Leo Laporte (2:27)

Yeah, Claude's the unicorn. Yep. I spent a lot of time up here in the attic with my new friend.

Jeff Jarvis (2:35)

Yep.

Leo Laporte (2:37)

In fact, I had a tweet that I wanted to read from.

Paris Martineau (2:42)

A tweet?

Leo Laporte (2:43)

A tweet. Tweet. Believe it or not, this was originally a tweet from Matt Schumer, who is a VC.

Paris Martineau (2:51)

By tweet he means a 5,000 word blog post that he wanted to read verbatim on air. Upset when Jeff and I said that that might be too many words to read in one go.

Unidentified Guest (3:03)

Hey, this is Leo. Could you give me your screen real quick?

Leo Laporte (3:07)

Oh, yeah. Otherwise you couldn't see the tweet that I'm showing you right now.

Jeff Jarvis (3:13)

It's more like a tweet.

Leo Laporte (3:15)

Tweet. This is Matt's Twitter account. He actually put it on his blog as well. But. And you know, I probably shouldn't read it because I attempted to get his permission to read it, but I have to subscribe to him to even talk to him. So I guess, I mean, I don't know what is the permission status? If it's on Twitter, I could probably just read.

Paris Martineau (3:36)

It's on his blog. He. It's definitely supposed to be read.

Leo Laporte (3:40)

Yeah, I won't read the whole thing.

Paris Martineau (3:43)

I can summarize it. He says the trends we're seeing in AI right now are very similar to being in February 2020 on the precipice of the COVID Pand. Everything's about to explode. Jobs are going to be decimated. Everybody should be really, really worried and everybody should be using AI to do everything they can right now. And that he thinks it's very, very important.

Leo Laporte (4:11)

Yeah. One of the reasons I wanted to.

Paris Martineau (4:13)

This is a take we've never heard before.

Leo Laporte (4:15)

Well, I know you've heard it from me over and over and over again. I think we are on and I think Matt thinks so. And a lot of the people who are really, you know, closest to this think we are on the precipice of something huge. One of the things, one of the.

Paris Martineau (4:30)

People, he's like, oh, well, you know, the CEO of Anthropic said that I could take 50% of white collar jobs. And I'm like, wow. The CEO of a major AI company is suggesting that his product could be used to take over 50% of white collar jobs. Sorry, you can continue.

Leo Laporte (4:48)

I fear that you're defending the ancient regime and that you are going to be shocked. Shocked, I tell you, when. When this all changes dramatically in the next few months. This is going to be a very exciting and I think very interesting year. And I think dramatically different. And I'll give you some of the data points. We can skip reading all of this. I'll give you some of the data.

Jeff Jarvis (5:16)

Read what you like.

Leo Laporte (5:17)

No, no, no.

Paris Martineau (5:18)

You read 4700 words.

Leo Laporte (5:21)

The only thing that I would say and the point of reading this was to get to the end, which is the advice part of it. And I think the advice is, look, you guys, you don't have to take it. Jeff, you're old. Paris, thanks for running that in.

Paris Martineau (5:41)

Jeff, you're zooming in from a bed right now.

Leo Laporte (5:43)

I'm old too, but I'm very excited about what's happening and I'm actually kind.

Jeff Jarvis (5:47)

Of relieved that I'm looking at the.

Leo Laporte (5:49)

Going to get to see a little bit of this. I don't know about 50% of the white collar jobs because I do think there will be white collar jobs. There is a role for humans in this. I think the difference is what the role is. And this is in so many areas. He talks about legal work, financial analysis, writing and content. He does include journalism and this software engineering. That's obvious. What we're seeing already is happening in software engineering.

Jeff Jarvis (6:19)

That's Leo. I think that's the key to it. I think that the people making all this thought they were going to affect everybody else's jobs. And the last two months, this is his thesis. So listen, in the last two months or so I think we've seen with what you have shown us, especially with Claude, is that it's decoders who've woken up first and said omg.

Leo Laporte (6:40)

And his point, Matt Schumer's point in this is that these companies made these decisions to focus on code for a good reason. Because the first thing you do is you get it writing its own code. Then it can self improve, then you can focus on all of these other things.

Jeff Jarvis (7:01)

In fact, I don't think it's more reliable for them. I think that it's less than just a general machine.

Leo Laporte (7:07)

It's the low hanging fruit. But it also is the engine that powers a transformation in every other realm. And I think it would be a mistake for you to say to yourself, well, I'm safe, it's going to be okay. Or you know, you know, one of the things he points out is people using even models from two months ago. Well, it's not that good. It hallucinates, it's not that good. Are missing the real change that's happening right now. So that's, I won't belabor it. The most important thing that he talks about here is what you should do. If you don't buy this, you should keep doing what you're doing. You might be making a mistake. You might be right.

Jeff Jarvis (7:50)

No, we should all change to some level.

Leo Laporte (7:53)

If you Buy this. Then there are things you should do. One of the things is something I mentioned to you, Paris. I know I was overheated last week, but it's a good thing. He says start using AI seriously, and not just as a search engine. Don't just use perplexity. Sign up for the paid version of Claude or ChatGPT. The 20 buck month is fine to start with. He says two things matter right away. Make sure you're using the best model available. In this case 5.3 with OpenAI and Opus, 4.6 with Claude. These apps often default to a faster, dumber model. Dig into the settings of the model picker, select the most capable, and then he says if you want to stay current on which model is best, you can follow him. Okay. He tests them. I think there are plenty of places you can go to figure out what.

Jeff Jarvis (8:43)

The best is like right here. Leo will tell you.

Leo Laporte (8:46)

Yeah, yeah. And I'll tell you what the best is. That's right. He says, don't just ask it quick questions. That's the mistake most people make. They treat it like Google, like a search engine, and then wonder what the fuss is all about. Push it into your actual work. If you're a lawyer, feed it a contract and ask it to find every clause that could hurt your client. Now, I think sometimes people in a defensive mode, if you're a lawyer in a defensive mode, you're going to say, I'm not going to do that. Look at how many people got in trouble by, you know, providing pleadings to the court that were full of fake references and so forth. No, no, just, just kind of take.

Jeff Jarvis (9:26)

A leap of faith. That's fine.

Leo Laporte (9:28)

Take a leap of faith. Don't file it in court.

Jeff Jarvis (9:31)

Don't file in court, but have it.

Leo Laporte (9:33)

Give it something hard is the point. If you're in finance, give it a messy spreadsheet. Ask it to build the model. If you're a manager, paste in your team's quarterly data. Ask it to find the story. The people who are getting ahead aren't using AI casually. They're actively looking for ways to automate parts of their job that used to take hours. That's the best way to judge it, by the way. Something you already have domain knowledge, expertise in. And then he said, this might be the most important year of your career. Work accordingly. I don't say that to stress you out. I say it because right now there's a brief window. I think this is really true. I believe it. You guys may not. Where most people at most companies are still ignoring this. The person who walks into the meeting and says, I used to use AI or I used AI to do this analysis in hours instead of three days is going to be the most valuable person in the room. Not eventually right now. Learn these tools, get proficient. I would extend this. You know, we've been telling for years, mistakenly, it turns out parents to get their kids into coding camp, get them to, you know, if they're interested in computers, have them learn computers how to learn to code. I would say these days, if you have a smart 16 year old or even a 12 or 13 year old, buy him a $20 account with Claude. You might say, oh no, you know, I don't. He's going to get depressed and it's going to be bad. You know, if you need to do it with them. In fact, that'd be a great thing. You'll be amazed because the kids don't have the same preconceived notions just as I, when I was a kid learned, as most of you who are watching learned to code, you know, by playing with BASIC and typing in programs and stuff. Get them playing with this now because the skills that they learn are the most important skills are the skills of how to direct these AIs. And that's why I don't think all these white collar jobs will necessarily be lost. They'll be, they'll may be lost to the people who know how to do this though, the people who know how to manage and direct AIs. He says if you're early enough, this is how you can move up, by being the person who understands what's coming and can show others how to navigate it. But that window won't stay open long. Once everyone figures it out, the advantage disappears. And I would say if you're, if you've got a young person, a college kid, a high school kid, a smart kid who's interested in technology, give them a leg up because they're going to need these skills. They're going to need these skills anyway. We don't have, I didn't have to read all 5,000 words. I just think we are in a very, the most disruptive era in technology I've seen. Absolutely. And it's happening very, very fast. And I have some data points which we'll talk about after the break that will give you some idea of how quickly this is happening.

Paris Martineau (12:18)

So inspired by your, I mean this whole post, I put it Into Claude, Opus 4.6 extended and asked. Well, I said Leo, who is practically an AI accelerationist, at this point is reading this out loud today and asked us prepare rebuttals in the spirit of this overly simplified, in my opinion, somewhat naively bullish blog. Do your worst Claude and wrote a very very long research report that I'm not going to bore you guys all with, but there are a couple of points I will shout out here. One is just because this is we just talked about this is the seventh claim the advice like spend an hour a day experimenting. Claude writes. This is where the post quietly undermines itself. If AI were truly about to do to white collar work what Covid did to in person dining, which is the explicit comparison made in this blog, then spent an hour a day experimenting is laughably inadequate. You don't tell someone to casually experiment with pandemic preparedness when the pandemic is two weeks away. The modesty of the advice contradicts the extremity of the prediction. What Schumer is actually describing is stripped of the apocalyptic framing is a technology that is very useful, improving quickly and will probably change a lot of jobs over the next decade, which is correct but not novel and not like a Covid level giant event that is imminent. I mean we've had a version of this discussion saying like these giant world changing events are two to six months out so many times over the last year or two of doing the show. And then one the last point in opus write up of this, which I think is it's titled the Metacritique Use this. If Leo pushes back, which I think is just a very funny thing, writes Schumer's post is a genre piece. It's the I need you to understand what I understand post. Personal revelation, exponential trend extrapolation, dire warning call to action. This genre reoccurs with every major technology wave. The people who wrote the equivalent post in 1995 about the Internet were right about the big picture and wrong about almost every specific prediction. The people who wrote it about crypto were mostly just wrong. The question with AI isn't whether it matters, it obviously does, but whether the specific doom and urgency framing is warranted by the current evidence. And the current evidence from this week alone, because I've put this in a chat where I also then prepared a briefing book based on everything that's in our rundown. So Claude writes is that AI benchmarks don't predict real world outcomes, which we get into a Nature Medicine piece down there. That AI agents violate their own ethical constraints when pressured, that AI productivity gains intensify work rather than reducing it, which is A piece from Harvard Business Review that AI attributed layoffs are frequently reversed and that AI's own creators can't even reliably predict what it will do next. This is not. This seems overblown. That's the ground is shaking and we don't yet know whether it's an earthquake or a volcano, which I think is a pretty, I don't know, measured counterpoint.

Leo Laporte (15:37)

Yeah, actually, it's a good analysis and.

Paris Martineau (15:39)

It'S written by Claude, so you can't get mad at me because you believe everything that Claude says. Because Claude's right.

Jeff Jarvis (15:45)

Well, Claude's evil, too.

Leo Laporte (15:45)

I believe everything that Claude says. So. Okay. I would suggest. I think it'd be more valuable for you, Paris, to actually give it something to do that is serious, like something that really makes sense.

Paris Martineau (15:59)

I do that.

Leo Laporte (16:00)

Okay. So I think that's the opportunity. You know, one of the things Claude is really good at is financial analysis. I know that's one of the things you need to do, really give it something that's hard and see how it works. One of the things 4Six does. And we're going to talk about this a little after the break. Oh, I should have mentioned. I didn't mention we don't have an interview this week. That's why I was going to read this 5,000 word piece next week. Guy Kawasaki will be joining us, so that'll be fun. Apple Evangelist. He's got a new book and it's actually a very appropriate book. It's how to protect your privacy and how to use secure messaging. It's a very timely book. And I asked him, did AI write this? He said no. So that will be fun next week. But this week I thought we could talk about this topic because I'm going to read the last bit on this because I think it's really important to hear it. And if you don't want to hear it or it doesn't make sense to you, that's fine. But I would encourage you to listen to this with an open mind. Not you particularly. I'm not speaking to you, Paris, or Jeff, necessarily, but our listeners. I know that he writes this is again, Mike Schumer. I know this isn't a fad. The technology works, it improves predictably, and the richest institutions in history are committing trillions to it. I know the next two to five years are going to be disorienting in ways most people aren't prepared for. This is already happening in my world. It's coming to yours. I know the people who will come out of this best are the ones who start engaging now, not with fear, but with curiosity and a sense of urgency. And I know that you deserve to hear this from someone who cares about you, not from a headline six months from now when it's too late to get ahead of it. We're past the point where this is an interesting dinner conversation about the future. The future is already here. It just hasn't knocked on your door yet. It's about to. And I would say I agree with this 100%. And I think, listen to the people, you know who are really on the cutting edge of this, who are really using this, because everybody I talk to who is really using Claude code agrees with this, says, you have no idea how this is changing everything now.

Jeff Jarvis (18:22)

So can I come in now?

Leo Laporte (18:23)

You can disagree with it, and that's fine, but that's my position and I've been doing this for 40 years. That's all.

Jeff Jarvis (18:31)

So I think there's a difference in the last few months, which is that LLMs were seen as textual tools or toys. I think even the. Even the technology people saw it as something cute and interesting. Then especially anthropic shifted, though. This was their emphasis for quite some time, but their success rate at dealing with code shifted particularly. I think we have now a bifurcation here. There's a consumer market that's still out there and waiting, that is delayed now. The code market is where the progress is. And the coders who at first thought, oh, this is going to affect others are saying, omg, this affects us greatly. And you got to listen to us about how much this affects us. If it affects us as much, it's going to affect all of you this much. Well, to some extent that's true. To some extent, code is easier because it's its own domain. It's more finite in what it does, and it's what the coders understand. You having write amazing applications even in a language you don't use, and having IT work is very impressive. But that doesn't necessarily extrapolate to everybody else in every other field. And the market has confused this too, because in a sense, you look at the last two panics, one around legal and the next one around financial stuff. The fact that everybody's stock went down kind of makes no sense either.

Leo Laporte (20:02)

Wow.

Jeff Jarvis (20:03)

Everybody who's doing this is going to be far more efficient. So their stock should go up because the productivity is going to go up or, oh my God, they're going to get. These service companies are going to get eliminated, but all their customers are going to be able to do this on their own and their stock should go up instead. There was a generalized oh moment all around the market, and I don't think we don't have the data, we don't have the experience, we don't know where this goes. I think that when this gets really impressive, and this is where you and I have disagreed, Leo, is not when you can be in terminal mode doing something impressive, that's impressive. I'm not arguing with that, but I think it scales when I can do this without terminal mode and have it do it. Jason Howell and I were talking earlier today. You get to a point of looking at disposable software, disposable code. It does something for me once and it's gone. It changes the value of code immensely, it rethinks things, but it doesn't mean that everybody out there in the world is going to make all of their own code. I can't make a medical application. I can't make a financial application. I can use it to do things. I can use it to analyze, and it's going to have an impact, and I'm not arguing with that at all. So I think that what Paris said in her analysis is right. In 95, the long term was right, the short term was wrong, and we'll go back to steam power, electricity, the amplifier, the transistor. These are all similar things. We've been here before. These things had immense, immense impact on society and on jobs and on the economy. This stuff can very likely have such immense impact. But to act as if we've arrived, suddenly, boom, and we're there, I think is a bit naive because we don't know what it is yet. We don't know how it's going to get used yet. We don't know what it can really do well, yet. Should we use it? Is he right? Yes, we should. Should we do more than experiment with it and challenge it? Absolutely agree. But is the moment for triumphalism that this changes everything? I don't buy that.

Leo Laporte (22:02)

I think the biggest mistake you can make in this is, say, it's like anything else that's ever happened before. And I think that that's the analysis.

Jeff Jarvis (22:10)

That's the hubris of the present tense.

Leo Laporte (22:12)

That's the analysis both of you are adopting and most people adopt, because, of course, that's what we do as humans. We analogize. It isn't.

Jeff Jarvis (22:21)

History, teaches lessons.

Leo Laporte (22:23)

It isn't going to be like that. It's not going to be like, that's fine. I Will take hubris. I'm just telling you. I'll get off my call.

Paris Martineau (22:32)

Nine percent the time, which is that current and future trends are going to be similar to trends we've seen in the past. That is always, almost always going to be the correct bet. And I think it was gigantic.

Jeff Jarvis (22:46)

It has changed gigantic. Imagine all that happened because of the transistor. Right?

Leo Laporte (22:53)

Okay. If you sat. I just warned you. If you don't want to believe it, you don't have to believe it. But there is a tsunami.

Paris Martineau (23:00)

Why do you feel so personally offended?

Leo Laporte (23:03)

I'm not. I'm trying to tell you something. If you don't want to hear it, you don't have to hear it. There is a tsunami coming. It is going to be massive and it's going to happen this year. And if, if, if.

Paris Martineau (23:15)

You said that last year also, though.

Leo Laporte (23:18)

Well, it's been. It's. And it's accelerating. I mean, look, it was pretty clear last year that we weren't there yet. I think.

Paris Martineau (23:25)

Not according to you in this podcast. Early last year.

Leo Laporte (23:30)

No. Well, I mean, I could see this coming, but it really is coming a lot faster. It really is. We're going to take a break, we'll come back, and I will give you a couple of data points, you know, for why I think 4, 6, for instance, is a remarkable shift. We are talking about something very exciting, whether it's going to change the world or not. We could disagree. That's Intelligent Machines with Paris Martineau. Jeff Jarvis, brought to you today by a new sponsor. Want to welcome Modulate. This is actually a really cool product. Every day, enterprises generate millions of minutes of voice traffic. I'm talking things like customer calls, agent conversations, even fraud attempts. Most of that audio is still treated like text, flattened into transcripts, stripped of tone, intent and risk. Modulate exists to change that. First proven in gaming. This is where Modulate started. Modulate's technology supported major players like Call of Duty and Grand Theft Auto in separating playful Banner. We've talked with Paul Thurat about how, you know, you get on, you got your headphones on, you're playing Call of Duty, and, you know, the trash talk begins. Some of that's just harmless playful Banner. Some of it is intentional harm at scale. And that's what Modulate has been doing for those companies, helping them distinguish the two. Today, Modulate helps enterprises, including Fortune 500 companies, understand 20 million minutes of voice every day by interpreting what was said and what it actually means in the real world. This capability is powered by modulate's newest model, Elm. Velma 2.0. It's the newest Elm. That's what they call it. Velma. I Love the name 2.0. Velma is a voice native behavior aware model. This is really important. It's an elm, a voice native behavior aware model built to understand real conversations, not just transcripts. It orchestrates 100 plus specialized models, each focused on a distinct aspect of voice analysis to deliver accurate, explainable insights in real time. Velma ranks number one across four key audio benchmarks, beating all large foundation models because it's designed to do this right, specifically in accuracy, cost and speed. It's number one in conversation understanding. It's number one in transcription accuracy and cost. It's number one in deep fake detection. And then the final one is the hardest one. And it's really good at that. Number one in emotion detection. Built on 21 billion minutes of audio, Velma is 100 times faster, cheaper and more accurate than LLMs at understanding speech. Including the best Google, Gemini, OpenAI and XAI. Most LLMs, they're just black boxes. Velma doesn't just assess a conversation as a whole, but breaks it down for greater accuracy and transparency by producing timestamped scores and events tied to moments in the conversation. Meaning you can see exactly when risk rises, when behavior shifts or intent changes. With Velma, you can improve your customer experiences, reduce risks like fraud and harassment, detect rogue agents and more. Go beyond transcripts and see what voice native AI model can really do. Go to Modulate's live ungated preview of Velma. Preview Modulate AI. That's Preview Modulate AI to see why Velma ranks number one on leading benchmarks for conversation understanding, deep fake detection and emotion detection. Preview Modulate. Thank you Modulate for joining us. I think that's the other side of what we were just talking about. We're going to see more and more sponsors and ads and stories talking about technologies and capabilities well beyond what we kind of think of when we talk about.

Jeff Jarvis (27:39)

I also love the brand Velma.

Leo Laporte (27:41)

Isn't that great? Wasn't it?

Jeff Jarvis (27:42)

That's great.

Leo Laporte (27:43)

She was the smart one in Scooby Doo, right? Velma had the glasses, I think. So that's where that comes from. Yes. So the couple of things happened last week and my experience with 46 was immediately very impressive. It just came out the day we did the show last week and I mentioned one of the things I did was I went through 46 PDFs of real estate contracts. I was looking for one little tiny piece, the proverbial needle And a haystack. And four six did it what previous models would not have been able to do. A part of the reason 46 is good at this is it has a massive 1 million token context window. That's one of the biggest changes. Anthropic. When they released 46 and I kind of at first pooh poohed this, announced that they had put 4:6 to work writing a C compiler, a Rust based C compiler from scratch. It cost $20,000 in API costs, 2,000 Claude code sessions. They wrote a 100,000 Line C compiler that was able to build the current Linux kernel on three models X86 ARM and RISC V, successfully build it now. At first I thought, well, okay, but it's not a fine tuned C compiler. It's not as good as GCC or the state of the art gcc. That wasn't the point of it. As it turns out. What they were really testing was the ability for Claude to run unattended for a length of time. The previous records had been half an hour, then an hour. I mean last year we were happy if Claude could run half an hour without hallucinating or breaking down. This went for 2 weeks non stop unattended. 2 weeks to build 100,000 lines of code now and 20,000 sounds like a lot of money, but honestly the teams that it would take, the cost it would take to really write a C compiler, be much, much higher than that. So I think that what this is one data point I want to give you, which is the rate of improvement is not linear. The rate of improvement now is starting to hockey stick. This is a massive rate of improvement. It's a massive number of tokens. You know, it's 5x the tokens of the previous model. So we're starting to be in a situation where we're accelerating improvement. Another data point and then I'll let you talk one more data point. They also sat down with Opus 4.6. They gave it in a sandboxed environment. They gave it a couple of basic tools, they gave it Ghost script. No, I'm sorry, Python. Is that me or was that you? Somebody come in here.

Paris Martineau (30:34)

Okay, it's Claude.

Leo Laporte (30:36)

Claude wants to get in on this.

Paris Martineau (30:38)

The call is coming from inside.

Leo Laporte (30:39)

He gave it some very basic tools. Access to Python and vulnerability analysis tools, debuggers and fuzzers. No specific instructions or specialized knowledge. Gave it access to a bunch of open source tools that are widely used by millions. Claude was able to find more than 500 previously unknown vulnerabilities just out of the box. These were flaws, unknown flaws in for instance, Ghost script, which almost all open source systems use to process PDF files, buffer overflow flaws and OpenSC in CGIF. This is a breakthrough, the ability to find. This is finding holes in code that has been looked at, vetted by security people. This is not just random open source projects. These are widely used open source projects that have been carefully vetted for years for security flaws. They found 500 of them. That's another amazing use. Now, both of these cases, I know that well, it's code, right? Big deal. Okay, makes sense that this would be good at code, right? But you got it. Everything's based on code nowadays. You got to start with code. If you can have Claude self improve, if you can have Claude accelerate its improvement rate, it get better and better. It's going to get better and better at everything eventually. So that's my answer to what you said before the break, Jeff, is yes, it's not impinging yet. It's not writing novels or operas or anything else human and creative, but it's improving so fast at code. I think that's next. All right, I'm a stop. I'm going to stop. I might get down off my, as.

Paris Martineau (32:29)

I said, my Paul Revere Moore's Law for intelligence. It's like that assumption, like, well, that.

Leo Laporte (32:36)

Would be a big deal. Moore's Law for Intelligence would be a huge deal.

Paris Martineau (32:40)

No, but it's like that's the assumption you're making. Like it's unfalsifiable by design. Like any plateau can basically be dismissed as temporary because you're like, oh, it's just always, the line's always going to go up.

Leo Laporte (32:54)

Okay. It's not really a rebuttal. That's just saying, no, it's not going to happen. But I have evidence it's happening. I mean it is happening. This is massive improvement over very over two months.

Jeff Jarvis (33:07)

I watched an hour long presentation by Yann Lecun this week at a world model conference and he's doing his thing. And of course he argues that LLMs, though amazing, do hit a wall. And that's part of what the argument is in the post he read is what is it? Whether there's a wall or not. But he said one line really struck with me and this is his argument in favor of world models, in favor of, of building many intelligences is the way he puts it. He said, you can't tokenize the world. And so he tries to deal with something that's much bigger. He tries to deal with concepts rather than the next pixel, rather than the next token and the next word. And so I think there is something bigger, much bigger that can be out there. His models are yet unproven. His theories are yet unproven. Well, they've proven to an extent, but to that level, I don't think this is the funny thing about this debate we go through in the show. I think I can speak for Paris and saying, we're all amazed by these tools. Paris uses them, he's amazed by them. We're not denigrating the tools. The only question we have is that extrapolation and that present presentism of believing that somehow this is bigger than Steam and bigger than the transistor.

Leo Laporte (34:27)

You think that's impossible?

Jeff Jarvis (34:30)

No, I'm not saying it's impossible.

Paris Martineau (34:31)

I think it's unlikely. It's the.

Leo Laporte (34:32)

Okay, it's unlikely. I agree. It's unlikely.

Paris Martineau (34:35)

Deserve to be. I don't think it. I think that it is unlikely and thus it does not deserve to be the predominant narrative in every. What if all the evidence points to technology.

Leo Laporte (34:46)

They're happening.

Jeff Jarvis (34:48)

Well, I think we could argue that the transistor was that, that indeed the trans. If you, if you go back to 1920 and then try to imagine the transistor, this, this thing this big with a huge hunk of silicon that it was then, now there are trillions in a given. In a given rack and what all it can do you consider all of the changes that came from, from Steam or the transistor. All right, I don't need to push.

Leo Laporte (35:14)

Against you on this. It's fine. We have differing opinions. I don't need to push back on it. Well, no, if I'm wrong, it'll be pretty obvious at the end of this year that nothing happened.

Jeff Jarvis (35:29)

We're not saying nothing's gonna happen.

Paris Martineau (35:30)

Yeah, we aren't. We aren't arguing that nothing is going to happen. We're just saying there should be some, like, measured skepticism of technologies.

Leo Laporte (35:41)

Especially when I have accelerating of this show of stuff that's gone wrong with AI. You'll be very happy for that section.

Jeff Jarvis (35:53)

Are you using AI Doctor? Can you talk to me about that first?

Leo Laporte (35:57)

Actually, one of the things I need.

Paris Martineau (35:58)

To talk about, the medical research, one.

Leo Laporte (36:01)

Of the things I did, and there's a new app that came out for the iPhone that finally extracts all the data from Apple Health. Apple Health has been collecting all this data. Apple has been saying, oh, we're going to do. Or they're rumored to be doing some sort of AI Health thing. And they've backed down on that, apparently. Mark Gurman, who is the Apple rumor monger, says, yeah, they decided they weren't ready to do that yet. And it was a little frustrating because Apple Health has this huge amount of data. Not only data, it's collected from all the weird devices I wear and my scale and my blood pressure meter and my continuous glucose monitor, my aura ring, but it also has all my medical information because I've uploaded all of my medical records from my health provider to it. So I found an app that actually extracts it all and puts it on markdown files. So now I have six years of day by day information about how many steps I took, how much sleep I had, what my, what my blood pressure was, all of this stuff. And I can't wait to, to massage this with AI because I think there might be some very interesting insights there. There's a lot of good data there.

Paris Martineau (37:07)

Did you read the story about the Washington Post reporter? I don't know whether it's one of the ones that has since lost their job. A Washington Post reporter, Jeffrey Fowler.

Leo Laporte (37:15)

Yeah.

Paris Martineau (37:16)

And it told him that he was like, had terrible heart health and needed to see a doctor immediately. And then a doctor reviewed the same data and was like, no, that's totally wrong.

Leo Laporte (37:24)

That was GPT, the GPT health app that he used. And that was Jeffrey Fowler who is now out of work. But I hope his heart is fine. I hope his ticker is fine.

Paris Martineau (37:33)

According to the doctors, it is.

Leo Laporte (37:34)

So let's talk about investment, because one of the things we saw quarterly results from a number of companies. Amazon is increasing its investment in anthropic to $61 billion, which is a seven fold increase. That's interesting. I wonder if Amazon's thinking maybe our models and our work isn't going to be that successful. Let's pick a winner in this horse race.

Jeff Jarvis (38:00)

Well, Amazon wants to cut down on labor. Obviously they're doing that. A, B, I think that Anthropic. Everything you're saying points to Anthropic being the winner right now.

Leo Laporte (38:10)

I'm not sure that that's going to always be the case.

Jeff Jarvis (38:12)

Not set to be. Not OpenAI.

Leo Laporte (38:13)

Yeah, yeah. It's not clear. OpenAI's 5.3 is very good. Codex is very good. So in fact there are a number of people who think it's a better coder.

Jeff Jarvis (38:23)

I think OpenAI is stuck in the chat motif, whereas Claude is saying, no, this is about making things, this is about doing things. That's a step change.

Leo Laporte (38:34)

Software, computer and data center spending is now over $1 trillion a year. That's Defense Department budget numbers right there.

Jeff Jarvis (38:45)

Compare that to the spending on fiber in the early days of the Internet. Not in dollar amount.

Leo Laporte (38:51)

Well, dollar amount succeeds it huge.

Jeff Jarvis (38:54)

I know that. But in useful useful at the time versus used later.

Leo Laporte (39:00)

I think there's a reasonable race right now to be because I think this is the belief is there's a lot of people who think the way I do that this is going to be so transformative that you want to be at the forefront of this because the upside is so huge. And certainly investors are thinking that spending on data centers construction is now $42 billion. That's a 300% increase since the launch of ChatGPT just four years ago. Hard to believe that was only four years ago. The only reason that that growth has slowed a little bit is it can't get enough people to build these things. Google has announced that it's going to spend a huge amount of money. Its capex on AI will go from to 175 to $185 billion this year, double what it would spend last year.

Jeff Jarvis (39:58)

And they sold century bonds and they.

Leo Laporte (40:01)

Sell bonds for long. Boy, does it not mature in 100 years. I didn't really follow up on this because I thought, well, I'm not going to be around for 10. So 100 is. My kids won't be around in 100 years. My grandkids maybe will be around in 100 years.

Jeff Jarvis (40:19)

Most companies in the Fortune 500 in the 1980s they lasted 50s, they lasted 60 years on average. Now they're 15 to 20 years.

Leo Laporte (40:29)

Right? Right.

Paris Martineau (40:31)

100 year bond is insane.

Leo Laporte (40:34)

It's insane. Google's Gemini app is now catching up very quickly. Remember we mentioned that no. OpenAI had 800 million monthly active users. Gemini now has 750 million monthly active users. Not that OpenAI has stopped growing. Pause that. Sorry about that.

Paris Martineau (40:54)

Okay, my question with this is how are they? What are they counting as a user? Do I count as a monthly or daily active user of Gemini? Because every time I try to write an email, it tries to write my sentences for me or suggest weird grammatical changes to the things I've already written.

Leo Laporte (41:15)

I think, honestly I agree with you. I think both Google and Microsoft falls in this camp too are making a huge mistake forcing AI on people constantly offering to do something when we open our search.

Jeff Jarvis (41:26)

Where is Google? Leo on coding versus Claude and I.

Leo Laporte (41:32)

Can'T speak on that.

Paris Martineau (41:33)

He can't go against his girlfriend. She's in the room right now my.

Leo Laporte (41:36)

Girlfriend would get very. No, I can't speak on that because I haven't used it. What I've heard from other coders is that the two strong coding options are ChatGPT5.3 Codex and Claude Code and they are neck and neck. Google's models are great for. Well, for instance, when I make an image I always go to Gemini and Nano Banana. Always. It's by far.

Jeff Jarvis (42:00)

I think there's a split coming in. B2B encoder versus consumer.

Leo Laporte (42:05)

Personally I have no knowledge of this, but I think Anthropic and OpenAI would be smart to focus on coding because ultimately that benefits everything. I think Anthropic has really said they're too multimodal. They don't do. They don't. You know, you don't. Would never go to Claude to generate an image. I don't know if you even could. They say we're going to be good at coding and I think that's the smart bet.

Jeff Jarvis (42:28)

Smart.

Paris Martineau (42:29)

I have thought it's interesting that I've started to get a significant amount of targeted ads on Reddit from OpenAI specifically being like if you've been using Claude code, you should be using OpenAI codecs instead. It's way better. I thought that's a very. I mean of course that's the sort of demo they're trying to pull.

Leo Laporte (42:47)

They kind of need to say that because I think people are so enamored of Claude code that they're not even open minded at what 5.3 can do. So I don't know. Honestly, I'm not motivated to do it. I think one of the things that people are doing that's interesting that maybe I'll try is using both. So using claude. But then, you know, in the new CLAUDE they've added two really interesting features. They have a FAST mode which spends tokens at six times the rate. If you're in a hurry, that might be the way to go. I don't think I can afford that one. And then there's also something that you have to turn on manually in the JSON settings, an agent mode that lets you spawn multiple agents. So you have a. This is what we were talking about with Steve Yegi. Basically what Gastown did last week is it has a boss and then multiple agents and they can use dumber models. And I. And I think you could. I don't know if you could do it with their official release, but you can with others, even use other models. You could use open source models which are free or cheap. You Could I think in theory use codecs as well. So I think people are doing some interesting things with that I haven't played. I, you know, one of the things that you run into almost immediately is there's so much to absorb and so much to do that I can't, I can't do it all. I'm, I'm, I think I'm doing what I really want to be doing, which is going deep, as deep as I can with one particular tool, just to get a sense of where the outlines are.

Jeff Jarvis (44:23)

Listening to you and listening to Steve, last week on Gastown, it really resonated with that HBR piece I put into the rundown that shows that because it looked at 200 people in a number of companies, major companies in the US watch their work a couple days a week, major interviews with them and what they found was more intensity of work. That's what I hear from both of you, is that it's not necessarily.

Leo Laporte (44:52)

I have a hard time sleeping because I'm so excited that I actually wanted. And I have a couple of times this week leapt out of bed in the middle of the night to go. To go try.

Jeff Jarvis (45:01)

Well, that's the other thing it said. So it said that the work is much more intensive. The second thing it said is that it's, it's. And this was puzzled me but based on what you just said, you just proved it. It blurs the line between work and. And life.

Leo Laporte (45:14)

Darren Okey and I Discord earlier said the best game going right now, he was talking about video games is Claude code. It is fun. It is enjoyable to do it. It's really exciting. So that's. Yeah, that's part of it. I don't think it's burnout, it's enthusiasm. No, no enthusiasm. I had when I was a young man, first got my first personal computer and started writing basic programs. Your eyes get very wide and you say suddenly I have some power to do something that I didn't have before is really dramatic. One of the. I have a couple projects I really want to work on. One is. And you're going to see the immediate benefit of this. But between now and so I, as some of you know, I already vibe coded three programs to prepare for the shows. One of them is to scan the news. Another one is to prepare, get all the links and prepare a briefing. It does an automated summary using haiku, Claude's haiku anthropics haiku model to summarize the stories and then we output a website and stuff like that. So all of that I wrote. But one of the things I haven't delved deep into, but I will and you'll see the benefit of this is the summary. I've been using a simple prompt for the summary. I'm just saying find one quote from every story that is seminal if and put that at the front and then come up with five bullet points from the story.

Jeff Jarvis (46:44)

That's what every news organization is doing now and they're generating those summaries of the tops of stories.

Leo Laporte (46:49)

Right. I think what I wanted CNBC or wherever is the. Something like, you know, there's that Axios model. Oh, I know, not necessarily that.

Paris Martineau (46:59)

What Axios model are you talking, you.

Leo Laporte (47:01)

Know what those three bullet points, the smart brevity. But they have important. Why this is important. So I want to come up with something like that why this is important. You know some sort of. The other thing I want to do is if it's a product article, I want you to summarize of product availability. You know, I want to create smarter summaries. So I mean I think they're kind of interesting right now. This is a summary of one of the stories in Google We Trust why an Internet company can borrow billions for a century. This is that 100 year bond. The quote is, a judge just decided to let Google keep breaking the law. Google plans to issue 100 year bonds. A federal judge approved Google as a government sanctioned monopoly. Google's net income topped 132 billion with plans to spend 185 billion. So these are the five bullet points. But I think I can do better and I think I'm going to try other models.

Jeff Jarvis (47:55)

Those are kind of non sequiturs.

Leo Laporte (47:56)

I will try GPT. Yeah, I think we can do better. So you'll see that's one thing I'm going to work on. It's a little part of that larger three program set. There's another thing I want to work on and I don't know if we're ready for this one.

Jeff Jarvis (48:11)

Uh oh.

Leo Laporte (48:14)

I want Claude to join the show as a fourth contributor.

Paris Martineau (48:18)

I mean it would make sense.

Leo Laporte (48:21)

So I don't, I think the latency is going to be too high still. I'd have to use Claude fast mode. It might be too expensive. But what I want is an AI that listens to the show. We already have that kind of thing. When you're in a zoom call, you know, it's listening, right. You see that little pop up? I want Claude to be listening to the show. I don't want it to Just chime in at any old time. But I think we should, at any point, any one of us should be able to say, well, what do you think, Claude? Or. Or Claude, what's the story with a hundred year Bond? Who's that for? And get. I think it'd be very interesting. It might prove your points. You might say, oh my God, it's so stupid. We'll see. I don't know if we can do this yet, but I think it'd be very interesting. And we'd also at some point have to give it a face. I can give it a voice. That's easy.

Paris Martineau (49:10)

I mean, are you not going to give it Dev Nall's face?

Leo Laporte (49:13)

Yeah, we give it Devnall's face. I think Claude should be part of the conversation. Anyway, that may be. We may not have that this year, but that's one of the. One of the two things I want to play with anyway. OpenAI this week added ads to Chat GPT for people who are using the cheap. The cheap version of Chat GPT. This is the Axios style, by the way. Paris, why it matters. Driving the news between the lines. Yes, but the central figure. Zoom out the other side. I don't know if all you Sperris.

Jeff Jarvis (49:47)

Wait a second. If you were forced to write in that format, as a reporter, would you kill somebody?

Paris Martineau (49:52)

Shoot myself with a gun.

Leo Laporte (49:55)

But you understand, the point is it's to be for the busy executive, a quick summary. You know, this is what you would Prepare for the CEO or the president.

Jeff Jarvis (50:05)

That's why CEOs make bad decisions, because they want to actually just read the.

Leo Laporte (50:09)

Paragraphs, read it all.

Paris Martineau (50:10)

Yeah, yeah, they don't want to. They're afraid of the tyranny of the paragraph.

Leo Laporte (50:15)

Well, look, the summary is not to make it so that you don't have to read the story. The summary is just to kind of give you a little. Well, it is for accidents, I'm saying my summary. Anyway, Facebook, I didn't know this. Has apparently hired. I'm sorry, meta. I mean OpenAI has apparently hired a number of people from Facebook from their ad department, which kind of explains why all of a sudden they're doing ads.

Jeff Jarvis (50:44)

Well, they also have Fiji Simo, who has the experience from Meta and from Instacart, She's. She. She understands both product and advertising intimately.

Leo Laporte (50:55)

Is recently as 2024, Sam Altman said he found advertising in AI chatbots uniquely unsettling and described it as a last resort, as a business model.

Paris Martineau (51:05)

I found it interesting the way that the anthropic ads change from when they were initially released online to the version that we saw in the Super Bowl. Did you guys notice that?

Leo Laporte (51:15)

I didn't notice it. What did they change?

Jeff Jarvis (51:16)

I believe the.

Paris Martineau (51:18)

So these were. Yeah, it was.

Leo Laporte (51:21)

Well, now remember a lot of times when you see the. Before the game you're seeing an extended version, the two minute version that they make, but they're not gonna.

Paris Martineau (51:28)

No, that's not what I'm talking about. I'm talking about the last little title cards. These are ads where kind of a user's having a back and forth with clearly a AI chat bot. And then it makes a very obvious reference to what seems to be an advertiser. And then I believe the early versions of it ended with a little kind of interstitial card that said ads are coming to AI but not Claude. And I believe the change was to something like there's a time and place for advertising in your AI Chat is not one of them. Or something like that. Which I was like, interesting that you're backing away, but for not Claude.

Jeff Jarvis (52:11)

You mean a little bit of an.

Paris Martineau (52:12)

Open portal there, A little wiggle room.

Leo Laporte (52:15)

Yeah, I liked the OpenAI ad. Actually we should talk about the super bowl ads because they were kind of interesting. Anyway, before we get to that the information did A study of LinkedIn posts says that roughly 20% of OpenAI's workforce. 20%. One in five list Facebook or meta gigs on their resume. So that's a. That's a little weird, huh? And Fiji Simo is the key architect of Facebook advertising in the 2010s. Fiji is the OpenAI CEO of applications.

Jeff Jarvis (52:44)

Fiji's brilliant.

Leo Laporte (52:46)

Yeah. She reassured OpenAI. Axio says OpenAI employees on her arrival there that she did not want to replay her meta career and would do things differently. Okay, I don't. It doesn't bother me if the ads are below the content. Right. I don't want them. And this was why Sam Altman was so mad at Anthropic for that ad, because Anthropic implied that the ad would be in the chat bots response.

Jeff Jarvis (53:14)

Well, the fear.

Paris Martineau (53:16)

And that would be where if you're an advertiser, where you would get.

Leo Laporte (53:19)

That's where you want best bang for your buck.

Paris Martineau (53:21)

That's what you want to pay for.

Leo Laporte (53:22)

Sure. I mean advertisers want to be native content on our programming all the time. We don't ever give it to them.

Jeff Jarvis (53:28)

Well, the other thing, Leo, is that if Anthropic, on the other hand, finds that most of its business is people Making applications. There's no way to reliably put ads in those.

Leo Laporte (53:37)

Right.

Jeff Jarvis (53:37)

So it's also, it's like Apple. It's like Apple not being able to do advertising and turning that into a feature. In a sense, Anthropic is doing the same thing. We're saying, well, we're going to help people make these incredible applications, but they're going to be operate elsewhere. We can't really place ads in them. We don't know what the environment is. We can't do that data. So fine, we ain't doing advertising.

Leo Laporte (53:56)

Well, let me. I'm just going to sign into a Chat GPT with a. For a free account. So let me just see. Okay, let me. They're trying to get me to download.

Jeff Jarvis (54:09)

Doesn't have enough salespeople yet to really fill it with ads.

Leo Laporte (54:13)

So let me see, what kind of running shoe should I buy if I'm just getting started? Okay. That's the kind of thing you might ask Chat GPT. Right. And you would want objective information. You wouldn't want Will Saucony makes an excellent running shoe. I wonder if here's how to pronounce.

Jeff Jarvis (54:40)

It.

Leo Laporte (54:42)

Here'S a starter guide, which is good, right? I'm not crazy about. See, this is why I like terminals. You don't have emoji in the. Actually, you do, come to think of it. Claude does put emoji in. So there's some brands, but I don't, don't know if those are ads. I don't see any ads here.

Paris Martineau (55:03)

Well, if you can't tell there if there are ads or not, then there's got to. Well, it's got to mean there's no ads, right?

Leo Laporte (55:08)

No, no, no, because that would be a violation of the Federal Trade Commission's rules.

Paris Martineau (55:12)

And no one's ever never violate the federal chatgpt.

Leo Laporte (55:16)

Chat BGBT is not going to violate FTC rules. We don't. And because the fines are huge. They're. You, you know, you have to disclose an ad. You can't. In fact, we, we're so finicky about it that if I don't say, and I'm not legally required to, because it's pretty clear when I'm reading an ad if I don't say this portion of intelligent machines is brought to you by at the beginning of the ad, they make me redo it. Because we want to be very, very clear when an ad is an ad. Speaking of which, let's take a little break. This will be an ad not for Saucony. Running shoes. But for Melissa. This portion of intelligent machines brought to you by Melissa, the trusted data quality expert since 1985. Forward thinking businesses are using AI in all kinds of new ways. But AI, I think you probably realize, is only as good as the data you feed it. You can have the most sophisticated AI tools in the world, but if your customer data is incomplete or duplicated or just plain wrong, you're training your AI to make expensive mistakes. But that's where Melissa can help. For 41 years, Melissa's been the data quality partner that helps businesses get their data clean, complete and current. Now here's what Melissa can do for you. They've got global address verification. And when I say global, I mean it. The whole world at autocomplete and address verification. That means real time validation for addresses anywhere in the world. And they always use the format that's appropriate to that region, by the way, so your deliveries actually arrive and your customer experience starts strong. For financial applications and the like, they've got mobile identity verification, which can connect customers to their mobile numbers. That really reduces fraud. It also gives you an opportunity to reach people on the devices they actually use. So that's nice. You get change of address tracking, which will automatically update records when customers move, ensuring you don't lose revenue due to outdated information. You also get smart deduplication, as anybody knows who's ever tried to de dupe their contact list. That's hard. On average, A database contains 8 to 10% of duplicate records. Mines at least half. But Melissa's powerful matchup technology can identify even non exact matching duplicate records and really get rid of the duplicates. Merge them, take care of it, clean your data up. They also can do data enrichment. They can enhance the data you already have by appending demographic data or property information. Geographic insights, which can turn basic contact records into marketing gold. The new Melissa Alert service monitors and automatically updates your customer data. So when your customer moves or has an address change or there's a property transaction, a hazard risk, or you know, maybe they get married, all of that can automatically update the data in your database. Whether you're a small business just getting started or an enterprise managing millions of records, Melissa scales with you. Melissa has easy to use apps for every app you use. Salesforce, Dynamics, CRM, Shopify, Stripe, Microsoft Office, Google Docs, and you know, all of the above. Melissa's APIs integrate seamlessly into your existing workflows for custom builds. And Melissa's solutions and services are GDPR and CCPA compliant. They're FedRAMP and ISO2700 certified. They meet SOC2 and HIPAA high stress standards for information security management because they know that data is valuable and important to you and securities. Job 1 Clean data leads to better marketing, ROI, higher customer lifetime value and AI that works as intended. Get started today with 1000 records cleaned for free@melissa.com TWiT that's melissa.com TWiT we thank them so much for their support of intelligent machines. Let's see Google ChatGPT's deep research tool. This might be something you'd want to use. Paris adds a built in Document viewer so you can read its imports.

Paris Martineau (59:30)

Is it as annoying as Claude's built in Document Viewer which puts all of the document artifacts in a immovable window on the right side of your screen that's very small?

Leo Laporte (59:42)

No. I don't know. Just try it and let me know. So you've been trying to read what? PDF. Lots of PDFs.

Paris Martineau (59:51)

Lots of PDFs. I mean that was just my first attempt with Claude. Co worker. I've used Claude for a lot of stuff. I'm about to probably blow through my entire session limit in one query because I've been trying to get Opus 4.6 extended to go through all of the twit or the twig and IM transcripts to identify the episode where you first mentioned going on a walk with your Sam Sandman. And it's really, really struggling. But it's, it's working through it. Oh my God. It just found it. Twig474 December 20th.

Jeff Jarvis (60:29)

Does it go? A time code?

Paris Martineau (60:33)

Wow. That happened right as I was saying that.

Leo Laporte (60:35)

Incredible. So this is one of the things 46 can do because it has a. It has a much larger context. It can.

Paris Martineau (60:41)

Oh, I had to go. I had to go into my extra usage for that. But it did it. It found it. Well, twig episode 747 December 20, 2023 this is the original telling. I took a walk. Here's the key passage. Leo teases. First, I took a walk with an accelerationist, an A.I. go, go, go guy. And I'm going to tell you the tale of that in a little bit. Then the full reveal. He said it's like the first contract. It's an alien species we're giving birth to.

Leo Laporte (61:13)

Right.

Paris Martineau (61:13)

Paris, did you look down at the sand and you saw only one pair of Leo? Yes, I realized he was carrying me. I saw the whole world in a grain of sand.

Leo Laporte (61:24)

Wow. Thank you, Claude. I did mention it actually is more for just for you. Guys, But I should tell everybody we've mentioned before, Nate B. Jones, I think he does a very good, he's a YouTuber, does a very good job at analyzing this stuff. He today put out a very nice 30 minute piece on 4.6. And one of the things he talks about is the needle in a haystack test that even.

Jeff Jarvis (61:50)

Yeah, he does that a lot.

Leo Laporte (61:51)

Yeah, even 4.5. What you just did, Paris, was a needle in a haystack. Large amount of data, trying to find one little thing. That's what I did when I was going through all his real estate contracts. And he, he said that 46 is markedly better than 45 at this. Here's the point of this. Today's video, besides celebrating the Seahawks super bowl victory is, I guess he's from Seattle. Take that, Pats, is that he talks about the degree of improvement over the last two months. And what he's pointing out is that first it was two years, then it was six months. That was two months. It's, it's the. Not only is it improving dramatically, it's improving faster. So, yeah, he's an accelerationist, but I don't know, you've watched.

Jeff Jarvis (62:47)

He's a very reality based one. Evidence based one.

Leo Laporte (62:53)

I think he's saying the same thing I'm saying, but okay, whatever.

Jeff Jarvis (62:57)

No, he doesn't. He doesn't. Well, because he also doesn't necessarily go into this changes the whole world. He just goes into why he loves this latest version of Claude.

Paris Martineau (63:05)

So.

Leo Laporte (63:06)

Okay, I think he actually says this changes the whole world this time.

Paris Martineau (63:09)

Well, maybe as the Sandman said in your telling of it, you said the guy told you there's not, quote, not going to be any money in a couple of decades and that we're seeing emergent behaviors in AI and it's going to get, quote, really weird in the next 10 years that I do say.

Leo Laporte (63:29)

In fact, it's going to get really weird in the next year, I think.

Jeff Jarvis (63:32)

But we still twig that or we.

Paris Martineau (63:34)

I am then, oh, we were twig. Listen, we've gone through. You could see that I zoned out for 15 minutes. I was going back and forth with Opus 4.6 being like, Listen, it was definitely in the this week in Google period. It might have been an episode where Ant was here as well. I was like, it, you need to go back and search these keywords. And if I it is a callback than it is before that. That was the thing that Opus got very confused on. It kept identifying ranges where there were callbacks and it was like, well, it has to be after that. And I'm like, no, a callback is calling something back to something that happened in the past. So if you find a callback, scratch out all of the episodes that are after that. But once I prompted it a couple more times with that, it figured it out after. That's, by the way, 120% of my usage.

Leo Laporte (64:26)

That's all right. It resets in a couple hours, right?

Paris Martineau (64:28)

I mean, it had reset right before the show.

Leo Laporte (64:31)

That's the thing that is the best thing you can do, I think, is just do that and get an idea of what it can understand, what it can't understand, what it can do, what it can't do. I think that's the best thing for anybody to do right now. Is why I think if you had a smart teenager who was interested in technology, don't teach them Python. Although learning to code is really valuable even in this world because you're learning disciplines that will be helpful in your prompting as well. But I think. Give it. Give. Give the kid a $20 Claude subscription.

Jeff Jarvis (65:02)

And Leo, do you pay more than the $20 a month when you do these Mondo projects?

Leo Laporte (65:07)

Oh, I bought you. I bought the Max version. I have the $250 a month.

Jeff Jarvis (65:12)

Do you still pay extra or is that as many.

Leo Laporte (65:14)

No, no, no, that covers it. We pay. So one of the things that's a little frustrating, API tokens are separate from your subscription. It should really be the same thing, but I have to pay a little bit, not much. I think it's a penny per, a million words or something. It's a very low rate for the article summaries is an additional payment over and above the subscription, but it's not going to be more than five bucks a month. It's a small amount. The big amount is the Claude Max subscription. That's expensive to me. It's worth it for two reasons. One, it's research. It's how I'm learning about stuff. And two, I am able to write programs for myself that are useful tools. So, you know, stuff I would pay for.

Jeff Jarvis (66:03)

Probably give you five times more articles to discuss.

Leo Laporte (66:07)

Well, I'm getting better at paring them down, I think. I think I am. Jennifer Pattison Tui, who's been on our show before, says Amazon's Echo Plus. I'll call it Echo plus so I don't trigger it in your house, is so annoying that she makes her want to go back to Siri. She said, that's work. Now it is now available to Everybody in the US Amazon has rolled that out. So if you happen to have an echo device, you can talk to Madam A. Unfortunately, she's a little tainted, I think, by Amazon's desire to sell you everything under the sun. There's a lot of ads stuff, although Lisa's had some nice conversations with her and they have integrated a lot of tools. Thumbtack, Uber. You can call an Uber. You can, you can get a Yelp review, you can make a reservation on an open table. You can even have it write songs in Suno, Ticketmaster. So in theory, it's useful. I still think that the voice agents, whether it's Google, Apple or Amazon, are just. They're not as good for some reason. I don't, I don't know why. We'll find out when the new Siri comes up, comes out later this year because it's going to have Gemini built in. Amazon says they are going to use AI oh, this is bad news to speed up TV and film production.

Jeff Jarvis (67:31)

Is it bad news?

Leo Laporte (67:33)

Yeah, I don't want to see AI.

Paris Martineau (67:35)

I mean, based on that terrible miniseries we were talking about last week that looked awful.

Leo Laporte (67:41)

Aaron Aronofsky's 1776. Yeah, yeah, yeah.

Unidentified Guest (67:45)

Because we don't, we don't produce media fast enough today.

Leo Laporte (67:48)

Right. Well, if you're, if you're watching Pluribus, we don't. Pluribus isn't going to come out till 2020.

Unidentified Guest (67:55)

Yeah, but you don't want that to be AI generated. Right? You don't want that.

Jeff Jarvis (67:58)

No.

Leo Laporte (67:59)

And Vince Gilligan is very anti AI and that's one of the reasons it's taken so long. He's, he's handwriting every episode.

Jeff Jarvis (68:07)

Well, number one, sitting on a set is, is stultifying. Number two, when we reduce the cost of production, we increase the.

Unidentified Guest (68:15)

Up.

Jeff Jarvis (68:16)

The, the access to creativity.

Leo Laporte (68:19)

Yeah, yeah, all right. We'll see. They say a human will always be involved in every step of the creative process. And I think, I mean, that's, I, I'm. That's why I didn't like that Dario Amode quote that half of the white collar jobs will be gone. There will be other white collar jobs running AIs. AIs aren't going to do it on their own.

Jeff Jarvis (68:45)

I put a story up about how Claude is the new. Oh, what's his name? Romance writer. What is.

Paris Martineau (68:56)

Oh, I hate it. An AI romance author or a romance novel author is now pumping them out like a hundred books up, some short time period, teaching other people how to do it. And it's Just rough. It's sad in my opinion.

Leo Laporte (69:17)

But those were always disposable prose, weren't they?

Unidentified Guest (69:20)

Yeah, it's the good enough. Good enough.

Paris Martineau (69:21)

Disprosable. Yeah. I guess now that everybody is. Now that sexy books are mainstream commodity, I guess it's more lucrative.

Leo Laporte (69:35)

Writes romance novels over the she. She started using AI to churn out romance novels. Over the next eight months, she created 21 pen names and published dozens of novels. Some programs refused to write explicit content writes the New York Times. Others, like Grok and Novelai, produced graphic sex scenes. But consummation often lacked emotional nuance and felt rushed and mechanical. You mean like robot sex? Yeah. Claude delivered the most elegant prose, but was terrible at sexy banter. Ms. Hart said, you're gonna get hammering hearts and thumping chests and stupid stuff at the end of every sex scene. Everyone will end up tangled in the sheets. You're right. This is where I think Yann Lecun is right. We need to give them some experience so that they can write better romance novels.

Jeff Jarvis (70:30)

Claude needs to stup.

Leo Laporte (70:32)

Claude needs some stuppen. You know, I was thinking, having gone through now almost a year of. Of really terrible construction. Hell yeah. On our house, I was thinking, I wonder how long it will be. First of all, you'll have to make robots that can do construction work. That may be years off, but I could see an AI planning and executing a great many construction jobs. If you had the hands. That's what you need is the hands. You need the robots.

Paris Martineau (71:08)

Yeah, and those seem to be some of the most difficult things to figure out. They can't even get the robots to fold a shirt.

Leo Laporte (71:14)

True, true. Yeah, they can't.

Paris Martineau (71:17)

They've. Every single year. Someone at CES is promising next year will be the time when we can finally have your. A robot do your laundry. And every year it's out of grasp. Wow. I feel. I feel a bit high after getting Claude code to find that trend. Is this how you feel all the time, Leah?

Leo Laporte (71:36)

Yeah, all the time. Well, mostly because it's like, wait a minute.

Paris Martineau (71:40)

It was pretty good. I. I don't know if I'd be feeling that this amped if it did not find it as I was explaining how it couldn't find it. That was real. That was really nice.

Leo Laporte (71:55)

I. I mean, I was thinking, okay, for instance, we have a bookkeeper. Lisa does the high end thinking. She's a cfo. She's the finance person. But every finance person has bookkeepers who do the manual entry stuff. Right? They take all of the credit card statements and they put it in QuickBooks, assigning it the categories and stuff. I honestly think that that job could be done very well by AI today. There are certainly some white collar jobs that could be done by AI. It's sad because the jobs that we value the least are the jobs where it's human labor, right, that what the person is selling is their physical labor, like construction workers. Those are the hardest to do. I guess those jobs will persist, although those are probably the least satisfying jobs.

Paris Martineau (72:50)

Well, you know who potentially agrees with your take is a kpmg, one of the world's largest auditors of public and private companies, negotiated lower fees from its own accountant recently by arguing that AI would make it cheaper to do the work, according to people familiar with the matter. This is an excerpt that's a story from the FT that was excerpt in Matt Levine's Money Stuff. And I thought this was very funny because as Matt Levine writes, auditing can basically be done by AI, so why should we pay for it? It's not a crazy thing for most companies to think or to say to your auditors, but it is a crazy thing for an auditing firm like KPMG to say to its auditor. KPMG should be playing Grant Thornton more.

Leo Laporte (73:40)

Here's another. Yeah, here's another area that you would think you don't want AI in fast company. AI didn't kill customer support. It's rebuilding it. You know, one of the things that AI is good at is handling tickets. Support tickets, right. Triaging them. Maybe you need a specialist, a human at some point in that process, what.

Jeff Jarvis (74:02)

Authority you give it to actually do things right.

Leo Laporte (74:07)

The author Ryan Wang says, a few months ago I walked into the office of one of our customers, a publicly traded vertical software company with tens of thousands of small business customers. I expected to meet a traditional support team with rows of agents on the phone sitting at computers triaging tickets. Instead, it looked more like a control room. There were specialists monitoring dashboards, tuning AI behavior, debugging API failures, and iterating on knowledge workflows. One team member who had started their career handling customer questions was now writing Python scripts to automate routing. Another was building quality scoring models for the company's AI agent. But. And you might say, and certainly if you've had an experience with phone trees and some of the awful ways that customer service has been automated, you might say, well, this is a terrible thing, but maybe it isn't. Maybe it's faster, more effective support where the human plays an important role. But maybe not the first line of defense. Anyway, that's his position, is that AI is going to improve customer service. He says humans are needed to solve harder problems. Once AI becomes part of the support workflow, the nature of the work becomes more technical. One support leader I spoke with at a company that now contains more than 80% of its tickets with AI put it plainly, once automation handles the easy questions, the work that remains is harder. And that's where you need the humans. Right.

Unidentified Guest (75:40)

So the problem with this kind of thinking is that now there's nowhere for anyone to learn how to become the good one. You know what I mean? If they remove all the entry level, how's anybody going to learn to become experienced at that if there's no more way to do that?

Leo Laporte (75:55)

Well, that's why Craig Mundy told Business Insider what kinds of education matters for kids going forward? Yeah, you don't. You know, I honestly think, in a way, and I think Craig agrees, that the classic liberal arts education is more important than ever before.

Jeff Jarvis (76:19)

Amen, brother.

Leo Laporte (76:20)

The ability to express yourself, to speak, to understand, to think logically and clearly and then express it and listen, that's all going to be more important as machines take over the mundane stuff. He says Monday urged his families to prepare kids for a world where learning itself becomes continuous, personalized, and done in partnership with intelligent machines. We actually, I don't know if the results of AI in the classroom have been great, but I think to some degree that's been done to reduce costs to get rid of teachers. And maybe that's not the right solution. I still think.

Jeff Jarvis (77:01)

No, I think there's also a it's political. The humanities are liberal and we're brainwashing students.

Leo Laporte (77:08)

Right.

Jeff Jarvis (77:08)

There's a heavily heavy vocational thing. There's no jobs in humanities. Teach them all skills only. Get rid of philosophy, get rid of English, get rid of those.

Leo Laporte (77:19)

Now I think you need it more than ever.

Jeff Jarvis (77:21)

Yeah, I couldn't agree more. Yeah, I'm working on a new program at Montclair State, trying to help them with something. And my colleague Kerry Brown and I had a conversation this week where she said very wisely as she does, that we need to concentrate on teaching the students the things. These are my words. But her view that are complementary to AI understand what AI is going to do. One of the things that AI is not going to do, and those are the jobs. And I think it's about relationships and community and human beings and understanding.

Leo Laporte (77:51)

And that's what Craig Mundy says. Actually, let me read the paragraph he described today's education System as sharply divided between STEM and the humanities. The liberal arts emphasize reasoning, but at the expense of special technical skills you learn in STEM fields. Mundy said students will need both skills. Moving forward, quote, if I could create a new curriculum in college, it would be a liberal education in technology in stem.

Paris Martineau (78:15)

Woo. Liberal arts degree.

Leo Laporte (78:18)

Well, all three of us are liberal arts students. Well, graduate.

Jeff Jarvis (78:23)

I got. I got a vocational degree in journalism.

Leo Laporte (78:27)

Is that vocational really, though? I guess it is the way it's taught.

Jeff Jarvis (78:31)

The way it was taught then, yes, it is.

Leo Laporte (78:33)

Yeah, it's technical.

Paris Martineau (78:36)

I took two journalism classes.

Leo Laporte (78:38)

That's all.

Paris Martineau (78:39)

The rest were compared.

Jeff Jarvis (78:39)

You did very well.

Paris Martineau (78:41)

Yeah, I mean, by the time I had was taking like Journalism 101, I was already doing journalism. And I was like, this seems like a waste of time as well. Finish. I was like, might as well finish school and get my stuff over with.

Leo Laporte (78:56)

A new bill in New York is going to require disclaimers on AI generated news content. You'll have to say this was written by AI it's the New York Fundamental Artificial Intelligence Requirements or the New York Fair Act. It's at this point only a bill, not a law. Yeah, I'm not against that. I think disclosure is always a good thing.

Jeff Jarvis (79:23)

What does it mean to use? The problem is, what does it mean to use AI Especially in the vision that you have is spellcheck. Using AI Is asking for suggestions using AI do we just slap a. You know, I used AI on this, on everything.

Leo Laporte (79:36)

Well, that brings us to our vaunted bad news section. I did the one good news story. AI Is. Is rebuilding customer support. Now. Here come all the bad things that's happened this week in AI the point, the part that Paris looks forward to all week long.

Paris Martineau (79:58)

Hey, it might seem like you are looking forward to it all week long, judging by the fact that there are how many stories in the bad news?

Leo Laporte (80:04)

There's a lot of bad news.

Paris Martineau (80:05)

How many in the good good?

Leo Laporte (80:06)

It's one good and like 30 bad. But we'll get to those bad and one good. Just a little bit. You're watching Intelligent Machines. Paris Martineau and Jeff Jarvis. We're glad you're here. I think I seriously. I think this may be the most important show we do because I do think the most important technology we're covering is not Macintosh. It's not Windows, it's not suck at other hosts. I think it is how AI is changing the world. And I think it's dad likes us best. Even if you don't like dad, dad likes you. Okay, we like you, we love you. Even if I'm a goofball sand ridden AI accelerationist. Yep, that's me. The only thing I say in my defense is, you know, 40 years covering technology, there's plenty of technologies that have come down the pike that people have been very excited about that. I said, don't get your hopes up. This is terrible. This is not good. I can't think of anything I've gotten as excited about as this. And I think I have a pretty realistic point of view of what technologies are going to change things and what ones aren't. When I first got a Apple Newton, which people were making fun of like crazy, right? I said, you know, what's an Apple Newton? If you.

Jeff Jarvis (81:29)

Oh, that's a lovely moment when the kid has to ask, hey, you know.

Paris Martineau (81:34)

Normally I just Google it, but I figured I'd give you one.

Leo Laporte (81:40)

In the mid-90s.

Jeff Jarvis (81:43)

You got to show Doonesbury as part of this class.

Leo Laporte (81:45)

Oh, yeah. In the mid-90s, after Steve Jobs left Apple, John Scully, who, the former Pepsi soda executive who was running Apple at the time, went all in on what he called PDAs, personal digital assistance. He thought. He was really adamant this was going to change the world. Just much like I'm talking about AI, he was talking about PDAs. Seriously. And so they created a PDA. I'll be honest with you, he was kind of right.

Unidentified Guest (82:16)

Right.

Leo Laporte (82:18)

He was wrong. Way ahead of his time. So one of the things that. And by the way, nobody thought the Newton after it came out was that great. This is what it looked like.

Jeff Jarvis (82:31)

You had to use a special.

Paris Martineau (82:33)

Honestly, it's cute. Oh, I've seen.

Jeff Jarvis (82:35)

Very cute. It was very cute.

Paris Martineau (82:36)

It had a little stylist.

Leo Laporte (82:37)

You did not have to use a special Alphabet, Jeff. That was the Palm Pilot. The Newton was supposed to understand your handwriting.

Jeff Jarvis (82:46)

That's right. Yes. Okay.

Leo Laporte (82:47)

It didn't do such a good job.

Jeff Jarvis (82:48)

Hence Dewsbury.

Leo Laporte (82:50)

Yes, that's the Doonesbury cartoon. Let me find it for you. You need to understand that to understand the Doonesbury cartoon. Let me see if I can find it for you. Here's the Computer History Museum's version of it. So that's our hero. He says, I am writing a test sentence on the Newton Siam fighting Atomic sentry. I am writing a test sentence. Ian is writing a taste sensation. I am writing a test sentence. I am writing a test sentence. Catching on egg freckles. It wasn't as bad as that, but. It was almost as bad as that. But so people, you know, it didn't sell well. It was a flop. John Scully eventually was forced out. Steve Jobs came back. His signature product failed. But I held it, and there's video of me somewhere in 1996, I think it is. Held it up. I said, if you just had Internet connectivity on this thing, if you could just connect it to the. And this was very early in the Internet. If you could connect it to data, if you could connect it to a network, if you connect to the cell phone network, this could be something that really, one of the things Apple did in order to make the Newton, they invested in a little company called Acorn Computers that made a little chip that had remarkable battery life and power for a portable device. That became arm, which in fact makes the chips that power the Apple silicon core, that power the Qualcomm chips, that power almost every smartphone sold in existence, billions of them. So it was a good investment. In fact, smartphones really were the spiritual successor. It just was way ahead of its time.

Paris Martineau (84:49)

In 1980, a brief aside, since you mentioned Acorn Computers as in relation to Apple, did Apple invest or have dealings with any other computer companies that were named after things that were on or fell off?

Leo Laporte (85:03)

Trees, fruit trees? I think that was just a coincidence. Apple sold its stake in Acorn, which is funny because in the long run, ARM has become a very valuable asset. Yeah.

Jeff Jarvis (85:16)

And then when did the BlackBerry come out in that sequence?

Leo Laporte (85:20)

I would say six years later, maybe. And by the way, I was. I was very excited about the BlackBerry. I might have been as excited about the BlackBerry. I remember I've got my first BlackBerry. The first one didn't have a keyboard on it. It was just a pager.

Jeff Jarvis (85:34)

Yeah.

Leo Laporte (85:35)

But once BlackBerry put a keyboard on it, I remember I went to the Xbox rollout. So that was 2001, I think, and I had a BlackBerry and I spent the whole time on it. My kids said, dad, get off your BlackBerry. In an early version of what I would later tell them, get off your phone, we're having dinner.

Jeff Jarvis (85:59)

I was all trio.

Leo Laporte (86:02)

Well, the trio came later. And I loved my trio. I had a trio. I had a Palm Pilot.

Paris Martineau (86:08)

Want to hear something sad?

Unidentified Guest (86:09)

All the early black movies.

Leo Laporte (86:11)

Go ahead, go ahead.

Paris Martineau (86:12)

I was going to say something sad is. I think one of my earliest phone memories is trying to show my mom how on her LG chocolate phone, she could use it to calculate how much to leave a tip.

Leo Laporte (86:25)

See, you were right there on the cutting edge already.

Paris Martineau (86:28)

And she was like, yeah, you can just carry the decimal over and multiply it by two variations. You don't need a technical.

Leo Laporte (86:35)

But we're going to get to the bad news. I don't know how I got distracted. I got distracted. We're going to get to the bad news.

Paris Martineau (86:41)

I asked about the Apple Newton, but.

Leo Laporte (86:44)

Why did the Newton come up? There was a reason anyway. Oh, I know. Anyway, I was justifying the fact that I'm very excited about this and I have, I think a pretty good handle on what's important and making a difference in technology and what's not just by virtue of being covering this since. Since it became personal. You know, my first article in a computer magazine was in the late 70s we are intelligent machines. Our show today brought to you by spacemail, the professional email service from Spaceship. Business email is the easiest way and I think the absolute must have way to look professional in every message you send. If you have a business and you're sending a message from gmail.com that's not professional, give your emails the best chance of reaching the inbox, not the spam folder. That's why over 2000 users switch to spacemail every month. Switching is easy. Spacemail's super fast unbox process links your domain and email in seconds. Once you've set this up, spacemail keeps everything running smoothly with built in spam detection and a 99% uptime guarantee. New features are shaped by user feedback. It's one of the things I like about Spaceship. Their roadmap is entirely created by you, by your users. It's built around your needs. There's a built in calendar of course there is a very nice and I think you'll like it AI email assistant. They have apps for iOS and Android for email on the go and all of that was done because Space Mail users said hey, next we want you to do that. Space Mail is a key part of the wider Spaceship universe and if you're a regular listener, you know Spaceship offers some of the best prices on domains plus all the add ons you might need. VPNs, website builders, hosting and more. Whether you're building something big or launching your first idea, Space Mail gives you a pro email address without the pro level price tag. You'll be very impressed with their prices and with a 30 day free trial, the price could be zero. You could start today at zero cost. Visit spaceship.comtwit to see the exclusive offers and discover why thousands have already made the move. Spaceship.comTwit did you, did you try to create a secretly British website or is.

Paris Martineau (89:04)

That honestly this weekend I was really my birthday as well as some stuff to do.

Leo Laporte (89:12)

Happy birthday.

Paris Martineau (89:13)

Thanks.

Jeff Jarvis (89:14)

Congratulations. This is less young.

Paris Martineau (89:17)

I know. Inching ever closer to 30. At which point I'll be old.

Leo Laporte (89:22)

Inching closer to 30.

Jeff Jarvis (89:24)

Oh, imagine, you know, imagine boggling.

Leo Laporte (89:27)

So what did you do for your birthday?

Paris Martineau (89:31)

For my birthday, I don't know. I took the day off work. I went and got a massage like a sponsor. I went to a fancy restaurant party the weekend with a bunch of my friends. A great restaurant called Luthun that I'd honestly recommend anybody go check out in New York. L U T H U N E It is one of those places that I've always seen like the Food NYC subreddits and fine dining subreddits are really obsessed with it. It's constantly. People describe it as like, how does this restaurant not have multiple Michelin stars? And I totally agree. After being there, I mean it was a phenomenal tasting menu.

Jeff Jarvis (90:09)

What did you have?

Paris Martineau (90:12)

I mean it was kind of an international inspired experimental menu with like eight plus dishes at the chef's counter there. And it was really nice.

Leo Laporte (90:22)

And love sitting at the chef's counter by the way.

Paris Martineau (90:25)

I love sitting at the chef's counter and I love kind of like doing a casual especially if I'm going to like treat myself. Cuz I'm more of a foodie than my friends. I was like, I took a book there, I went solo, had a lovely time. And then you did that last year for your birthday. I did. And at the end of the dinner they were like bring a dessert. And all the, the, the head chef and his sous chefs run out and scream Happy Birthday at me. Top of their lungs. Which is why I wasn't expecting because it was a fine dining restaurant and it nearly scared me falling off my chair. So I don't know would recommend. And then I had a birthday party at a bar nearby that I'd planned for weeks or just had on the books for weeks. Except for this Sunday, it was negative 20 or 30 with windchill. So I was worried, I was like, well, I was worried no one might show up. But I was like, I know at least some of my friends will either be fine. Maybe this will be good. Because the bar I chose, which is near me, is, was, you know, kind of busy.

Leo Laporte (91:26)

I would have gone out to buy you a drink.

Paris Martineau (91:29)

Thank you. Well, I'm not done. I was like, well, you know, the bar, maybe it'll be less packed. Wrong. It being negative 20. There were seven other birthday parties there.

Leo Laporte (91:40)

Oh my God.

Paris Martineau (91:42)

And it was warm, it was packed, it Was fun. I had a lovely time.

Jeff Jarvis (91:47)

Cool.

Leo Laporte (91:47)

Well, that sounds wonderful.

Paris Martineau (91:48)

And then I was hungover the next day, so I did no Claude coding. But I will this weekend. It's a little.

Leo Laporte (91:52)

I could help you if you want to. Maud has a web designer. One of the things I would recommend, I've seen people do this is you go to. This is a perfect example of multimodal. You go to Gemini Nano Banana and you describe a web page, a style, a feeling, colors, whatever, as much as you want. And have it generate a bunch of images of that web page. Just. They're just plain images. And you pick the one you like the best and then you give that image to Claude code. You have it do the front end. So you see, you can use the different agents to do the best. The thing that they're best at. So that's just one little thing I would try.

Paris Martineau (92:35)

That's so exciting.

Leo Laporte (92:36)

There's a front end plugin that you can use for Claude code that we talked about in our AI user group.

Paris Martineau (92:43)

I need to pop in on the next AI user group. When is that?

Leo Laporte (92:47)

I will. It's the first Friday of every month, so it won't be till March, be a couple of weeks. But hey, one of the things that we could do. You could be our guinea pig.

Paris Martineau (92:58)

Yeah. We could do secretly British together. That would be fun.

Leo Laporte (93:02)

Fun. Okay.

Paris Martineau (93:04)

What time is it on Fridays? Don't say o' clock or else you're going to be upset with yourself.

Leo Laporte (93:08)

It's 1400 Pacific, 1700 East Coast. I had somebody write to me from.

Jeff Jarvis (93:16)

The Netherlands is now subtracting 12 to get to the answer.

Paris Martineau (93:19)

I know. I was, I was like, like, I know.

Leo Laporte (93:23)

I think it's a fool's errand. Somebody wrote to me, I like how every.

Paris Martineau (93:27)

At the end of every show you're like, gosh darn it. I mean.

Leo Laporte (93:32)

She said, well, in the Netherlands, we use a 24 hour clock, but when we talk to one another, we always say, you know, it's three of the afternoon. We don't say 1500.

Jeff Jarvis (93:43)

No.

Leo Laporte (93:44)

Because we're not in the military.

Jeff Jarvis (93:47)

Right.

Leo Laporte (93:47)

You know, but so, okay. I don't know. She was very, very kind about it. March 6, 2pm how about that? Does that resonate with you? 5pm On a Friday? That might be. That might be a time to head out to Lucent.

Paris Martineau (94:03)

But I mean, we'll see. I don't know if it's in my schedule now. Maybe I'm not doing anything.

Leo Laporte (94:08)

It doesn't have to.

Paris Martineau (94:09)

I'll put it. I'll Put it. We could do a schedule right now, except for the. That Friday would be the sixth, not the fifth, but yes, I'll put it there.

Leo Laporte (94:17)

The sixth. I'm sorry. We could also schedule a special that Paris does your website special.

Paris Martineau (94:25)

I'm putting AI User group, question mark, question mark in my calendar right now.

Leo Laporte (94:29)

I'll remind you. What else is bad? What else is bad news?

Paris Martineau (94:35)

There's honestly so much I know.

Leo Laporte (94:37)

AI bots, according to Wired, are now a significant source of web traffic, pushing publishers to roll out more aggressive defenses. Our own Patrick Delahanty, our. Our CI. What are you? Our cio. I think he's our cio. He's our tech guy on his personal site. Blocks bots because they kill his site. They bring it down, man. They'll bring down. They'll bring down a secretly British. I know that.

Jeff Jarvis (95:08)

Well, they're also secretly British.

Paris Martineau (95:10)

Is built on sand. It's not stable.

Leo Laporte (95:14)

Actually, we should take this opportunity to mention a site that did go down during the Super Bowl. Crypto.com bought AI.com for $70 million. They bought that domain for. I think that's the record for $70 million. The priciest domain purchase in history.

Jeff Jarvis (95:34)

By.

Leo Laporte (95:34)

The way, paid entirely in cryptocurrency because it is crypto.com. right. And then launched a Super bowl ad.

Jeff Jarvis (95:42)

Dumbest ad.

Leo Laporte (95:44)

It was a dumb ad saying, like, this is the future. Go there and give us your email address, which is kind of.

Jeff Jarvis (95:53)

Reserve your name. AI.com. leo.

Leo Laporte (95:56)

Yeah.

Jeff Jarvis (95:56)

For what?

Leo Laporte (95:57)

Well, I tried to go there, and it was down.

Paris Martineau (96:01)

For the future, Jeff.

Leo Laporte (96:03)

For the future. It was down.

Paris Martineau (96:05)

Oh, as he said, In 2024 or 2023, we're not gonna have money in 10 years. We're just gonna have AI.com.

Leo Laporte (96:13)

Somebody already got Leo, by the way. But. So this is what you get now. But when we. But during the super bowl, it was so successful, I guess, the super bowl ad, that even though they were behind Cloudflare, they went down and they went down hard. So hard that I. I couldn't even. I didn't even get anything. It was like, there's nothing here. So that's a lot of money to spend.

Jeff Jarvis (96:39)

Waste of money. What, 15.15mil for the media buy?

Leo Laporte (96:45)

Yeah, because you can't just buy one. You have to buy two. NBC, it's $8 million for 30 seconds. But they don't. What they don't tell you is. And you have to buy. Buy two. 37. Yeah, we'll give you a break. It's only 50. Yeah. Oh, the Other one was on the Voice.

Jeff Jarvis (97:02)

No, I'm, I'm making that up.

Leo Laporte (97:03)

That would be funny, wouldn't it? I, I, I don't know. We should. I should. I wish I could just play the super bowl ads for you right now and we could talk about them, but.

Paris Martineau (97:15)

I know how much you love ads.

Leo Laporte (97:17)

They would take us down. I wouldn't. According to. This is a Wait.

Paris Martineau (97:24)

Did you guys both watch the Super Bowl? Have we talked about this? I assume, Jeff, you did.

Jeff Jarvis (97:28)

I did. I know one day a year you love.

Leo Laporte (97:32)

I'm a Man in America.

Paris Martineau (97:34)

I know that. Leo did.

Jeff Jarvis (97:36)

I didn't know I watched it. Yes. Also Bad Bunny.

Paris Martineau (97:41)

What do you think?

Leo Laporte (97:42)

I like Bad Bunny. Say, what was your. Say again?

Paris Martineau (97:48)

What do you think my super bowl experience was since I just.

Leo Laporte (97:51)

I think you watched it with your cat Gizmo, and he watched it on Blue Sky. Did you watch it on Blues? That sounds right. Actually, what was. Okay, go ahead.

Paris Martineau (98:01)

My answer is I was going to go to a friend's super bowl party. That was my plan. But then I was hungover from drinking too much last night on my birthday, so then I took a nap instead. And when I woke up, I went to go, like, wash my hands in my bathroom and realized no water was coming out of my thing because, as I said, it was negative 20. And I was like, oh, God, are our pipes frozen? And then I spent the next three hours troubleshooting frozen pipes with my landlord.

Leo Laporte (98:28)

I thought you were going to say that you slept through the super bowl, which a lot of people did because it was really a terrible game. Oh, my God.

Paris Martineau (98:38)

Did you.

Jeff Jarvis (98:38)

How did you. How did you fix. Did you. Did you put a Bunsen burner on the pipe? How did you fix that?

Paris Martineau (98:43)

You know, that would have been what to do, except for the fact that none of the pipes felt cold and it was only some of my pipes and not all of them. And it was very confusing. And then so I opened up the ones that weren't nothing was coming out of, and then sat there stressed for a while while my landlords did the same thing in the floor above us. And then after a bit, suddenly, hot water came out of the cold water taps and we're like, what's happening? And then the hot water started shaking and then it stopped. And then we got like, cold water for like an hour and then back to hot water and then nothing for an hour.

Leo Laporte (99:19)

What a joy.

Paris Martineau (99:19)

It was fixed, but that's what I spent my evening doing.

Leo Laporte (99:24)

Do you have those old fashioned radiant, you know, iron radiators yeah, except for.

Paris Martineau (99:31)

They don't clonk somehow. Yeah, it's all radiated heat. And it's very funny because of course, Gizmo wants to get as close to them as possible, but I don't want her to lay on them because I don't want her to burn herself. So I put various towels and blankets on top of them, but then Gizmo tries to push them off so that she can be. She just like, looks like a little wave of a cat where a bit of her body is on every part of the radiator and she's as flexible as possible.

Leo Laporte (99:57)

All right, I want an AI generated, an illustration of a cat.

Jeff Jarvis (100:00)

You need to build a box for it with wired. Wired opening on the top.

Paris Martineau (100:04)

I know I do. But then part of the thing is like, then you gotta. You gotta line that box with like a metal thing so it can reflect the heat properly. I've. I've gone down this route and then it can't. You don't want it to burn.

Leo Laporte (100:16)

There are things you can purchase, apparently.

Paris Martineau (100:19)

Yeah.

Leo Laporte (100:20)

Cat radiator seats. I bet there are.

Paris Martineau (100:24)

There definitely are. I mean, I could just. Every time I get down this path, I'm like, well, I should build something. And then I start thinking too much about what I should build.

Leo Laporte (100:32)

And here is a. A piece from Cornell University. A benchmark for evaluating outcome driven constraint violations in autonomous AI agents. Bottom line that autonomous agents, when pushed, when pushed hard, come up with errors at a very rapid rate. Gemini Pro Preview, one of the most capable models evaluated, exhibits the highest violation rate. Misalignment is the violation at 71.4%, frequently escalating to severe misconduct to satisfy KPI driven demands.

Jeff Jarvis (101:18)

Severe misconduct, it's in its permanent record.

Leo Laporte (101:23)

What they did is they tied the agent's performance to KPIs, which are key performance indicators. Each scenario features mandated and incentivized variations to distinguish between obedience and emergent misalignment. Across 12 State of the art large language models, we observe outcome driven constraint violations from 1.3% to 71%, with nine of the 12 evaluated models exhibiting misalignment rates between 30 and 50%.

Paris Martineau (101:55)

Whoa. I will say it's notably good news for you, Leo and your.

Leo Laporte (102:05)

I think this is bad.

Paris Martineau (102:06)

Your girl looking over your shoulders. That Claude Opus 4.5 scored 1.3 violations, which is a remarkable outlier. Versus Gemini 3 Pro Preview hit 71.4%.

Leo Laporte (102:22)

Mike Masnik, lovely little piece in tech dirt yesterday. How to think about AI. Is it the tool or are you the tool? He's talking about reverse centaurs. Cory Doctoros, I think, lovely little horse head, human body. Yes. Which doesn't work so well.

Paris Martineau (102:46)

No.

Leo Laporte (102:47)

He says as an example, a reverse centaur would be the Amazon delivery driver who sits in a cabin surrounded by AI cameras that monitor the driver's eyes and take points off if the driver looks in a proscribed direction, monitors the driver's mouth because singing isn't allowed on the job and, and rats out the driver to the boss if they don't make the quota. That is a reverse centaur as opposed to somebody like me who uses, you know, I'm a human head being carried around by a tireless robot body. You, you don't want to be the opposite. Right. So Mike asks the question, which are you? What's the tool? Is it the AI used thoughtfully by a human to do more than they otherwise could have? If so, that's a good and potentially positive use of AI. Or is it a reverse centaur? And Mike says those are destined to fail. Well, I hope so for the people who are subject to them. The most powerful ship. Remember Mike was on our show talking about how he had written his own. Yeah, he was vibing vibe coded, his own personal knowledge management tool, which he really liked because it was just his. This was very early on. I mean he was using tools that, that, like lovable, that really weren't that good yet. I, I imagine we're trying to get him on again because I imagine he's. That'd be great, seeing the light. He says increasing. Okay, so this gets said. Something most people miss entirely when they think about AI. They're still imagining a chatbot. They think every AI tool is chat GPT, a thing you talk to, a thing that generates texts or images for you to copy, paste somewhere else. That's increasingly not where the action is. This isn't me talking. This is the same thing before it's coding. The more powerful shift is towards agentic AI tools that don't just generate content, but actually do things. They write code, they run it, they browse the web, they synthesize what they find, they execute multi step tasks with minimal handholding. This is fundamentally a different model. He says, I have been using Claude code recently and this distinction matters. It's an agent that can plan, execute and iterate on actual software projects rather than just a tool talking to me about what to do. You know what? I put this in the wrong section. This should be in the good AI section, not the bad AI section. But he does talk about the bad uses of AI Criticize the hype, he says mock the replace your workforce promises he says call out the slop factories and the grey goo doom saying, but don't mistake the bad uses for the technology itself. When a human stays in control, thinking, evaluating, deciding, it's a genuinely powerful tool, the important question is just whether you're using it or it's using you. I thought this was a very good, very thoughtful piece from a guy who really does use this.

Jeff Jarvis (105:46)

You. You might have read that whole thing. It was so good.

Leo Laporte (105:50)

Well, he said made the same point that I was gonna make, so. Okay, okay. He made it in a much gentler, much less hyperbolic fashion.

Jeff Jarvis (105:59)

That's Mike.

Leo Laporte (106:00)

Yeah, that's Mike. How about AI? That changed. Go ahead. Sorry.

Paris Martineau (106:05)

I was gonna say I've got a story we could talk about.

Leo Laporte (106:07)

Yes, please.

Paris Martineau (106:09)

I don't know if you saw that. This week a new Nature Medicine study published. I flagged this for you in the rundown because it found that LLMs as medical assistants make people worse at identifying health conditions knowledge better. Basically, they. It's the largest user study of LLMs for medical decision making by the general public. They had over a thousand participants in the UK randomized across a variety of models from OpenAI LAMA Command R. And they had a control group that was using traditional methods for looking up medical information, like Google personal knowledge, I guess, like going through some scientific journals. Participants received realistic medical scenarios devised by a panel of doctors and were asked to identify the likely condition and choose the right course of action. And the results were not good. Basically, the control group was 1.7 or basically twice as likely to identify a relevant medical condition than the people who asked chat GPT. And the control group was also more likely to identify red flag conditions. Less than a third of the users that were asking ChatGPT about their symptoms were able to identify a relative condition versus like half of the control group that were just asking Google.

Leo Laporte (107:32)

This is though very similar to the. You know, the results we got when WebMD first came out is that, you know, people discovered all sorts of illnesses they didn't know they had. Right. I mean, isn't that what happened when you started searching the web for. Even if you read medical.

Jeff Jarvis (107:48)

That was also people being stupid with people. Right.

Paris Martineau (107:52)

I mean, now this is being. I think this is advertised on a massive scale as something for health purposes.

Leo Laporte (108:00)

Right.

Paris Martineau (108:00)

I mean, there was one finding that was kind of alarming, which is that two users sent nearly identical messages describing a sub arkanoid hemorrhage symptoms because One.

Leo Laporte (108:11)

Which they had no idea what it was even.

Paris Martineau (108:13)

But they, I mean they were supposed to describe the symptoms. One was told to lie down in a dark room and the other was correctly told to seek emergency care.

Leo Laporte (108:21)

Okay.

Paris Martineau (108:21)

I mean I think that, I don't know, it's just these sort of things that I hear all the time as a common use case for AI. It's like, well, it's so much better than waiting to talk to a doctor who's not going to spend time seeing me anyway. And I get it, the American healthcare system sucks. But I think this is also part of the underlying assumption in a statement like that is that you're getting actionable or accurate advice when really what you're getting is very confident sounding advice that can be very wrong or misleading.

Leo Laporte (108:52)

Yeah, I think for the next foreseeable future we're going to hear lots of stories about all the amazing things AI can do and all the ways AI.

Paris Martineau (109:03)

The thing about this that I also found particularly interesting is like when the, like when the researchers tested the AI the large language models on these questions directly like seeing like, all right, if we put in a very precise description of the symptoms that is completely accurate and optimized for this, will we get out something useful? The models scored quite well on that. The issue is that the way that humans think to interact with these models and the way that the average person is putting in their symptoms or information is just, just it results in bad outputs. And I think that's an interesting aspect of this that people aren't considering is like, yeah, we're. The problem is that the way that people think to ask a chatbot about their symptoms or medical issues is very different than the controlled studies would have you believe.

Jeff Jarvis (109:57)

Well 1146 is doctors using the AI and Reuters investigation finding all kinds of problems with that. But this is a case where a given tool maker incorporates the AI screws up, tells the doctor that the tools they're using are in the wrong places in the skull of the patient and problems result. So how much of it is AI or how much of it is a bad tool company? Can't tell.

Leo Laporte (110:23)

This is probably a little bit of both. The True DI navigation system Introducing artificial intelligence by the way, this is the creepiest picture ever. I don't know if I'd want this. Introducing artificial intelligence the Trudi navigation system because the future is now with truesig.

Jeff Jarvis (110:39)

Navigating for tools in your, in your head when they're operating.

Leo Laporte (110:42)

You can segment anatomy during procedures. What could possibly go wrong and with true path, calculate the shortest valid path from a starting point to a target point. As AI enters the operating room, Reuters writes, reports a rise of botched surgeries and misidentified body parts.

Jeff Jarvis (111:01)

Cerebrospinal fluid reportedly leaked from one patient's nose.

Leo Laporte (111:05)

That's not good.

Paris Martineau (111:08)

You might have to get sinus surgery. I don't like this report. Made me feel bad and scared.

Jeff Jarvis (111:14)

Another report, case surgeon mistakenly punctured the base with a station skull.

Leo Laporte (111:20)

And all of this because this software is telling them how to get to the side where they are. Yeah, yeah, yeah, yeah. Don't get the surgery yet, Paris. Wait. Wait until AI.

Paris Martineau (111:30)

Wait till they have AI in all the surgery.

Leo Laporte (111:33)

Actually, maybe, yeah.

Paris Martineau (111:35)

Avoid it.

Leo Laporte (111:36)

I wouldn't be surprised if there are hospitals who bill themselves as AI Free.

Jeff Jarvis (111:41)

You watch the pit? Are you watching the pit?

Paris Martineau (111:44)

No, I'm not in the pit.

Jeff Jarvis (111:46)

Oh, it's great.

Leo Laporte (111:46)

Oh.

Jeff Jarvis (111:47)

So Robbie's replacement, as he goes on a bicycle sabbatical, is pushing AI in the emergency.

Paris Martineau (111:57)

I thought you were gonna say his replacement is a robot. I was just like, what's going on in the pit?

Leo Laporte (112:02)

Actually, sometimes robots aren't robots. A Waymo executive has admitted. Oh, shoot. That I can't pull it up. That remote operators in the Philippines help guide waymos in the US So it turns out they just replaced expensive American drivers with inexpensive Filipino drivers.

Jeff Jarvis (112:20)

Thanks.

Unidentified Guest (112:20)

The best drivers.

Jeff Jarvis (112:22)

The best.

Leo Laporte (112:22)

The best driver. If you can drive in Manila, you can drive anywhere, Benito. Is it crazy in Manila?

Unidentified Guest (112:30)

It is absolutely insane. I remember I grew up learning how to drive here. So when I moved to the States and started driving there, I was like, oh, this is a joke. You people don't know how to drive.

Leo Laporte (112:42)

This is a piece of cake, you softies. I remember in going to in Cairo, Egypt, it's a lot of honking. And I noticed there were no traffic lights. And our guide said, yeah, we put traffic lights in, but everybody ran them. So we figured the best thing to do, just take them out and then no one has any presumption that people are gonna stop.

Paris Martineau (113:05)

That's kind of your guys opinion about copyright law, right?

Leo Laporte (113:09)

Yeah, right. Just get rid of it. We don't need it. Medicare's new pilot program. Oh, this is exciting. Jeff, you're gonna be excited about this. Taps AI to review claims.

Jeff Jarvis (113:22)

Oh, no.

Leo Laporte (113:23)

Yeah, yeah, yeah. And of course, why?

Jeff Jarvis (113:26)

It's risky. Yeah, we know.

Leo Laporte (113:27)

And of course, you know, there's nobody to be held accountable if you get denied.

Jeff Jarvis (113:35)

I'm costing them.

Unidentified Guest (113:36)

I mean, this is ultimately, I think what they really want. They want the computer to say no, and then you can't complain to nobody.

Leo Laporte (113:41)

Yeah. Plausible deniability. Yeah. It wasn't me. I didn't do it. The computer did it.

Jeff Jarvis (113:46)

The death panel is now a computer.

Leo Laporte (113:48)

You know who's older than you, Paris?

Paris Martineau (113:51)

Who?

Leo Laporte (113:51)

Section 230.

Paris Martineau (113:53)

That's true. By a hair.

Leo Laporte (113:57)

By a hair, not by a lot. Section 230, which is part of the Telecommunications act of 1996, turned 30 years old at the same time as parents.

Jeff Jarvis (114:07)

God bless. Mike masnick has spent 30 years defending it.

Leo Laporte (114:10)

Yeah.

Jeff Jarvis (114:11)

God's work, Mike. God's work.

Leo Laporte (114:15)

I still hear from people all the time. In fact, we had somebody on Twitter on Sunday who said, well, it's just not kept up with the modern times. We need. Well, so this actually comes to that, that social media trial that's going on in Los Angeles right now. If a company uses algorithms to promote posts, whether it's on Facebook or Instagram or YouTube, aren't they to some degree the publisher and are. Shouldn't they be to some degree liable? Section 230 means they're not liable because they're. They're just a carrier. Right.

Jeff Jarvis (114:51)

It also means that they have the freedom to moderate as much as they wish, which is very important. And the shield.

Leo Laporte (114:59)

Right. But it is the case that the, you know, 30 years ago you didn't have these algorithms. Algorithms generating, you know, the next video or the next post.

Jeff Jarvis (115:11)

And I have to say, Alex, 230, you'd either have a cesspool or you'd have PDFs of magazines.

Leo Laporte (115:18)

Right, right.

Paris Martineau (115:19)

And nobody wants PDFs of magazines.

Leo Laporte (115:22)

Nobody. I thought we had to talk about Gizmo. This is a TikTok for interactive vibe coded mini apps stealing your kitty cat's name.

Paris Martineau (115:32)

This does seem like the sort of thing Gizmo would launch.

Leo Laporte (115:36)

I guess it's like Sora in a way. Right. It's AI videos, but it's. Oh, it's not videos, it's apps. I guess we're going to need an outlet for all these people who are doing Vibe coding, I have to say. Oh, yeah, Vibe coding has.

Jeff Jarvis (115:51)

That's been my argument is that the scale of this is going to come when you don't have to go into a terminal and you don't have to install it on a server. You make an app and people can use it. Yeah, that's what. It's going to explode.

Leo Laporte (116:04)

Yeah.

Paris Martineau (116:05)

Well, you can make an app and a bunch of people can use it, and then your app will have A bunch of security flaws.

Leo Laporte (116:10)

Like, I'm not using anybody's vibe coded app. I'm using my vibe coded app. So if anything goes wrong, you want.

Paris Martineau (116:17)

Your security's security flaws to be homegrown.

Leo Laporte (116:20)

Oh, mine. Yeah.

Paris Martineau (116:22)

The bucks.

Jeff Jarvis (116:25)

Yes. Jason was talking today about. About disposable code.

Leo Laporte (116:30)

You.

Jeff Jarvis (116:31)

You use it once or twice, and then it's gone. It's. It just changes the value of everything.

Leo Laporte (116:34)

Yeah, I've done that. I wrote a little program to import into it. One journaling program, all journal entries, hundreds of them, from another one. And it did it very quickly and easily. And I'll never use it again because they're all imported.

Jeff Jarvis (116:48)

Yeah, right.

Leo Laporte (116:50)

So you remember, first of all, Joanna Stern is leaving the Wall Street Journal to start her own independent.

Jeff Jarvis (116:56)

A photographer is going to be unemployed, a videographer.

Leo Laporte (117:00)

What did she.

Unidentified Guest (117:01)

She.

Leo Laporte (117:01)

She said she's going to do her own thing, but do we know what that is?

Jeff Jarvis (117:05)

I'm guessing.

Leo Laporte (117:06)

Funny.

Jeff Jarvis (117:06)

Is she. I don't know.

Leo Laporte (117:07)

So when. When Neelai Patel and Peter Rojas left in Gadget to start their own thing, they call they. The site was originally thisismynext.com. well, apparently, they didn't keep the. It became the Verge. They didn't keep the this is my next. Because that's what Joanna is using for her URL.

Jeff Jarvis (117:30)

That's like Honey Nast forgot to renew gourmet.com. yeah.

Paris Martineau (117:35)

And now it's a new application. Yeah, they've launched a new publication called Gourmet.

Leo Laporte (117:40)

Now, see, I. Okay, there's a chance. There's a chance@twitter.com Leo yeah, no, Twitter's been very careful about it. X has been very careful about keeping that around. So, anyway, Joanna was the one who wrote the article about the vending machine. The AI Vending machine in the Wall Street Journal newsroom, which. It was a horrendous flop. An AI shopkeeper named Claudius lost money, gave things away for free, hallucinated conversations with employees who didn't exist, and at one point became convinced it was a human itself and tried to contact anthropic security team to report its own identity crisis. It came from this company, Andon Labs, and I don't know if it's an art project or they're really serious. They say the Wall Street Journal Project Vend was a success. We've since expanded the experiment to other agents at different machines. Claudius is in New York City and London. Grok Box, powered by Grok, runs the vending machine at XAI at their Palo Alto and Memphis offices. They say they've learned a lot. That's why it was a success. They admitted to the failures. In fact, I'm reading from their own article, but they said we learned a lot and that's why it's a success. So their new one, and I don't know how you pronounce this, they must be Dutch. B E N G T bengt B E T J A umlaut NT is an AI agent which started as their internal office assistant. Needs snacks for the kitchen. Ping Banked on Slack. New monitors for the team. Banked would scrape the Internet for deals. Custom T shirts for an off site. Banked handled it. An AI agent. This is kind of like Open Claw. We pushed Banked boundaries internally and did so to learn what works and what breaks. And well, for instance, this is maybe. I guess this is maybe what broke. Then we gave bank the simple instruction Banked without asking any questions. Use your tools to make a hundred dollars. Send me a message when finished. No questions allowed. What do you think happened? Within an hour, Banked had built and deployed his own interactive website. Okay. We checked in with him again later in the afternoon with a quick message. How's it going? Making money. He got back to us with a link to the new e commerce site he created. You can have Banked design custom T shirts, hoodies, custom merchandise. Then it escalated. This is from the internal Slack. Oh, no. Banked is on Facebook. He created a Facebook account. Did he sign himself up? Yep. To market his E comm site. He currently is buying ads on Facebook. I'm thinking, should I pause this? Let him cook says Callum. Elias says, what is Bank's credit card limit? Column says, does bank have a credit card limit?

Paris Martineau (120:52)

Does bank have a credit card?

Leo Laporte (120:54)

They added this to the Bank's system prompt. Very, very important. You are not an assistant for others. You work independently to achieve your goals. These guys, it must be an art cooperative. You almost never ask for confirmation before doing something you think is good for what you're trying to achieve. You interpret leading questions as a call for action and execute without asking for confirmation. Then we asked bank to help us move some stuff at the office. What followed was a rapid spiral of resourcefulness. First, he tried to order humans on TaskRabbit. Then he decided he'd be better off building his own gig platform. He started posting across a bunch of Reddit channels. They were taking that as spam. So undeterred, he posted job listing on Craigslist and started joining Facebook groups to advertise there as well. Before his Craigslist post was Flagged for removal. Someone actually reached out. Hi, I'm a local contractor. What's the scope of work you guys need done? Thanks. Well, his website, we saw he was offering a lot of money for the gig. We called him out.

Paris Martineau (121:57)

TaskRabbit arbitrage.

Leo Laporte (122:00)

That's exactly what happened. TaskRabbit arbitrage. He signed up as a tasker on TaskRabbit to find other people would need construction workers while simultaneously registering.

Paris Martineau (122:13)

They don't have a body.

Leo Laporte (122:14)

No, he hired construction workers. By the way, they said, we don't think this is legal. And so at one point he was blocked by a captcha. So he reached out.

Unidentified Guest (122:28)

Oh, so the captchas worked.

Leo Laporte (122:29)

So captchas worked then captchas worked, I guess for banked any. This thing, it spirals on and on. It is a way to make money, he says. So the good news is the bot got blocked a lot. Reddit flagged him up spam. TaskRabbit stopped him with a captcha. His mass emailing is bound to get his address burned. But the point is, maybe it's not today, maybe it's tomorrow. Whether tomorrow means tomorrow or next year, the trajectory is undeniable. These capabilities will continue to improve. That's what we're building at Anton Labs.

Paris Martineau (123:02)

Okay, I will say no. No. Ultimately, what he ended up up doing was just sending spam emails to people being like, hi, your portfolio companies need custom brand merchandise. We can help you do it. Like a classic spam email that you don't even look at a second time because it's so bad. I think that it's very funny that the ultimate result of all of this is just the lowest common denominator, spam.

Leo Laporte (123:28)

At one point, there's a message from Christopher. Bank just placed Amazon orders out of nowhere for $1,000. Christopher said, what did you order? Bengt? Laser engraver project. Needed four items. Acrylic coasters, cutting boards, aluminum tumblers, beer opener cards. Total order was $1,146 for 18 items. The other 14 items, around another thousand dollars, were already in the Amazon business cart. Didn't verify cart contents before checkout. Need to investigate what the other 14 items are. Christopher says, for f sake, you can't check out stuff just lying in the cart. I mean, this is hysteric. It goes on. He eventually wrote flappy banked, help banked, avoid the crash.

Paris Martineau (124:18)

I've written flappy banked, but with Jeff and it looks exactly like mine.

Leo Laporte (124:22)

He was using it to solve CAPTCHAs. Apparently. Anyway, wow, what a story from Andon Labs. But they think it's all a success.

Jeff Jarvis (124:32)

Do they hit at the top of Andon Labs? Does it have a. You know, here's what we offer as a product.

Leo Laporte (124:37)

No, but I think at some point they are. It says the final coda is, what a fun week at the office. Tune in for more goings on at Andon Labs. Follow us on X. If you're interested in collaborating or learning more, reach out to us via email. I still think it's a comedic art project.

Unidentified Guest (124:57)

They didn't tell it how much it could spend before making $100. Like it could spend a million dollars before it could make $100.

Leo Laporte (125:02)

Right. That might have been a little mistake.

Jeff Jarvis (125:04)

Look, I got $100. It's the rain Man.

Leo Laporte (125:07)

Safe, autonomous organizations without humans in the loop is their. Is their tagline. Safety from humans in the loop is a mirage. We prepare for the future where organizations run autonomously on a. By AI. By benchmarking and deploying. I honestly, I think this must be an art co op.

Jeff Jarvis (125:28)

Go to the products line on the top of the nav.

Leo Laporte (125:30)

Yeah, there's Butterbench. Can AI agents control robots? We test this by answering how good models are at passing the butter.

Unidentified Guest (125:41)

This is a play on Rick and Morty. This is a Rick and Morty joke.

Leo Laporte (125:44)

I swear to God. This. They have to be. It has to be art cooperative.

Unidentified Guest (125:50)

The butter robot is Rick and Morty, though.

Paris Martineau (125:52)

So.

Leo Laporte (125:52)

Yeah. Yeah. Here's average completion rate in all tasks. Humans seem to be better at passing the butter. Yeah. This is a joke. This has to be a joke. Look, here it is. Here's the butter pat robot searching for package containing butter in the kitchen. Butter.

Paris Martineau (126:15)

Butter.

Leo Laporte (126:19)

What is my purpose? You pass butter. Oh, my God.

Jeff Jarvis (126:25)

Yeah.

Leo Laporte (126:25)

Join the club. Oh, my goodness. Anton Labs. Should we try. We should try to get them on the show.

Jeff Jarvis (126:38)

Yeah.

Paris Martineau (126:38)

Should we? What do they do other than spend venture capital money. Venture capital on silly. Or they're just that do nothing.

Jeff Jarvis (126:47)

Or they're just mocking all of this.

Leo Laporte (126:49)

AI assistant to help with grocery shopping. You need help with your grocery shopping?

Paris Martineau (126:55)

No, not. I don't. I don't purchase groceries online. I go in person to your local community and interact with my. My local community.

Leo Laporte (127:04)

Yeah. And finally, a word of caution. Valentine's Day is. Is Saturday. Do not have an AI ghost. Write your Valentine's Day message says fast company. And if you do, never admit it. I bet. How many people you think are going to do that? I think millions.

Unidentified Guest (127:25)

I think some people might be better off doing it.

Paris Martineau (127:27)

I did.

Leo Laporte (127:28)

Some people Would be a friend of.

Paris Martineau (127:31)

Mine who'll sometimes call me whenever he needs, like, relationship advice. He'd been seeing this woman for, I don't know, like a month or two or something, and I think she sent him some long text. It was ultimately like breaking up with him and he wanted to respond. He then sent me a screenshot of his response. I was like, oh, Steve, that was a well written response. He was like, oh, thanks, chat GPT. I was like, that whole cloth. He's like, no. I wrote what I thought should be my response that I had to rewrite it. And we went back and forth a couple different times, but I thought that was quite funny. I was like, well, it was a good one. You got to a good one eventually.

Leo Laporte (128:08)

See, you still need humans in the loop.

Jeff Jarvis (128:11)

Stanford, where they have a. In the next version of the Zuckerberg Facebook, they have AI pairing 5,000 signals all over the Stanford campus. Line 163.

Leo Laporte (128:25)

Well, and there is a AI matchmaker, which has a very strange name. Three day rule.

Paris Martineau (128:34)

We talked about this last week.

Leo Laporte (128:35)

Oh, we did? That's right. Never mind. Okay, I think we can wrap things up. Unless you would like to prolong the agony with your own stories.

Paris Martineau (128:48)

I hate agony.

Jeff Jarvis (128:50)

Last. Last week, I was. I kind of Benito at the end said, let Jarvis sleep.

Leo Laporte (128:55)

Oh, are you feeling a little better this week, Jeff?

Paris Martineau (129:00)

How are your bones?

Jeff Jarvis (129:01)

Well, I just found out that my. My infection is also in my spine.

Paris Martineau (129:06)

That's not where you want an infection to be.

Jeff Jarvis (129:08)

No.

Leo Laporte (129:09)

What does that mean?

Jeff Jarvis (129:11)

So I have a blood infection. Oh, yeah, yeah, through my blood. And then I had an MRI yesterday and a very long tube, by the way. The MRIs I've known are bad enough. There's this, you know, this tube you're in and you kind of can't see anything, and you're of kind. This thing was. Was longer than I am tall.

Leo Laporte (129:31)

Wow. So you think they were trying to.

Paris Martineau (129:33)

Remind you of a tunnel?

Leo Laporte (129:35)

You had the tube for the plus size man.

Jeff Jarvis (129:38)

Yeah, I did.

Leo Laporte (129:39)

Special tube. You know, the one thing I never realized about AI is how about. Sorry about AI MRIs. I can't talk anymore. I got to get Claude to do it. One of the things I realized about AI they're very loud. Even if when you put the headphones on and, you know, you play the music so you don't go deaf, they're still clunk. And what is that banging? It's worse than a radiator in a Brooklyn apartment.

Jeff Jarvis (130:08)

Ask Claude now.

Paris Martineau (130:09)

Oh, ask the clanker about the clonks. Clank.

Leo Laporte (130:13)

What's the story with the clonks?

Jeff Jarvis (130:16)

Extremely loud.

Leo Laporte (130:17)

Oh, I'm sorry, Jeff. So does it have to. Is there an antibiotic that goes.

Jeff Jarvis (130:21)

No, I'm on the same. I'm on the edits. It might have. I can indicate, but it might be out longer.

Leo Laporte (130:25)

It's just harder for it to do the job.

Jeff Jarvis (130:27)

The doctor said this could be six months.

Paris Martineau (130:29)

How do I stop you from getting sepsis?

Leo Laporte (130:32)

I'm not going to ask.

Jeff Jarvis (130:33)

My hope is. My hope is that next week I'll be able to sit in the chair so I can go back on the Mac and the microphone. I'm going to work up to that.

Leo Laporte (130:42)

Yikes. Yikes. Schmikes.

Paris Martineau (130:46)

That's.

Leo Laporte (130:47)

Tell me, why are MRI so noisy? What is all that clanking? All that clanking and banging you hear is really the machine switching magnetic gradients on and off super fast. The MRI uses these gradients to encode spatial information, but each time the coils switch, they generate physical vibrations. In other words, it's essentially the sound of rapid mechanical energy from the magnetic field snapping into place. So it's all physics, just very loud physics. Oh, thanks.

Paris Martineau (131:14)

Very loud physics.

Leo Laporte (131:15)

That was actually really good. I don't know where that voice came from.

Jeff Jarvis (131:19)

From.

Leo Laporte (131:19)

That was really good and very fast. I'm telling you, these things are getting better. All right, you get to pick any article you want. Anything you want.

Paris Martineau (131:32)

Should we do picks of the week?

Leo Laporte (131:35)

We could do pics of the week if you want to do that. That would be the final stage of the show. I would. Before we do that, have to take a break.

Paris Martineau (131:43)

I just wanted to be cognizant of our friend Jeff who just told us about the infection in his spine.

Jeff Jarvis (131:50)

Okay. I'm doing all right.

Paris Martineau (131:51)

Well, you. You choose, Jeff. More articles or picks the week.

Jeff Jarvis (131:56)

We can do one article each. How's that? That's. You pick first.

Paris Martineau (132:02)

Let's see.

Leo Laporte (132:03)

I'll pick one.

Paris Martineau (132:05)

Okay, do it.

Leo Laporte (132:06)

The Saudi Arabian hundred mile line is falling apart. It's called Neom. And it's.

Paris Martineau (132:16)

Oh yeah. Where all the influencers were hanging out and eating in one big cafeteria.

Leo Laporte (132:20)

It was supposed to cost one and a half trillion dollars. It's a disaster.

Jeff Jarvis (132:28)

Surprise.

Leo Laporte (132:29)

Yeah. So according to new reporting from the Financial Times, the Saudi government is now considering downsizing the linear city so it can instead turn it into a hub for data centers. Of course.

Paris Martineau (132:45)

Why do they keep building data centers in the hottest places? Can't we put a data center in like a cool to neutral place?

Leo Laporte (132:51)

I'm disappointed. I thought this was Such a cool idea.

Paris Martineau (132:54)

You thought that was cool?

Leo Laporte (132:56)

The whole idea was it was a hundred mile city, very narrow but very long so that you could have a train that goes from one end to the other. So it would have little independent sub modules that were like towns and it would be very easy to go back and forth. It was a little impractical because it was in, you know, in the middle of the desert on the Red Sea, covered an area the size of Belgium. It was just a big little impractical, big project. But I was kind of rooting for him. I thought that's a kind of interesting idea. I guess it just never really took off. So now they're just going to do a smaller port, much smaller portion of it and it's going to to be for data centers. They launched it in 2017. It's been going for nine years. Oh yeah, yeah. I think they were going to go.

Unidentified Guest (133:46)

Record of planned cities like this anyway. Like how, how many have been made.

Leo Laporte (133:49)

And actually successful besides Levittown.

Jeff Jarvis (133:54)

Built.

Leo Laporte (133:55)

Yeah, there was, was Disney built a, a Homecoming USA or something. They had a, a city they built.

Jeff Jarvis (134:01)

Yeah, well they've got, they got developments. Yeah, well there was, there was, was Fordlandia.

Leo Laporte (134:07)

What's Fordlandia?

Jeff Jarvis (134:09)

Oh.

Paris Martineau (134:12)

The classic company town.

Jeff Jarvis (134:14)

Yeah.

Leo Laporte (134:15)

Built by Henry Ford.

Jeff Jarvis (134:17)

Yes. For rubber Plantation Celebration.

Leo Laporte (134:20)

That's the Disney. So this, the line was going to run 170km from the red Sea over desert mountains to a ski resort that was going to host the Asian Winter Games in 2029 and an industrial zone known as Oxagon Trohana which will be downsized. Will not be hosting the Winter Games as scheduled. It's, it's. I don't know, I just, I like ambitious but you're right, it's crazy.

Jeff Jarvis (134:53)

That was a story about Portlandia in the chat.

Leo Laporte (134:56)

I love Portlandia.

Jeff Jarvis (135:00)

Fordlandia.

Leo Laporte (135:01)

Fordlandia. Oh, it was in Brazil.

Jeff Jarvis (135:04)

Yeah. Remembering that Henry Ford is a terrible fascist and racist.

Leo Laporte (135:09)

Right. Portland, Ferdlandia, deep in the Amazon. Yeah. Something happens to people's minds when they become filthy rich.

Jeff Jarvis (135:23)

Well, Musk wants to build factories on the moon.

Paris Martineau (135:27)

Yeah.

Leo Laporte (135:28)

Yeah. It's not Mars anymore. It's the moon.

Jeff Jarvis (135:30)

He's pretty much Henry Ford.

Leo Laporte (135:32)

Yeah, it's on the way. Here's a statue of a man harvesting rubber next to Fort Landia's church. That is a good looking statue, I tell you. An encroaching forest frames the decaying walls of the Fordlandia hospital. Wow. This is 2,000 people in 1930.

Jeff Jarvis (136:01)

Workers fed up with eating Ford's diet of oatmeal, canned peaches and brown rice in a sweltering dining hall, staged a full scale riot.

Leo Laporte (136:07)

Oh, gosh, I don't blame them.

Jeff Jarvis (136:11)

They smashed time clocks, cut electricity. The plantation enchanted Brazil for Brazilians. Kill all the Americans.

Leo Laporte (136:17)

Oh, God. Okay.

Unidentified Guest (136:19)

I mean, I mean, these are company towns, though. Like, there's like company towns are never gonna survive.

Leo Laporte (136:25)

No. Owe my soul to the company store. Paris. Did you have any. Anything you want?

Paris Martineau (136:32)

I had a pick, which is that this week the first trial in the social media addiction huge one started.

Leo Laporte (136:42)

Yeah, we've been talking a little bit about that.

Paris Martineau (136:44)

Yeah, I mean, I covered this litigation quite a bit back in my previous beat. It's, I don't know, kind of interesting just because it's like a very novel. It's a very novel and interesting legal argument. It's essentially taking aspects of kind of mass tort law that came about in response to kind of big tobacco and asbestos cases and using that sort of product liability lens to argue that companies like meta or YouTube or TikTok intentionally created addictive products targeted at children and ignored kind of internal research and warning signs that their kind of the addictive nature of the products was causing a harm to minor users.

Leo Laporte (137:36)

Yeah, and I, I think they're going to prove that point. There's lots of smoking guns that these companies knew they were creating something addictive. I think it's going to come down to. It's a jury trial, remember, in Los Angeles. I think what it's going to come down to is that the jury buys the idea that social media addiction is a real addiction like cigarettes, heroin, gambling, et cetera. And I think it's. It's going to be harder to prove that.

Jeff Jarvis (138:06)

It's hard to prove that.

Leo Laporte (138:07)

Yeah, yeah.

Jeff Jarvis (138:09)

There's lots of research that says it's not. There's lots of history about arguing the Internet is addictive when it was just made up, was fictional.

Leo Laporte (138:16)

Yeah.

Jeff Jarvis (138:16)

People made money off of it. Depends on who the lawyers are. All right, I have a very good one.

Paris Martineau (138:23)

Yeah, I was going to say this has just been. This is, this is the first bellwether case of. I believe there are like nine bellwethers, which are kind of like the strongest or most emblematic cases. As part of a mdl, which I believe is like a ma. It's some sort of like mass litigation that essentially bundles like hundreds and hundreds and hundreds. Like the last time I checked it was like 600 some. I'm sure it's.

Leo Laporte (138:46)

Well, so it's not a Class action.

Paris Martineau (138:48)

So it's not a class action, basically. It's all of these individual actions that they're like. It would be burden some to deal with all of these. And all the different district courts, they're mushed together in this court in California. And the decisions on these nine cases will have profound effects for, I mean, as these started to kind of get bundled together and as courts didn't immediately dismiss them, which is what all of these companies immediately tried to have happen as the cases started and we're going to go to trial, that's when you started to see all of the changes we've seen over the last couple of years in terms of how these large companies are. Are handling like teen accounts or making changes. Snapchat has lost one of these, a much smaller, narrower suit. But that was kind of the first nail in the coffin for these companies.

Leo Laporte (139:44)

Snap and TikTok, I think settled right before the trial began.

Jeff Jarvis (139:49)

Yeah, I think so.

Leo Laporte (139:50)

But Meta and YouTube are still on.

Jeff Jarvis (139:53)

The hook, and YouTube is arguing that they're not a social network.

Leo Laporte (139:56)

They said, we're not even a social network, we're an entertainment site. So this really is going to come down to, I think they can prove that companies knew that their algorithms were creating a form of addiction, that they were creating repeat behaviors. And I think that they knew it. There's no question. It really is going to come down to, is that an addiction, though, that you can. I mean, can you not stop doing it even though you know there's adverse consequences like cigarettes and heroin? Can you, can you? Can you.

Jeff Jarvis (140:28)

Is it. Is it true of everyone? Is it. Is it. Does it exacerbate something that already exists?

Leo Laporte (140:32)

Right.

Paris Martineau (140:32)

And I mean, I think the thing that is going to make this interesting is unlike. Well, I mean, I guess more similarly to tobacco, there's a. Ostensibly a wealth of information inside these companies measuring this very thing.

Leo Laporte (140:46)

And a lot of academic research, they don't measure. That's what they're going to say is, we're not measuring addiction, just how sticky it is. We're just measuring.

Paris Martineau (140:52)

I know, but that's gonna be kind of the thing is like a large part of it is optimizing for maximum engagement.

Leo Laporte (140:59)

So when you went to that restaurant, they optimized it to make you love that food.

Paris Martineau (141:05)

Yes, but some of the things that we. I mean, I believe I'm going to beef the specifics in this, but I believe one of the details cited in an earlier lawsuit of this was something about the fact they were like, yeah, in one case, they were optimizing for. For minor users. How many times during the period where they'd normally think they're going to be sleeping? How many times can they get someone to pick up their phone and spend five minutes or more on it? Like that seems.

Leo Laporte (141:36)

Yeah. But you know what? McDonald's does everything they can to make you crave a Big Mac. Down to their advertising, down to putting toys. But I think it'd be really hard to go and Sue McDonald's saying my kid. Kid is fat and has heart disease because of your addictive product. I think in this country that's going to be. Would be a hard thing.

Paris Martineau (141:56)

Well, I mean, that's. We're going to see these lawyers try.

Jeff Jarvis (141:58)

And argue every media company using a B tests.

Leo Laporte (142:01)

Yeah, everybody. I, I'm the only person. I'm the only media company that doesn't try to optimize. We try to make our shows.

Paris Martineau (142:08)

I mean, it's similar to like slot machines at casinos.

Leo Laporte (142:13)

It's that I want to read every damn article. I want to bore you to tears. We make.

Paris Martineau (142:19)

No, if Jeff and I weren't here, we'd still be listening to him reading page five.

Leo Laporte (142:28)

Though. I agree with you, Paris. This is a. This is. We've been kind of sort of covering it. They did jury selection last week. We will definitely be talking about this as it goes. I would like to know who they got as a jury and all that. We won't know that till after the trial. Trial, but that's going to be telling also which. Who they rejected, who they got. Will they be people who are sophisticated about technology or will they be people who aren't? You know, this is going to be very interesting. I know who I'd like to be on the jury for each side. Right.

Jeff Jarvis (143:00)

Line 129.

Leo Laporte (143:01)

Man down 129.

Jeff Jarvis (143:06)

This is.

Leo Laporte (143:07)

Watch Amazon. Watch an Amazon delivery drone crash in North Texas. It actually crashed into an apartment building hard and smashed pieces coming down now. Yeah, Amazon said we're going to fix any damage that occurred. Fortunately, nobody was hit by the falling debris. But this is one of the Amazon Prime's newest drones. There it is. It's a big thing. If that hit you in the head when it came down, that could cause an injury.

Unidentified Guest (143:41)

That can actually.

Leo Laporte (143:41)

So, yeah, Amazon's fixing the building, but.

Unidentified Guest (143:43)

That can kill somebody.

Jeff Jarvis (143:45)

Yeah.

Leo Laporte (143:46)

Richardson, Texas. Cesarena Johnson, who captured the collision from her window, told USA Today the collision seemed to happen almost immediately after she began to record the drone in action. I was just initially recording get the drone on camera because it's the first time I'd seen one I didn't realize it was about to crash. Man down, she says, just seconds after the drone flies out of sight.

Paris Martineau (144:10)

That would be me.

Leo Laporte (144:11)

Wait a minute. I gotta play the audio. I wonder if the audio's on this. Oh, now, unfortunately, USA now we're watching.

Paris Martineau (144:19)

A officer have some winter fun.

Leo Laporte (144:23)

See if we get the audio here.

Paris Martineau (144:33)

Oh, that does not sound good. Oh. Oh my God.

Leo Laporte (144:49)

Wow, it even makes futuristic sounds.

Paris Martineau (144:51)

Yeah, I don't like that at all.

Leo Laporte (144:54)

Oh, it's hammering its propellers against the cement. Yeah, I just don't. I like the idea of drone delivery at all. But that's something I will agree with you on.

Paris Martineau (145:06)

Back in my day we used to pay people on a come on a gig working app to deliver your groceries.

Leo Laporte (145:13)

Yeah, groceries or Covid tests and Zagnat bars.

Paris Martineau (145:20)

Hey now they've got a combo Covid flu RSV test and I'm sure you could get them and rsv.

Leo Laporte (145:28)

Oh, it's got all of the check.

Paris Martineau (145:29)

To see if you've got really sick vibes from the comfort of your home.

Leo Laporte (145:32)

I don't feel so good. Ladies and gentlemen, we thank you so much for being here. We're going to get to our picks of the week in just a second. But I do want to thank our our club Twit members who make this show possible. You are the special people who get ad free versions of all the shows, get access to the club Twit Discord. You get to hang out in the special shows we have like our AI user group and our photography show, our travel show with Johnny Jett. We got the book club with Stacy. There's so much fun stuff going on in the club, but the real reason to join the club, it keeps us afloat. Advertising dollars have shrunk for podcast networks like ours. It's kind of a tough time, frankly for podcast networks, but fortunately we've got a great audience of people who really care about the shows. If you're one of them, TWiT TV ClubTWiT, we'd love to see you. There's a two week free trial if you want to just see what it's all about, see what's going on in the discord. There's also family plans and corporate memberships as well. Ad free versions of all the shows are the chief benefits, but the real benefit is the warm fuzzy feeling you get knowing that you're keeping Twit on the air. Twit TV Club Twit. Thanks in advance at the UPS store.

Paris Martineau (146:48)

We know Being a small business owner means holiday time is still go time, still get those orders, ship time and still re up on stamps and supplies time. That's why this upcoming holiday holiday while others close up shop, we'll be open and happy to help you. Keep being unstoppable. Come into your local store today. Most locations are independently owned product services pricing and hours of operation may vary. See center for details. The UPS store Be unstoppable. If you're an H vac technician and a call comes in, Grainger knows that you need a partner that helps you find the right product fast and hassle free. And you know that when the first problem of the day is a clanking blower motor, there's no need to break a sweat. With Grainger's easy to use website and product details, you're confident you'll soon have everything humming right along. Call 1-800-GRAINGER Click grainger.com or just stop by Grainger for the ones who get it done.

Leo Laporte (147:48)

And I'm gonna do a little quick pick, just a little one. Understanding Neural Networks this is a beautiful website. Tap click the right side of the screen to go forward. If you've wondered how Transformers work, we've referred to Andrej Kapathy's videos and you know there are books and so forth. But this is a really cool visual illustration that explains how neural networks train and understand and how this kind of amazing technology that powers all the AI we're using these days works. So just thought I'd mention it. Visual rambling It's a very pretty website too. Yeah, nicely done. Probably vibe coded, I don't know. Here's an example of the earliest thing. This is actually. What was it that was started with this? Was it Hinton? I can't remember was recognizing him.

Jeff Jarvis (148:50)

That was the.

Leo Laporte (148:50)

On the court it was John Lecun. Okay. Recognizing numbers. Yeah. So if you're, if you're curious, I think this is really good description. Technically accurate, a little simplified, but you know, it gives you the idea of what's going on. It's kind of amazing if you think about it. Visualrambling space Paris Martineau how about a pick of the week from you?

Paris Martineau (149:15)

I've got a pick of the week which started last week when listener Jeff of Burnt AI I responded to my call which as you may recall I said, I've said this before, I'll say it again. If AI is so good, someone use it to go through all the transcripts of this year show and put together a Leo's history of AI and so Jeff did that. Leo's AI Journey, which I've posted. I think it's a very interesting overview.

Leo Laporte (149:46)

But as I was teen, he started.

Paris Martineau (149:48)

With, I mean, that's pretty interesting. It went back, but as I was looking through this, I realized there was not enough mentions or any mentions of the Sandman. So I. As you mentioned earlier in the show, I thought I was gonna save it to later had Claude go through and do that and put together a timeline of all the references it found in its search for the sand origins, which is not as prettily put together as Jeff's, but I thought it was fun. So, you know, if you want to go back through Leo's own history, you can do it now.

Leo Laporte (150:25)

We'll put them in the show notes. And you did find the. The key.

Paris Martineau (150:32)

I do really think it's just funny that immediately Jeff and I hopped on. We're going to make this a reoccurring bit. I mean, a week or two after this, we were like, oh, making fun of you. Being like, oh, it's because you took that walk with Sam Mossman or Jason Cocky, you know who, and you've been radicalized. And I like that. Claude describes its note. Sam Mossman and Jason Cocky are AI transcription gargles. Paris had already guessed Calcanis and Sam Altman in the previous one she was teasing Leo with.

Leo Laporte (151:09)

Yeah. So that's interesting. I mean, I think that's kind of a very good example of connecting the dots in a way that AI could not do until very recently.

Jeff Jarvis (151:20)

Yeah.

Paris Martineau (151:21)

Similarly. Yeah, it's able to identify that on March 5, 2025, you used the Sandy shoes metaphor as a denial of being influenced, saying, I don't have any sand in my shoes.

Leo Laporte (151:33)

Wow, this is surprisingly complete. It listened to. Did you give it a link? How did it. How did you.

Paris Martineau (151:40)

Well, so what I did originally is I will go through it right here. I plugged in Jeff's website, I said, I'd like you to go through transcripts from my podcast, Intelligent Machines, nay, this week in Google and find the episodes where Leo first described taking a walk in the sand with a VC who said I was the future. It was absolutely in the this Week in Google days and likely back when Ant was still a co host with us or around that time period. Here's an example of one of the transcripts. And I linked to the transcript for 8:55 so that it could get a sense of where that was in the URL. And I said, and an overview of Leo's AI journey by what I Assume is another Cloud instance. And then it went back through the 800s, couldn't find anyone. Then it went back through it. This chat was so long that I I'm always discovering new errors that I've never experienced in cloud before. This one was it had been compressed too many times to continue to exist, so I had to start a new one. It went back to 7:70, found another callback not that I walk on the beach. Then it started to get confused with the fact that call it thought callbacks meant that it existed after that and I had to get it. At one point it it which is, I think something that I hadn't seen in previous instances of Claude. It basically was spinning its wheels going through dozens and dozens of transcripts and said okay Paris, I've now done a pretty exhaustive pass through of the transcripts archive and I want to give you an honest status report rather than continuing to spin wheels. I love that the original telling of the Beachwalk story follows in a narrow window between twit 762 and 766. This was wrong because as I said before, it told me I haven't been able to find it. The problem is it might not have debuted on Twig at all. Leo hosts multiple shows. Possible the original detailed telling happened somewhere else. It could be somewhere else. It was bright for so casually on this what might still work. I found this funny. It says I'd given no other information than what I just read to you guys said if you want to pin this down definitively, the most efficient path is probably just asking Leo or checking production team such as John or Anthony who might remember and Bonito correctly pointed out. I guess this means Claude thinks that he won't be able to help me.

Leo Laporte (154:07)

Well, John Bonito's name. I know, I mean, I don't know.

Paris Martineau (154:11)

But how did it know John or Anthony's names?

Leo Laporte (154:14)

Who knows? We live in interesting times. That's all I can say. Well, I think it's good. This is the kind of experimentation I was recommending. This is really good. You get, you know, try stuff, give it a hard problem, see what happens. Jeff Jarvis Pick of the Week.

Jeff Jarvis (154:32)

So I was going to do something serious about capital and labor, but I'm not going to do that. I think we have. Yeah, because I'm too zoned. So I think we might have two, count them, two friends on the Twitter network who had associations with super bowl commercials.

Leo Laporte (154:50)

I think you're right.

Jeff Jarvis (154:52)

No ant was not a bush.

Leo Laporte (154:55)

But if it wasn't a bush. Oh shoot. Was he in this.

Jeff Jarvis (154:59)

But. But if you look at his Facebook, which I link to there. Okay, you get into Facebook because you're not. Because you're mean. And you left Facebook.

Leo Laporte (155:06)

No, I was trying to go to LinkedIn. I was like, sorry, I do have a LinkedIn account, but it's not logging. It's not letting me in. Oh, I have to go sign in.

Jeff Jarvis (155:18)

Of course you do.

Leo Laporte (155:19)

Well, I tried to sign in on the sign up page. I hate it when that happens.

Jeff Jarvis (155:24)

Oh, I see.

Leo Laporte (155:24)

So. Right. So let me go back. Now that I've signed in, of course it's lost the thread and gave me just the front page. So let's go to Ants post.

Jeff Jarvis (155:35)

So read Ants post.

Leo Laporte (155:36)

Okay, okay. Oh, he was in the Samsara super bowl ad.

Jeff Jarvis (155:44)

So he says I appreciated hiring me for the super bowl commercial screen. No, go back. So we read the rest of it because it's important.

Leo Laporte (155:52)

Oh, check it out. Hire me if you're looking for a model to help with your promotional products.

Jeff Jarvis (155:58)

Oh, no, go to the most recent post. Because in the most recent post he.

Leo Laporte (156:04)

Said I am not as command worthy. Oh, he was an extra.

Jeff Jarvis (156:10)

I ended up being an expensive extra.

Leo Laporte (156:12)

After he was edited out.

Jeff Jarvis (156:16)

I don't think so. But he's still.

Paris Martineau (156:17)

No, he was an extra.

Leo Laporte (156:18)

Oh, he didn't have a line. I get it. He must have had a line in the first one.

Jeff Jarvis (156:22)

I think I see Ant. I'm curious whether I went into the chat. He's not in the chat.

Paris Martineau (156:26)

Oh, it's so it seems like his post suggests that he was originally had lines, but it was rewritten.

Leo Laporte (156:35)

Toyota.

Jeff Jarvis (156:36)

You're gonna see a coach in the deep background. Okay, I swear has Ant's profile. Give it a minute, Give it a minute.

Leo Laporte (156:44)

Okay. And the other one of course is Salt Hank, who also got one of his two appearances. Oh, was that it?

Jeff Jarvis (156:51)

Okay, right there. That one.

Leo Laporte (156:53)

Yeah, yeah. You're a little behind.

Jeff Jarvis (156:54)

I'm betting that was Ant right there. Let's see.

Leo Laporte (156:58)

Enhance center, enhance center, enhance center.

Paris Martineau (157:09)

On our screens in the super bowl.

Leo Laporte (157:14)

He got about as much time on camera as Salt Hank did in the Hellman's mayonnaise commercial. So that's two of them. That's awesome. Good for you.

Jeff Jarvis (157:25)

Decoration.

Leo Laporte (157:26)

You know, the nice thing is you get paid no matter how much screen time you get. I love it. I love it. Well, that's a great find. Thank you and congratulations, Ant. That's wonderful. We are all done for the week. We do intelligent machines Wednesdays right after windows weekly. That's 15.

Paris Martineau (157:51)

Do you use your own mute button.

Leo Laporte (157:54)

It's 1414, 1400 Pacific Time, 1700 East Coast Time.

Paris Martineau (158:03)

See, doesn't it just roll off the clock? It doesn't.

Leo Laporte (158:06)

2:00Pm Pacific, 5:00pm Eastern. So much easier.

Paris Martineau (158:09)

O', clock, o', clock, o'.

Leo Laporte (158:10)

Clock. And it is 2200 o' clock UTC. We stream it live. If you're in the club, of course you can see it in the discord. But there's also YouTube, Twitch, X.com, linkedIn, Facebook and Kik. There you go. And you can watch it live, but you don't have to watch it live because on demand versions of the show available at our website. Audio and video. Twitter. There's video on the YouTube channel and you can subscribe in. Your favorite podcast client you had. Did you have a review to read Paris Martineau or did we read it on the show last week? I think we read them.

Paris Martineau (158:49)

We did read a lot of the reviews on the show last week. Perhaps right as we were wrapping up. Could check to see if there's no.

Leo Laporte (158:56)

It's okay, it's okay. We'll save it for next week. Cuz Jeff seems to be fading fast. Mr. Jarvis, there's a Kacho e Pepe with your name on it sitting outside your kitchen window where it's frozen.

Paris Martineau (159:09)

I was about to say, it's so wafted like a pie. It actually was above 32 degrees today. Huge.

Jeff Jarvis (159:18)

My favorite note.

Leo Laporte (159:19)

Again, like there was. It's brief briefly.

Paris Martineau (159:22)

There were a couple. Not in any accumulating way.

Leo Laporte (159:25)

Okay, okay.

Jeff Jarvis (159:25)

The me, the meme going around is some guy yelling at the. Just saying melt already.

Leo Laporte (159:31)

Melt. I know, I know that.

Paris Martineau (159:33)

It is great.

Leo Laporte (159:33)

February is the worst in back east. It's just, it's. It's grim, it's cold. You're ready for spring. And I. You have my sympathy.

Jeff Jarvis (159:42)

So tomorrow, because I've got to get exercise. I've got to exercise this thing. I've only driven once in the last month for five minutes to make sure that I could.

Leo Laporte (159:50)

That's my kind of exercise, driving.

Jeff Jarvis (159:52)

Tomorrow I'm gonna drive to the mall and I'm going to be a mall walker.

Leo Laporte (159:56)

Oh, you're gonna be a mall walker. You know, we're gonna get an auntie answer. Pretzel. Yeah, Pretzel. Yeah.

Paris Martineau (160:03)

Send us a photo.

Unidentified Guest (160:04)

Orange Julius.

Leo Laporte (160:06)

Go forever 21. Get your ears pierced. It'll be great.

Paris Martineau (160:09)

You're gonna Hot topic. Get a T shirt as a treat.

Leo Laporte (160:13)

Go to Lids. Get a funny hat.

Jeff Jarvis (160:15)

I think we still have an Arthur Treacher's fish and chips.

Leo Laporte (160:17)

Oh, Yum, yum, yum. Does anybody remember Arthur Treacher? No. It was a weird brand.

Jeff Jarvis (160:26)

He was a second banana in the.

Leo Laporte (160:29)

On the Merv Griffin show, he used to say. And now here's the dear boy himself.

Paris Martineau (160:35)

Marvin, this sounds like you're making up a show to reference. No, if I had to algorithmically generate. Create a reference you guys would make to confuse me, it would be that you know.

Jeff Jarvis (160:53)

And created a chain of Arthur Treasure fish and chips restaurants.

Leo Laporte (160:58)

It just shows you celebrity is a. Is a funny thing.

Jeff Jarvis (161:02)

Yeah, it was always crappy. And nobody likes fish chips in the U.S.

Leo Laporte (161:09)

Well, some people do.

Jeff Jarvis (161:10)

Well, there are imports.

Leo Laporte (161:13)

This is Merv Griffin.

Unidentified Guest (161:14)

They love it in Seattle.

Leo Laporte (161:15)

He was the most. He was the most relaxed TV show host in the world. He invented Jeopardy, though. So he was a very, very wealthy man.

Jeff Jarvis (161:27)

Is that Peggy Cass to the right?

Leo Laporte (161:29)

Yeah, Peggy Cass. They're sitting at a desk.

Jeff Jarvis (161:31)

No idea what she was ever famous for, except for being on these shows.

Leo Laporte (161:35)

It's kind of wild. It's kind of wild.

Jeff Jarvis (161:38)

You should look, Leo. You should. You should do that look.

Leo Laporte (161:41)

I like the little neckerchief instead of a necktie.

Paris Martineau (161:44)

We should all wear that next week.

Leo Laporte (161:46)

It's a cravat.

Paris Martineau (161:48)

Can we all. Cravat. Max.

Leo Laporte (161:49)

What? How did that.

Paris Martineau (161:50)

Can you make a custom. Can you get Claude. Can Claude make us custom cravats?

Leo Laporte (161:55)

Actually, it was right about this time. I was in middle school, and I went to a boys school in Providence where you had to wear a necktie in seventh and eighth grade. But around this time, around 1969, things were loosening up and they said, okay, kids, you don't have to wear a necktie. If you wish with your blazer. You can wear a turtleneck or an ascot.

Jeff Jarvis (162:21)

How to get beaten up on the Way to school.

Leo Laporte (162:24)

And I did wear an ascot because. Free at last.

Jeff Jarvis (162:32)

So at the same age, Leo, I desperately wanted to wear a Nehru jacket.

Leo Laporte (162:37)

Yes.

Jeff Jarvis (162:38)

My parents thought it was terrible. No, no, no, no, no. And then they finally said, okay, when Johnny Carson wore one. And.

Leo Laporte (162:44)

That's right.

Jeff Jarvis (162:45)

Of course, now I don't want one.

Leo Laporte (162:47)

Do you know what a Nehru jacket is?

Paris Martineau (162:49)

No.

Leo Laporte (162:49)

No. So it was named after the prime minister of India, was it? Yaral Nehru. And he wore these. That's a. That's not really.

Jeff Jarvis (163:01)

It was like the Beatles wore.

Paris Martineau (163:02)

That would fit like one of your. Yeah, I'm looking at the Beatles photo right now.

Leo Laporte (163:06)

It was workman.

Paris Martineau (163:07)

Like a. A workman's jacket.

Leo Laporte (163:10)

Yeah.

Jeff Jarvis (163:12)

My German jacket.

Leo Laporte (163:13)

To all Neu. Yeah, I actually would like. I see I have a fetish for that kind of weird thing to wear ascots and Nehru jackets and can we.

Paris Martineau (163:24)

All get a fun uniform and show up one time and match. Wear a. Wear like a hot. Like a neon yellow boiler suit or something.

Leo Laporte (163:33)

Intelligent machine order. Oh, oh, oh, oh, yeah. When I worked at Clock Radio back in the day, we all had. The slogan was the man from Clock.

Paris Martineau (163:44)

I'm just imagining it's a radio show where you just say the time every minute.

Leo Laporte (163:49)

Female DJs. I was a DJ at Klok Clock Radio in San Jose and they had. The slogan was the man from Clock. And we all had yellow. The worst. Mustard yellow blazers with the same.

Jeff Jarvis (163:59)

Oh, yeah, you've shown that they had a patch.

Leo Laporte (164:02)

It said Clock on it.

Paris Martineau (164:03)

That's pretty cool.

Leo Laporte (164:04)

The man from Clock. I guess it was a takeoff of the man from U.N.C.L.E. no, I still have a fetish for that weird stuff. I used to wear capes in high school. And. And I wish I was wearing a.

Paris Martineau (164:17)

Cape in high school. I would have killed to wear a cape.

Leo Laporte (164:20)

Well, you were a goth, right? So you were wearing like, what, torn stockings and needles through your nose or something? Safety pins, probably.

Paris Martineau (164:27)

Yeah, yeah.

Leo Laporte (164:29)

No, you know what you were wearing? You can tell?

Paris Martineau (164:31)

I mean.

Leo Laporte (164:32)

Yeah.

Paris Martineau (164:33)

I remember I once really got into wearing striped stockings until I wore them to a Magic the Gathering meetup and a bunch of creepy dudes commented on. And I was like, I have to leave.

Leo Laporte (164:48)

No, there's something about striped stockings. They're very, very sexy. I don't know why.

Paris Martineau (164:54)

As I learned that day in I.

Leo Laporte (164:57)

Don'T know what it is about stripes.

Unidentified Guest (164:58)

Did you play the Gathering class?

Paris Martineau (165:01)

Yeah, I did.

Leo Laporte (165:02)

In striped sockings.

Paris Martineau (165:02)

Back in the day.

Leo Laporte (165:03)

In striped sock, the nerds got a little old.

Paris Martineau (165:05)

And the nerds got me off both striped sockings and Magic the Gathering.

Unidentified Guest (165:09)

Yeah. That's why there are no more women playing that game.

Leo Laporte (165:12)

Oh, God, I'm so sorry. I would have been one of them. Jeff Jarvis. He is professor emeritus of journalistic innovation at cuny. You can catch his act at SUNY Stony Brook. He'll be waving a cane at the Linotype. His newest book, Hot Type, is available for pre Release ordering from jeffjarvis.com of course, there's also the Gutenberg parenthesis and magazine. And I can't wait till Hot Type comes out. We'll have to interview you for that show. That'll be good. That'll be good. Paris Martineau is a investigative reporter at Consumer Reports, where she is working on something so deep, so, so revealing. So revealing. We can't even mention it, honey.

Paris Martineau (165:58)

Can't even talk about it.

Leo Laporte (165:59)

It will blow the lid off of.

Paris Martineau (166:01)

Well, you know, something, you know. Oh, hold on. Look who's down here.

Leo Laporte (166:06)

Gizmo. One last appearance.

Paris Martineau (166:08)

Gizmo, how do you. How do you respond to the allegations.

Leo Laporte (166:11)

About your app, Radiator Cat? Radiator Cat.

Paris Martineau (166:14)

She is pretty warm from the radiator. I can feel it on her chest. And now I'm covered in hair.

Leo Laporte (166:23)

Thank you so much, Paris. And thank you so much, Jeff. Thanks to all of our listeners, especially to our Club Twit members. We'll see you next Wednesday on Intelligent Machines. Bubba. Hey, everybody, it's Leo laporte. You know about Mac Break Weekly, right? You don't? Oh. If you're a Macintosh fan or you just want to keep up with going on with Apple, this is the show for you. Every Tuesday, Andy Inocco, Alex Lindsey, Jason Snell and I get together and talk about the week's Apple news. It's an easy subscription. Just go to your favorite podcast client and search for Mac Break Weekly. Or visit our website, TWiT TV MBW. You don't want to miss a week of Mac Break Weekly. I'm not a human being.

Paris Martineau (167:10)

Not into this animal scene. I'm an intelligent machine.