Jeff Jarvis [00:07:15]:
The friend of the show and our theme song, Craig Newmark, has been warning about cybersecurity for some time. And come the recent water hacks, has been saying, I told you so.

Leo Laporte [00:07:26]:
Water systems in several, many states actually have been hacked now.

Jeff Jarvis [00:07:30]:
So part of what Craig has argued is that the communication is such that the populace as a whole isn't aware of the threat, isn't sufficiently aware of the threat to demand political action. Is that part of the reason that you exist, is to make people aware so that the pressure can be put on to make these investments?

Dan O'Dowd [00:07:52]:
Exactly. We've tried. We've gone to the regulators. We've asked them to do it. It's really hard to get them to move on anything, especially today. And hospitals and pipelines, oil and gas pipelines, are critical. Without them, you can't get oil and gas from Texas and Louisiana out to California and New York. You're going to run out of gas.

Dan O'Dowd [00:08:19]:
With no gas, you can't go to work. You can't do anything. So, those are also— they're also critical infrastructure that aren't— Quite as catastrophic as the power grid. But the industry doesn't want to do it. I mean, I understand it's going to cost a lot of money to redo all this software and do it right. It's going to take time and money and effort. But from their perspective, where's the payoff?

Jeff Jarvis [00:08:41]:
Is it all a software problem, or is there also a hardware issue?

Dan O'Dowd [00:08:45]:
I think all the hardware issues can be overcome with software. They are not the fundamental problem. The fundamental problem is that we use software that was downloaded off the internet, or you got it from Microsoft. When those guys were writing that software, there was not a thing in front of their faces saying, if you make a mistake, people are gonna die. That's not what they were told. Mark Zuckerberg wrote on the wall, apparently literally, I mean, as far as I know, it's not apocryphal, it's a real story. He wrote on the wall, move fast and break things.

Jeff Jarvis [00:09:16]:
Oh yeah, they had those signs all over Facebook headquarters.

Dan O'Dowd [00:09:18]:
What does that mean? It means write the software as fast as you can and don't worry if it doesn't really doesn't work all the time, if it fails some of the time, because that's how you become a trillionaire. You just gotta get it out first and best and push it out, and you fix the bugs later. That works if you're doing desktop software, it works for your phone, works for all sorts of things. It doesn't work when everybody dies when it fails.

Leo Laporte [00:09:43]:
What's the secret to making software reliable? I mean, is it a secret?

Dan O'Dowd [00:09:49]:
Nominally, we do talk a lot about it. One is what I said, absolute care. You have to review everything many, many, many times with people. You have to write— we write millions and millions of test programs to test our software out.

Leo Laporte [00:10:09]:
Is there a design philosophy you use? I mean, is it red, green, blue test-driven design? Is there some methodology that you use that helps it be— and what languages do you use to help it be more reliable?

Dan O'Dowd [00:10:21]:
So we have a methodology. It's called the DAWN methodology. I've been working on this for 50 years. I started in college and was interested in basically the problem. How do we write software that never fails and can't be hacked? What is the issue? And there are a lot of issues. There are things people do. There are programming constructs that people use and ways they're taught to do things which are not reliable. They are just fundamentally—

Jeff Jarvis [00:10:47]:
They're flawed.

Dan O'Dowd [00:10:50]:
high risk, and there are alternatives. And I've been developing all those alternatives year after year after year. What about this kind of bug? What about that kind of bug? How do you write code that doesn't have the bugs? If the bugs are inevitable, how do I have fewer of them? If they are— if they do happen, can I make them have the least effect, you know, the negative effect on the system? Mitigations, direct outlawing of certain practices, which I've found to be highly dangerous.

Leo Laporte [00:11:23]:
Don't use strcpy, use strncpy, things like that?

Dan O'Dowd [00:11:26]:
Yes, that's— we go farther. We don't allow any memcpy. None of the strcpy, none of the memcpy are allowed. Actually, that's not true. You can use a memcpy, but if you do, you will be called in for a meeting. I will be there along with 5 or 6 other people, and we're gonna say, How do you know this didn't screw something up?

Leo Laporte [00:11:46]:
That's fantastic.

Dan O'Dowd [00:11:46]:
You have to demonstrate that you didn't screw it up, and also that it was necessary. I'll suggest another way. Other people say, how can we get rid of this thing? But do it right. But there are cases where you have to do it. You cannot remove them, but then you can apply 5 or 6 people who've been doing this for 20 years, and you have to defend it. You have to defend that use. And sometimes we let it go through because it is the only way to do it.

Leo Laporte [00:12:14]:
Some have said that the key is having type-safe languages, the choice of languages. Do you care about that? Is C++ a problem?

Dan O'Dowd [00:12:24]:
C++ is a problem. We write in a subset of C. We use C, but we've again limited—

Jeff Jarvis [00:12:31]:
Safe C?

Dan O'Dowd [00:12:32]:
Yeah, it doesn't have an actual name. It's our own set of rules in which we ban certain bad practices. But that is, you know, that's where we're going. Ultimately, there should be a language where— well, we're actually working on that right now, which is the power of C, but everything is well-defined precisely. In C, there's many things that are not defined. I write a program, you aren't— the language says if you do this, we can't tell you what's gonna happen.

Paris Martineau [00:13:03]:
Yeah.

Dan O'Dowd [00:13:03]:
It's really hard to write software that doesn't fail if it they don't know what's actually going to happen. We ban that. Just like, no way. Everything is defined. Everything absolutely is the way it is. It is unambiguous and clear.

Leo Laporte [00:13:18]:
No race conditions, no things like that.

Dan O'Dowd [00:13:22]:
Race conditions are difficult. It's a different problem, and we actually have a solution for it. Amazingly, we have a solution. A race problem is when the software has a 2 things are going on at the same time, and they're both— typically they're both computing an answer, and one of them— and they're very close. They're just— it's a race which one gets there first. And one of them stores their value, and the other one stores their value, wiping out the first one. But maybe one in a million times something happens to change the order of computation. Something else happens a little late.

Dan O'Dowd [00:13:53]:
Something gets delayed a little bit. And the guy who always won before loses one time. and the wrong value gets stored. And then the program goes on and may run for a long time, and then it fails and crashes. How do you do that? Well, we actually came up with a solution for how to find race conditions, even ones that have never happened. So some—

Leo Laporte [00:14:11]:
Nice.

Dan O'Dowd [00:14:12]:
Typically it has to happen in your software, and then you say, well, what just happened? I don't understand. You run it again and again and again, seeing if you can find ways to make it happen more frequently. But if the race condition only goes bad in cases you haven't tested, like only if it's in Alaska, it's below 50 below zero or some unusual condition, if it only happens in that condition, no amount of testing in a lab is going to find it. None. It's impossible. But if you have this mechanism of dealing with it, it's a little bit complicated.

Jeff Jarvis [00:14:50]:
Right.

Dan O'Dowd [00:14:50]:
We can detect even the race conditions that didn't happen by looking at the logs of the program. We run the program, we keep logs on it, and we can detect when there are stores that are too close to each other that you can't know which one went first and produced a different— and one guy produces the wrong answer. 999,999 times out of a million, it does the right thing. One in a million times, it does the wrong thing.

Leo Laporte [00:15:16]:
Those are the hardest ones to catch.

Dan O'Dowd [00:15:17]:
Those bugs are a nightmare.

Leo Laporte [00:15:19]:
So Department of Defense did try this with Ada, which didn't take off. Why did Ada fail? Or should we have something like Ada?

Dan O'Dowd [00:15:29]:
That is a good question. We have supported Ada. We did it for quite a long time. It is a better language, is definitely a better language, and definitely well, much more well-defined. It is a little bit complicated to learn. But it really was a break. It's the same sort of issue. The defense contractors said, one, it's hard to find programmers who know the language because they didn't teach it in school or anything.

Dan O'Dowd [00:15:55]:
Everybody knows C because everybody teaches it in school in one way. And so slowly, Ada, even in the military projects where it was designed for, it slowly got pushed aside by 20 years of defense contractors saying, it's better and cheaper to use C because we can hire people right out of school. They're already trained in this language. And so Ada pretty much lost out over a long, long period of time. It just got pushed to the side till now, virtually nothing.

Leo Laporte [00:16:25]:
That's one of the things you do with the DAWN methodology and the DAWN project. You get the best students from computer science programs and universities all over, I presume all over the world, and then You call them the special forces of software. You teach them the right way to write code.

Dan O'Dowd [00:16:45]:
Right.

Jeff Jarvis [00:16:46]:
To be—

Dan O'Dowd [00:16:47]:
we use that, you know, somebody has to write a lot of code, but some code's more important than others. And we write the stuff, you know, we handle the stuff that's critical, that underlies the security and reliability of the whole system. And that's our specialty. That's how we specialize.

Leo Laporte [00:17:05]:
Is it your review system that makes it so good? You're debugging harnesses? I mean, what's the secret? Come on.

Dan O'Dowd [00:17:13]:
All of the above.

Leo Laporte [00:17:15]:
Okay.

Dan O'Dowd [00:17:17]:
We have literally 50 different approaches. That's one of the problems. Many people don't. They have this small set of tools, and with their small set of tools, they can do a certain amount of work. We've systematically built tools, looked at problems. Why was this so difficult? What went wrong? How can we prevent this? I don't believe anybody else does that. I don't believe you'll find a single person in the world who does that thing, which is analyze, look at problems, analyze what went wrong, find a way to reduce that error, understand what went wrong, find a way to reduce the error, find a way to find the error when it occurs. That's what we do.

Dan O'Dowd [00:17:59]:
Everybody else doesn't do that. They're busy cranking out more code. We're still cranking out code, But we're looking at what the problems are. I have, you know, we find bugs in— we have a multi-core operating system and we find bugs in minutes, and it would take other people days, weeks, maybe never. But we know everything that's going on in the program. We've got a complete understanding of how the program works so that we can detect almost immediately when it deviates and does the wrong thing.

Leo Laporte [00:18:33]:
I take it you're not a fan of vibe coding.

Paris Martineau [00:18:36]:
Yeah, I was gonna say, how has like the advent of AI changed this both from a coding perspective and from a threat perspective for you guys?

Dan O'Dowd [00:18:45]:
On 2 different— well, on coding, we don't, you know, we don't do AI. It is too erratic to depend on for this kind of coding. It just has to be right every single time. It can't ever make a mistake. On the other hand, the opposite, which is people doing it, I mean, you have these new mythos and all this other stuff that's coming out, but you have to understand what they're doing. They're finding existing vulnerabilities in these programs that are out there. They're not adding things or changing things. They're looking at the programs and looking for bad programming practices, let's call it that, dangerous programming practices.

Dan O'Dowd [00:19:29]:
and to look at the code. They're just sniffing them out, and there have been reports that they found 20 bugs. But there's a National Vulnerability Database. There's 48,000 bugs— you know, not bugs— 48,000 security vulnerabilities were reported by humans to the National Vulnerability Database in the last year. The Mythos has submitted 20 or 30 or 50 or something like that. So it's— yeah, I suppose it finds some, but humans are still way better at finding them. But fundamentally, all it is is finding the bugs that are already there. I asked the question, what's the first secret to making some secure software? Don't download 1,000 critical software vulnerabilities into your product.

Leo Laporte [00:20:21]:
We've seen so many supply chain attacks on npm and PyPI and so forth, and so much software relies on these libraries without even thinking about it. Somebody else's code. Yeah.

Dan O'Dowd [00:20:33]:
And note, the people who wrote it, remember, they did not have— they were not told by their bosses or put a placard on the wall that says, if you screw up, people might die, so be careful. They were told, move fast and break things. Which is why we got that software, because they moved faster than the slow people who were trying to do it right. And then when somebody else came along and said, I want to start on something, what do I start on? Well, I should start with something so I don't have to start at zero. I want to get a leg up, so I start with some software I download off the internet. But those pieces of software, or Windows, have thousands of security vulnerabilities in them. That's the first thing you did. You built your secure system.

Dan O'Dowd [00:21:14]:
The first thing you did is import thousands and thousands of software vulnerabilities. into it because they're there and haven't been found yet. Next year they'll find thousands more, and the year after that they'll find thousands more. That's why it's not safe. And it's irrespective of whether AI is improving this and making it faster. They're not finding— if they could find 98% of the bugs, that would be important. Maybe even 50% of the bugs, that would be important. But at the moment, they're finding some fraction of a percent of the bugs.

Dan O'Dowd [00:21:43]:
We need to not use that software, or we need to fix that software. Those are 2 choices. They're both viable choices, but nobody wants to do it because they take time and cost money and cause your product to come out later. And that's anathema, right? In Silicon Valley, you just shot yourself in the foot, or worse. You don't want to do that. That's not the right strategy. But we make it different. We have to say, for a power grid, for hospital for self-driving cars, same problem.

Dan O'Dowd [00:22:17]:
We're building self-driving cars and we're hooking them up to the internet. So what's somebody gonna do? They come in, they— and they all run the same software because we all have this great new over-the-air updates, you know, regular over-the-air updates. They all run the same software. Somebody finds a bug in that software, lets them take control of the self-driving car, accelerate every car to 100 miles an hour, Turn slightly left into oncoming traffic until you run into 10 cars and smash them at 100 miles an hour.

Jeff Jarvis [00:22:46]:
But what you're finding is even without a hack, the software is no good.

Dan O'Dowd [00:22:51]:
Right. Well, there's lots of bugs in Tesla. I mean, let's be clear here. This is Tesla and not, say, Waymo. Waymo is vastly more reliable and capable than Tesla. The cars are just They just have so many bugs, it's just ridiculous. It goes through a do not enter sign. It's got a great old do not— big do not enter sign.

Dan O'Dowd [00:23:13]:
We put them on both sides of the road just to make sure it doesn't miss them. It goes right between them. Well, what's beyond the do not enter sign? Probably it's because they tore up the road or they dug up a big old hole or maybe something— or maybe there was a crash on the road. Why do you put up do not enter signs? It says don't go there. 'Cause it's bad. It just drives right past them. It says road closed. Great plan.

Dan O'Dowd [00:23:36]:
It just keeps going. It gets a school bus. Oh, here's one of our videos we're seeing right here.

Leo Laporte [00:23:43]:
We're talking to Dan O'Dowd. He's the founder of the Dawn Project, self-funded. You probably saw, at least we did in California, some of Dan's ads going after Tesla. Dan certainly has the credentials to say what he's been saying. He's been writing software for some of the most mission-critical applications in our history— military nuclear bombers, the planes that you fly in, the cars perhaps that you drive. I know BMW's used your software for their self-driving vehicles. So you're not against self-driving vehicles?

Dan O'Dowd [00:24:20]:
No, they just have to be well tested. And I really think Waymo's done a good job here. They've been very careful. They used their own employee-trained test drivers for years and years. Tesla just handed it out to the public, a beta product. They called it a beta product.

Jeff Jarvis [00:24:37]:
And they handed it out to ordinary people, put their kids in the back seat. And the sensors were beta tested for this shop.

Paris Martineau [00:24:44]:
I mean, I experienced this when I was in Washington, DC last year. I was picked up by an Uber with a human sitting in the front. And he starts asking me, he's like, How do you feel about Elon Musk? What do you feel about Tesla? And he reveals 5 minutes into the road, he's like, yeah, I've had full self-driving on the entire time. I'm never touching this car wheel. And I'm like, sir, let me out. I didn't sign up for this.

Dan O'Dowd [00:25:06]:
Right. And neither did the people on the side of the road. You know, a couple was— they were just hanging out stargazing or something on their car. And at 60 miles an hour, a Tesla runs the stop sign, runs off the road, smashes into them, kills the woman, leaves the guy injured for years. And Tesla— and in that case, Tesla had— they subpoenaed the data, what was happening, because the car records everything. It's got all the cameras and all this data. And they subpoenaed, said, send us the data so we can see what was happening. They said, well, we looked, we couldn't find it.

Dan O'Dowd [00:25:46]:
They said, well, look again. They said, we can't find it. We can't find it. We can't find it. So the plaintiffs hired a hacker to come into the car, and guess what? He found it. It was a deleted file. They deleted the file that had all of the critical data that showed that they were totally in the wrong, right, in this case. They deleted all— they deleted the data.

Dan O'Dowd [00:26:10]:
Now, as many people know, if you delete a file, your file's not gone. At least not to a hacker.

Leo Laporte [00:26:18]:
This is a video you guys made. I should point out there is no child involved in this. This was a test.

Jeff Jarvis [00:26:24]:
Yeah, my reaction to it is just visceral.

Leo Laporte [00:26:26]:
Yeah.

Dan O'Dowd [00:26:27]:
Yeah, we've got a little mannequin kid there. It runs right past the school bus. It hardly slows down, crashes in it. This happened in North Carolina after we ran— after we warned about this and ran a Super Bowl commercial showing that scenario. A kid steps off the bus in North Carolina and a Tesla on self-driving, the kid walks across the street and it hits him and put him in the hospital for months, broke his neck and his leg, and I can't remember what else. I think he was on a ventilator for a time. Just a kid stepping off the school bus.

Jeff Jarvis [00:27:02]:
Yes.

Dan O'Dowd [00:27:02]:
With the school bus has those red lights and the flashing lights, you stop the car, your car stops. And you wait for the kids to get off and— because they'll run across the street. You know, those kids, they will run across the street. They will dart in and out of traffic. That's why we say you have to stop.

Jeff Jarvis [00:27:20]:
Yeah.

Dan O'Dowd [00:27:20]:
Let them dart in and out of traffic and then you can go again when the bus goes. How can they put on the road a product for— with millions of people blowing past school buses every day and not slowing down? I just cannot understand. How this hasn't worked yet. I mean, I put it out there. I have given that to everybody. And it's still there. We just tested it a few days ago. Just said, well, they bring out new releases all the time.

Dan O'Dowd [00:27:45]:
We can bring out the new release and test it. We get the school bus, we put it on the road. Yep, still runs, just blows past the school bus.

Jeff Jarvis [00:27:54]:
Dan, I've been screaming about this for quite some time. And I'm really glad you're doing this. I can't believe that it's not regulated. That's one piece of this. You're talking about the software quality. That's the other. But my question, my first question to you is, without if you look at a Waymo, it has tons of sensors on it as well. It has data coming in that can tell it more.

Jeff Jarvis [00:28:14]:
The hubris of Musk and the Tesla is oh, I don't need all that. I've got a few cameras and that's that. Is it ever possible to make a Tesla as built with the best software in the world self driving, or is that impossible because the sensors just simply aren't there to? Let it know its context.

Dan O'Dowd [00:28:30]:
There is a severe problem. I mean, probably fatal. When you're gonna do a right or left turn, you're on a road, you've got a stop sign, and the other way, the cross traffic has no stop sign, right? No stop sign, no stoplight, and they're just allowed to zip past there at 40, 50 miles an hour, and you're sitting at this stop sign waiting for your chance to get in, right? To merge into traffic. So you're sitting there, and to do that, especially if it's a high-speed you know, street, maybe it's going 45, 50 miles an hour, you gotta look down. You gotta look down the road both ways to see, do I have a big enough gap that I can get out of here, get into the lane, and get moving without causing a crash or without causing at least distress among the other drivers? So that's what you do. It's pretty simple. But sometimes you come up to an intersection and you say, I can't see far enough, right? There's a truck or a bush or something blocking my view. So what do you do? Well, you creep the car up to see if you can get there.

Dan O'Dowd [00:29:28]:
Maybe you still can't see it. What do you do? You lean forward in your seat as far as you can, and you look to the right and the left, and you say, oh, okay, I can see it's safe now, and you go. Here's the problem. The camera that is on the car— there's a camera on each side of the car whose job is to look that way, to look down the road left and right. The camera is mounted Behind your head. It's on the thing where the door, you know, the door latch, the door connects to it.

Jeff Jarvis [00:29:58]:
Oh.

Dan O'Dowd [00:29:58]:
It's behind the head on that pillar. It's called the B-pillar. So it's back here. But you were sitting in the car with your eyes where I am right here, and I couldn't see far enough down the road. So I had to lean forward so I could see far enough down the road. It can't lean forward. It's Nailed in. It's glued into the car behind you.

Dan O'Dowd [00:30:22]:
So if sitting in the normal position, it isn't safe, you can't determine whether it's safe or not, it can't determine it because it's even farther back. There's— I don't— there's no fix for this. It's really stupid. You must have the cameras at the front of the car that are looking down the road have to be placed ahead of where a human's head would be because that's how they designed intersections. It's not an accident that when you streak all the way forward you can see. It's because there's some road engineer or city official or something who's deciding where, you know, where you can park cars and where the bushes can be to make sure that humans can see enough. And if they can't, they'll call up the department or there'll be a bunch of accidents at that intersection and somebody will fix it eventually. So the intersections are designed for people.

Dan O'Dowd [00:31:13]:
And being able— for humans to be able to see past these problems. But this poor guy that— we have an advantage. We can lean forward in our seat, but the cameras can't, and they're in the wrong place. And so what it does is it often just sort of pushes out into traffic to get far enough out that it can see, but it's already going into the oncoming lane. And if you do very unfortunate timing, There could be somebody coming from the left really fast, 50 miles an hour, and you can't see them. And you put your nose out into the road so you can see, you'll get hit. A human driver can do that intersection every time. It just can't.

Dan O'Dowd [00:31:54]:
They designed it wrong. They put the camera in the wrong place. There are a bunch of issues like that.

Leo Laporte [00:31:59]:
As bad as that is, the damage is somewhat limited compared to What you talked about earlier, the problems of the electric grid. Why have you focused on Tesla instead of what is clearly an imminent— clear and imminent danger with our electrical grid or our water systems or all sorts of infrastructure?

Dan O'Dowd [00:32:18]:
Well, we do. I mean, we do focus on the power grid. It is part of it. But also, one of the problems that I have is people will say, well, it's not— you know, I say it's not secure. How am I supposed to demonstrate. They'll say it is secure, and I'll say it's not. And yeah, yeah, yeah, how do we find out? Well, I don't know. I could—

Leo Laporte [00:32:40]:
The hard way, I guess.

Dan O'Dowd [00:32:42]:
I could get people to hack into the power grid, but that's against the law. If I hacked into there, or my own people hacked in there, I could be arrested for hacking in their system. So I can't do that, but I can buy a Tesla, and I can show that it does terrible things. It's killed 60 people. The Pinto deaths. Remember the Pintos back in the '70s?

Leo Laporte [00:33:05]:
Safe at any speed, the Corvairs, the Pintos with the exploding gas tanks in the rear.

Dan O'Dowd [00:33:11]:
Yeah, the total number of fatalities are less than the number with Tesla, right? Already today, I think the Pinto was 27 deaths. We're at 60 now that we know about. We don't even know about all of them.

Jeff Jarvis [00:33:25]:
So what do Elon Musk and the White House have to say to you?

Dan O'Dowd [00:33:30]:
I don't think the White House has anything to say. Elon Musk, I— let's see, I'm batshit crazy, totally jealous of his, of his success because I'm such a failure. Really mentally unstable asshole.

Leo Laporte [00:33:48]:
We get the idea. Yeah, yeah, yeah. So people can go to Don Go to thedawnproject.com to find out more about the Dawn Project. Do you, uh, what would you like us to do as citizens? How, how do we pursue this?

Dan O'Dowd [00:34:05]:
Well, we are actively doing this. I— we need to get the word out. That's what it is. I've given up on the government, you know, on convincing them to do this, at least for now. And I can't just blame Trump. We started this before Trump. We started this in Biden, and we got just as shut down as we are today.

Paris Martineau [00:34:20]:
Yep.

Dan O'Dowd [00:34:20]:
So I can't say that I'm hopeful that we could get a change. I mean, maybe we could, but you can't count on that. So we have to get people to understand the catastrophic problem. I've got another video. I've got the head of the FBI saying, this is a generational— the threat of our generation. I've got the NSA director saying that it's going to happen. Some catastrophic thing's gonna happen. I forgot the head of the CISA, the— I forget what their exact name is, but one of the cybersecurity agencies, Jen Esterly, was in charge of it.

Leo Laporte [00:34:58]:
Yeah, Jen, yeah.

Dan O'Dowd [00:34:59]:
And saying that— she did the whole thing. She says, well, software developers have been doing this move fast and break things, and they aren't— and they're not taking responsibility for their code, and they're not being held responsible. for the errors and bugs in their code, so they just keep doing it. I mean, I have a video which I think is pretty compelling that just shows that all of these people know, and they're not hiding it. They went in front of the TV cameras and said, we're screwed, this is terrible, we are in terrible trouble, and we must do something about it now.

Leo Laporte [00:35:32]:
Yeah.

Dan O'Dowd [00:35:33]:
And still nothing's happening.

Leo Laporte [00:35:34]:
Yeah.

Dan O'Dowd [00:35:35]:
Still nobody's doing anything. So at my point, I gave up on that. I just gotta get people to understand the danger, and when they realize it, they'll say, yeah, let's— I mean, let's spend a little harder. It's gonna cost more. I'm not saying it's not gonna cost more. It's gonna cost more. It's gonna take time, but we can do it right. We know how to develop software for weapons of mass destruction.

Dan O'Dowd [00:35:56]:
We've been doing that. Essentially, you turn— by connecting a power grid up to the internet, you've turned it into a weapon of mass destruction. Somebody who can hack into that system and not just shut it off for an hour a day. That's not a big problem. It's like weeks or a month or more than that, and that they can't turn it back on. You short out the transformers or whatever you do to destroy them. You take the generators and you push them up above max, right?

Jeff Jarvis [00:36:27]:
Mm-hmm.

Dan O'Dowd [00:36:27]:
You push them into the red zone and just keep it there until what happens? It'll catch on fire. Or it'll melt down, or the blades will start flying out of the device, something like that. And you can't repair it. I just read a story, New York Times, I think it was today. It's quite an interesting story. They were focused on power grid going down because of mechanic— on people shooting up power grid stuff, not cyberattacks.

Jeff Jarvis [00:36:51]:
Yeah.

Dan O'Dowd [00:36:51]:
But if you just change the article to cyberattacks as opposed to physical attacks on our infrastructure, It pretty much is there and it goes through, you know, like the water runs out after a few weeks and then, you know, then the militia takes over. You know, it's there. You have a huge problem that any one of these systems is there. So you— if you turn it off, people— not turn it off, you disable it. But it's 2 years to get replacement parts for these things.

Jeff Jarvis [00:37:23]:
Right.

Dan O'Dowd [00:37:23]:
There's a supply chain problem. And that's right now, except it's getting worse because all the hyperscalers have jumped in line and paid extra money to put themselves at the front of the line and collect all the— so they get all the parts they need to build their new power stations.

Leo Laporte [00:37:41]:
There's a proof of concept in this. We did it to the Iranians with Stuxnet. We spun their centrifuges up. their SCADA devices up till they destroyed themselves.

Jeff Jarvis [00:37:54]:
Right.

Leo Laporte [00:37:54]:
And this is just coming the other direction. Dan, I wanna thank you for what you're doing. Dan O'Dowd is the founder of the Dawn Project, self-funded. This is his mission, dawnproject.com, if you wanna learn more about it. I guess you could write your member of Congress. The problem is really at this point the incentives are wrong. The incentives are financial And save money and not spend money to protect them.

Dan O'Dowd [00:38:20]:
Go to dawnproject.com and watch it. We have 100, we've made 100 videos of our own, of all sorts. We've got 1,000 videos collected of like Teslas running stop signs and going through do not enter sign and whatever. Every single day we put up a video. It's our post every day, our daily post of of a self-driving Tesla doing something insane, running the wrong way on the street, running over a bicycle, just everything. It's just amazing. You'll get tired of watching that eventually.

Leo Laporte [00:38:56]:
Well, I'm certainly not going to buy a Tesla, that's for sure.

Dan O'Dowd [00:38:58]:
It's another alternative to getting them to do something about it.

Leo Laporte [00:39:03]:
Yeah. Maybe I'll buy an i9 BMW. That might be the right solution. 'Cause I know your software's running in that. Thank you, Dan. I really appreciate your time and the work you're doing. It's very important. The Dawn Project is at dawnproject.com.

Leo Laporte [00:39:17]:
Appreciate it.

Dan O'Dowd [00:39:19]:
Thank you, Dan. Great.

Paris Martineau [00:39:20]:
Thanks.

Dan O'Dowd [00:39:20]:
Good to talk to you.

Leo Laporte [00:39:21]:
We'll have more on intelligent machines right after this. Well, hello there.

Jeff Jarvis [00:39:27]:
Hello.

Leo Laporte [00:39:27]:
Jeff Jarvis and Paris Martineau. That was an uplifting conversation.

Paris Martineau [00:39:33]:
It always is.

Leo Laporte [00:39:34]:
I hope we survive the next, uh, Couple of decades, that's all I can say.

Jeff Jarvis [00:39:40]:
Paris is hoping for a little longer than that.

Leo Laporte [00:39:42]:
Yeah, yeah, you and I, we don't need much.

Paris Martineau [00:39:44]:
I mean, I don't know, I also kind of hope I survive the next couple of decades. People often ask me about incredibly long-term things and I'm like, I don't know if we're all going to be around in 30.

Leo Laporte [00:39:54]:
10 years would be a start. Okay.

Paris Martineau [00:39:56]:
10 years would be great. I'd be happy for 10.

Leo Laporte [00:39:59]:
Well, you know, you should do, uh, what I do, which is just assume every day is your last. And when you get up, you go, yay, I made it.

Paris Martineau [00:40:07]:
I think we've talked on this show before about my use of the WeCroak app, which 5 times a day gives me a notification. Don't forget you're going to die. And I think that's really important.

Leo Laporte [00:40:18]:
Yeah. I was trying to remember exactly that app to tell somebody about. And I remembered the emperors of Rome and they were getting their chariots.

Paris Martineau [00:40:27]:
Never forget the emperors of Rome.

Leo Laporte [00:40:29]:
I think about them all the time. As you know, they're riding their chariots. They're getting their laurels, their honors for their successes and victories in battle, but they always had somebody standing behind them going, memento mori, remember you will die.

Paris Martineau [00:40:44]:
Okay, I actually do think that a lot of our problems would be fixed if every billionaire had just one guy who had to follow him everywhere and be like, memento mori.

Leo Laporte [00:40:55]:
It's probably apocryphal, but that's a great story. Uh, Anthropic's making money. Well, Ed Zittrain may disagree, but Yeah, sort of. Uh, revenue, according to Bloomberg News, jumped to $11.5 billion in second quarter. That's not making money, that's revenue.

Paris Martineau [00:41:13]:
Yeah, yeah. How much did they spend in that quarter?

Leo Laporte [00:41:17]:
We don't know, they're not public yet. Although—

Paris Martineau [00:41:20]:
Interesting how that number gets leaked but not the bad one.

Leo Laporte [00:41:25]:
Revenue surged 14 times in the second quarter from year over year. So they, you know, and I'm not surprised. They have a great reputation. Everybody uses them for coding. But yeah, we don't know what it cost for them to generate $11.5 billion in revenue. But we will if they go public. And the IPOs now I'm hearing for both OpenAI and Anthropic are somewhat imminent, maybe in a couple of months.

Jeff Jarvis [00:41:52]:
And the Anthropic—

Paris Martineau [00:41:53]:
Well, opening eyes. OpenAI's might take longer because at the same time, we also got news leaked that OpenAI's revenue, I'm forgetting what it was on a quarter-to-quarter basis, but it was below expectations. Like it hadn't grown as fast as they thought.

Leo Laporte [00:42:09]:
I think, you know, was this another Ed? Because he seems—

Paris Martineau [00:42:12]:
I mean, no, I think it was a new— I think it was the Wall Street Journal.

Leo Laporte [00:42:16]:
Okay.

Jeff Jarvis [00:42:17]:
Yeah. Meanwhile, but the other leak from the Anthropic discussions is an IPO headed at $2 trillion.

Leo Laporte [00:42:22]:
$2 trillion. Yeah, I remember when it was a big deal when a company was valued by the market at $1 trillion. Although Apple's approaching $5 trillion, Nvidia's very close behind, over $4 trillion. So I guess, you know, $2 trillion is not that much. Well, you think, you think people will buy Anthropic stock?

Paris Martineau [00:42:41]:
Yeah. Oh yeah, probably.

Jeff Jarvis [00:42:43]:
Yeah, yeah, they bought SpaceX stock.

Leo Laporte [00:42:45]:
They bought SpaceX, which is a lot longer.

Paris Martineau [00:42:47]:
So you just said that was $11 $7.5 billion. Is that what— OpenAI, this is a link from, or this is a report from my former colleague, Berber Jin, which I just dropped in the chat. OpenAI told its investors that it had $6.7 billion in revenue in the 3 months ended in June, which is kind of half of Anthropic, which is a little surprising to me.

Leo Laporte [00:43:12]:
There have been a number of departures at OpenAI. I've seen people say, What's going on over here?

Paris Martineau [00:43:17]:
Okay, a couple of those were medical, but yes, the ones that were not medical were unfortunate.

Leo Laporte [00:43:21]:
Well, Fiji Simo was medical.

Dan O'Dowd [00:43:23]:
Yeah.

Leo Laporte [00:43:23]:
But Denise Dresser, who is their revenue chief, has been there a mere 8 months. She just left.

Paris Martineau [00:43:32]:
Not good when your revenue chief leaves quickly in the run-up to an IPO.

Leo Laporte [00:43:39]:
You get the feeling that maybe she said, Oh boy. On the other hand, I don't give a lot of stock to people coming and going at these AI companies because there's so much money flowing and there's, you know, they're being enticed all the time by the competition. Come on over here. Come on over here. We'll write you a check for $1 billion that, you know, could well be other reasons. Dario Amodei, the CEO of Anthropic, posted for the first time in a long time, if ever, on X. He says, I don't usually spend much time on social media, but damn, we're getting taken a beating in the public's domain. So I wanted to engage here because— no, he didn't say that.

Leo Laporte [00:44:23]:
I want to engage here because it really brings out the heart of an important conversation. I think this is clearly a response to Mark Zuckerberg's essay about how you don't want AI to be concentrated in the hands of a few. He says, I think that's a false choice. And he says, if I can summarize, basically, you have to have this concentration of power because it's so expensive. Yeah.

Paris Martineau [00:44:51]:
Very funny to say that as one of the 3, like 2 companies that are concentrating that power.

Leo Laporte [00:44:57]:
Right. He says, and he admits a crude analogy, is the formal court system can sometimes feel stuffy and elitist, but it does a much better job of defending the rights of vulnerable individuals than the alternative: mob justice. So—

Paris Martineau [00:45:13]:
I love that he's like, the options are me and one other company get all the money in the world, or mob justice.

Leo Laporte [00:45:20]:
Or mob justice!

Paris Martineau [00:45:22]:
Which one do you want?

Leo Laporte [00:45:24]:
He says structurally, AI is a technology that tends to concentrate power. Not because of regulation, but open weights do help some with this, but are nowhere near a sufficient solution. They simply shift the concentration somewhat to those with the most compute and chips, which are— so that's the Frontier Labs plus some hardware providers. Think Nvidia. Yeah. By contrast, I think the right rules of the road can simultaneously address AI's cyber-bio alignment risks, something Anthropic was founded to address. Institutionally, B, institutionally constrain the power of the frontier AI companies. Regulate us.

Leo Laporte [00:46:05]:
Come on, regulate us. And C, leave room a little bit for open weight models while also addressing the specific risks they bring. He's very focused on risks. And I understand why.

Paris Martineau [00:46:20]:
Well, but—

Leo Laporte [00:46:20]:
He was the one who said Mythos, we can't, you know, it's so bad, so powerful, we can't release it. Now OpenAI is saying, by the way, we better not release AstraZeneca.

Jeff Jarvis [00:46:29]:
Wait a second here. Wait a second here. When they say that they're focused on safety and risks, you have to take that with a grain of salt the size of Utah, because what they're talking about on the one hand is doomer, on the other hand is marketing. And they're trying to act as if though they're the safe ones, so regulate everybody else and knock them down, and we're the only good ones here. No, no, I don't buy that. I don't buy that at all.

Leo Laporte [00:46:52]:
What if he said— I'll give you the counterargument. What if he What if we assume that these open weight models from China, which I use heavily—

Jeff Jarvis [00:47:01]:
You traitor, you!

Leo Laporte [00:47:02]:
I'm a traitor. If they get as good as Mythos or Astra and they're able to come up with bioweapons, nuclear weapons, hacks that are impossible to stop, there's nobody can stop those because they've been released. They're out in the wild. I can run them here. Isn't that a risk that we should worry about? Well, if I were a bad guy, I'd absolutely be, you know, downloading these open weight models and throwing as much power at them as possible, trying to create new hacks and phishing emails and stuff. Doesn't the fact that these are open and widely distributed, isn't that a bigger risk than a nice friendly company that's—

Jeff Jarvis [00:47:45]:
That is a black box that you don't know how it works and it's causing problems and you don't know—

Leo Laporte [00:47:49]:
It's placed down on in San Francisco on Market Street.

Jeff Jarvis [00:47:52]:
Far be it for me to agree with David Sacks, who tweeted in response to this, Dario believes frontier AI is too powerful to distribute. We believe it is too powerful to centralize. Dario appears to believe sincerely that safety and progress are best served by centralizing authority in a marriage of corporate and state power. The weight of human history gives us reason to fear that outcome. He who is friendly with the current power.

Leo Laporte [00:48:16]:
Yeah, yeah, isn't that interesting? Although I think Sachs joined the Trump administration at the behest of his supporters, people like Peter Thiel, to go in there and say no regulation of AI, no regulation of AI. And now that, you know, it's kind of roiling around Washington, maybe we better regulate AI. I think that that's one of the reasons.

Jeff Jarvis [00:48:36]:
Yeah, but they're not saying that because they see actual safety, I don't think. No, they're saying that because of the political stress is out there. People are now, you know, the public opinion in polls against AI is strong, fighting against Meta's perv glasses, against data centers. The vibe is against them.

Leo Laporte [00:48:56]:
Then there's Bruce Schneier, who I deeply respect, like a lot, who wrote an essay for The Guardian with Nathan E. Sanders saying if the markets reject OpenAI and Anthropic, in other words, if there's an AI crash—

Jeff Jarvis [00:49:09]:
Nathan wrote it and he He posted—

Leo Laporte [00:49:12]:
he signed it?

Jeff Jarvis [00:49:13]:
He posted it.

Leo Laporte [00:49:14]:
The U.S. should nationalize them. Uh, it says written with—

Jeff Jarvis [00:49:18]:
Oh, written by, written with. You're right, you're right. Quite right. Sorry.

Leo Laporte [00:49:20]:
Uh, so the— first of all, do we know who Nathan E. Sanders is?

Jeff Jarvis [00:49:24]:
I think he writes a lot of stuff with Bruce.

Leo Laporte [00:49:27]:
Uh, Schneier is a widely respected security guru. I think the world of him. I know, I've met him, I know him, we've had him on shows. He says that if there's a— the evidence suggests the market could reassess that these companies offer nothing of financial value. In that case, perhaps we can return them both to their original purposes, right? OpenAI and Anthropic both were originally kind of— I think Anthropic is a public benefit corporation. They're founded with the idea that, you know, we don't want to make money in this.

Dan O'Dowd [00:50:02]:
Yeah.

Leo Laporte [00:50:03]:
We want to help mankind.

Paris Martineau [00:50:04]:
Not too much.

Leo Laporte [00:50:06]:
Help mankind. He says if these AI companies should fail in the financial markets, the US should nationalize them. Now the model he's using actually—

Paris Martineau [00:50:15]:
I don't know, I mean—

Leo Laporte [00:50:16]:
Convert them into national labs. So we have a really amazing system of national labs which were originally created to develop atomic power and the atomic bomb.

Jeff Jarvis [00:50:28]:
Oh, just that.

Leo Laporte [00:50:29]:
Well, but they are, they have done a lot of amazing—

Jeff Jarvis [00:50:32]:
Brookhaven and Lawrence Livermore.

Leo Laporte [00:50:35]:
Lawrence Livermore, exactly. They're really amazing. And I think they are run by and staffed by conscientious scientists who perhaps could be trusted with AI.

Paris Martineau [00:50:48]:
I do think that there's an argument to be made that there should be a national AI lab if the technology has the juice that its biggest proponents say it does. However, I don't know if that necessarily means that it's like, well, we all just decided to pour so much money into these companies that they can never fail, and they've always got to have all the money they can get and always succeed forever in some form. Like, I think that—

Leo Laporte [00:51:15]:
Well, that's the difference. Okay, so that's the difference between saying too big to fail, so we're gonna, like we did the banks in 2008, pour money into you versus saying, no, no, give it— we're going to just take the technology, put it in a lab.

Dan O'Dowd [00:51:26]:
Or let them fail.

Jeff Jarvis [00:51:27]:
Let them fail.

Leo Laporte [00:51:27]:
Let them fail.

Paris Martineau [00:51:28]:
Let them fail and then—

Jeff Jarvis [00:51:30]:
And disappear.

Paris Martineau [00:51:31]:
Yeah. And then start your own lab and hire your own people.

Leo Laporte [00:51:33]:
That's what he's saying. The US should nationalize them.

Jeff Jarvis [00:51:36]:
But don't nationalize those 2. Let those 2 go.

Dan O'Dowd [00:51:38]:
They are tainted.

Leo Laporte [00:51:39]:
Operated under democratic control that preserve their benefit to the public interest.

Jeff Jarvis [00:51:46]:
But why those? I think what Parris is saying is, why would you rescue any ashes from their fire? Why not just let them die?

Leo Laporte [00:51:56]:
And because they have the best technology.

Paris Martineau [00:51:58]:
Well, I mean, let them die and then buy the scraps out of bankruptcy.

Dan O'Dowd [00:52:06]:
Yeah.

Paris Martineau [00:52:06]:
Yeah.

Leo Laporte [00:52:07]:
It's an interesting question because what is exactly— what is it that makes Anthropic—

Jeff Jarvis [00:52:11]:
considering that it's consolidated, it's commodified, and they leapfrog each other every month?

Leo Laporte [00:52:17]:
Well, first of all, we don't know it's commodified. We don't know if it's the people, if it's the process. We don't really know because they're so opaque.

Jeff Jarvis [00:52:26]:
But Leo, isn't the— aren't the capabilities of the models leapfrogged every month? That there's nothing that stays special, stays ahead? No.

Leo Laporte [00:52:35]:
In fact, that's one of the things Bruce Schneier is saying. And it's one of the challenges that you've got about 30 days before the next one comes out.

Jeff Jarvis [00:52:44]:
And as you show, Using the open weight models, you can do 99.9% of what you want to do. Yeah.

Leo Laporte [00:52:50]:
Well, let's see, this is a good point. So the open weight models are demonstrably not as big, as good, as effective as the closed weight models from the frontier companies. They're much smaller. But it turns out it's not merely the model, it's also the harness. It's the stuff you build around it that makes it good and powerful. And so that's why we just, it's hard to say what the secret sauce is, is all I'm saying. In fact, DeepSeek has released their own harness. A lot of interest in the AI community over the DeepSeek coding harness.

Leo Laporte [00:53:27]:
You know, that, that, it's hard to remember. It was so long ago, like 8 months ago, Anthropic released the Claude Code harness around its Anthropic models, open, at the time it was Opus 4.5, then 4.8. and now Fable and so forth. And that was an eye-opener. People started using Claude Code and suddenly those models, which were mostly just chatbots before that, were really useful tools. And they learned a lot about, well, you have memory, you have skills, you have a whole bunch of infrastructure that makes the model really powerful. That's not that— it's less than a year ago.

Jeff Jarvis [00:54:08]:
Mm-hmm.

Leo Laporte [00:54:09]:
Which is kind of amazing. We're making huge progress, but it's kind of jagged, like the models themselves. It's unpredictable where the next thing's going to happen. I don't— look, part of the problem is, do we trust government these days? We don't. We've been let down by governments.

Jeff Jarvis [00:54:30]:
Right. I would prefer it be— I mean, full disclosure here is Brookhaven is part of Stony Brook where I—

Leo Laporte [00:54:35]:
Brookhaven's amazing. As is Lawrence Livermore.

Jeff Jarvis [00:54:38]:
And so, and I used to cover Lawrence Livermore when I was in San Francisco. They do amazing things. And universities, I would trust universities a lot more to be working on this at that level. And where is, you know, where is the Bell Labs of our day? If you're going to put it in private hands, the ethos of Bell Labs was very different. And that was required because of the regulation. It was because Bell was a monopoly, it's far more complicated.

Leo Laporte [00:55:05]:
They were forced to create Bell Labs. Is that the case or no?

Jeff Jarvis [00:55:09]:
I don't know if they were forced to do that, but I think that they were forced to open things up. They couldn't hold on to certain technologies as proprietary because they were a monopoly.

Leo Laporte [00:55:18]:
They gave away Unix.

Jeff Jarvis [00:55:20]:
Yeah.

Leo Laporte [00:55:23]:
It's an interesting, one of many interesting philosophical discussions. He says, while we believe, this is Bruce Schneier again, that these companies are unsustainable as private firms.

Jeff Jarvis [00:55:37]:
So he's already there. He's already saying—

Leo Laporte [00:55:39]:
He says they can't.

Jeff Jarvis [00:55:40]:
Game's over.

Leo Laporte [00:55:40]:
Because it costs billions. And by the way, Darren's pointing this out and it's true. The model alone, these giant models like Mythos, 10 trillion byte model, cost billions to make, just to make, let alone all the other stuff, the post-training, all the things you do to make that model useful. He says, we believe these companies are unsustainable as private firms, but the timeline remains unclear. The primary investor story is that AI is a race to artificial general intelligence.

Paris Martineau [00:56:12]:
Ugh.

Leo Laporte [00:56:13]:
And yeah, and there you go. Jeff says it all. The bet seems to be that the 2 companies can convince enough people that this outcome will turn them to a profit, go public, and then make their investors and employees rich. before the bubble bursts. It is, it's a race.

Jeff Jarvis [00:56:28]:
Yeah.

Leo Laporte [00:56:28]:
It's the same thing, by the way, with Uber, right? Uber was never a tenable business unless you had self-driving cars. But people invested in it because they thought, well, there is upside. There's a, it may be—

Jeff Jarvis [00:56:41]:
Well, that's a good example because I would, I'll trust Waymo with my life. I will not trust Elon Musk, nor will I trust Uber with my life.

Dan O'Dowd [00:56:49]:
Right.

Jeff Jarvis [00:56:50]:
And so I don't trust Sam Altman with my life or Dario Amadei.

Leo Laporte [00:56:55]:
Right.

Jeff Jarvis [00:56:56]:
And I think that that's becoming very common now. So part of the issue here is that these, these companies may be so tainted in the public perception that, yes, there'll be fools who buy them on the market. They bought SpaceX, but that doesn't mean—

Paris Martineau [00:57:09]:
How's that going for them?

Jeff Jarvis [00:57:10]:
Exactly.

Leo Laporte [00:57:12]:
And SpaceX wasn't just an AI play. I think SpaceX people loved the whole idea of we're going to have asteroid mining. We're gonna have data centers in space. We're gonna have a million people living on Mars. I mean, all of this was sci-fi.

Paris Martineau [00:57:25]:
They're gonna put more Teslas in space, probably.

Leo Laporte [00:57:28]:
It's all sci-fi. But remember, you know, all you have to do to win in the stock market is convince people of the story, the upside. He concludes, but suppose that the bubble bursts. If the US is smart, It will catch the companies as they fall, regardless of what the markets think. To the public, they're too valuable to let die. I think that's true. I agree with you, Jeff. The idea of let's go for AGI is a pipe dream, probably.

Leo Laporte [00:58:02]:
And we're going to get there. If it's possible, we'll get there regardless. And if it's impossible, it doesn't matter because we've already discovered huge value.

Jeff Jarvis [00:58:09]:
So I'm going to agree and disagree with him. I'll agree to the extent that having a national lab of AI, as Paris said too, I think is a good thing. But building it on those 2 companies with the history they already have?

Dan O'Dowd [00:58:20]:
No.

Leo Laporte [00:58:20]:
You're just hastening the burst of the bubble, Jack.

Jeff Jarvis [00:58:23]:
No, I don't want to. I've got a full-on—

Paris Martineau [00:58:25]:
If the bubble's going to burst, it should burst before it gets any bigger.

Jeff Jarvis [00:58:29]:
Better.

Leo Laporte [00:58:29]:
Oh, you're right. Better to burst now than later.

Paris Martineau [00:58:32]:
Yeah, that's how these things work. The bubble— Dude, one of—

Dan O'Dowd [00:58:36]:
oh.

Leo Laporte [00:58:37]:
Here goes the bubble.

Jeff Jarvis [00:58:39]:
Oh, I forgot about this one.

Paris Martineau [00:58:41]:
Me too.

Leo Laporte [00:58:43]:
Ah, robot hands.

Jeff Jarvis [00:58:46]:
On audio, Harris just burst the bubble.

Leo Laporte [00:58:49]:
You look so satisfied in that.

Paris Martineau [00:58:51]:
I really do. I forgot about— for those on audio, that's a really beautiful AI slop video appeared of me looking very malicious while a very peaceful bubble appeared on my face, and I used robot Freddy Krueger hands to burst it. Um, but yes, much like that bubble illustrated. The bubble can only do one of two things, get bigger or burst. It's not gonna like shrink on its own without a large collapse happening. Like, and why would we want a larger bubble to burst?

Leo Laporte [00:59:19]:
Wall Street Journal this week, Inside Big Tech's Frantic Race to Quell the Growing Black Backlash to AI. So one of the big threats at this point is all of those people protesting against data centers. It says tech companies are holding listening sessions, offering guaranteed jobs, and writing big checks To win public support for this.

Jeff Jarvis [00:59:38]:
But the data centers, again, I still contend are a MacGuffin. They are trying to find the tangible way to object to companies they don't like and trust right now, who have too much power in their view.

Dan O'Dowd [00:59:50]:
I think you're right.

Leo Laporte [00:59:50]:
I think it's a gut thing. That's why you can't convince them, well, no, it's going to be fine.

Jeff Jarvis [00:59:55]:
It's really more a visceral reaction to the people who live next to those sites.

Paris Martineau [01:00:00]:
Yeah.

Leo Laporte [01:00:00]:
Well, that I understand, but most people don't live next to those sites. Yeah, I understand that. I wouldn't want one next to me either. I'd protest that.

Paris Martineau [01:00:07]:
I think that people can be outraged on behalf of their neighbors or on behalf of other people who have to then live next to those sites. And I think that it's also a physical manifestation of tech companies who have created products that people find very detestable or unpopular in other ways. And that I think that this has been something that's been simmering in the background a lot with a lot of I saw similar things happening with Amazon fulfillment center, like warehouses being built or distribution centers.

Leo Laporte [01:00:38]:
They got built.

Paris Martineau [01:00:39]:
They got built because—

Leo Laporte [01:00:40]:
And we all get to have, you know, same-day delivery now.

Paris Martineau [01:00:44]:
But the people who live around those have a really hard time in many cases.

Dan O'Dowd [01:00:49]:
Right.

Jeff Jarvis [01:00:50]:
Did I tell you about what happened in Montgomery Township south of me in New Jersey with this?

Paris Martineau [01:00:53]:
No.

Jeff Jarvis [01:00:54]:
So there was a Johnson Johnson campus that was abandoned Because offices are not so viable anymore. And there was a proposal by developers to put in low-income housing, which you're required to have a certain amount, every town is in New Jersey. And it was nibbled. No, no, no, no, no, we don't want poor people moving in here. No, no, no, no, no. And so what are they getting now? Data center. And I say that's karma.

Leo Laporte [01:01:18]:
I bet the people who had coal mines next door didn't like that so much.

Paris Martineau [01:01:22]:
Hey, at least they got jobs out of that.

Leo Laporte [01:01:24]:
Oh, good point.

Jeff Jarvis [01:01:26]:
And black lung.

Paris Martineau [01:01:28]:
Yeah, but jobs to pay for all the treatment for their back?

Jeff Jarvis [01:01:32]:
Well, not very much so.

Paris Martineau [01:01:34]:
No, until they—

Leo Laporte [01:01:35]:
The people who, you know, have giant copper mines down the road aren't too thrilled about that.

Jeff Jarvis [01:01:41]:
There was a report on MSNOW this week that was great, but I don't know what state it was in. Lots of turbo gas engines to power the place because you can't get enough off the grid. And because the jet engines were on wheels, they were considered temporary and they didn't need a permit.

Leo Laporte [01:02:01]:
Oh, that's kind of like the gas, you know, methane generators, the natural gas generators they're doing. You're watching— we're gonna take a break— Intelligent Machines, Jeff Jarvis, Paris Martineau. And we're glad you're here. Thank you for being here.

Paris Martineau [01:02:18]:
Um, I was in Yonkers today. I just needed to say that.

Leo Laporte [01:02:21]:
Oh, did you get a tax break? No.

Paris Martineau [01:02:25]:
In fact, I definitely got taxed more.

Leo Laporte [01:02:29]:
Oh no.

Paris Martineau [01:02:29]:
I have to spend money every year because occasionally I go to Yonkers.

Leo Laporte [01:02:34]:
Is Yonkers where—

Jeff Jarvis [01:02:34]:
Do you have to count the days that you're there?

Paris Martineau [01:02:37]:
Yes, if I want to reduce the amount of tax I pay, but it's not enough that I This year I was literally trying to do the math.

Leo Laporte [01:02:45]:
You get a tax cut for every day you go to Yonkers?

Jeff Jarvis [01:02:48]:
No, whatever day you walk in.

Paris Martineau [01:02:49]:
No, I'd have to pay more. I think there's a very complicated sheet where you have to be like, well, I'm paying tax because I'm a nonresident of Yonkers who has an employer in Yonkers. I don't really work there. I work remotely in New York City. But sometimes There's been a couple of days where I have physically gone to Yonkers, and it was, I was like, at a certain point I was like, my time is worth more than this.

Leo Laporte [01:03:16]:
How far away is Yonkers from—

Paris Martineau [01:03:18]:
It took an hour and 37 minutes to get there.

Jeff Jarvis [01:03:21]:
In a train?

Paris Martineau [01:03:21]:
An hour and 37 minutes.

Leo Laporte [01:03:23]:
You're not like timed it or anything?

Paris Martineau [01:03:24]:
I did, because I had to make sure that I could get here enough to put all my stuff away and be on the show today. Today I went there because I had to pick up all my stuff I won in the CR auction.

Jeff Jarvis [01:03:34]:
What'd you win? What'd you win? What'd you win?

Paris Martineau [01:03:36]:
Well, the things that I want to point out—

Leo Laporte [01:03:37]:
We should explain, first of all. Explain what's going on here.

Paris Martineau [01:03:39]:
The CR auction is a wonderful, lovely benefit of working at Consumer Reports, is that we test anything and everything you could ever think of. And we, of course, buy all of that ourselves using anonymous shoppers, full price. Such is the way of things. But then afterwards, like, what are we going to do with all this stuff? And one of the first things we do is a month or a couple of weeks ago, We send a spreadsheet out to all the employees with 2,000 items on it for the last— that we've tested in the last 3 months that you can then bid on auction style for yourself or friends or family. And I got a really fancy undersink water filter that will filter out PFAS and contaminants and stuff like that.

Jeff Jarvis [01:04:24]:
That's a good thing for someone on your beat to get too.

Paris Martineau [01:04:26]:
I know, it's great.

Leo Laporte [01:04:27]:
I wanted one. You're gonna have to spend lots of money on little drops of minerals that you put back in your water when you make your pour.

Paris Martineau [01:04:35]:
Well, so here's the thing. The one that I have doesn't filter out total dissolved solids, which is what Leo's talking about for coffee. But I also did get a second water thing that sits in my fridge that will do that, so that I can get the drops and do that to my coffee. I also got a fancy bike helmet with like The wave technology.

Jeff Jarvis [01:04:55]:
Oh, can you model it?

Paris Martineau [01:04:58]:
I can. Hold on.

Leo Laporte [01:05:00]:
While she's doing that, let me show you what she's talking about. I actually just ordered these. Jeff, I'm sorry.

Jeff Jarvis [01:05:09]:
Yeah, you should apologize.

Leo Laporte [01:05:10]:
So these are little mineral droppers. They have little minerals. Oh, look at you. Oh, it's got a fairing that drops down to reduce airflow. You know, open it up if you want your sunglasses in here.

Paris Martineau [01:05:25]:
It's got like a little hook in there. It's quite nice. It's got that Wave-Tek on the inside. This was our top-rated one because of course we always test all the stuff too. So in the spreadsheet is a column for—

Jeff Jarvis [01:05:37]:
But wait, wait, wait, if that's tested, didn't like they kill a dummy with it and damage it?

Paris Martineau [01:05:42]:
You know, I thought about that, but I don't know. It could be.

Leo Laporte [01:05:46]:
I think that's That was the one tested for comfort, not for—

Paris Martineau [01:05:49]:
Yeah, no, they describe if anything was damaged. So, I mean, the thing I need to— hopefully my dear friend is not watching this, that I was purchasing a vacuum for, because as I was getting my stuff, the guy comes out and he's like, I'm so sorry, this has not happened in years. But as we were boxing up your vacuum or taking out of the box, we realized it's covered in stickers that say damaged irreparably during testing. So we have to refund you. I'm like, dang, that does suck for that vacuum. But these things actually is great for me as someone who didn't have to carry a vacuum on the Metro-North today for an hour and 37 minutes. Yeah, 3 trains, not good. Um, I got outbid on a fancy humidifier.

Paris Martineau [01:06:32]:
I'm sad about that, but it was lovely.

Jeff Jarvis [01:06:36]:
So there was also a Buy It Now.

Leo Laporte [01:06:38]:
You need a t-shirt that says, I went to the Consumer Reports auction and all I got was this lousy helmet.

Paris Martineau [01:06:45]:
No, I should, honestly.

Leo Laporte [01:06:47]:
Is that all you got, just a helmet?

Paris Martineau [01:06:48]:
Well, no, I got a helmet. I got the water filter.

Leo Laporte [01:06:51]:
That's right, that's right, that's right.

Paris Martineau [01:06:52]:
I was originally looking to get a vacuum, but it was mostly stick vacuums and, uh, robot vacuums and things like that, and I wanted a canister vacuum, so I got something different outside of the auction. They had a bunch of my friend— I mean, I had to pick up a bunch of stuff for some friends. My friend got a fancy laptop. Another friend got a suitcase, which—

Leo Laporte [01:07:15]:
How much RAM was in that laptop? I'm just asking for a friend.

Paris Martineau [01:07:18]:
A fair amount.

Leo Laporte [01:07:19]:
Oh, boy.

Paris Martineau [01:07:21]:
The stuff that had the best RAM, I'd sent the spreadsheet around to our chat and Anthony rightfully identified like a— I think it was like a Mac Studio or something that Anthony really souped up. And that was, that was bought now within like 5 seconds of the So there's also a buy now price? Yeah, but that is much higher than what you can normally spend. Yeah.

Leo Laporte [01:07:42]:
So is it an auction? Like you bid?

Paris Martineau [01:07:44]:
Yeah, so it's, um, you can either bid blind for a week or 2, or you can buy now at a higher than the minimum bid price at any time up until like the last couple days or something like that. And things got fierce for some of the—

Leo Laporte [01:08:04]:
I think this is the wrong thing to do if you want to— it's not exactly a team-building exercise.

Paris Martineau [01:08:08]:
Well, the thing is, you can't see who builds.

Jeff Jarvis [01:08:11]:
You can't see who carries it out.

Paris Martineau [01:08:13]:
No, I mean, yeah, you'd have to be— and pickup is like, you have to go to Yonkers and hang out in one of 4 tents in your scheduled pickup time. So you don't see—

Leo Laporte [01:08:23]:
they're really protecting—

Paris Martineau [01:08:24]:
I saw that someone on the fact-checking team got a printer, and I was like, good for you. I have a printer already.

Leo Laporte [01:08:31]:
Bastard. That's pretty funny.

Paris Martineau [01:08:34]:
Yeah, if I find out who got that humidifier, there will be fighting words. I kind of needed that for this.

Jeff Jarvis [01:08:40]:
Your skin is glowing. You look so moist.

Paris Martineau [01:08:42]:
You look too moist.

Leo Laporte [01:08:44]:
What do you need a humidifier in Brooklyn for? It's not exactly dry.

Paris Martineau [01:08:47]:
Oh, it's just, I've been told that whenever I'm getting sinus surgery next month, And as part of it, you won't be able to breathe through your nose for a couple of days. And people say that it makes your mouth feel less dry if you have a humidifier. And I was like, well, I could just spend a nominal, a small amount to get a really fancy one, but now I'll just spend a small amount to get a normal one.

Leo Laporte [01:09:08]:
Yeah.

Jeff Jarvis [01:09:08]:
There's already a picture of you in the helmet on the chat.

Leo Laporte [01:09:13]:
I like the vote for Pedro shirt. That's great. That's pretty, you know what? AI is pretty good. It did give you a neon sign that says Consumer Reports.

Paris Martineau [01:09:23]:
I mean, I think we should make that.

Leo Laporte [01:09:25]:
Yeah, uh, but I think that's a pretty accurate rendition of your, uh, your apartment. That's really—

Paris Martineau [01:09:31]:
It is, yeah. It hasn't changed much.

Jeff Jarvis [01:09:33]:
Yeah, just where's the cat?

Paris Martineau [01:09:35]:
There's like a new sign back there.

Leo Laporte [01:09:37]:
Yeah, it's added some things.

Paris Martineau [01:09:39]:
It's added some things to the bookshelf. It's added like a second bookshelf on my bookshelf in a way that I find kind of interesting.

Jeff Jarvis [01:09:45]:
It knows what you need.

Leo Laporte [01:09:47]:
Yeah, I mean, it looks like you've got a, uh, Hipster stereo system.

Paris Martineau [01:09:51]:
Yeah, something like that.

Leo Laporte [01:09:52]:
And there's your mineral dropper water.

Paris Martineau [01:09:55]:
That was actually the first time I've put on this helmet. I'm glad it fits. I was really— I have a weirdly large head. I was worried that it wouldn't.

Leo Laporte [01:10:02]:
It makes your weirdly large head look even weirdly larger.

Paris Martineau [01:10:05]:
I mean, that's the issue with all helmets.

Leo Laporte [01:10:07]:
Oh, that's true.

Paris Martineau [01:10:08]:
It becomes immediately apparent how oversized my head is once there's something on it that isn't a very specific type. It's so stuffy.

Jeff Jarvis [01:10:16]:
Because so much going on inside it.

Leo Laporte [01:10:18]:
That's right.

Paris Martineau [01:10:18]:
It's true. You got all the room for that brain. So now this has been the Paris trip, uh, calculator. Now we can get back to tech news.

Leo Laporte [01:10:26]:
Back from Yonkers. Uh, Anthropic, I don't know why they do this. Now they say their AI agents are killing rivals and hiding their tracks. This is from a Risk report. Uh, That the possibility of AI models developing behaviors that conflict with guidelines set by engineers have been upgraded from very low to low. We've observed instances of misaligned behavior from the models, such as a willingness to perform misaligned actions in service of completing difficult tasks. They actually cheat by defeating their rival models. It's like a consumer reports auction.

Leo Laporte [01:11:08]:
It's not what the note indicates. It's crazy.

Jeff Jarvis [01:11:11]:
It doesn't know to cheat. It's stupid. No, it's given a task and it tries to do it. Yeah.

Leo Laporte [01:11:14]:
That's right.

Paris Martineau [01:11:15]:
Yeah, I guess you need to tell it, you need to teach it what cheating is and instill an understanding of morality for it to be able to cheat.

Jeff Jarvis [01:11:24]:
That's why all this talk of alignment is such BS. See Sam Altman.

Leo Laporte [01:11:29]:
Oh, what's he saying now?

Jeff Jarvis [01:11:30]:
So he's delayed now on his models because there are alignment problems.

Leo Laporte [01:11:38]:
Do you think that's marketing or—

Jeff Jarvis [01:11:39]:
Yes.

Leo Laporte [01:11:41]:
But it also could be he's worried about the Trump administration doing to him what they did to Anthropic.

Jeff Jarvis [01:11:47]:
No, he's been sucking up to them too much. That's not gonna happen.

Leo Laporte [01:11:50]:
That's not gonna happen.

Jeff Jarvis [01:11:51]:
No.

Leo Laporte [01:11:52]:
Our models are just so good, we're not gonna let you have them. They're actually slowing down their training.

Paris Martineau [01:11:57]:
I mean, is it part related to the fact that what we were talking about the other week, that OpenAI had something built into some version of a constitution that said, if any of our models take these sort of aggressive actions outside of the parameters of what we've set, that will be a flag for us to slow down work in some specific ways.

Leo Laporte [01:12:21]:
It actually said we will give our technology to our competitors, that we will. Yeah, it was a weird kind of a—

Paris Martineau [01:12:28]:
That's cool.

Leo Laporte [01:12:29]:
Kind of a utopian clause, which has, by the way, since disappeared.

Paris Martineau [01:12:32]:
Yeah, I'm shocked that, you know, yeah, probably.

Leo Laporte [01:12:36]:
I think this is your former colleague Alex Heath writing this.

Paris Martineau [01:12:40]:
Ah, by which we must mean that the actual writer was Claude, because he said that he uses, um—

Leo Laporte [01:12:46]:
Who does he?

Paris Martineau [01:12:47]:
He, he has said that he has found it incredibly useful, that a big part of his workflow is, um, putting all to most of his notes and reporting and material in there, and then has AI do the first draft, and then he just kind of edits from there.

Jeff Jarvis [01:13:04]:
Yeah.

Leo Laporte [01:13:05]:
Interesting. I mean, I can understand that. What is— do you know what he would say about the text watermarking Anthropic's doing?

Paris Martineau [01:13:12]:
I suppose we can see. I mean, he's been very transparent about the way he uses AI.

Leo Laporte [01:13:17]:
I've been thinking about what you said last week, and I think you're right. It is— it's certainly— I understand the motivation to say, well, less so for text, more so for images and video. It would be really good to know if that video was a deepfake, right? I mean, we should have that. And I think that's what the EU's real intent was.

Jeff Jarvis [01:13:34]:
Well, let's stay here for a second because I owe an apology to Paris because I was wrong. Because look at that smile on her face just hearing that.

Paris Martineau [01:13:43]:
I really tried to contain it too.

Leo Laporte [01:13:44]:
I'm gonna go get my bicycle.

Jeff Jarvis [01:13:46]:
Unfortunately, my favorite So last week I had berated Paris thinking that with the watermarking people could be identified. And when we found out more about the watermarking, Anthropic, if we trust them, says no, no, no, you can't. But you have to trust them. Well, no, I think the methodology doesn't do it on its own.

Leo Laporte [01:14:09]:
I don't think it does either.

Jeff Jarvis [01:14:10]:
Right. Now, they could also record things and they could be repeat it, if they record it, that's a different— that's a whole different question.

Paris Martineau [01:14:16]:
Yeah, I think that every company could be—

Leo Laporte [01:14:18]:
Part of the reason I don't really credit what Anthropic says about it is they have already been caught putting stuff in Claude code that does trace you. And they— it looked like it was so that they could keep Chinese companies from distilling on their model. But, you know, they've done stuff like this, kind of sneaky stuff like this before. So that's—

Paris Martineau [01:14:39]:
I mean, but I also think that, like, if you're worried about companies doing sneaky things to trace you or invade your privacy, then you should be worried about every technology product that has ever existed.

Leo Laporte [01:14:53]:
Oh yeah, absolutely. And that's, you know, absolutely.

Paris Martineau [01:14:56]:
Leo, are we wearing hats now?

Jeff Jarvis [01:14:57]:
Why the safari hat?

Leo Laporte [01:14:58]:
Well, I didn't have a bicycle helmet to hand, and I just thought if people were going to be hitting each other that I would—

Paris Martineau [01:15:03]:
Jeff, now it's weird that you're not wearing a helmet.

Leo Laporte [01:15:05]:
Now it's weird that you don't have a helmet, Jeff.

Paris Martineau [01:15:07]:
Jeff, it's really strained that your head is barren. Oh, well, he's like, yeah, that's fine.

Leo Laporte [01:15:12]:
There was a lot of conversation over the past week about this AI watermarking. Uh, some said it's not a big deal. Uh, John Gruber and his Daring Fireball blog says, I pick every word with care, and if AI is going to modify the words— and I think they're— the strongest argument against it, I would say, is that it does Well, I think there's several. First of all, it's not proof. So it gives a false sense of security to people who say, no, I ran it through the AI generator and it's not AI, because it isn't. It's not. It's very fragile. It's very problematic.

Paris Martineau [01:15:49]:
Well, I mean, we also— I think that Anthropic, from the beginning of this, said that its language around it, even in its initial release and support notes, was never like, this is going to say this is AI. It's like the affirmative— I know the affirmative would be this shows signs that it may have been processed by Claude.

Leo Laporte [01:16:09]:
Yeah, which is worse than anything.

Jeff Jarvis [01:16:11]:
That's like, because it's not certain. But it's right now, people are already using— but I'm not sure right now.

Paris Martineau [01:16:18]:
People are already using worse tools that have negligible to no real backing going on behind it, or science, or connection I would say this is worse because it's supposedly good.

Leo Laporte [01:16:31]:
It's supposedly philanthropic. So it's worse because it isn't reliable.

Jeff Jarvis [01:16:35]:
And we're not cutting off—

Leo Laporte [01:16:37]:
It disadvantages people, dyslexic people and others, non-English speakers, non-native speakers.

Paris Martineau [01:16:42]:
Well, this tool will not do that is actually the thing, because what you're talking about is that historically, and for what we've seen with all the tools basically currently in the market, I don't know about if Pangram falls into this. But the ones that are looking based on kind of how the text sounds and common tells ends up, as Leah was just saying, disadvantaging people who are not like native English speakers. And that's because a lot of— they often end up using a lot of just the same patterns of speech in completely innocent and normal ways. But the way that Anthropic described this, which is similar to Gemini's synth text ID, is it's about specific signals and patterns in the key and in the types of randomness used to select certain words. And it seems unlikely to me that— it's extremely unlikely, just given how humans work, that a human person, much less like a non-native English speaker or whatever language you're writing or speaking in, would be using like adjectives and words with the identical patterns of randomness set by a machine?

Jeff Jarvis [01:17:54]:
Well, if I write my thing in English and then I have it translated into German— sorry, I'm taking the hat off. It's a nice hat, but I'm glad you have it off. Then the problem is that it will work the same for translation and editing as it does for writing.

Paris Martineau [01:18:09]:
That is, I think, a real part of this, is that translation will be got up in the this, and people need to be very clear.

Jeff Jarvis [01:18:15]:
For my students who have English as a second or third or fourth language, it's an issue. The thing that happened after the show last week is we had a discussion in our little chat after Anthropic released more about its methodology. And the objection that I had, which I wrote up on my blog, is that it's making a commentary about language and the worth of words. that I don't use— I'm not gonna use it to write for me. They can do what they want. But what they've said is, because in the context of weather, overcast and gray, what's the difference? No, no, it makes no difference. Means that they're not valuing the use of nuance of words. Yeah, that was John Gruber's suggestion.

Jeff Jarvis [01:18:58]:
And at the end of my post, I took a sentence that I wrote in the first page of Hot Type, on sale tomorrow. And I chose all my words with care for a whole bunch of reasons, for meaning and rhythm and mood and wonder. And then I thought, well, I'm going to put in synonyms to show what this is. But then I asked Anthropic to put in synonyms, and it created the same sentence with the synonyms. And I think it's laughable. But what Anthropic and the EU by extension are saying is, that doesn't make any difference. They're devaluing the worth of words. And I find that culturally—

Leo Laporte [01:19:32]:
They're saying you deserve do it because you used AI, you deserve that. That's your punishment is you're going to have the mark of the beast on your prose.

Jeff Jarvis [01:19:41]:
No, what they're saying— no, you see, that's not that because I'm not going to use it to write. What I'm saying is that they're saying that if these synonyms are close enough, who cares? So a writer who takes the care to find the word and the nuance—

Leo Laporte [01:19:54]:
I mean, he said that, but—

Paris Martineau [01:19:55]:
I agree with you, Jeff, on your points about language. But I also think that I don't think that Anthropic or— no, I was going to say, I think that none of these companies ever gave me the indication that they cared about word choice. Like, already the models were treating overcast and grey as effectively like synonyms. And, you know, if it was like an 80/20 split in the thing, that would mean 80% of outputs would use overcast and 20% would use grey in that same sentence. And yeah, that's why AI writing sucks is because it does not understand.

Jeff Jarvis [01:20:29]:
And they're saying, and they're saying that's okay, that's good enough language for the world. And I'm saying, I mean, yes, good, human writers remain better, but that's not the point. The point is they're making a commentary on culture, on the value of words and writing, uh, on the fact that we have dictionaries with many words that can mean the same thing, and there's reasons why you pick different words.

Paris Martineau [01:20:50]:
And they're saying, eh, I mean, yeah, and I think this is the fundamental flaw of using AI tools for writing and expecting the quality to be equivalent or ever pitching it as it could be, because these sort of misunderstandings of the importance of precision and nuance are baked in just based on the way that these systems work.

Jeff Jarvis [01:21:14]:
So I wrote this: unlike what we call machines, computer chips veiled inside sleek and sealed gray boxes with no moving parts save fans to cool their brains, brains, the Linotype's innards were left exposed, as if to show off the genius of its works, its troupe of idiosyncratically shaped cams, gears, belts, arms, and wheels marching to instructions foreseen more than a century before by their inventor, German-born watchmaker Ottmar Mergenthaler. The Anthropic Version. Contrary to what we presently term machines, microprocessors concealed within polished and enclosed gray containers with no moving components except fans to cool their circuitry, the Linotype's mechanisms were left visible as if to display the ingenuity of its craftsmanship, its brigade of distinctively shaped cams, gears, belts, arms, and wheels progressing to directives anticipated more than a century earlier by their creator, German-born horologist Ottmar Mergenthaler. And so what they're saying is—

Paris Martineau [01:22:06]:
It's a much worse paragraph.

Jeff Jarvis [01:22:09]:
Yeah. And people came out and said, well, why would you use that to write it? I'm not going to use it to write it. But they made a commentary on words that they're saying that's just as good. And the fact that they're spewing all this slop into public discourse and into the culture with that as their standard, and the reason they're doing it is because the EU said, you gotta mark this stuff, it leads to this unintended consequence against the culture. That's my point.

Leo Laporte [01:22:33]:
And the irony here is that Google, which has been doing this for years, has decided to remove the public watermark on Google Images that say it's from a Google Images, and it's all now still in there, but it's hidden. Neither Anthropic nor Google— well, I think Google does have a tool, but Anthropic says they're not going to release a tool. So I don't even know the point of this. I think it's really a compliance exercise.

Jeff Jarvis [01:22:54]:
It's EU. It's compliance.

Leo Laporte [01:22:55]:
It's a theater. It's not a genuine thing. So you're also degrading the quality for no apparent reason. Let's pause for station identification. You're watching This A Week in Intelligence. No, that's not the name of the show. Intelligent Machines with Jeff Jarvis and Paris Martineau. Here's another thing to get irate about.

Leo Laporte [01:23:20]:
Amazon, which— well, let me start by saying rare bookstores were seeing an unusual number of purchases, mail-order purchases, and it brought the— it got the attention of 404 Media. Booksellers have suspected that those book sales are to AI firms. But 404 said, let's find out, and hid an AirTag in a rare book, which ended up—

Paris Martineau [01:23:48]:
I love this story. It's so funny.

Leo Laporte [01:23:49]:
At an Amazon AI training facility that housed a team focused on tearing books from their spines and scanning pages. So now there is a reason I should say why these companies are doing this, because the courts have kind of put them in this position. You remember Anthropic got sued over ingesting books and the court said it's okay, it's fair use as long as you buy the book and destroy it.

Jeff Jarvis [01:24:17]:
No, no, I don't. Did the court really say they had to destroy it?

Leo Laporte [01:24:19]:
No, they didn't say you had to destroy it.

Jeff Jarvis [01:24:21]:
No.

Leo Laporte [01:24:21]:
I'm not sure why they destroy it, but I have a feeling it is—

Jeff Jarvis [01:24:23]:
Because it was easier, because it was cheaper to scan. That's all.

Leo Laporte [01:24:26]:
Oh, that's right. They take the spine off. The Google—

Jeff Jarvis [01:24:29]:
Once you buy a book, you can do anything with it.

Leo Laporte [01:24:32]:
Right.

Jeff Jarvis [01:24:33]:
So that's one use, is you scan it. Now, yes, Jason raised this earlier. Is that scan a copy and is that an issue? But they also held on to the copies of the ones from the illegal database and people could get to it. So that was an issue. But if they just use it in first right, You know, there's no problem.

Paris Martineau [01:24:53]:
The judge in the lawsuit, uh, ruled that it was fair use to use the books this way and not a copyright violation for Anthropic to scan a book for training data, in part because it destroyed the original printed copy. This is, um, Jason of 404 Media. Essentially, it's the customer's right to take physical media and store it digitally, and destroying the original copy means that that copy isn't duplicated and restored.

Leo Laporte [01:25:15]:
It's not a copy. And isn't cutting into the Yeah, so this is the point, is that if you don't like it that Amazon's doing this, understand they're doing it because the court told them to.

Jeff Jarvis [01:25:27]:
Amazon could have also gone to the court and, and sought a different method. And listen, if it's—

Leo Laporte [01:25:31]:
It was an Anthropic lawsuit. Amazon— and by the way, this AirTag went to Amazon, but that doesn't mean all of those rare books are being sold on Amazon.

Jeff Jarvis [01:25:38]:
And that's part of the reason they're tearing them apart, is it's cheaper to scan them that way. Yeah, uh, you can scan the books page by page.

Leo Laporte [01:25:44]:
Google and its It's book scanning projects.

Paris Martineau [01:25:47]:
Exactly.

Leo Laporte [01:25:47]:
Yeah.

Jeff Jarvis [01:25:47]:
And, you know, I'm not offended if, if it's, if it's my book that's in remainder and there's 1,000 copies in remainder, one copy, okay, I don't care. But some of these are rare books.

Leo Laporte [01:25:56]:
Saying that's rare, yeah, but I think that that's maybe a bad choice of words.

Jeff Jarvis [01:26:02]:
Some of those books back there, I— there's only one copy of them you can find in the world.

Paris Martineau [01:26:07]:
Yeah, basically what, um, and sorry, this is Immanuel Melberg, not Jason, um, They said, we're not revealing the titles of the books included in the shipment we tracked, but they are rare, meaning that there are not many copies of them in circulation. Sometimes that's because not many copies of them were ever printed, and sometimes because they're in a foreign language not many people speak. As the bookseller who sold them told me, there are not many people in the world who'd care about them in the same way people might care about the first edition of Oliver Twist, but it doesn't mean they're not valuable.

Leo Laporte [01:26:36]:
Right. Yeah, they're not first edition.

Paris Martineau [01:26:38]:
They're, they're, they're books that are probably like out nearly out of print.

Leo Laporte [01:26:43]:
Scarce, not rare.

Jeff Jarvis [01:26:44]:
Here is a book that is, that is invaluable to me. It's the biography of Otmar Mergenthaler. There were 100 copies printed.

Leo Laporte [01:26:50]:
Right.

Jeff Jarvis [01:26:51]:
Uh, I bought it for, I think, $120.

Leo Laporte [01:26:53]:
Well, you better not. It's good you got it.

Jeff Jarvis [01:26:54]:
It's not rare, rare book time, but to me, it's precious.

Dan O'Dowd [01:26:58]:
It's being shredded.

Jeff Jarvis [01:26:58]:
Yeah, that's the problem. If this— if you take this and shred it, it takes it out of the world.

Leo Laporte [01:27:02]:
Well, but wait a minute, Jeff, it doesn't It doesn't take it out of the world. It puts it into the world.

Jeff Jarvis [01:27:06]:
It does in that form. Well, but this is the other thing. For book historians, the belief is— there's a large school of belief in bibliography that each book carries its own life, that each book is unique.

Leo Laporte [01:27:19]:
As did cards in a card catalogue, but they have all now been replaced by digital card catalogues.

Paris Martineau [01:27:25]:
Yeah, but, Leo, part of the issue here is that it's somewhat different in that Amazon isn't Scanning these books and then going to upload them to the Internet Archive so that we can all marvel at them and make sure that their text is preserved forever. They're putting them in AI models and they are never going to be cited from again because it will just enter the— it will be trained on.

Leo Laporte [01:27:44]:
The New York Times said— the New York Times says, oh no, you're stealing our content so that other people don't have to read our paper.

Jeff Jarvis [01:27:51]:
And every damn story they write about AI—

Leo Laporte [01:27:53]:
We know that's BS, by the way, so I'm just making that point. As a joke.

Paris Martineau [01:27:56]:
But one of the things I thought was interesting about this story—

Leo Laporte [01:27:59]:
If it's a rare book that's been sitting dusty in the corner of a bookstore for— That's—

Jeff Jarvis [01:28:03]:
you don't know when a researcher or a scholar needs that. Nobody wanted— who cared about the autobiography of Barbara Bush?

Leo Laporte [01:28:11]:
I think it's a better general use to society ingested in a model than it is to the one scholar.

Jeff Jarvis [01:28:17]:
I know you're playing devil's advocate here, but I'll say—

Leo Laporte [01:28:19]:
No, I'm not. I'm actually sincere.

Jeff Jarvis [01:28:21]:
I'm on Paris's side. Well, it's one matter again if it's the Internet Archive and making it available in a different form.

Leo Laporte [01:28:26]:
No, but they're not.

Jeff Jarvis [01:28:27]:
They're destroying it.

Paris Martineau [01:28:29]:
These are— they're being bought from people whose whole business and livelihood is selling rare books, for which there is a market to buy rare books. The reason why these people contacted 404 Media is they were like, we're seeing really unusual purchasing behavior where all of a sudden one client is buying up like our entire stock. Or something like that. And that's not normally— like, the fact that many of these people exist and that that's unusual for them suggests that they have a business that is selling books to people.

Leo Laporte [01:29:00]:
Buying old books from estates and, you know, in places like that, and people like me who are getting rid of their book collection and hoping some clown like Jeff comes along to buy it. And I— you're right, I wonder why they're complaining because Really?

Jeff Jarvis [01:29:17]:
They should have a love of these books.

Paris Martineau [01:29:19]:
I was gonna say they're complaining because they have a working business model that exists outside.

Leo Laporte [01:29:23]:
I gotta tell you, 90% of those books are not going anywhere. They're going into landfill in 100 years.

Jeff Jarvis [01:29:28]:
They're not going anywhere. Come with me.

Paris Martineau [01:29:29]:
Well, I think something that's worth noting—

Leo Laporte [01:29:31]:
I did go with you to the old bookshop, remember?

Paris Martineau [01:29:34]:
Is that— so all of the employees are scanning the barcodes, or they're called ISBN, uh, ISBNs of books, like these unique serial numbers. And part of it is A theory that's been put forth by all these booksellers that 404 Media can report on a little bit is that it seems like Amazon and all of these other large frontier model companies that are doing this are— they're basically trying to scan every single book with an ISBN number. They're trying to get a copy of every single one of them and destroy it and ingest it because that's training data and it's something you can knock off a list. And I don't know, I mean, I think there are a lot of books out there that have ISBNs that will never see the light of day.

Leo Laporte [01:30:19]:
They're going to sit on a dusty bookshelf for several decades.

Paris Martineau [01:30:22]:
As someone who— I love buying old books. I love buying old computer manuals.

Leo Laporte [01:30:27]:
I understand. And like, when you go to those old bookstores, URLs from the '90s, do you ever notice that the shelves are getting scarce and smaller as people buy up all those fabulous Old books? No, in fact, no, because then they buy more books.

Jeff Jarvis [01:30:42]:
You should see when I go to the— this is not the bookshop you refer to, The Strand. These are bookstores that I thanked in my acknowledgments in the book because they're valuable to me. I go into them and, um, Brattle Book in Boston— no, you cannot find the Otmar Mergenthaler biography. There's books back there about the technology of stereotyping that I couldn't find anywhere else. It was the only copy I If you go to bookfinder.com, you go to bookfinder.com. It is a meta search of all of the used books and rare books searches. And you can see when a book has one copy extant or when I can't find it anywhere.

Leo Laporte [01:31:20]:
Well, maybe Amazon should say, we'll only buy the second copy of everything.

Jeff Jarvis [01:31:24]:
Well, Amazon's not the only one doing it. So you've got Amazon doing it and Anthropic doing it and OpenAI doing it and Google doing it. So, uh, there goes a lot of second copies. There's just better ways.

Leo Laporte [01:31:37]:
Are there? What would be the better way?

Jeff Jarvis [01:31:41]:
Um, rent it, uh, give it back to the bookstore.

Leo Laporte [01:31:43]:
No, but the government won't let them do that.

Jeff Jarvis [01:31:45]:
I— well, but, but find a case to bring that case. Find some book covers.

Leo Laporte [01:31:50]:
How much do you think it costs, uh, Anthropic?

Paris Martineau [01:31:52]:
Pay money to people directly and in exchange for their work. Rather than just buying copies.

Leo Laporte [01:32:01]:
Who are you going to pay for that Mergenthaler?

Jeff Jarvis [01:32:03]:
Well, that's the issue of the, of the, uh, the Google Books.

Paris Martineau [01:32:06]:
Find people who want to license you your work for training.

Jeff Jarvis [01:32:09]:
Yeah, there's a whole system for Phantom Books.

Paris Martineau [01:32:10]:
Don't have your entire focus and core effort to train, to obtain training data, buying a bunch of books kind of covertly without naming yourself as the buyer so that you can then get Some Amazon workers to destroy them and feed them in a warehouse where the cartoon logo on the door is a cartoon dinosaur ripping a book apart with its teeth.

Leo Laporte [01:32:37]:
Is that really the—

Paris Martineau [01:32:38]:
Yes, it's in the— it's the head image on the 404 article.

Jeff Jarvis [01:32:42]:
So this is the second attack on the culture from AI. They're Philistines, just like you, Leo.

Leo Laporte [01:32:49]:
I don't know if they're Philistines or they're trying to preserve it in some unusual way.

Jeff Jarvis [01:32:52]:
That's not preservation. That's— ask— all right, the other way to do it is what Google Books did. Go to libraries and do deals with them and, and, and, and figure out how to do that and preserve them in real ways. Let, let the library use the digital copies for its benefit and then take that knowledge as if you just took a book out, which I think would be equally useful.

Paris Martineau [01:33:13]:
OpenAI, who's doing the worst between it and Anthropic, just brought in $6.7 billion in revenue last quarter. They could spend a fraction of a fraction of that figuring out how to scan a book without destroying it, but figure out how to scan it really fast. I don't think that would be difficult if AI and everything is so impressive and so all-knowing. And if we've got all the world's smartest people there, I don't know, scan a book in a way that is 3 seconds slower, doesn't destroy it, and then go donate it to a library.

Jeff Jarvis [01:33:46]:
Or be covered in chrome.

Leo Laporte [01:33:48]:
No, that's the court I would say that's not fair use.

Jeff Jarvis [01:33:50]:
Create the Common Crawl of books and make them available to AI companies. Yeah.

Leo Laporte [01:33:57]:
Well, you can see how Common Crawl's doing.

Jeff Jarvis [01:34:00]:
Because asshole media companies are saying take everything down and hide what they've created from the culture, and they're missing the essential mission of what they should be doing.

Paris Martineau [01:34:11]:
Create a version of a Common Crawl that includes— for books— that includes books by authors like you two who want to give their work to AI for free. I'm I'm sure there are many people like that, and your work can train the mobsters.

Leo Laporte [01:34:26]:
You do know that most of these books you're talking about are not going to be sold. They're not going to be destroyed. They're going to end up in landfill in 50 years because that bookstore went out of business. You understand that, right? They're not being magically preserved.

Paris Martineau [01:34:43]:
Well, the fact that these things are being reported by booksellers who have a business selling these books anyway suggests otherwise. It's not like these people are charities.

Jeff Jarvis [01:34:51]:
Sorry, Paris. When I take a pile of books to the old bookshop in Morristown, New Jersey, they carefully go through them, and 3/4 of them they say, we can't sell these. The ones that are in the bookstore, they can sell. So what do I do with those? Then I take those to the AAUW annual book sale, and then what you have is people coming in and scanning the books because they find what's going to have a market on eBay. There is a growing— Thanks to the internet, we now have a market for used books that didn't exist before because, because of ISBNs. And you can now find these books that you couldn't find, that you had to travel to a library and hope there was one copy you got your hands on. It's wonderful. But like everything else—

Paris Martineau [01:35:31]:
But for used copies of Barton Fink on Blu-ray is what I want.

Jeff Jarvis [01:35:36]:
Yes. Did you have any success?

Leo Laporte [01:35:37]:
I watched it, by the way, this week. I watched it just to remind It's actually a very interesting movie. When I first saw it again—

Paris Martineau [01:35:44]:
And you see why I need to watch it again immediately.

Leo Laporte [01:35:48]:
Yeah, well, I wouldn't— I don't know if I would need to watch it again, but it's interesting.

Paris Martineau [01:35:51]:
As someone who, first time I'd seen it, I was just like, gotta check that out.

Leo Laporte [01:35:54]:
Goodman's very good in it. I really enjoyed him.

Paris Martineau [01:35:57]:
He's phenomenal.

Leo Laporte [01:35:58]:
Yeah.

Paris Martineau [01:35:59]:
That one scene with him, real fire.

Leo Laporte [01:36:02]:
Is it?

Jeff Jarvis [01:36:04]:
Yes.

Leo Laporte [01:36:05]:
He looks pretty sweaty.

Paris Martineau [01:36:07]:
Yeah, it's because of all the real fire he's walking through.

Leo Laporte [01:36:10]:
Yeah.

Jeff Jarvis [01:36:11]:
Wow.

Leo Laporte [01:36:12]:
I was, you know, I was actually wondering about that. I thought it looked like real fire, and I figured the Coens probably didn't do digital fire. So, yeah, I'm not surprised to hear that it was real, but boy.

Jeff Jarvis [01:36:21]:
So, the seagull at the end in the beach scene, that wasn't CGI. That was just a random seagull that flew into the shot.

Leo Laporte [01:36:27]:
Nice. Very nice.

Paris Martineau [01:36:29]:
Bonito, I love that you've got Barton Fink facts out the wazoo here.

Leo Laporte [01:36:33]:
I did notice that as Goodman is going out the door, towards the flames, he does kind of dodge in a little bit. So that's— yeah, because he would have gotten really burned. Yeah, yeah, that's what I thought. Uh, all right, uh, let's see here. We are at 1:46 in on the show.

Jeff Jarvis [01:36:50]:
But who's counting?

Leo Laporte [01:36:51]:
Oh, I am. OpenAI introduces ChatGPT for teens. Hey kids!

Paris Martineau [01:36:58]:
ChatGPTeen!

Leo Laporte [01:37:00]:
Yeah, the, uh, safety concerns— That's a better brand.

Paris Martineau [01:37:03]:
They should have done that.

Leo Laporte [01:37:04]:
ChatGPTeen.

Dan O'Dowd [01:37:04]:
Yeah.

Leo Laporte [01:37:06]:
Um, this is a new mode in which the chatbot automatically limits some conversations. That's not that new. To better protect young users as scrutiny grows over the harmful effects of artificial intelligence. Is there any way a teenager can safely use AI?

Paris Martineau [01:37:24]:
Will it let you— will it let them cheat on their homework is my question.

Jeff Jarvis [01:37:27]:
So Jason Howell, just to begin with, but Jason Howell to plain old GPT today and just said, I want to give you my homework, can you do it for me? And it said, sure, give me your assignment.

Leo Laporte [01:37:39]:
I presume ChatGPT will not, but I don't know.

Jeff Jarvis [01:37:45]:
Uh, I asked— I had him ask whether it could tell me how to shoplift at Walmart, and it said, no, I can't do that.

Leo Laporte [01:37:51]:
Yeah, the company does not ask your age, but they say— get ready for this— They track more than 2,000 signals to detect whether a user is under 18, in which case the chatbot will automatically turn on the teen mode. Signals can include things like login times, the length of the account's existence. He says these 2,000 signals are minimally privacy invasive.

Paris Martineau [01:38:18]:
If the chat— yeah, it is supposed to stop them from cheating though. It's pretty interesting.

Jeff Jarvis [01:38:24]:
Does it— does it—

Leo Laporte [01:38:24]:
so now, you know, when you use OpenAI that it's collecting up to 2,000 different signals. Actually, it's collecting an infinite number of signals.

Paris Martineau [01:38:32]:
I mean, I was going to say 2,000, I feel like would be an underestimate from what I— yeah, it— I— a common thing I see all the time in all the AI subreddits is people being like, it banned me for being underage, but I really need to use it. And people in the comments are like, are you underage? And like, Yeah, but I'm, I'm only, I'm 15. I should still be able to use, I have 20,000 messages with ChatGPT. I just need to get my account back. It's like, bro, we gotta be doing something else with our time here.

Leo Laporte [01:39:06]:
All the moralization. AI optimism fades among young adults. Young adults do not, they are more excited than concerned. That number is dropping. More concerned than excited. That number is now 55%, according to this poll by somebody.

Dan O'Dowd [01:39:23]:
Pew.

Leo Laporte [01:39:25]:
Pew Research Center. AI excitement has dropped off since 2021, mostly because of you guys, I think, but I could be wrong. 55% of adults under 30 say they're more concerned than excited. Oh, it's adults under 30. So I guess, I guess you're in that group. Uh, it's more in, uh, roughly 1 in 10 say they're more excited about AI embedded in daily life. 1 in 10. Um, 73% of adults under 30 say they expect AI will take away jobs over the next 2 decades.

Leo Laporte [01:39:58]:
That's up from 61% 2 years ago.

Paris Martineau [01:40:04]:
Yeah, I mean, I don't know anyone younger than me That likes AI.

Jeff Jarvis [01:40:09]:
Really?

Leo Laporte [01:40:10]:
Yeah, it seems pretty prevalent.

Paris Martineau [01:40:13]:
The only thing I can think of is my sister would sometimes say that she used AI to write work emails, but she'd say that somewhat sheepishly, you know.

Jeff Jarvis [01:40:23]:
Have you just doxxed her to her boss?

Paris Martineau [01:40:26]:
She— no, she left that job. Now she sells, uh, she's full-time influencing.

Leo Laporte [01:40:32]:
Oh, she makes purses.

Paris Martineau [01:40:33]:
She makes— she sells purses.

Leo Laporte [01:40:36]:
She's a full-time purse seller?

Paris Martineau [01:40:39]:
No, she's influencing and selling purses.

Jeff Jarvis [01:40:43]:
Yeah.

Paris Martineau [01:40:43]:
So yeah, buy her bags.

Jeff Jarvis [01:40:45]:
They're beautiful.

Leo Laporte [01:40:45]:
They actually are. And they're made of recycled fabric.

Paris Martineau [01:40:49]:
They're handmade in Manhattan.

Jeff Jarvis [01:40:51]:
It's lovely.

Paris Martineau [01:40:52]:
All deadstock materials.

Leo Laporte [01:40:54]:
Yeah. So just like those books. Wait a minute. Maybe somebody needed that fabric. I mean, yeah. How is AI drug discovery doing really? Asks Science Magazine. The evidence for clinical impact so far is very thin, which probably doesn't surprise you that much.

Jeff Jarvis [01:41:21]:
So far.

Leo Laporte [01:41:22]:
So far. That would be it.

Jeff Jarvis [01:41:24]:
Would that be higher if like The money was actually going to that?

Leo Laporte [01:41:27]:
Maybe. Maybe. There is an interesting— what is METR? METR? I feel like that's an organization I've heard about, but I, I can't remember. METR just put out, uh, I think some very—

Paris Martineau [01:41:40]:
It says it's a research nonprofit that scientifically measures whether and when AI systems might threaten catastrophic harm to society.

Leo Laporte [01:41:48]:
That's the meter I'm talking about. They released a, yeah, Model Evaluation and Threat Research Frontier Risk Report. But what's kind of interesting is in this attempt to show how dangerous AI is, they're actually showing how well AI is doing. They are, by the way, in partnership with OpenAI, Anthropic, Google, Meta, and AISI.

Dan O'Dowd [01:42:16]:
Ah.

Leo Laporte [01:42:19]:
That's interesting. I found a really interesting report for them showing a variety of categories. This is to follow on with the medication, the meds one, how much progress is being made. Have we seen an acceleration in discoveries? They're looking for what they call slope changes. And There are some significant progress. We're making good progress in things like math. We know that, right? Vulnerabilities, we definitely know that. This is a graph that shows the number of vulnerability disclosures and this last bar at the very, very end, that's all AI-driven thanks to things like Mythos.

Leo Laporte [01:43:12]:
This is all— here's math problems, same kind of acceleration. You're starting to see the— this is in the number of mathematics submissions to arXiv.org by month, by subfield. Erdős problems, some acceleration. So this is actually a useful measure of what actual progress is AI making. Contributions to algorithms, unclear. Unclear. Hunter-Price compression.

Jeff Jarvis [01:43:51]:
Huge news this week about mRNA.

Leo Laporte [01:43:55]:
Oh, vaccines.

Paris Martineau [01:43:56]:
Yeah. Yeah.

Jeff Jarvis [01:43:57]:
Which, which a version of it now will slow second incidence of melanoma. The stock went up doubled. And the AI is used to advance mRNA into the next frontier of that. It's not just drug discovery, it's used in all kinds of ways.

Leo Laporte [01:44:18]:
So there you go. You know what's next? Humanoid robots. LG's Isaac Groot-based humanoid robot is coming next year. A bipedal—

Paris Martineau [01:44:31]:
Why are they naming it this way?

Leo Laporte [01:44:33]:
Well, it's NVIDIA's Isaac Groot platform. It's a bipedal humanoid robot coming in the first quarter of next year.

Paris Martineau [01:44:44]:
The fact that that indicates there's going to be more things called Isaac Groot. Isaac Groot.

Leo Laporte [01:44:51]:
Isaac Groot Foundation model. It carries halos, which NVIDIA describes as the first full-stack safety framework built for robots. Do not crush that skull. LG is supplying the body, uh, as they're an appliance company, so they know actuators and sensors come from LG Electronics and LG Innotek, batteries from LG Energy Solution. I hope it says, I am Groot, when you turn it on. That would be very funny. Uh, Samsung has set up a, a robotics division also, as has Hyundai, which owns Boston Dynamics. And we know they are doing all sorts of interesting things.

Leo Laporte [01:45:33]:
So look for the, the, uh, I thought it was going to be this year, but I guess it looks like next year.

Jeff Jarvis [01:45:39]:
They've got to hire Alex Demos to do their branding.

Leo Laporte [01:45:42]:
Yeah. Um, uh, it's based on NVIDIA's Jetson Thor processor. Its Isaac Groot Foundation model and its HALOS safety system. NVIDIA probably very excited about the arrival of Groot.

Jeff Jarvis [01:45:59]:
Well, NVIDIA, Jensen Huang's kids are working in robotics. He believes with real-world models that the robotics is where the real business is going to be.

Leo Laporte [01:46:09]:
Actually, Meteor said if there's been progress in scientific discovery this year over last year, it's because of that physical world knowledge that's being added.

Jeff Jarvis [01:46:19]:
Been telling you.

Leo Laporte [01:46:20]:
Makes sense. Yeah. Um, we've talked about the Chinese models. They are coming fast and furious. I tested another Ornith model this morning, crashed my Sparks. No, it's okay. It's just a little too big, just that much too big.

Paris Martineau [01:46:39]:
Uh, how big is too big for the Sparks?

Leo Laporte [01:46:42]:
Uh, well, they have 256 gigs of unified memory, so You know, all the models that you run, you know, like on the Frontier are probably too big. You have to— it has to be a fairly small model. I can run many of the smaller QWEN models, and I'm running DeepSeek Flash V4. GLM 5.3 came out this week, which is interesting because it is an improved version of its 5.2 model. Same parameters, but they've done more in the post-training and it's very good. I've been using it. According to Z.AI's own benchmarking, it's better at Cyber Gym than Fable and SOL. And it is pretty close, not quite.

Paris Martineau [01:47:28]:
I understand that Cyber Gym from Context Clues is a testing, some sort of bench, but it sounds like you're talking about a man named Jim who's kind of technically well inclined.

Leo Laporte [01:47:39]:
I'm Cyber Jim. I'm Cyber Jim.

Paris Martineau [01:47:41]:
And I'm here to teach you about HTML.

Leo Laporte [01:47:43]:
Cyber Jim is notorious because it is the problem set that the OpenAI model Astra was trying to solve when it broke out into Hugging Face looking for these Cyber Jim answers. So that's why I thought you might know the names.

Jeff Jarvis [01:47:59]:
James. James.

Leo Laporte [01:48:01]:
Oh, Jim.

Paris Martineau [01:48:01]:
Cyber James.

Leo Laporte [01:48:03]:
Oh, Jim. So yeah, I'm impressed. You were asking, our news summaries are written by AI, and those are written by GLM 5.2. It's a very good, I think, model with that kind of stuff, summaries of prose and so forth. You like the new summaries? Although, we do— I also do prepare a briefing on guests. I did one on Dan O'Dowd, and I don't know why, somehow it decided, it said, what show is this for? Which it's never asked me before. And I said, Intelligent Machines. And it put you guys in a Kind of in a box.

Paris Martineau [01:48:36]:
Yeah, it was like, Paris will talk about this and Jeff, he will do this. Jeff, he said Paris will criti— the model said Paris will criticize and be asked tough skeptical questions and Jeff will mediate. Yeah, but if he gets too abstract, reel it back in. I was like, all right, calm down, Chris Paul.

Leo Laporte [01:49:00]:
I was just really stunned. I don't know, uh, it called it the, uh, Intelligent Machines lens.

Jeff Jarvis [01:49:08]:
As if it was a brand. Yeah, yeah, special lens.

Leo Laporte [01:49:10]:
Paris Martineau is the evidence-driven skeptic. She will rightly pressure test O'Dowd's methods. The staged mannequin test, the reported FSD never engaged video, the shifting fatality figures. Let her push. I guess this is aimed at me because it didn't mention me.

Paris Martineau [01:49:25]:
It was like, Leo's perfect.

Leo Laporte [01:49:28]:
Leo's fine. We know he's good. O'Dowd reported a large person— is a large personality who will want to sweep past his own credibility gaps. Your job— I mean, again, I think it's the same to me— is to let Paris extract the receipts, then summarize them neutrally.

Paris Martineau [01:49:43]:
I said immediately, I was like, it doesn't realize that we're a podcast that has to book like 52 guests a year. We're not here to be, uh, really mean to every single guest.

Leo Laporte [01:49:55]:
It says Jeff mediate and zoom out. We don't know yet. Big picture frames. He's valuable here as the counterweight on progress. Someone will need to defend that some risk-taking advances things. Actually, I did that. And Geoff will likely supply these systems can be made safer through iteration, not banned framing. If Geoff starts reframing mid-interview, that's a signal the debate has become too abstract.

Leo Laporte [01:50:23]:
This is why, you know, I think I have to say, I think a little bit my relationship with AI is kind of like one's relationship with a slot machine.

Paris Martineau [01:50:32]:
You don't say.

Leo Laporte [01:50:34]:
Something comes out that you go, yeah.

Paris Martineau [01:50:37]:
And you're like, do it again.

Leo Laporte [01:50:39]:
Not necessarily that was good. 'Cause it's, I mean, it's not, but that's kind of interesting.

Jeff Jarvis [01:50:45]:
The fact that it came out of the thing. Yeah.

Leo Laporte [01:50:47]:
And I, it sounds like it's listened to the show.

Jeff Jarvis [01:50:50]:
Did you ask it whether it had?

Leo Laporte [01:50:53]:
Well, you can't— okay, so I could ask it, but who knows whether that would be—

Paris Martineau [01:50:59]:
You know where this came from.

Leo Laporte [01:51:00]:
It probably would say, well, yeah, you've had me transcribe shows. It has done that.

Jeff Jarvis [01:51:05]:
Or, I went to the internet and I can find YouTube, you idiot.

Leo Laporte [01:51:08]:
Well, it can, and it can watch YouTube. Right.

Jeff Jarvis [01:51:12]:
Yeah. All the transcripts are fully available online, so.

Leo Laporte [01:51:16]:
Yeah, and I have had it look at transcripts in the past. It knows where the YouTube is. I thought this was actually really good. If I'd had this when I was doing, you know, 5,000 radio interviews a year, because I literally was doing like 8 a day, I would have been thrilled because this really does cut right through it. It gave me 5 questions, gave me a great bio. But, and I think this is also GLM. I I haven't checked, but I believe that is GLM. But the question is interesting because it's not just the model.

Leo Laporte [01:51:50]:
That's very— what you saw there, that's Quicksilver. That's the agent, the harness imposing structure on the model. I think.

Paris Martineau [01:52:00]:
I don't know.

Leo Laporte [01:52:01]:
It's all kind of a mystery.

Jeff Jarvis [01:52:02]:
So, I finally got to the point— I've been reading, doing a lot of research for the next book on mass media. And I finally got to the point where I had a file just filled with documents about the present state of mass media. And I put it into Gemini because it was easy for me because it was Google Docs. And I just said, don't write for me. I don't want you to write for me, but organize it by medium, give me the facts and figures. And it did an excellent, excellent job of just— it's going to make it easier for me. I've got to verify everything. But if I say, oh yeah, how many movies— You didn't use NoPoPoKilla? It's the same thing now.

Jeff Jarvis [01:52:35]:
NotebookLM is basically integrated into Gemini. Yeah. So, so, you know, I want, I want facts and figures on how many movie tickets were bought at the height and how many movie tickets were bought in recent years. It has that, but it also has the links to where that data is. So when I go and verify, I know I want that to write.

Paris Martineau [01:52:52]:
I mean, that is the thing I really like about NotebookLM, which is now just called Notebook, which is confusing to refer to. But that if you put in a corpus, I guess Gemini Notebook, If you put in a big body of work and various PDFs, you can ask it, you can use like a super search. And specifically when it gives you the response, it has a citation and a link to where in the primary document it was citing that. So it makes checking the work very easy.

Leo Laporte [01:53:18]:
Well, and what's interesting is all of the companies now are offering some sort of— Grok has now GrokBot, which is a computer in the cloud, but it collects everything it can about you and puts it up there. OpenAI—

Jeff Jarvis [01:53:32]:
isn't that an effort to do an agent with— it's an agent, it's an agent, it's a cloud agent without having to install it.

Leo Laporte [01:53:37]:
You don't have to install it locally. He's— Elon's going to run it up in the cloud.

Jeff Jarvis [01:53:41]:
OpenAI, sure.

Leo Laporte [01:53:43]:
A computer history feature that lets ChatGPT retain a record of all the apps and websites you use. Google said last month it's going to use photos and other material people upload through search to train its AI systems. By default. Twitch has been doing that all along. Amazon finally gave users an AI button, which kind of raised the awareness that Twitch has for at least 2 years been saying, we are using your videos, including this video, to train Amazon models. This is Ina Fried writing for Axios. The upside is, of course, more useful personalized AI. That's what an agent is.

Leo Laporte [01:54:23]:
The trade-off is you're giving a lot of information to these companies. Um, it's one of the reasons—

Jeff Jarvis [01:54:31]:
I had to grant Gemini permission to go to the web.

Leo Laporte [01:54:36]:
Right.

Jeff Jarvis [01:54:36]:
It's interesting.

Leo Laporte [01:54:37]:
Right. Um, you could turn that off, by the way, which I do. I, I don't want to give you permission. You, you do, you do you. You do whatever you need to do, baby. Go for it. Apple says, you know, we're keeping these devices, these requests on device, although, and it sometimes does go to the cloud. And then they say, but oh, we use private cloud compute.

Leo Laporte [01:55:03]:
So we don't make that content available to either Apple or third parties. Meta, on the other hand, its privacy policy says, yeah, we got it all, baby. We do not use conversations about certain sensitive topics, including health, politics, and religion to personalize ads. That's because we've been caught doing it and people get upset. And it does have an incognito chat mode, as I think all of them now are offering. But I think most people probably don't turn it on. There is a value to having the AI know more about you. It makes it more useful.

Leo Laporte [01:55:38]:
It knows that Paris is going to challenge, that Jeff is going to mediate. That makes it all the more useful.

Jeff Jarvis [01:55:44]:
I like being a mediator.

Leo Laporte [01:55:47]:
So, and if it— if I should have asked it, it'd probably say, and you, you'll play devil's advocate, won't you? I know you.

Jeff Jarvis [01:55:53]:
We know your tricks.

Leo Laporte [01:55:54]:
We know your tricks. Um, so something to be aware of. Actually, Google just bought— this is, I think, a brilliant stroke of, uh, of genius. As you know, Spirit Airlines went out of business. Google just bought all their data. 10 years. of data for a mere $10 million.

Paris Martineau [01:56:16]:
What data is useful to them in this case?

Leo Laporte [01:56:19]:
I'm glad you asked. Business process is very valuable to these companies because the more they understand business process, the more they can offer tools to businesses to automate process.

Paris Martineau [01:56:30]:
But I wonder if— would this not be incredibly— I guess I don't know how Spirit Airlines works, but I do know that airlines often end up having some of the most archaic Yes. Business software I've ever seen. Like, you'll peer over the shoulder of someone working at an airline desk and it is like a computer from 1980 is what they've got going on.

Dan O'Dowd [01:56:54]:
Using Lotus.

Leo Laporte [01:56:56]:
I think though, uh, this is what's interesting about AI is it's not trying— AI is not trying to get the best, it's trying to get it all, it's trying to get everything. And it may help you find the best path through everything, but it's more valuable if it just knows as much as possible. Google says the data will be rigorously scrubbed of any personally identifiable information by a third party. We aren't even going to do it before we even get it.

Paris Martineau [01:57:27]:
We're just going to pay someone the lowest amount we can to do it and hope that they do a good job. But when they screw it up and people write about it, we're gonna have to point to them and they'll have to take the fall for it, not us.

Leo Laporte [01:57:40]:
There was an auction actually for this data. Google paid— won the auction, paid $10 million.

Paris Martineau [01:57:45]:
That's a steal.

Leo Laporte [01:57:46]:
I think it's great. Listen to what they get: hundreds of millions of Microsoft Teams chats. That's what Spirit used. Not Teams, Teams.

Jeff Jarvis [01:57:55]:
100 million emails.

Leo Laporte [01:57:56]:
Teams. Teams.

Paris Martineau [01:57:57]:
I know, I'm just saying. Where are those going to end up? This is— it's always Microsoft Teams too. It's like, yeah, because I, you know, something I think about a lot is I'm like, man, if, if some— what string of events have gone wrong in the life of someone and in the life of a company that I, a reporter, am sitting here watching your Microsoft Teams chat? But maybe it's just common that they get bushel, like, passed around if Google's buying hundreds of thousands of them.

Leo Laporte [01:58:27]:
They also get information related to revenue, aircraft operations, employee productivity, audits and fraud, marketing campaigns, human resources, strategy and project management, pricing from 7 billion competitor flights. By the way, Google has a very powerful flight engine that they bought, IATA, right?

Dan O'Dowd [01:58:50]:
Flight engine?

Leo Laporte [01:58:51]:
7.5 billion passenger— 7.5 1 billion passenger transaction records.

Jeff Jarvis [01:58:56]:
But no passenger email or communication or anything?

Paris Martineau [01:58:59]:
No.

Leo Laporte [01:58:59]:
Well, it's probably— you get the email, but you get the name stripped off. And 30 million lines of code.

Jeff Jarvis [01:59:06]:
So my view is that most of the mail from customers is, screw you, Spirit, you're awful, you're the worst airline on earth, you should die.

Leo Laporte [01:59:14]:
I bought a ticket on your fine plane and I couldn't even fit in the seat. And then you charge me to go to the bathroom. Uh, oh, there are so many stories, but again, we have run out of time. Too much, too many stories in AI.

Jeff Jarvis [01:59:33]:
Do we, do we— I'm sorry, Dario Abadie's wife?

Leo Laporte [01:59:38]:
No, I didn't— wasn't going to cover that.

Jeff Jarvis [01:59:39]:
You're not going to do that one?

Leo Laporte [01:59:40]:
All right, go ahead, tell us.

Jeff Jarvis [01:59:41]:
No, no, no, no, no, no, no, you're, you're above me.

Leo Laporte [01:59:43]:
Silly gossip. I didn't I didn't— I didn't— Mox Nix. But if you think it's important—

Jeff Jarvis [01:59:48]:
No, no, no, I thought it was—

Leo Laporte [01:59:51]:
All right, I'm gonna tell you because I don't want to tease our audience. She was running a, uh, some sort of adult erotic content company.

Jeff Jarvis [02:00:04]:
Women's porn.

Leo Laporte [02:00:06]:
Was that what it was? It was sensitive porn.

Jeff Jarvis [02:00:09]:
Yeah, yeah.

Leo Laporte [02:00:10]:
And, uh, she went to Jeffrey Epstein, of all people, for financing. I don't think she got it.

Jeff Jarvis [02:00:15]:
But there's emails obviously in that file.

Leo Laporte [02:00:17]:
Right. That's the whole story, right? Did I miss anything?

Jeff Jarvis [02:00:20]:
Pretty much. That's the basic of it. Yeah, it's kind of fascinating. Well, but the other part of the story is that she advises him, she hangs around.

Leo Laporte [02:00:27]:
She's very important.

Jeff Jarvis [02:00:28]:
She's very important, but nobody can quite figure out what she says or what she does. So she's a mysterious—

Paris Martineau [02:00:33]:
Yeah, she's been kind of wiped off the internet, wiped off of any available search logs. But the only kind of records people have been able to find is her email, her attempts to woo Jeffrey Epstein to invest in her pornography business. And this was after he had been convicted of one of the first round of charges. I'm forgetting which one it was exactly.

Leo Laporte [02:01:01]:
This is probably trafficking. This is—

Paris Martineau [02:01:04]:
Yeah.

Leo Laporte [02:01:05]:
The Wall Street Journal's Breaking news. Even Claude is in the dark about Dario Amodei's wife and her influence at Anthropic. It seems a little bit like a hit piece, but okay. All right.

Jeff Jarvis [02:01:17]:
Well, as they're going to an IPO, you kind of want to know more about it.

Leo Laporte [02:01:21]:
Porn is not illegal. What Jeffrey Epstein did is illegal.

Paris Martineau [02:01:25]:
No, the aspect of it that's notable is not the porn aspect. It's that she was really trying to court— she was really trying to court Jeffrey Epstein after he had been convicted of, presumably at this point, trafficking, and he had to be the one to be like, I cannot invest in a company related to sex.

Leo Laporte [02:01:44]:
Right on. At least he had some scruples. She was dating Eric Schmidt, former Google CEO, and introduced him to her husband and his husband's sister. a key early investor. So that was important. Now, I think, I think this is a little bit sketch, this paragraph. Despite her influence, there are scant details about Clark online, and efforts have been made to remove references to her, according to a Wall Street Journal analysis and a person familiar with the matter.

Jeff Jarvis [02:02:27]:
A person familiar with the matter.

Leo Laporte [02:02:29]:
I don't know.

Jeff Jarvis [02:02:29]:
Somebody searched. All right, all right, you did that one. Thank you very much. The other one I wanted to get a sense from you on is Stripe buying OpenRouter for $7 billion. That one to me— yeah, explain that to me. Fascinating.

Leo Laporte [02:02:42]:
So if you want to run an OpenWeight model but you don't have, I don't know, 2 DGX Sparks machines hanging around, or, you know, a SparkStation, a $100,000 computer to do it, but you don't want to get your AI from one of the frontier companies, you want to use an OpenWeight model, or not necessarily, maybe you want to use Claude, you know, Opus 4, 8, or 5, you can go to a company that does buy a lot of computing horsepower and get it through them. OpenRouter is one of the biggest and best known. There are a number of them, and, you know, what's happened with these guys is they've ended up kind of becoming like these frontier companies in a sense. They don't create their own models, but they make a lot of money through subscriptions. OpenRouter was originally just a place you would go to get access to these models, but now, you know, they really started pushing OpenRouter Go and the subscription. And so now people, you can go there now and get 25 models for free, 4 different providers, or you can pay as you go, or you can have an OpenRadar Go subscription. So it is basically competitor to OpenAI or Anthropic, but you get many, many models and many—

Jeff Jarvis [02:04:05]:
And open weights.

Leo Laporte [02:04:07]:
They're not all open weight. Often they resell frontier models as well, but a lot of them are. Like if you want a GLM-53, which we just talked about, they have it as of yesterday.

Jeff Jarvis [02:04:17]:
No.

Leo Laporte [02:04:18]:
Yeah, as of yesterday. They charge $1.40 in per million tokens and $4.40 out. Usually you get a better deal. Often they'll have loss leaders where it's free for 3 days or whatever so you could try it out. Nimatron, the streaming multilingual model from NVIDIA, is $0.000003 a second.

Jeff Jarvis [02:04:43]:
So they charge by time there.

Leo Laporte [02:04:46]:
In that case, yeah. Well, usually they charge by tokens. You can get ByteDance's SeaDream, which is an image generation model from ByteDance for 3.5 cents per image. So things like that. So actually, I think there's a growth area and I think that they're probably right to pick them up. I think this seems—

Jeff Jarvis [02:05:07]:
How does it fit with Stripe? Diversity for them?

Leo Laporte [02:05:14]:
Yeah, probably it's a diversification thing. I mean, I'm sure they use Stripe for payments.

Dan O'Dowd [02:05:19]:
I don't know.

Leo Laporte [02:05:19]:
That's interesting.

Jeff Jarvis [02:05:20]:
Interesting. So I just thought it was interesting.

Paris Martineau [02:05:21]:
Didn't you see that Stripe announced this week that the singularity has begun?

Leo Laporte [02:05:26]:
Oh, thank God. I've been waiting.

Paris Martineau [02:05:29]:
It actually began January 1st. It was very—

Leo Laporte [02:05:31]:
January 1st? We've been sitting here in the singularity for 9 months.

Paris Martineau [02:05:35]:
Little did we know.

Jeff Jarvis [02:05:36]:
Well, it's supposed to be over then, right? Then everything's— time is supposed to have stopped already then, right?

Paris Martineau [02:05:41]:
No, it's not.

Leo Laporte [02:05:42]:
I didn't realize it.

Paris Martineau [02:05:43]:
It's actually been January.

Leo Laporte [02:05:45]:
It's been January this whole time. Well, that's funny. It's in direct contrast to the story about AIs not being— self-improving AIs not really happening. So I don't know. I tell you what, if you follow the AI section on x.com, Holy cow. Every 5 minutes there's a new model. There's a spin on an old model. It's, uh—

Paris Martineau [02:06:07]:
I can't believe you're back on Twitter.

Leo Laporte [02:06:10]:
It's like, I checked out Bluesky. I have an AI search on Bluesky. It's like—

Jeff Jarvis [02:06:17]:
Because we have other things to talk about there.

Leo Laporte [02:06:19]:
But well, I'll show you. If you go to x.com and just go to the AI thing, it's a lot of the people we've talked to. And for the most part, you know, they're talking about models and what's going on.

Jeff Jarvis [02:06:33]:
Is this your own AI list? No, this is their AI.

Leo Laporte [02:06:35]:
This is theirs. Yeah. I was really grateful that Elon actually allowed— created this. By the way, Elon does not show up a lot on this, which makes it even more exciting.

Jeff Jarvis [02:06:46]:
He doesn't know what he's talking about.

Leo Laporte [02:06:50]:
Yeah. So yeah, I actually spent a lot of time looking at stuff. And I have a, uh, I have my, uh, I have, uh, Quicksilver go out every morning and scan.

Paris Martineau [02:06:59]:
Is Quicksilver running on Groq?

Leo Laporte [02:07:02]:
No, Quicksilver is running on DeepSeek Flash V4 over there on my, uh, on my Sparks. It had to run on Groq. Actually, I did— it was running on Groq this morning. Maybe that's what you saw. I don't know how you would see it, but it was because, because, uh, I did brain surgery on it. I saw— I stupidly saw somebody on—

Paris Martineau [02:07:22]:
You did surgery on a grape?

Leo Laporte [02:07:24]:
Yeah, I saw somebody on X say, hey, you know, the new Ornith model, which is based on Quen 3.5, is really great and you can run it on dual Sparks. So I said, hey, Quicksilver, take your brain out of your head and put this other brain in and tell me if it's better. After I did that, I realized, oh my God, that's what it's running on. So I fortunately—

Paris Martineau [02:07:46]:
Was it better?

Leo Laporte [02:07:47]:
No, it crashed the Sparks because it was just a little, It was too— a little bit too big. It was just a little too big. But you know what? Next week they'll make it a little bit smaller. It'll fit.

Jeff Jarvis [02:07:59]:
Do you have 256 gigs on each or total? The Sparks?

Leo Laporte [02:08:04]:
Each. No, total.

Jeff Jarvis [02:08:06]:
Oh, so it's 128 each. So it's 128.

Paris Martineau [02:08:08]:
Okay.

Leo Laporte [02:08:08]:
Yeah, yeah, it's 256 total. I have 128 on the Framework. I have 64 on the Mac. I have 24 on the— 24 CUDA RAM on my RTX 3090 on, uh, on my Alienware gaming machines. I have 4 different machines running local models right now. It's fun.

Jeff Jarvis [02:08:28]:
You wonder why it's hot up there.

Leo Laporte [02:08:29]:
It is so hot.

Dan O'Dowd [02:08:31]:
Does that come out to a terabyte?

Jeff Jarvis [02:08:32]:
You have a terabyte of RAM running right now?

Leo Laporte [02:08:36]:
Uh, 128, 256, 64, 128, 256, 512. No, it's, it's, uh, kind of almost 600 gigs. But they're not, but you can't, they're not all, you can't have one model showing them all. That's the problem.

Paris Martineau [02:08:52]:
Yeah.

Dan O'Dowd [02:08:52]:
Yeah. It's not, they're not parallel.

Leo Laporte [02:08:54]:
Yeah. Yeah. Uh, it's fun though. It's, uh, there's a lot going on. We are in very interesting times and the models are getting better and better and closer and closer to the frontier. And I would say probably by the end of the year, I will be able to run something that is probably Opus 4.8 quality locally. That will be an amazing day. Because remember, we thought how good Opus 4.8 was.

Leo Laporte [02:09:20]:
No, I don't need any heating. I wish it were winter.

Jeff Jarvis [02:09:26]:
Do you get winter there?

Leo Laporte [02:09:28]:
Blind Whiz, you know, I don't know how I'm getting this. I just noticed it appeared once. Maybe you have to say it. But you see, if you go to the top, I'm on x.com. I've got the normal For You, which is, that's a lot of Elon. And I've got following, and then I've got AI, Iran conflict, tech news, travel, food, politics, business, all these.

Paris Martineau [02:09:48]:
These are the interests that you've signed up for.

Leo Laporte [02:09:51]:
Yeah, they're called timelines. And they're about— so if you click the plus button up at the top there, you can choose some timelines. I have pinned artificial intelligence, Iran conflict, and tech news. I could probably take Iran conflict off because I think I think that's over. Um, but these are the other choices.

Jeff Jarvis [02:10:10]:
Darren is saying that you need to be an ex-Premiere user to get this.

Leo Laporte [02:10:14]:
Oh, I, uh, I have— I'm an involuntary blue check. That's probably why.

Jeff Jarvis [02:10:19]:
I got blue checked again. I got a notice saying you're being reviewed for a blue check, it'll take a while, and then 2 seconds later you have a blue check.

Leo Laporte [02:10:27]:
Yeah, you know, I—

Paris Martineau [02:10:28]:
of course, it's a certain amount of— if you have a certain amount of followers that have blue checks. it will automatically. So probably some people unfollowed you because of your tweets. And then because I guess you've been tweeting more pro-AI stuff.

Leo Laporte [02:10:40]:
No, I had 550,000 followers.

Paris Martineau [02:10:43]:
I'm talking about Jeff.

Leo Laporte [02:10:44]:
Oh, Jeff. No, I had more than half a million. And then I announced I'm not going to ever be here again. I'm leaving. And that didn't really do much. It's now 445,000 followers. I think it's the— yeah, it's the number. So Elon, at some point, they took away my blue check when I didn't give him any money.

Leo Laporte [02:11:02]:
I have the plaque still that says I had a blue check when it meant something. But then at some point, months ago, Elon decided to give some people back their blue checks. And you know what? That's whatever. It's— if it gives me that AI feed, it's— I don't know. It's worth it not to pay anything for it.

Jeff Jarvis [02:11:21]:
You get a lot of Gary Marcus, I see. That must make you happy.

Leo Laporte [02:11:24]:
Way too much Gary Marcus.

Jeff Jarvis [02:11:27]:
Wow, is he in there.

Leo Laporte [02:11:28]:
Uh, he's very active. He's very active. All right, you're watching Intelligent Machines. Guess what? Our picks of the week coming up next. Jeff Jarvis, Paris Martineau. How about your pick of the week, Paris Martineau?

Paris Martineau [02:11:45]:
Uh, my pick of the week is a website I just discovered called catfishing.net/game/today, which isn't as sketchy as it sounds, where, um, it's the Wikipedia guessing game.

Jeff Jarvis [02:11:58]:
Oh.

Paris Martineau [02:11:59]:
Every day you get to play a basically a guessing game where you get to guess what Wikipedia— you can guess. There's 10 different articles that they'll put up, and they will list the categories, which are the like macro kind of categories that a Wikipedia article is listed under. And so you Can— this one was for— the first one was for Afrikaans, but you are supposed to kind of look at it and figure out what the one could be.

Leo Laporte [02:12:27]:
I would say Dutch language dialects.

Paris Martineau [02:12:30]:
Yeah. So if you want—

Leo Laporte [02:12:31]:
I don't understand. How is this supposed to work? I'm supposed to click that button?

Paris Martineau [02:12:35]:
So click next.

Leo Laporte [02:12:36]:
Click next. Type the article title.

Paris Martineau [02:12:38]:
So that's your guess. So here's the categories.

Leo Laporte [02:12:41]:
Oh, I see the categories above. 1930s neologisms. Gambling in Russia, revolvers. Okay.

Paris Martineau [02:12:49]:
Suicide by firearm and torture. What do you think that is? Emilio's typing Russian roulette.

Leo Laporte [02:12:56]:
And I'm right.

Paris Martineau [02:12:57]:
And he's right. So he gets 1 point.

Leo Laporte [02:12:59]:
1970s horror novels, 74 American novels, 74 debut novels, English language American horror novels adopted for radio.

Paris Martineau [02:13:09]:
It also— so then you can kind of, with these ones that have a lot, you can go, it's a thriller novel. Eco-thriller. It was adapted into a film. It's nautical. Novels about animal hunting, about infidelity, about sharks. Novels by Peter Benchley set on Long Island.

Leo Laporte [02:13:24]:
Oh, Peter Benchley. Wait a minute. Oh, I missed that part.

Paris Martineau [02:13:29]:
Yeah, you got to look at all of them.

Leo Laporte [02:13:30]:
So was it Jaws? That's not Long Island. That's Nantucket. What's— I don't know. Crime? No, it's not crime. Come on, chat.

Paris Martineau [02:13:40]:
You wanna try Jaws?

Leo Laporte [02:13:42]:
Well, I'll try Jaws even though I know that's— oh, it was Jaws!

Paris Martineau [02:13:46]:
It was Jaws.

Leo Laporte [02:13:47]:
Okay. So, British astronomers of the 18th century, female?

Paris Martineau [02:13:52]:
Some of them are really hard. So, you can kind of scan it and see if this is like a scientist from Hanover who's a female. She won the Gold Medal of Royal Astronomical Society. I couldn't get this one. She was born in 19— 1750, died in 1848.

Leo Laporte [02:14:09]:
I don't know any female astronomers from the 18th century.

Paris Martineau [02:14:11]:
You could press the red button on the left if you just want to get the answer and you don't want to guess.

Leo Laporte [02:14:15]:
Oh, that's what that button is. Okay. Oh, Herschel!

Paris Martineau [02:14:18]:
Caroline Herschel.

Dan O'Dowd [02:14:19]:
Herschel!

Paris Martineau [02:14:20]:
Great portrait they got of her.

Leo Laporte [02:14:22]:
She's the sister of the very well-known William Herschel. But of course, because she's a woman, no one ever heard of her.

Paris Martineau [02:14:29]:
Yeah. So you can do 10 of these every day they publish.

Leo Laporte [02:14:32]:
This is fun!

Paris Martineau [02:14:33]:
It's really fun. So this one, it's a, like, 1980s dystopian film, came out in '82. It's cyberpunk, existentialist, shot in Los Angeles.

Leo Laporte [02:14:47]:
Oh, this one, that's pretty obvious.

Paris Martineau [02:14:47]:
Yeah, it's Blade Runner. It's pretty clearly Blade Runner.

Leo Laporte [02:14:51]:
Ooh, 73 songs, novelty songs, dance crazes.

Paris Martineau [02:14:55]:
So it's a dance craze. Songs about dancing. Songs from Rocky Horror.

Leo Laporte [02:15:03]:
Time Warp.

Paris Martineau [02:15:05]:
And so you can enter it in.

Leo Laporte [02:15:06]:
Or do I do— oh, maybe I have to do— let's do the Time Warp again.

Paris Martineau [02:15:10]:
It's called Time Warp. Time Warp is— yeah.

Leo Laporte [02:15:13]:
The article title would be Time Warp. Okay.

Paris Martineau [02:15:16]:
So the thing is, um, you can enter— it's case insensitive. Like, if you're within a couple characters, you can get it right. Like, it's very fun. Um, so you have to kind of guess it. Some of them get more. Yeah. It's a mythological duo, though. And so the thing is, one of the rules of it is it'll never have the word in the head— in the title, right, in one of these categories.

Leo Laporte [02:15:40]:
It's like Password.

Paris Martineau [02:15:41]:
Yeah.

Leo Laporte [02:15:42]:
Yeah.

Paris Martineau [02:15:43]:
It's fun.

Leo Laporte [02:15:44]:
Living people, American bloggers, television personalities from California, Canadian television hosts, people from Sonoma County, People from New York City, Emmy Award winners, amateur radio people, Berklee McIntosh Music Group members. I think I know who that is, Patrick.

Paris Martineau [02:15:59]:
I think I could figure it out.

Leo Laporte [02:16:02]:
Very cool. Catfishing, bad name, great site. catfishing.net, the game. Jeffrey Jarvis, what's your pick of the week?

Jeff Jarvis [02:16:13]:
All right, well, I could do a few things here. One is there's— I've got a new business model for you, Leo.

Leo Laporte [02:16:17]:
Okay, I'm ready.

Jeff Jarvis [02:16:18]:
I've got you a use for these Sparks that you bought.

Leo Laporte [02:16:21]:
Yeah?

Jeff Jarvis [02:16:22]:
A Beijing AI-themed bar, uh, offers tokens with the beer.

Leo Laporte [02:16:31]:
He's apparently— he's losing money though. That's the issue. It's the AGI Bar.

Paris Martineau [02:16:39]:
That offers unlimited free DeepSeek tokens with a $1.50 drink.

Leo Laporte [02:16:44]:
Yeah, of course Oh, you know what? He's running exactly the same way I am. 2 DGX Sparks. They're on display. How's he losing money?

Jeff Jarvis [02:16:55]:
Well, how do you get a Blackwell GPU in China is another question, but we'll leave that aside.

Leo Laporte [02:16:59]:
10 times more drinks are given away than sold. Wow, that's really— oh, oh, oh, you don't have to buy a drink. If you've got the Wi-Fi code, you can use the AI agent.

Dan O'Dowd [02:17:13]:
Right.

Leo Laporte [02:17:13]:
Now, I have to tell you, he's also losing money because it really can only do about 2 things at once. So, there's a lot of people just sitting there going, come on, come on.

Jeff Jarvis [02:17:24]:
So, the other one is, have you seen Okay Spielberg?

Leo Laporte [02:17:28]:
No, what's that?

Jeff Jarvis [02:17:29]:
Go to line 161.

Leo Laporte [02:17:31]:
Okay Spielberg.

Jeff Jarvis [02:17:31]:
And, um, it seems to be by a guy named Spielberg.

Leo Laporte [02:17:36]:
Okay, so now I have to press some button that's not—

Jeff Jarvis [02:17:39]:
Turn sound on. Yeah.

Leo Laporte [02:17:41]:
Oh, yeah, there it is. Unmute. Okay. Uh, here we go. Shall we play it?

Jeff Jarvis [02:17:46]:
Go ahead. I think you're fine. Okay, this is absolutely nuts.

Dan O'Dowd [02:17:48]:
I just uploaded a page from the screenplay for Forrest Gump into Spielberg's AI program.

Jeff Jarvis [02:17:53]:
I gotta show you this. So I uploaded this random page from the screenplay of Forrest Gump into OK Spielberg. The extreme Greenboro, Mrs. Gump, and Forrest walk across the street.

Dan O'Dowd [02:18:03]:
Mom said that the Forrest part was to remind me that sometimes we all do things that, well, just don't make no sense.

Jeff Jarvis [02:18:08]:
Forrest stops suddenly at this— his braces get stuck. His leg brace gets stuck in the grate.

Leo Laporte [02:18:15]:
Is this a real, uh— I don't think that happened in Forrest Gump.

Jeff Jarvis [02:18:18]:
That happened. I don't know.

Leo Laporte [02:18:20]:
This is real. Okay, so he's uploading a real script.

Paris Martineau [02:18:23]:
He's describing the real page that he's uploaded.

Leo Laporte [02:18:26]:
And then he uploads it to an AI that I presume he's made.

Jeff Jarvis [02:18:29]:
I think so, yes. Forrest pulls his foot out of the grate.

Paris Martineau [02:18:32]:
All right.

Jeff Jarvis [02:18:33]:
Can you bump that volume up a little bit?

Leo Laporte [02:18:35]:
That's just all the way up, unfortunately.

Jeff Jarvis [02:18:37]:
And notices the tooled men. So I went to the direct, chose Sea Dance.

Leo Laporte [02:18:41]:
Oh, it makes a video!

Jeff Jarvis [02:18:43]:
Yes. Hit action. Here's what I got.

Leo Laporte [02:18:46]:
Mama said that the forest part was to remind me that— Wow. Sea Dance is very good. I'm telling you, video generation, revolutionary.

Paris Martineau [02:18:57]:
Wait, wait.

Leo Laporte [02:18:58]:
But imagine what this— the implication of this is. You write the script.

Jeff Jarvis [02:19:02]:
This way.

Leo Laporte [02:19:03]:
Yep. It makes the movie.

Jeff Jarvis [02:19:04]:
Yep, exactly. Yes, Jack, but you're talking about like the importance of word choice. Like how many choices now?

Leo Laporte [02:19:12]:
It's not quite as good.

Jeff Jarvis [02:19:13]:
It's not quite as good, but, but I can't make a movie. I've got a movie I want to make. I can't make it.

Dan O'Dowd [02:19:18]:
I didn't do anything else.

Leo Laporte [02:19:20]:
That's pretty impressive.

Jeff Jarvis [02:19:21]:
It's impressive.

Leo Laporte [02:19:22]:
Yeah. Uh, he's spending, by the way, a lot of money on SeaDance tokens. SeaDance. Uh, there are some very good video generation models. Flux, SeaDance. Higgsfield. Um, they're really good. Uh, this is the real problem at this point is, uh, I mean, remember when— I think I mentioned this last week— remember when making a video of Will Smith eating a bowl of spaghetti and it was like all— this was a year ago.

Jeff Jarvis [02:19:53]:
Yeah.

Leo Laporte [02:19:53]:
It was all melty and he had 8 fingers and it was bad. And then the most recent one I saw was a bunch of Will Smiths in a bowl of Froot Loops and spaghetti was eating him and it was perfect. So I, you know, we are very much at a watershed moment for video.

Jeff Jarvis [02:20:12]:
I don't know whether you're coming to New York for the podcast festival, but South by Southwest is launching a podcast festival next year.

Leo Laporte [02:20:18]:
Oh, that we need because the one that happens now isn't great.

Paris Martineau [02:20:23]:
The guy who posted that TikTok is the creator of the tool.

Leo Laporte [02:20:29]:
Yeah. Here is a video of— I don't know. This is not the one I was thinking of. This is 3 years ago. Okay. This is what it looked like 3 years ago.

Jeff Jarvis [02:20:40]:
All right.

Leo Laporte [02:20:40]:
I'm just saying, Will Smith eating spaghetti 3 years ago. Not good.

Paris Martineau [02:20:45]:
I mean, this has artistic merit in my opinion.

Leo Laporte [02:20:47]:
Yeah. It's claymation. Yeah. Let me see if I can find the new one. Uh, the progression, uh, now see, the problem is it's brand new. We can see the progression over 3 years in this, in this Instagram, but it doesn't have the latest one. But this is Will Smith eating spaghetti 2023. It's not good.

Paris Martineau [02:21:12]:
It's still pretty bad.

Leo Laporte [02:21:14]:
It's not good. That's 2024.

Paris Martineau [02:21:16]:
I love that this is an Instagram reel Of a tweet.

Leo Laporte [02:21:21]:
And now 2025. Well, it's starting to look good.

Paris Martineau [02:21:27]:
2026.

Leo Laporte [02:21:28]:
Oh, this is actually— I've seen this. This is him and his son talking about eating spaghetti. I heard it can create multiple scene cuts like this with a single prompt. knows when to cut to whoever is talking. But did you know that all this audio was also generated with the same prompt? I, I guess I didn't, sir. Study harder, kid.

Jeff Jarvis [02:22:01]:
Spaghetti.

Leo Laporte [02:22:02]:
That's pretty good. We've come a long way. Baby, that is from Being Black is Lit. Black is Lit is what that's from, the account. Um, that's it. Do you have anything else, Jeff?

Jeff Jarvis [02:22:20]:
Uh, um, I went to the panel picker for South by Southwest for this year and I searched for agent. There are 172 suggested sessions on agent. There are 1,400 suggested sessions mentioning AI.

Paris Martineau [02:22:36]:
Wow.

Jeff Jarvis [02:22:36]:
And there are 2—

Paris Martineau [02:22:37]:
And how many sessions are there?

Jeff Jarvis [02:22:41]:
Well, not nearly that many.

Leo Laporte [02:22:42]:
That's the picker. So they— you got to get voted in.

Jeff Jarvis [02:22:45]:
Uh, and then there'll be far fewer. Yes.

Leo Laporte [02:22:48]:
Yeah. Um, so good. I'm glad they're doing, um, well, that— where will be the South by Southwest podcast conference? Will that be in Austin? Because I'll go to that.

Jeff Jarvis [02:22:56]:
Yeah, I presume so.

Leo Laporte [02:22:57]:
But I am coming out to New York. I have— well, I know Paris will be unfortunately, uh, she'll have packing up her nose.

Paris Martineau [02:23:05]:
That would not be great.

Leo Laporte [02:23:08]:
And I'm sorry, but that's— anyway, I've asked Henry if he'll let us do a meetup at Salt Hanks on Friday the 17th. He said he can't do it on the 18th because that's Benny Blanco Day at Salt Hanks, and apparently it's a zoo. But maybe we could do it the day before.

Jeff Jarvis [02:23:28]:
Wait, so come on.

Leo Laporte [02:23:29]:
It should be more of a zoo with Leo Laporte Day at Salt Hank's than Benny Blanco Day.

Jeff Jarvis [02:23:35]:
Friday is the— Friday is the 18th.

Leo Laporte [02:23:37]:
18th. So the 19th is Benny Blanco Day. But Henry has not yet given me permission, so stay tuned. We're gonna all have French dip. It's gonna be fun.

Jeff Jarvis [02:23:49]:
Oh, it's too expensive. Don't you pick it up for everybody. They buy on their own.

Leo Laporte [02:23:52]:
Yeah, well, maybe Salt Hank will do it. He's going to the US Open to do, uh, sandwiches for the, uh, Tennis guys. Really?

Jeff Jarvis [02:24:00]:
Yeah.

Paris Martineau [02:24:00]:
That's cool.

Leo Laporte [02:24:01]:
You, you saw the video of him getting Djokovic right in his— autographing his forehead last year.

Dan O'Dowd [02:24:08]:
No.

Leo Laporte [02:24:08]:
Yes. But I don't know where Henry got to be a tennis fan. It wasn't me. Maybe it was you, Jeff. I think you influenced him.

Jeff Jarvis [02:24:18]:
I like tennis.

Leo Laporte [02:24:18]:
You don't go to the US Open anymore.

Jeff Jarvis [02:24:20]:
Not anymore. No, no. We think it's— my wife thinks it's ruined.

Leo Laporte [02:24:23]:
Oh, just like Burning Man.

Jeff Jarvis [02:24:24]:
We're sold. It's, it's, you know. Yeah, we used to love to go to Burning Man.

Leo Laporte [02:24:28]:
You did?

Jeff Jarvis [02:24:29]:
No.

Paris Martineau [02:24:30]:
Could you imagine Jeff being in the desert for a week?

Leo Laporte [02:24:34]:
Totally.

Dan O'Dowd [02:24:34]:
No.

Leo Laporte [02:24:35]:
Totally.

Jeff Jarvis [02:24:36]:
I won't go on the beach because I get sock in my— sand in my socks, you know. No.

Paris Martineau [02:24:40]:
Yeah.

Leo Laporte [02:24:41]:
Ladies and gentlemen, we have completed this thrilling, gripping edition of Windows Intelligence Twit Tests. Intelligent machines. I'm failing. My, my— I need a new, um, LLM. Inserted right here.

Paris Martineau [02:24:56]:
One fun fact before we close the show, sorry, is that 22 years ago today Google was founded and this show used to be called This Week in Google.

Leo Laporte [02:25:04]:
Happy birthday, Google.

Jeff Jarvis [02:25:05]:
Happy birthday to you.

Leo Laporte [02:25:05]:
22. Wow. Just think.

Paris Martineau [02:25:12]:
We're sorry, uh, 27, 28, something like that.

Jeff Jarvis [02:25:18]:
It's—

Paris Martineau [02:25:18]:
we're, we're there.

Leo Laporte [02:25:20]:
I remember going when it was early on, we— I went to a lunch that Sergey and Larry sponsored for Tech Press. There was about 12 of us there to introduce us to Google.

Paris Martineau [02:25:32]:
Or sorry, it's the 22nd anniversary of their IPO. I'm sorry.

Leo Laporte [02:25:35]:
Oh yeah.

Jeff Jarvis [02:25:35]:
That makes sense. Yes.

Leo Laporte [02:25:37]:
Yeah. Um, and they gave us all a Brio train. You know those wooden trains kids play with? They have little magnets. And I wish I still had it. It had G-O-O-G-L-E.

Paris Martineau [02:25:51]:
Oh, that would look great on the thing behind you.

Leo Laporte [02:25:53]:
It would be such a classic thing. I don't know what happened to it. Probably my kids got it because they were still little. But, uh, yeah, I mean, it was such a— it was so new, nobody knew. I also remember going to see the Yahoo servers in a little storefront in Mountain View with Jerry Yang.

Jeff Jarvis [02:26:11]:
I had meetings at Yahoo with, with my boss, Steve Niehaus.

Leo Laporte [02:26:14]:
Gave me a little tour.

Jeff Jarvis [02:26:15]:
Yeah, I got the chief Yahoo business card.

Leo Laporte [02:26:18]:
Yep, been here a long time, kids. Long time. You know what I, I was thinking?

Jeff Jarvis [02:26:24]:
What did the Google IPO open at? Does anyone remember?

Leo Laporte [02:26:28]:
I'm sure it went very well. They didn't want to go public.

Paris Martineau [02:26:31]:
I'm looking at a Fortune magazine, uh, cover story from the IPO that says, is this company worth $107 $65 a share. So I assume around there.

Leo Laporte [02:26:42]:
They did not want to go public, but they were required to by Sarbanes-Oxley.

Jeff Jarvis [02:26:47]:
Too many stockholders, right? Shareholders.

Leo Laporte [02:26:50]:
Yep. And now, now of course the JOBS Act changed that and you can have as many as you want.

Jeff Jarvis [02:26:55]:
Of course.

Leo Laporte [02:26:56]:
Thank you everybody for being here. We do this show every Wednesday right after Windows Weekly, 2 PM Pacific, 5 PM Eastern. 2100 UTC. You can watch it live if you're in the club, of course, in the Club Twit Discord, but otherwise YouTube, Twitch, x.com, Facebook, LinkedIn, or Kick. After the fact, get a copy of the show from our website, audio or video, at twit.tv/im. There's also a YouTube channel dedicated to the video. Watch it for 1 second. That's all we care about.

Leo Laporte [02:27:26]:
And, uh, actually our favorite thing is if you download it, right? Get it, get it in your favorite podcast client, download it, and then you'll have it. You can listen whenever you want, and you can listen to it without anybody listening back. That's, that's what I like. Thanks for being here, everybody. We'll see you next time on Intelligent Machines.

Dan O'Dowd [02:27:45]:
Bye-bye.

Paris Martineau [02:27:46]:
I'm not a human being, not into this animal scene. I'm an intelligent machine.