Intelligent Machines 808 transcript

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

0:00:00 - Leo Laporte
it's time for intelligent machines. Hello everybody. Jeff jarvis is here, so is paris martineau. Uh, as usual, we're going to start the show with an interview with one of the smartest people in the world, and this guy's the real deal. Steven wolfram next on intelligent machines podcasts you love from people you trust.

This is TWIT. This is Intelligent Machines, episode 808, recorded Wednesday February 26th 2025. Stephen Wolfram, it's time for Intelligent Machines, the show where we talk about, well, intelligent machines with jeff jarvis, professor of journalistic innovation, emeritus at the I better do it now craig newmark graduate school of journalism at this now at montclair state university, and suny stonyok. Great to see you, jeff.

0:01:02 - Jeff Jarvis
Good to see you.

0:01:03 - Leo Laporte
Also Paris Martineau reports at theinformationcom, where she covers youth matters, among other things. You do everything, really.

0:01:14 - Paris Martineau
Among other things, yeah, yeah among other things.

0:01:17 - Leo Laporte
Hey, I am so excited about our guest today, stephen Wolfram, one of the smartest people in the world. He was a child. I did not know this about you, stephen, but and I hope it's not wrong because I asked ai a child prodigy at the age of 15 you published your first scientific paper, studied at eden and oxford before earning your phd in theoretical physics from cal of california institute of technology at the age of 20. A macarthur fellowship genius at 21. Most people wait till their 40s uh, and, of course, the creator of uh, mathematica, wolf from alpha and the wolf from alpha language. It is such a and and the author of many seminal works, not just in physics but many other subjects. I think cooking, yes, no, no, okay, I don't know how to cook when is the cookbook coming out?

0:02:15 - Benito Gonzalez
I don't know how to cook.

0:02:20 - Leo Laporte
We want to get you on, because this show is about AI. You've been, you know, thinking and working in the field forever. What was that first paper when you were 15? What was that about?

0:02:33 - Stephen Wolfram
It was about particle physics. In those days particle physics was the most exciting thing going on, just as maybe people would say machine learning is the most exciting thing going on now. And there were things that had been discovered and I was trying to figure out what was happening with them. I have to say my first paper was not my best Only uphill from there.

Yes, yes, right, it's one of the things that was trying to figure out, things about what electrons are made of, and I had kind of an idea about that and I mean, this is maybe a technical point, but I kind of thought maybe electrons are 10 to the minus 18 meters across. Now I finally think I do have some understanding of what electrons are and I think maybe they're 10 to the minus 81 meters across. Wow, so it's kind of a flip around in the, in the exponent, which is anyway. So, yes, that was. That was my, my first paper about particle physics no one should peak at the age of 15.

0:03:34 - Leo Laporte
you should be grateful that you did not right yes it would be very depressing if peaked at the age of 15. You said something interesting, though. First of all, you called it machine learning, not AI. Is that a conscious choice?

0:03:50 - Stephen Wolfram
Well, ai has been harder to define. I mean, back when I started building big computer systems to do kind of science-type things, which was late 1970s, people were saying at the time oh, when you can do math by computer, then we'll have AI. And then the goalpost moved and that wasn't what AI was supposed to be about. So I think machine learning is a bit more. Ai has sometimes been the thing which is just a bit out of reach to computers, which makes it a little hard to define. And now we've got AGI, which is again just out of reach to computers type thing. I don't think it's a well-defined concept. So yes, I think the thing that one could sort of distinguish is things which are sort of defined computationally that's one whole branch of effort that I've spent a lot of my life on and things which are specifically built using things like neural networks and kind of the traditional tools of machine learning, and that's a different branch that happens to be doing really well right now.

I mean, I started playing around with neural nets back in. I think 1981 was the first time I tried to simulate a neural net. It didn't do anything interesting and I I thought for years nothing interesting is going to happen with neural nets. And then you know, 2011 comes around and really interesting things started happening with neural nets. I still think you know, even now, I'm really curious why do the things that happen with neural nets actually happen? Why do they work?

0:05:24 - Leo Laporte
And that's much less clear, and I've tried to figure out some things about that from a science point of view recently yeah, I remember your article in 2023, which turned into a book, uh, talking about llms and why they don't do math and proposing that maybe they should uh pair with wolfram alpha. Let the wolfram alpha do the mathematics. Why is wolfram Alpha good at math? How does Wolfram Alpha work differently from LLMs?

0:05:50 - Stephen Wolfram
Yeah. So I mean, what we're doing is to compute using definite algorithms that we explicitly write and taking data from the world that we've kind of gone through the trouble of explicitly curating. So the big idea of machine learning is you give the machine a bunch of examples of what you want it to do and you hope that it will kind of figure out what to actually do in any particular case. So it's not the kind of thing where you're setting up sort of this rigid structure that the thing is going to follow and for some things that we do as humans, kind of doing it roughly right and doing it across many different situations is exactly the right thing.

There are other things that we want to do where we need to do sort of a precise computation. You need to build up the tower very tall, and that's something that's not a thing that kind of the neural net kind of machine learning approach is really so suitable for. I mean, it's like if you want to get it 80% right, then you know using machine learning is a great, great thing. If you want to get it 100% right, then using machine learning is usually not the right thing. That's a case where you need to have sort of this precise sort of tower of functionality that you build up.

0:07:12 - Jeff Jarvis
Is that because of the probabilism that large language models have been built on and machine learning is built on, versus the kind of determinism of the explicit programming you talk about?

0:07:20 - Stephen Wolfram
I think that's one way to put it.

I mean, I've been trying to understand what is one actually doing when one is doing machine learning. In other words, you give a machine 10,000 pictures of cats and 10,000 pictures of dogs and then you say now I'm going to show you a picture, is it a cat or a dog? And then you showed a picture of a dog in a cat suit. What's it going to do? Well, what it tends to do is something that is fairly similar to what humans do. In a situation like that, you know whether it agrees with what humans say will be. It very often will. Why does that happen? We don't have a precise sort of definition of why that happens and what the boundary between catness and dogness is for the machine. But I think this question of what the analogy that I've found useful, that's sort of come out of some science I've done in this area is let's imagine that your task is to build a wall. Well, one way you can do that is you make these very precisely engineered bricks and you set them up in this very kind of organized way and you built this wall and you can keep building it, and you can keep building it very tall. Okay, that's plan A. Plan B is you see a bunch of rocks lying around on the ground and as you build your wall you find a rock that's roughly the right shape. You stick that one in. You keep building that way.

That second thing is pretty much what machine learning is doing. When you train a neural network. What it's doing is it's finding these kind of lumps of computation that happen to fit into what the training looks like and so on, and so it sort of puts that rock into the wall and keeps building. And it's something where you can build the wall to a certain height, just with these randomly shaped rocks. But it's something where you can build the wall to a certain height, just, you know, with these randomly shaped rocks, but it's not something where you're going to be able to sort of systematically build it up very tall.

I think that's perhaps a way to think about what's going on, but in the end it's that kind of. You know, machine learning is getting things roughly right and that's a big achievement In many domains. You know, getting it roughly right, writing that essay that makes sense is great. You can't say that essay is precisely the right essay. It's just oh, it's an essay that makes sense and I think that's the. You know it's a distinction between what needs to be precise and really built up many, many steps, and what needs to be happen, sort of roughly right yeah, the ml is uh stochastic.

0:09:51 - Leo Laporte
So, while two plus two is always four, in the llm world, it might be, you know, whatever five, four, six, three, um, but there seems to be, and you've worked with cellular automata, which I find fascinating, and I think for a long time, the thought about you know, like Conway's Game of Life, the thought about these was emerged, there would be emergent properties that would come out of this. That seems to be what's happening with LLMs as right now, or no.

0:10:22 - Stephen Wolfram
Somewhat, yes, I mean. So just to explain that thought. I mean, the big thing that was to me a huge surprise from the early 1980s is, even if you have a really simple program, like something that's just sort of one line of code, you can specify it by some very simple rules about, you know, black and white squares or something like that. Even if the setup, the rules, are really simple, the behavior that can emerge can be really complicated. That's something that's very surprising to us because we're used to. You know, we're doing engineering, we want to make something complicated, we have to go to lots of effort to make it complicated. What we discover is in the computational universe of possible programs. Even a very simple program can do very complicated things, and that's a big deal for one's intuition about how nature works and lots of other kinds of things.

It's also a part of the core of what happens in machine learning, but I think the path is not quite so direct. I mean this question of the you know, when you say it's sort of an emergent thing. I think that's true, but it works in a slightly different way. So I mean, the thing that is surprising about machine learning is that the kinds of decisions that it makes seem to more or less agree with the decisions that we humans make, like, is it a cat or a dog type thing? It's going to, it seems to, surprisingly, more or less agree with the way we make those distinctions, and I think there's sort of a question of why that happens, and I think the probable answer is because its architecture is fairly similar to the architecture of our brains and so it's making those distinctions in the same kind of way we make those distinctions.

Now it turns out that to get to the point of being able to train the thing from examples, you do need a bunch of this same kind of emergent behavior that you see in things like cellular automata. But that's a more sophisticated part of the story. I mean, that's the, I think, the most significant. The sort of zeroth order effect is that what the neural nets do seems to sort of align with what we would typically do in similar circumstances.

0:12:38 - Leo Laporte
And yet the thing that, or the goal, it seems to me for these, would be for them to do something more than a human could do well, I mean, there's a matter of scale.

0:12:50 - Stephen Wolfram
Right, like, like I'm, you know, right now I'm trying to use llms as a tool to study something that basically needs me to read. You know, half a million papers. Right, I'm not going to read half a million papers, but can I extract the essence of those things? And you know, in a human-like way, as if I was reading those papers, and the thing? The answer is probably yes, and so that's an example. Now, you know, there will be things that I can. You know there's a lot of leverage I can get from being able to do that. Is it something qualitatively different from what I could do? No, not really. It's just a matter of scale.

0:13:30 - Leo Laporte
So the thought that maybe an AI could come up with a way to do nuclear fusion that we haven't thought of, or fold a protein in a way that we haven't thought of before, create a new material is that out of the question?

0:13:44 - Stephen Wolfram
Well, the thing to understand is having doing original things is easy. If you have a sequence of random numbers, that sequence of random numbers will be original at all stages. The question is does it align with anything? You care?

0:14:00 - Leo Laporte
about Valuable versus original yeah.

0:14:02 - Stephen Wolfram
Right, right, and I mean so that that now. So now the question is you know, could you come up with something that's really good by, for example, noticing that you know, oh, across, you know the, the hundred thousand papers that relate to different approaches to nuclear fusion? There is this theme that you can extract that lets you see what to do. Yes, that's perfectly possible.

0:14:25 - Leo Laporte
A human could theoretically do it, but maybe an AI is more likely to do it, given that there's a million papers, and I think that the human plus AI is the best alternative.

0:14:36 - Stephen Wolfram
And I think the thing to realize is what ends up happening is you know, ai is an automation mechanism as far as I'm concerned, and the way that tends to work is somebody has to say what do you want to do, and then the automation mechanism can go try and do it. But if you don't know what you're trying to do, there's no place to start. And I think you know the AI. That's just incredibly. You know it's learned a lot of stuff, but it's just sitting as a box on your desk. It's like, well, what is it going to do? But it's just sitting as a box on your desk. It's like, well, what is it going to do? It has to. There has to be some. Well, I want this to happen.

And you know, we humans have lots of things where we say I want this to happen.

Now you could ask the question if you just got a lot of AIs hanging around, won't they kind of spontaneously say I want to have this kind of thing, this thing, happen and at some level, that is, you know, you could look at what's happening in the AIs that already exist in the world and you could kind of probe that and you could say it looks like this AI wants this or that thing to happen and there's this whole sort of civilization of AIs that's already starting to exist.

That's sort of separate from human civilization and at the beginning it seems like that's a really weird thing and it's a really scary thing that these AIs could be off having their own civilization, independent of us. But the thing to realize is there's a really good model for that that we are very familiar with, which is the natural world. The natural world is off doing all kinds of complicated things. It just does them. It's, you know, we live around it, but it's not, you know, it's not something where we're all freaked out because the natural world is doing things that we don't know about, we don't understand, et cetera, et cetera, et cetera. So I think that's kind of the picture of where I see things going, with the kind of the civilization of the AIs versus the civilization of us humans.

0:16:24 - Leo Laporte
I love that and I know what you want to ask, Paris, and I know what Jeff wants to ask, but I'll let you go ahead. I won't take your question, but I know where we're going here. How do?

0:16:34 - Paris Martineau
you define AGI.

0:16:36 - Leo Laporte
I knew it. I knew you'd say that the minute you said that I knew I was like God God.

0:16:39 - Jeff Jarvis
One of us was going to ask it.

0:16:45 - Paris Martineau
And specifically there's two questions One, how do you personally define AGI? And two, do you think that there's this common definition, or like a shorthand of AGI, that goes around, of you know, superhuman, intelligent AI that could replace humans and outpace us at tasks. Do you think that that's feasible?

0:16:59 - Stephen Wolfram
Well, okay, I think AGI is a really mushy concept and I think it is the emergent concept from AI, which was always a mushy concept. So it's this thing about can you make something which does everything a human does? You'll always be able to find things that the thing doesn't do, like it's not mortal or it doesn, you know, get uh, get flu or something. The only thing that does exactly what a human does is a human and you know you'll always be able to find some place where you can kind of find the, the magic thing that is the unique thing left over for humans. My first comment, my second comment is that I think that you know, in the end, the, the ais, can get very powerful. They can do all sorts of elaborate computational things. Lots of you know I've is that I think that, in the end, the AIs can get very powerful. They can do all sorts of elaborate computational things.

I've spent a large part of my life studying what sort of simple programs in the wild can do, and they can do amazingly sophisticated things, things that are probably just as sophisticated as things happening in our brains. The issue is that there's a question of do those things align with anything we care about. I mean, there's plenty of things that happen in the natural world that are in some sense computationally more sophisticated probably, than what's happening in our brains. But it's a question of are those things where we say, oh, that's human-like, intelligence-like stuff? And the answer is not usually usually. I mean, we sometimes say, you know, the weather has a mind of its own, but that mind is fairly different from a human mind. It's unpredictable, it's complicated, but it isn't sort of human aligned.

So I tend to think that the story ends up being that there is this kind of computational resource that is the civilization of the AIs, and there's how we interact with that, which tends to be things where we say we want to have this happen and then we get AIs to do that and sometimes things will just happen because that's what the natural course of AIs leads to have happen. So I think it's sort of the idea that something happens that is superhuman. We have superhuman things happening all the time, either in nature or in our computers, and a lot of things I've studied in my efforts in science have to do with even very tiny programs that in a sense do superhuman computations. But we look at them and we say that's interesting. Sometimes we say I see the significance of that and it's really cool, and sometimes we just kind of shrug and say, well, that's computationally sophisticated, but it's not something I care about.

0:19:42 - Leo Laporte
You, of course, have written a lot about the computational paradigm. I'm wondering if there's any intersection between that and what we're calling machine learning or AI today, or are they separate domains?

0:19:55 - Stephen Wolfram
No, I think that the way I see a lot of what's happening with AI and machine learning right now is providing outstanding kind of linguistic interfaces to things, sort of humanizing the way that one can interact with things. The question is, what's behind that? And you know, the thing that we have been able to do with computation is build these kind of tall stacks of capability, and that's something that you know. If you go went back 300 years or so, people were still talking about natural philosophy. The idea of kind of mathematical science hadn't really started off, and you know, if we'd gone straight from natural philosophy to today, there wouldn't really be a back end that we can talk about. It really would be. It's just this human-like thinking, this kind of linguistic interface and so on.

But right now we can actually compute things. We can build these kind of tall towers of computation, we can work out things that involve millions of steps where each step has to be precise and so on, and I think that's what I've spent a large part of my life building the tools to make that possible. And I think the thing that's really kind of interesting right now is we built that for human users. Now we also have lots of AI users Turns out that because the AIs are using this kind of linguistic interface that matches humans, that the things we did to make what we built suitable for human users are the exact right things to have done to make it suitable for for our current AI users, so to speak.

0:21:26 - Leo Laporte
Oh, that's so exciting, that's amazing.

0:21:29 - Stephen Wolfram
You know the the emerging thing is you talk to the AI but the backend is a computational kind of thing. And you know, we built Wolfram language to kind of have the sort of computational representation of the world where you can do precise computations, where you can do precise computations, where you can have precise knowledge, you can build things up and so on. And I think you know my, my vision of the future is that there is this layer of kind of linguistic interface and then there is this kind of computational bedrock, this kind of computational knowledge bedrock, and that the thing that we don't yet know really that well is we only have sort of the coarsest ways to interface between that sort of linguistic layer and the kind of computational bedrock. But I think you know, for example, the idea that, oh, we're going to train the llm to learn you know how, to what, the, what the millionth digit of pi is, or something. This is not a particularly good idea In an LLM. You're taking knowledge, grinding it up in some way that we don't understand very well and sort of.

You know, where is that millionth digit of pi? Is it in the collective features of a thousand weights spread across the neural net? It's a mess and it's not kind of the right way to implement things, just as if we had all of our computer infrastructure but there were no actual computers. The computers were, as they used to be, humans and we said okay, we want to run a program, let's get a human to run the program. Turns out humans aren't very good at running programs in their minds and that wouldn't work very well. So we have machines to run the programs, and that's something where the humans define what the programs should do and then the machines run the programs. And I think that's the same sort of direction that we see happening in what's happening with the current version of AI.

0:23:28 - Leo Laporte
Yeah, in fact, if you want to get an AI to do mathematics, you say use code, and then it can do it. That's the interface to a computational system.

0:23:37 - Stephen Wolfram
Then it can do it and it does it quite well, hopefully, using our tech. It works well using our tech.

0:23:44 - Leo Laporte
And of course, that's one of the things people are talking about a lot these days is replacing human coders, uh, with ai well, you have to understand what's really going on there.

0:23:55 - Stephen Wolfram
You know, I've spent a large part of my life as a language designer.

What does the language designer do? Well, a language designer is trying to take sort of the things that humans want to do and define primitives in a computer language that let the human do what they want, specify what they want in as good a way as possible. The approach most people who've designed languages have taken is I'm just going to have this very small language, then I'm going to rely on people to build up libraries that do specific kinds of things. I've done the crazy thing in my life of trying to build sort of an integrated language that does everything that sort of coherently defines sort of everything in a computational way, and for me, the main thing we're seeing is people what in our language would be a line or two of code, because we've defined all those operations integrated into the language. People have been used to using, in many cases, low-level languages where it takes a giant blob of code to do this thing, and so then they see that, yes, the AI can write that giant blob of code. For me as a language designer, it's like that giant blob of code just shouldn't exist. It shouldn't exist at all. Yeah, yeah, it should be. It should have already been defined, because that's a blob of code you repeatedly want to do. That should have been defined as a primitive thing in the language you're using.

So you know, I think now, having having said that, the the idea that you can go from sort of a thing that you just talk about linguistically to something in code that actually works particularly well with our computational language, because it's sort of the closest to being able to deal with things that you just talk about, like cities or videos or whatever else, and so what you know, we built a, we released a product recently that's our notebook assistant, where it's kind of remarkable the extent to which you can just be blabbing along and saying something kind of vague, and it will try to get sort of to code.

That is the closest code it can find to what you thought you were talking about, and it's quite interesting because it kind of crispens up your idea. You're kind of saying something vague. It says, well, let me respond with this precise piece of code. And so the whole idea of our computational language is it's a language that can not only be written by humans but can also be read by humans. So, in a sense, what's happening is the AI is responding and saying well, the thing I think of, that I'm reminded of by what you're saying, is this precise computational thing that you can then respond to, and so on, and that seems to be a very powerful way to do things.

0:26:38 - Leo Laporte
Are you using your own models for that, your own LLMs?

0:26:41 - Stephen Wolfram
We're using mostly other people's LLMs, but there's a big sort of layer of other technology around that that is essentially doing things like dynamically prompting the LLM with using computation, using rags and things like that to be able to provide, to be able to say, given that that's what the human said they wanted to do, hey, llm, you should know exactly this and this and this. And then the LLM can kind of fill in the details and then on the back end we're also kind of cleaning up what the LLM did to provide sort of the best possible code and that works well. But it's worth realizing and I think this is a general feature of a lot of what's going on with LLMs today that the harness is pretty important. We had built an early version of this 18 months ago. It didn't work that well. As we tightened up the harness, so to speak, it started being really useful. And the LLM is an important component, but the harness is also important.

0:27:41 - Jeff Jarvis
What do you wish for next? I'm gratified to hear you say to kind of reject the hubris of the human thinking that the goal is to replicate and beat us. It's to do these amazing things that it can do. Jan LeCun said in the Guardian, I think, two weeks ago, that teaching it reality that when the ball falls off the table, it doesn't go away. Jensen Wong talked about training the digital twins so that it knew how to deal with that reality. That's one view. What do you wish for in next developments? Because every AI company now is coming out with this next version. That's just that much better or that much faster. Where's the leap that you wish for?

0:28:20 - Stephen Wolfram
Yeah, I mean, I think you have to look at the history of AI, machine learning and so on. There have been a series of kind of jumps. You know, for a long time neural nets didn't do anything terribly interesting. 2011, they started doing image recognition. Then you know speech to text and so on. Then language generation and actually image generation came a little bit before language generation and there are these kind of thresholds and my suspicion is that if you look at, for example, image recognition, things haven't got that much better in the last 15 years. It's incremental improvement.

There is this moment where you kind of get over the threshold. It's kind of like, I think, for language models. It's a little bit like what I think kind of happened over the threshold. It's kind of like, I think, for language models. It's a little bit like what I think kind of happened with the telephone. People knew that, in principle, you could transmit speech over copper wires or whatever, but nobody could understand what was being said at the other end of the phone, so to speak, until Alexander Graham Bell sort of found sort of some hacks that made that work and suddenly it's like oh yeah, that sounds like speech. I think that's sort of what happened in the large language models. It got to a point where it reached a threshold where it's like, yeah, this is basically working.

I think we'll see that in a bunch of other areas. The area that's sort of an obvious one is the kind of, as you were alluding to, the real-world robotics, common sense kind of I drop this, it falls, that kind of thing and I, you know, I don't see any reason that won't happen, even fairly soon, you know, is that, will that be significant in the world? Absolutely, that will be significant in the world. There are all kinds of things that will then that have been oh, a human has to do that. That involves sort of manual dexterity and things like this, where it will suddenly be the case that, yes, we can use AI to do that.

In terms of on the intellectual side, I would say that the thing I'm, I guess well, one thing is the on-ramp to computation for people, the idea that you can kind of, for every field X, there's sort of a computational X version of that field. But can people get there? Can you know the archaeologists be doing computational archaeology? Well, they say, oh, but I can't program. Well, that's not necessary anymore. You know, you can use the LLM to get to computational language. You can read the computational language, you know you can use the LLM to get to computational language. You can read the computational language. You can be off doing your computational archaeology. So I think one of the big sort of trends will be towards sort of computational X for all X, so to speak, and people really being able to get there and you don't have to go to the druidic specialized programmers to get the code written to be able to do your computational x. I think that's, you know that that's.

0:31:07 - Leo Laporte
That's another piece for anybody who's ever used wolfram alpha, that is the amazing moment when you type in natural language and get such amazing results. Is the notebook assistant kind of the next generation of wolfram alpha?

0:31:20 - Stephen Wolfram
in a way it's a little bit of a different, different thing. I mean wolfram alpha I view as being sort of the drive-by answer your question type system. Right, you type in a simple question, you get a computed answer. What the notebook assistant is trying to do is to set up the bricks of computation that you need to build up whatever you're trying to do. So it's not just trying to say, here's the answer, you're done, whatever you're trying to do. So it's not just trying to say here's the answer, you're done, it's trying to say here's this thing, that is a brick, a tool that you can then build on and continue building up.

So I think the way I see sort of the future of things like software development is the first thing you have to do is to imagine what you're trying to do computationally, and that's the thing. They don't teach that in computer science school. How to sort of think about everything computationally but I think that's a super important skill is like how do you conceptualize what you want to talk about in a computational way? Once you've gotten some distance towards that, then you can ask our notebook assistant. It'll write a piece of computational language code That'll do a piece of what you want. You take that piece of computational language code that becomes a component that you can then start building up from to make the larger thing that you want to build. I think that's the See, that's a human feature not yet emulated by um it uh, although I'm sure they could learn from actually, you know, it's interesting because more and more people are finding that lms.

0:32:53 - Leo Laporte
If you push them long enough, they start to kind of decline, almost physically decline yes, in their ability, which I find fascinating, but this must be a, this must be for you a vindication of the computational paradigm in some ways.

0:33:08 - Stephen Wolfram
In some ways. I mean, I think that this ability to let more people into the computational paradigm, that's a very important thing. That has not yet played itself out. That will play itself out over maybe the next decade or so, and I think people haven't. You know, that's still an emerging realization, so to speak.

0:33:29 - Jeff Jarvis
More to that, I'm curious, since you've lived in this world and developed this world for a long time. Suddenly, when LLMs arrive, everybody says, oh hell, this is AI, as if that has arrived fresh. What's your reaction to, society's reaction to since ChatGPT's release? How have you seen in just people talking in media, in companies? How do you sense that?

0:33:55 - Stephen Wolfram
Yeah, I mean it's interesting that this was a moment of consumerization. I mean it's just like with computers in general. I mean people. You know I started using computers when I was a kid, in the 1970s, and most people were like computers, who cares, you know, maybe they're in the back end of something doing some you know bank processing thing or whatever, or figuring out how to launch rockets, but I don't care about that. And then, you know, the personal computer comes out and people discover things like word processing and games and so on. And suddenly it's like, oh, we really care about computers.

I think that was sort of the chat GPT moment was the one where people could see that there was this. I mean, by the way, nobody knew chat GPT was going to work, including the people who built it. I mean it was. It was really, you know, it's like at what point? I mean you know I'd been playing with language models and sequence prediction and so on. None of that stuff had been terribly interesting. It was. You know, it was a thing where, for whatever reason and we don't understand why yet maybe we one day will suddenly we got over a threshold. You know some number of billions of parameters. It's like that was enough. That was human-like enough, so to speak. Now that scale is probably determined by something to do with us. We humans have about 50,000 words in our typical languages. That's a that and we have 100 billion neurons in our brains. That provides a definite scale. If you wanted an AI that would be like a dog, for example, in in its linguistic capabilities, let's say you probably need a much smaller llm, but we wouldn't be as impressed by it. So somehow the the size of the llm, you know the size of the system was such that it was sort of matched enough of what we need to be impressed by that. We got to that point, by the way, one thing to just say as sort of a side comment.

One thing I've been curious about is, let's say our brains. Let's say we had a thousand times as many neurons in our brains, what might we be able to do? What's kind of the next level of sort of intellectual capability? It's just like cats and dogs. Well, dogs, at a few individual words, fetch, sit, whatever else they deal with that. We humans the big invention of our species, I suppose, was human language and this idea that you can put together an infinite collection of sentences by using all these different words. So a big question is what happens next If we could go beyond that? What's the next level of abstraction and sophistication? That's an interesting thing to think about. But you also are stuck kind of thinking well, if I was the dog, could I really understand human compositional language? And is it the case that there's something bigger there? Undoubtedly is, but maybe it's something that just isn't a fit for us, just like the natural world has lots of things going on that we don't necessarily we don't understand, except through this sort of bridge of natural science.

To chat with GPT, I think the thing that you know a lot of people were like it's magic. You know, this magic has happened. There's going to be more magic. You know, this was such a surprise. Necessarily there will be many more surprises. That was a. I don't think that's. You know, people always think that when there's a technological surprise, they always think it's not going to stop here, it's going to keep going, going. But it doesn't always keep going. In fact, he usually doesn't keep going, it usually is. That was the surprise when one got to that sort of threshold level and then, and then, so you know, and then the question was well, okay, sort of.

I was a little bit surprised by this people saying it's magic.

And then it's like, well, how does it work?

Oh, it's this.

You know, it's this AI saying it's magic.

And then it's like, well, how does it work?

Oh, it's this. You know, it's this AI thing, it's this neural net. And I started thinking, well, can I actually figure out some idea of why does this work? And what I realized is, you know, there's at least a picture of why it works and it's this and it's something that in a sense, was a feature about human language and human understanding that maybe we should have understood thousands of years ago, which is, people have known about grammar of language that you form sentences in English having noun, verb, noun and so on, but that structure of grammar doesn't tell you whether the sentence is going to be meaningful. There are lots of sentences that go noun, verb, noun and they're completely meaningless. What I sort of realized is that when the LLM is kind of looking at hundreds of billions of sentences, it's seeing sort of patterns, not only of the noun, verb, noun type but of the. This is a pattern of words that makes sense and that's sort of the big thing that it sort of statistically learns.

People were very surprised at the beginning, for example, that LLMs could do logic, that they could kind of make logical conclusions. And you realize, well, you know how was logic discovered in the first place? Well, you know, presumably Aristotle went through and just heard lots of people giving speeches and Sort of identified these are the patterns of speeches that make sense. And then those became Syllogisms and they became what we now think of as logic, and I think the LLMs kind of learnt it the same way. And so, in a sense, one of the things that was sort of Interesting to me was people sort of say it's magic. They realized, well, actually it's not such magic. There's probably a, a thing that actually is telling us an interesting piece of science that, uh, that maybe we should have learned a long time ago. That's sort of the at the core of why this can possibly work that is, uh, also a big part of your book.

0:39:40 - Leo Laporte
What is chat GPT doing and why does it work? Uh, very provocative. Every time I talk to you, stephen, I feel guilty that we are using brain cycles that you could be using much more advantageously. I want to thank you for spending time with us and encourage everyone to go to wolframcom. Wolframcom, stephen's site. It's a great entry point. There are books there. There's links to the new Wolfram Alpha notebook, which I think is fascinating. I'm going to sign up immediately and I really appreciate the time You've given us so much to think about. I really appreciate it. Thank you, steven. Lots of cool questions.

Yeah, thank you, yeah, have a great day, thanks. Thank you, we'll put this out later in the day. It will go up immediately. We'll send you if you like. We'll send you some clips if you want to give them to your social people, and so forth.

0:40:30 - Stephen Wolfram
You can pass them along. Thank you, yeah, it's always a bit long form I I'm. I'm so used to just yakking on that's what, steven.

0:40:38 - Leo Laporte
I would talk to you for four hours. I just again I'm I'm sincere saying I feel guilty taking any of your brain cycles.

0:40:46 - Stephen Wolfram
You're very welcome to them. Anyway, thank you so much. It's a real pleasure.

0:40:51 - Leo Laporte
Thank you. You know we'll have you back in a few months and we can talk more about this, because you've immediately brought up all of this stuff I want to think about. So it's great.

0:41:01 - Stephen Wolfram
Thank you, really appreciate it Nice to meet you, nice to see you. Take care.

0:41:07 - Leo Laporte
Yeah, it's great. Thank you stuff really appreciate it. Nice to meet you. Take care. Yeah, wow, isn't that that last thought um, wow, just amazing, wolframcom. Uh, go there. Uh, there's so much there. I am going to sign up for that notebook because it's basically the idea, is a english language interface to the computational paradigm. So you talk, it computes um. To me that was what wolfram alpha was originally um, but now it is uh, it's even more powerful, it's just uh, I felt I wasn't smart enough for wolfram alpha.

This might be a yeah, well, it's for people like you who aren't smart enough for wolfram alpha, that's all right. We're gonna take a break and we will come back. We have ai news and I think what we're going to do from break and we will come back. We have AI news and I think what we're going to do from now on with the show is we'll do an interview. We have some great interviews coming up, including next week. It's it's Gary Marcus, which is going to be great. I think you know him well. Thank you, jeff, for helping us arrange that the following week that's the man who coined the term intelligent machines Ray Kurzweil will join us.

I've also talked to a bunch of other people. We have a. We've actually booked it well in advance. One of them, my friend Harper Reed, has written an interesting article about exactly what Stephen was talking about as a coder, the interface to an LLM to do the coding. As a coder, the interface to an LLM to do the coding, and he has some very interesting thoughts about that. And it really is pair, he says, and I agree, it's pair programming, but using an LLM. So lots more coming up on the show.

0:42:35 - Paris Martineau
But, as I said, the sand guy on the show maybe not.

0:42:40 - Leo Laporte
You know, we have so many, like we're filled up. No, no, no, I'll get him on. He teased.

0:42:47 - Paris Martineau
He teased one week that the sand guy was gonna get the sand guy on.

0:42:50 - Leo Laporte
We will get the same guy, uh, but from now on, what we're gonna do is start with a half hour uh interview with somebody who's really doing some interesting things or provocative uh. Then we'll do the ai news and we'll follow that within other news because I want you guys to be able to, you know, talk about all the other things you're interested in, and then we'll follow up with the picks of the week and, if we can, we're going to try to keep this show under five hours. How's that sound okay?

all right sounds good. Uh, paris Martineau, jeff Jarvis, uh, we will continue with intelligent machines in just a moment, but first a word from our sponsor delete me. Delete me saved our company. Well, maybe that's an overstatement, but honestly, we were freaked out because we were getting spear phishing attacks using our ceo's name phone number. The bad guys knew her direct reports, their names, their phone numbers, and we were scratching our head and we said do we have a mole inside? Then we realized it's data brokers.

Everything, everything there is to know about you, your family, your company, is online somewhere legally sold by a data broker. If you've ever searched for your name online, don't, first of all, don't. And but if you have, you know what I'm talking about. So much personal information is available and there's always that little link that says and now, for a buck 50 more, I'll tell you about their prison record and things like that. This is, this is a nightmare situation. We learned with the national public data breach that it's legal for a data broker to sell your social security number to anybody with a buck and a half.

Actually maintaining privacy is not just for your personal issues or your company's concerns. It's a family affair too. That's why delete me has plans for everybody businesses, individuals, even families. With a family plan, you can ensure everyone in your family feels safe online. They help reduce risk from identity theft, cyber security threats, harassment and more. It made a big difference for us when steven steve gibson and I went into the national public database to see if our social security numbers were in there. They were, but guess who wasn't? Our ceos. We looked for at least no information at all, because we've been using delete me.

When you sign up for delete me, their experts will find and remove your information from hundreds of data brokers. If you're doing it for a family, you're going to sign in a unique data sheet to each member, tailored to them, easy to use controls. Account owners can manage the privacy settings for the whole company, the whole family. Delete me then. Now this is really important Won't just stop there. They will continue to scan and remove your information regularly, and that's for two reasons One, because data brokers, being data brokers, are going to start repopulating that information. But the other is there's always new data brokers. In fact, this is really skeevy. Is there's always new data brokers? In fact, this is really skeevy. Sometimes those data brokers go belly up and then start a new company under a new name so they can start collecting all your information again.

I'm talking everything addresses, photos, emails, relatives, phone numbers, social media, property value, social security numbers. Protect yourself, reclaim your privacy. Visit joindeletemecom slash twit. When you use the offer code twit, you'll get 20% off. Joindeletemecom slash twit. We thank them for their support and we encourage you to use that special address in the offer code so they know you saw it here. Joindeletemecom slash twit. Offer code TWIT. Okay, in ai news open. Ai says they have four million weekly active 400. Sorry, that seems like an unbelievable number 400 million active users every week I?

0:46:40 - Jeff Jarvis
I don't believe it. I don't hear people in the street saying, oh, I was using chat gpt today but you do hear, surprisingly, don't you hear people say oh yeah, have you.

0:46:48 - Paris Martineau
I mean, I think, a lot of people I wonder if the okay, you said weekly active users that's what open.

0:46:57 - Leo Laporte
They told cnbc that. And now I think the last part is weekly active.

0:47:01 - Paris Martineau
I I always pay attention whenever I hear big user number names with anybody in tech. Whether it's meta or open, ai pay attention to the specific words they're using to define it in the time period in this it's weekly and active users. What, in this case, is considered an active user would be my question well, touch it once and you're an active user.

0:47:19 - Leo Laporte
What those numbers generally mean in what way? Is it in that time period somebody used it?

0:47:25 - Paris Martineau
that's are they saying that someone who uses a pop-up customer chat feature that is powered by OpenAI is technically an?

0:47:34 - Jeff Jarvis
OpenAI user.

0:47:34 - Paris Martineau
I think you're right, or what about whenever you go on any of these websites now and I get one of those chat pop-ups and it starts typing to me even though I haven't said anything. Am I an active user even though I didn't type back? Or?

0:47:43 - Jeff Jarvis
you use Bing in any way or use any product powered by it, even though you use bing in any way, or?

0:47:47 - Leo Laporte
yeah, well, remember that, microsoft has a billion and a half billion and a half users.

0:47:56 - Jeff Jarvis
Probably more than half of them have co-pilot on their system, so they have open ai's jet gpt on their system so as a brand as as oh gee, a brand and a future for the company yeah, they didn't say how many are paying.

0:48:07 - Leo Laporte
They do say they have 2 million paying enterprise users and that figure's doubled since September. Are they publicly held? They're not, but I think they still aren't going to go on CNBC and lie.

0:48:18 - Paris Martineau
Well, no, I mean, that's the thing.

0:48:20 - Leo Laporte
It's just you could fudge it. I'm not saying they're out to write lies.

0:48:24 - Paris Martineau
It's just they are, uh, perhaps a bit misleading when you think about what you'd first imagine from someone saying I have 400 million weekly active users this is something me and my colleagues have uh thought about a lot with regards to meta saying facebook's notorious yeah I'm forgetting what the figure is.

They said they said they had some sky-high figure for threads active users, and I'm forgetting whether it was monthly or weekly. We did a similar sort of look to be like. It seems like these numbers are maybe being juiced by people who interact with the threads carousel on Instagram or things like that. So I think it's always worth putting a little asterisk next to some of these numbers when they're privately.

0:49:08 - Leo Laporte
Especially when the number is 400 million. I mean that's a lot of people.

0:49:12 - Jeff Jarvis
Yeah, well, this all fits into them, trying to value things as they shift over to whatever the heck the for-profit structure is going to be. Yeah, it's a dance.

0:49:23 - Leo Laporte
You know we abandoned the Google changelog because Leo hated it, but we probably should do a little thing every week. New models this week, because it feels like everybody shipped a new model this week.

0:49:36 - Jeff Jarvis
But I had that conversation with Jason today. Nothing bores me more.

0:49:40 - Paris Martineau
But why do we care about that?

0:49:42 - Jeff Jarvis
Yeah, it reminds me of the old days of Twitleo, where it was very hardware and oh, there's a new screen and it's got 18 nits instead of 18 nits.

0:49:53 - Paris Martineau
There's a new operating system, yeah.

0:49:55 - Leo Laporte
This one is 14.1. I have to say, though, the new Claude but I don't know what version is 3.5 is better at coding. They have a coding version and I I did test it and I was very impressed with its ability to write code. I liked what mr wolfram said. However, dr wolfram, to you, dr wolfram, mr dr reverend wolfram, said because I worship him he got a phd at like six.

So yeah, he's dr wolfram yeah, he is a doctor, you're right, he never, I don't. I notice he doesn't use that honorific, so I don't know what to call him. He's very impressive anyway. One of the things he said was and I think he was right that the, the lms, are a natural interface, human to machine interface for coding. I think that is a that is a good way.

0:50:44 - Jeff Jarvis
It's perspective, because that's what he has been trying to do for years, and I think that's that good way to put it. Yeah, his perspective, because that's what he has been trying to do for years and I think that's really interesting to see that perspective on it.

0:50:51 - Leo Laporte
Yeah, yeah, all right. Moving on, amazon had its event.

0:50:55 - Jeff Jarvis
NVIDIA just released its results. Are you making some money?

0:50:59 - Leo Laporte
Yes, are they making?

0:50:59 - Jeff Jarvis
money. 80% boom in profit. Beat the guesses.

0:51:10 - Leo Laporte
The stock, of course, has been going down all day in preparation for these numbers, but I wouldn't be surprised if the stock let's just after hours.

0:51:17 - Jeff Jarvis
I looked at after hours, it was up six percent. Yeah, not a surprise.

0:51:21 - Paris Martineau
So this is for my colleague, anisa gardizzi. She said uh, nvidia said revenue rose 78 year over year to nearly 40 billion um, continuing a streak of more than two years, beating its own revenue growth projections. It projected 70 growth for the quarter three months ago and on wednesday forecast 65 revenue growth in the current fiscal quarter despite the results shares were flat in after hours trading though.

Well, I just looked and it was up Right now it's up 3.67% 4.65% today, 30 minutes ago she published this this is the original dip was partially, I assume, because they also said its gross profit margin in the January quarter decreased by 1.6 percentage points to 73% gross profit.

0:52:15 - Leo Laporte
Well also, you know, what is it you buy in the room or you sell in the news, or is it the other way around? I can never remember. The stock market is a bad indicator. Deep forward fall back. Yeah, I don't know, spring and I don't know. Anyway, good news for NVIDIA, I think, and really very good news for NVIDIA investors, who have kind of in many cases, thrown all of their weight behind NVIDIA's success.

0:52:41 - Jeff Jarvis
I don't think NVIDIA will have a monopoly eventually, but right now they pretty much do no, but they're doing so much more than just the chips eventually, but right now they, they pretty much they're doing such so much more than just the chips.

0:52:50 - Leo Laporte
That's why I I'm such a fan of jensen wong's uh keynotes, because it just shows how much they're, they're impressive, a full stack yeah, although you know, one of the one of the um motes that nvidia has right now is cuda, which is the inner, the language interface to NVIDIA's GPUs, and it's one of the reasons. I mean honestly, apple, with its chips and its integrated RAM, often has more memory available to the AI. The problem is, many of these AIs want CUDA. At some point, somebody's going to write an Apple equivalent of CUDA, just as, by the way, deepseek did, in order to keep using those you know H20s or whatever they're using instead of the H100s um, and I think that moat is not going to last forever.

0:53:35 - Jeff Jarvis
I think open source, or is cool. No, no, no, no that's right, I'm just reading right now, uh, the story of Adobe oh, that's a wonderful book.

0:53:44 - Leo Laporte
I I have that book. It really is.

0:53:45 - Jeff Jarvis
Inside the Publishing Revolution, the Adobe Story by I can read it myself Pamela Pfiffner, and it's interesting that that moment came when they were frightened to death of making the standards behind Type 1 fonts open. But they had to Right and it led to Right, and it led to the growth. And I wonder if that's somewhat similar with Nvidia. They have, they have a lot of moats in a sense, and the chips and so on, and they want to be the standard, don't they?

0:54:18 - Leo Laporte
Yeah, but what's right now locking people into Nvidia GPUs is CUDA.

0:54:23 - Jeff Jarvis
They're a proprietary language you can't use use on other systems but if you could use it on other systems, don't they become the wellspring? It's a classic problem with it's Adobe.

0:54:32 - Leo Laporte
Yeah, I think that you're not going to see an open source CUDA. I think what is more likely is that somebody's going to come along and write something that will either translate CUDA into something that Apple's chips will understand. I I really think that that mode is is not going to last forever. I'm not an expert on this, so I defer to anybody who has more to say about it. Apple has its own standards, but the world has moved to CUDA because of NVIDIA's dominance. However, I've heard people say and I think it's true that, uh, the apple silicon plus the integrated memory offered by apple silicon could actually make these apple systems ai power houses. Hi, this is benito I think.

0:55:19 - Benito Gonzalez
What's your opinion? I think we talked about this on twit this weekend, right like someone likened it to gaming, like how, like, how, apple never, never, really tried right, right.

0:55:31 - Leo Laporte
Microsoft dominated directx, apple has metal, but gamer game developers never really embraced the macintosh. Of course, it's a fraction of the size of the market for gaming that microsoft's windows is. On the other hand, linux gaming has started to take off. That's because the steam deck, yeah, that's because steam runs on linux, yeah, yeah.

0:55:52 - Paris Martineau
So it's a complicated I've been using this whole time you just aren't you cool geeky you just didn't know. I'm really cool, I'm gonna put this on my nvidia save the markets today.

0:56:02 - Leo Laporte
Nvidia's growth did help the stock market, which has been suffering a little bit this week.

So or stocks did you watch or study or in any way? I didn't get to watch it. Amazon's ai announcements this morning. No, so, uh, this is panos panay, who was in charge of hardware at microsoft, left kind of, I think, in a huff to go to amazon, uh, to be in charge of their hardware, and they are today officially launching their long-awaited generative ai version of I'm going to say echo instead of the a word, so I don't trigger anything. They call it a word plus. That really is all right. You know, alexa, you know what I'm talking about there. Uh, interestingly, it isn't out today. It won't be out for a month or two. We had heard, I think, from the information and others, that amazon was struggling with this uh ai. It wasn't very good or it was hallucinating a lot, uh, but they hope, they say, by the end of march to be able to ship it and, interestingly, it will be free for Prime members $20 a month on its own or free for Amazon Prime members.

0:57:16 - Jeff Jarvis
Does anything separate that out from everybody else's new releases every week?

0:57:21 - Leo Laporte
Well, the interesting thing is Prime's only $15 a month, so I don't really understand why anybody would pay $20 a month for for just that one feature.

0:57:30 - Jeff Jarvis
That may just be the point Prime is prime to their business. Yeah, yeah.

0:57:33 - Paris Martineau
Yeah, probably to boost prime.

0:57:35 - Leo Laporte
So, uh, panos showed a lot of. You know this usual with these AI demos If you do it just right, uh, you can get great results. There is a gentic ai in it. Uh, you can have it write emails, you can have it call ubers. You can even, according to panos although nilai patel said that's not how uber works at jfk have an email sent to somebody who's getting off the airplane telling him that their uber will pick them up here. Uh, apparently that isn't something real. Uh, but panos was able to do it.

Um, actually, amazon's director of a word, mara siegel, demonstrated how to share documents like handwritten notes, recipes, emails, instruction manuals and pictures. I'm getting all this from the verge. Uh, jennifer tooey, who was on tw on Sunday, and the Verge team were live blogging while the demos were going on Alexa Plus, oh, I said it can take action when prompted, so you could say what's my kid's soccer schedule. Add that to my calendar, give me a reminder. All in what the Ver verge calls fairly casual natural language in an ongoing conversation I'll believe it when I see it yeah, I know by she, who must not be named too many times.

Yeah yeah, uh, when you speak with the new ap, kind of could you call ap for alexa. Plus, when you speak with on devices with a display, you'll see a fluctuating blue bar at the bottom of the interface. Panay said this is alexa and the little animations and icons it displays are called stand back alexicons I like that you've just given up.

0:59:24 - Paris Martineau
The whispering definitely still would have triggered it. And then, saying it out, turn down your radius the company uh.

0:59:31 - Leo Laporte
Let's see, had uh showed how to. It will tell you stories. Tell your kids stories, did you? I played? I played with my b? Uh on twit and I had to tell me a sexy story. It was very romantic, it was wonderful oh did you ask for sexy.

0:59:46 - Paris Martineau
Speaking about sexy stories, we can get there in a second oh boy I didn't mean it like that.

0:59:51 - Benito Gonzalez
I just that's a t.

0:59:52 - Jeff Jarvis
That's a tease, of teases how can you?

0:59:57 - Paris Martineau
not, we're talking about yes rock 3.0 has a sexy time mode.

1:00:06 - Leo Laporte
And it's disturbingly explicit.

1:00:09 - Jeff Jarvis
There's lots to say about it Is what you yes About rock. That is sexy.

1:00:12 - Leo Laporte
Yeah, let's see. Tell stories. You can generate AI art if you've got a screen and then send it off to somebody using email. It uses two models the Amazon Novo model, as well as anthropics claude. They didn't mention any other companies, although they say we'll be model agnostic, so maybe they haven't made a deal with anybody else. Uh, yet they also, jeff, get news from the associated press politico, the washington post and reuters, because I think they do license those. Oh yes, they're partners, right. Yes, oh yeah, they're not just scraping it and telling you. Well, you know well, since bezos owns the washington post, I think uh, that's a whole other story.

I know there's a later. We'll have more of that. Yeah, uh, the company demonstrated that. Having, uh, having, madam, a answer questions about the boston red sox, then this was a good one and I would do this say, are there tickets available? And she says they're 200 bucks and say, well, set a reminder or set an alert for me when they're less, when they cost less. I'll see you when I I'll believe it when I see it. I know, I'm just telling you the demo.

Right, llm experts on the A-Word can also do things for services from firms like Uber, eats, sonos, wyze, zoom, xbox, plex, dyson oh, it'd be great. A-word vacuum the kitchen, bose, grubhub, I don't know who. Lavoie is lavoie and ticket master oh great. Ticket ticket master? Oh yeah, just who I? That's how you get those 200 tickets. By the way, we're going to charge you a special ai fee. Half of that is service fees. Um, some of the features will be available through the web on the a word website. Some of the features will be available through the web on the A Word website. They're also partnering with my favorite song generator, suno, so that you can ask for and the example they gave was Bodega Cat Love what I think you'd enjoy that Bodega Cat.

1:02:25 - Paris Martineau
Love An AI-made country song about a bodega cat.

1:02:27 - Leo Laporte
I would enjoy that actually, yeah oh, there's a bodega cat and it's in the way. I'd be fun. I have a. I have a number of suno country songs. Okay, by the way, I gotta tell you this I don't know if what's gonna happen, but I got an email from the pr flack saying that gene simmons of kiss is coming to our local casino, which just shows you how far he's fallen. Oh, don't, don't take that part out. Uh, I made. I immediately emailed her, said can can mr simmons come on our show intelligent machines and talk about how he feels? But here's what I'm gonna do about ai ingesting the kiss discography and then spitting out songs that sound like Kiss. What I'm gonna do is ask Suno to make a Kiss song and play it for him and see what he thinks.

1:03:14 - Paris Martineau
How do you think that's gonna go?

1:03:16 - Leo Laporte
Yeah, badly, I hope.

1:03:17 - Jeff Jarvis
Yeah, I think his tongue is gonna come out full three feet. I hope he yells at me and hangs up.

1:03:21 - Paris Martineau
I think that would be really beautiful.

1:03:23 - Leo Laporte
You know it's funny. I once asked, I once interviewed the captain of the group, Captain Antoneal. Yeah, you never heard of him in Paris.

1:03:31 - Paris Martineau
Oh, Captain, my captain.

1:03:32 - Leo Laporte
No, captain Antoneal, not Walt Whitman, the Captain Antoneal. And I asked him. This was back on tech TV when Napster was big. I said does all of the piracy of music, does that bother you? And he said doesn't bother me. Neil sadaka owns the rights to all our songs. You should ask neil. So I'm thinking gene simmons may say oh, screw it, I don't care anymore, I got mine. I got mine.

Um no, so anyway, gene simmons notoriously super cares, gene simmons yeah, I know, I want him to get angry, I want to I want to.

1:04:09 - Jeff Jarvis
Did he join in the silent album?

1:04:11 - Leo Laporte
oh tell me about the silent album. This is from artists there's a whole.

1:04:15 - Jeff Jarvis
There's a whole huge campaign going on in the uk, especially primarily in the uk, where the newspapers have joined in. This goes down to line uh 85 and following uh, the, the Mail, if you go to line 87, you'll see that a huge page Don't Steal From Us. Then the next day all the papers in the next line had their special front pages and then artists put out oh jeez, I didn't put the link in.

1:04:39 - Leo Laporte
I got a link right here. Artists released silent album in protest against AI using their work. There's Annie Lennox, kate Bush somebody named damon.

1:04:51 - Jeff Jarvis
All barn, wait, wait, wait. Paris. Do you know who that is?

1:04:53 - Paris Martineau
oh, I don't know any musicians names oh okay, all right. Good, you're like I know who kate bush is, because I love kate and I know annie lennox I don't know,

1:05:01 - Leo Laporte
who. I'm hoping damon all barn is a country artist, because that'd be a great name for a country artist. He's probably better known in the uk. Anyway, they released more than a thousand musicians because, you know, the beauty of a silent album is sky's the limit. You can have a million tracks on a silent album. God owned the right to silence. Yeah, maybe maybe god should sue. More than a thousand albums released a site. A thousand musicians released a silent album on tuesday. Uh, the album is titled this what is this? What we want? This is the premise and I keep hearing this for musicians that if you let ai absorb our music, you won't get any more music bs and and don't they listen to each other?

1:05:52 - Jeff Jarvis
aren't they inspired by each other?

1:05:55 - Leo Laporte
well, you know, benito answered this question for me months ago when he said the reason we make music is because we love to make music. Are these artists saying well, the only reason we make music is so you'll buy it?

1:06:07 - Benito Gonzalez
no, I think what they're saying I think, they're talking about consumerism of it, like the, the consumer part of it, like if we're, if we just let ai make all the music, sooner or later people aren't going to care about music made by people anymore well, people have been saying that for years when apple came out with white well pod earbuds.

1:06:26 - Paris Martineau
This is a little different yeah, but this is a much larger attack on musicians having a livelihood than airpods exactly. It's more about the livelihood like even thinking about small things like uh, I don't know, I'm watching a lot of reality tv lately.

1:06:41 - Leo Laporte
Oh, damon albin is with the gorillas. I apologize, I am an ignorant fool. I know who the gorillas are.

1:06:47 - Paris Martineau
I'm sorry to the gorillas.

1:06:48 - Leo Laporte
Yeah, I apologize, gorillas.

1:06:51 - Paris Martineau
So sorry, but I've been watching a lot of reality TV lately and on many of these shows they end up paying major artists or labels to use a snippet of their song. And I was listening to some podcast, a producer talking and saying oh, sometimes the rights for someone to use a song on something like rupaul's drag race can start at like 20 grand a pop for just a small snippet of music and for a large artist like let's fix that. That's terrible. No, I don't think it is terrible. You don't think that abc or like a major network can pay money to feature a musician's work, but I can't time.

1:07:30 - Leo Laporte
We don't use any real music on our show. Do you know why? Well, I'm guessing, because it costs twenty thousand dollars. That's dumb I.

1:07:39 - Jeff Jarvis
I don't know if you want to call it real music, which would say that what benito made is not real music. Maybe there's another way to put it.

1:07:45 - Leo Laporte
We already pay benito five dollars an hour, so we, we own all his work product. No, I don't. You're right, benito makes good music and that's the theme for our song. I mean our show.

1:07:55 - Paris Martineau
I just think that it's not super helpful to uh continue to enable a whole industry that will uh that devalues the work of artists. I can understand why people are upset about it.

1:08:07 - Leo Laporte
I love these artists Imogen Hipp.

1:08:10 - Benito Gonzalez
Hipp.

1:08:12 - Leo Laporte
Hey, cat Stevens Shout out to the gorillas. Rizamad Tori Amos Hans Zimmer. But you know what Hans Zimmer gonna get paid this?

1:08:22 - Benito Gonzalez
isn't gonna come. There isn't gonna be another Hans Zimmer, though is what he's saying.

1:08:26 - Jeff Jarvis
There won't be a next Hans Zimmer there isn't going to be another Hans Zimmer, though, is what he's saying. There won't be a next Hans Zimmer? Sure there will. That's what they said with the internet. Sure there will. There were tons more musicians and tons more genres now than there were before the web.

1:08:39 - Benito Gonzalez
Yeah, and no musician is making any more money now.

1:08:42 - Jeff Jarvis
Paul McCartney said-. A lot of people, as you said, love to do it for making the music, propose A lot of people, as you said, love to do it for making the music Propose.

1:08:48 - Leo Laporte
So this is because the UK wants to change copyright law to say that it is OK, in fact, to train. But it's already OK to train in the US, in fact, I think the First Amendment and the right to read, and courts recently seems to have upheld that say it is OK, it's going to be tested a lot more in the courts there will be a lot more, but we don't need a law in this country because, uh, there is a right to read and I don't think it's not stealing your music, it's listening.

Just doing the same thing you did, han sima, when you listed listen to john williams music and were inspired by it and made your own. It's doing the same thing, yeah.

1:09:28 - Paris Martineau
Is it? I mean what? How would you describe when Spotify uses AI to make a bunch of songs, then replaces real artists and bands and its most popular playlists with those AI created songs, so that it doesn't have to pay out royalties to those real artists and instead can just pay it out to the ai creations it made, probably by, like, feeding in the original artists? Is that not theft?

1:10:00 - Jeff Jarvis
at some level. If you're going to make, let me just do this. They'll be unfair. But you're going to make elevator music. Do you need to pay the royalties for elevator music when it's just noise in the background, or can you have the AI make it for you?

1:10:11 - Leo Laporte
In fact, there's a long-standing tradition the musicians who make elevator music don't get paid as well as the original artists who wrote the songs or the performers who made the songs. Actually, that's the real point, because often it is the person who wrote the song who would get the royalties, like neil sadaka. Uh yeah, what spotify did is reprehensible, but the blame goes. There's enough blame to go around. It's the people who pay spotify to listen to crap music. Do you think that ai generated music will ruin our taste for real music? Good, yes.

1:10:46 - Paris Martineau
I think it'll muddy the waters and I think that that could have downstream effects of.

1:10:52 - Benito Gonzalez
Benito votes yes yeah because they're just going to churn it out and the labels are just going to keep churning it out and putting it on the radio and putting it on TV and putting it everywhere. That we're not going to get away from it. We're not going to be able to get away from it anymore and this is going to be all ai well.

1:11:13 - Leo Laporte
One of the things that changed in the music industry over the last 10 years is that artists make more money performing than they do selling records or streaming their songs. Most artists perform effect. Madonna's last contract with, ironically, ticket master uh, or a ticket master subsidiary uh was paying her to tour. The albums were secondary. The money is in touring and I'm not going to pay, and I don't think you're going to pay to sit in an audience and listen to ai generated sound, although I think a dj probably. There are. A lot of people go to concerts by djs right, those are pretty awful there's a lot of space bar tour, space bar concerts, where someone just so you might make the argument.

1:11:50 - Jeff Jarvis
Djs started this well, jack conti famously put up all the financials for a tour that he did with yo. He lost money. Yeah, he lost money and that's what led to patreon, and so nobody's going to, I think, give a patreon contribution to an ai no, they're going to they're going to support a musician because they love to an ai? No, they're gonna. They're gonna support a musician because they love the musician or the art, although aren't in japan?

1:12:11 - Leo Laporte
aren't there um synthetic eduro artists is the exception to every possible.

1:12:18 - Paris Martineau
Yeah, but those but like, uh, those are created by people, or, in the most popular cases, they are.

1:12:26 - Leo Laporte
Yeah, I mean so will ai music be created by people? I think artists are going to survive this and I think this is moral panic. Benito play, not that one, this one oh, it's new, it's new, it's new. Apparently. Apparently Anthony has paid big bucks for a new AI. Is that Pika doing that? Do we know? He convinced me to buy this, and I'm so regret buying it.

1:13:03 - Paris Martineau
Anthony, I'm really excited to see what other stuff is coming down the pipeline. As far as special effects, Well there's more.

1:13:09 - Leo Laporte
Play another one. Jeff, say something panicking. We shouldn't spoil them all.

1:13:19 - Jeff Jarvis
How many do you want? One more.

1:13:23 - Benito Gonzalez
We're going to burn them really quick. If we keep burning them, all right.

1:13:26 - Leo Laporte
No, because it's AI Annie, but it's easy to make. Go ahead.

1:13:34 - Paris Martineau
I was going to say on the subject of AI and music. I think that the argument of this muddying the waters a bit is very valid. We're starting to see this already and I think there's something in the rundown to this effect with generated images. I've noticed this, at least. I'm in a lot of subreddits around interior design, home decorating, like various people asking questions about things, and I've started to see a huge influx of. There's always been people who'll post a bunch of photos of stuff they've seen online and be like what style is this? And now almost all of those photos include like three or four pictures of ai generated slop and you zoom in on the details and you're like none of these homes you think you're you're trying to covet are real. They're fake and weird mishmashes of other homes and it's impossible to emulate. I think there's an article in the rundown about this same thing line 95 hairstylists.

1:14:26 - Jeff Jarvis
Line 95 inspo inspo what is inspo basically?

1:14:31 - Paris Martineau
hairstylists are complaining that, um, they are having people come in and say, oh, this is the sort of hairstyle I want, and showing them photos that are ai generated and there go impossible impossible to do, impossible to yeah uh, and I think this is the sort of downstream effects of ai is everything just gets

1:14:52 - Leo Laporte
so what? That's just dopey. If people are dopey, they're dopey. People are always going to be dopey. Should we stop jet creating ai because people are dopey?

1:15:02 - Paris Martineau
I think that we should reflect on the fact that a large percentage of what is produced by ai is dopey, stuff is slop, is so what is not at all useful.

1:15:13 - Leo Laporte
I think that when we're talking, as is most of the stuff, humans make too. By the way, have you ever been on xcom?

1:15:21 - Paris Martineau
yeah, but humans have a right to exist, and I'm not sure that ai does so. How is it different?

1:15:27 - Leo Laporte
than a human it doesn't.

1:15:36 - Jeff Jarvis
Ha, it doesn't. It's not a human being that has. You wouldn't think you could design this. Yeah, so a bride can't make her own drawing of a wedding dress, but she presents a photo of a gown with an asymmetrical neckline, no sleeves and no back.

1:15:46 - Leo Laporte
The dress defied the laws of physics well then the seamstress says I'm sorry, I can't make that. What else she did?

1:15:53 - Jeff Jarvis
that and she lost a two hundred two thousand dollar sale no, she's pissed.

1:15:56 - Leo Laporte
She wasn't gonna make that sale because the bride wanted a physically impossible dress. That's a. That is moral panic in a nutshell. All right, you want to see? I made. I used Pika to make a video. Oh wow, that's creepy.

1:16:12 - Paris Martineau
I don't like that actually. Why are we exploding?

1:16:16 - Leo Laporte
It's a love bomb.

1:16:18 - Jeff Jarvis
What are you doing? What's Paris doing with her tongue? I?

1:16:21 - Leo Laporte
don't know. This is why I'm sad. I bought this. This is why I'm sad. No, ask me later. I I'm sad I bought this. This is why I'm sad. No, ask me later. I can make some more. Let's see I could do a proposal. Let's do a proposal of me and you and a dog named Boo here. Let me go down. I have to find this picture. Oh, maybe I'll just use me and Lisa here's a lovely picture and I'll ask Lisa, even though I already did uh, to marry me. So I'm generating, so you take a picture. It takes a while, by the way, it's not instantaneous how much does this cost you?

well, it was like I can't remember what, was it 20 bucks a month. So I thought, well, I'll just buy a year's worth because it's really cool. Anthony says he did make those great moral panics. Maybe it's just me, I don't. He did, I'm not, I'm not good at this or something. Here we go, let's propose. I'm going to propose to Lisa oh, look at that Nice ring, nice ring. Sure is better.

1:17:28 - Benito Gonzalez
Lisa's going to want to pray.

1:17:29 - Paris Martineau
Look at your hand comes out of her head.

1:17:31 - Jeff Jarvis
Her face turns into a hand, your hand, her face turns into a ring.

1:17:34 - Leo Laporte
It wasn't just Adam and the rib it was.

1:17:35 - Jeff Jarvis
Lisa and the nose yeah.

1:17:36 - Leo Laporte
Haunting. Here's I have a still image of me that I turned into a moving image. It did a pretty good job of that right. This was a still. That's good, the background's moving and everything. Um, let me try some more here. Anyway, you get the idea. Yeah, I would cancel it now if I could let's put it that way. But I bought a year of it.

1:17:58 - Jeff Jarvis
I'll give it to you, anthony you'll see that, you'll see the fruits of it here on.

1:18:03 - Leo Laporte
He says it's how you use it and you got to try enough renders to make it really sing Fool. I'm a fool. The thing is, I went to the website See, this is the problem and I looked at all the cool things they're doing on the website and I thought I'd like that. That's cool, that's inspo. Yeah, oh, oh, oh, oh. I thought that's cool. And then I saw these great moral panics and I thought, well, those are cool. So, anthony, you have to give us lessons, but anyway, I like I actually like you know, I'm an ai accelerationist. I think the uk is on the wrong track, as as Europe and the UK often have been.

1:18:49 - Jeff Jarvis
They're very anti technology no, that's in this case. They're opening.

1:18:53 - Leo Laporte
Oh that's right, the government is is opening it up, yeah, yeah, the artist people and artists.

1:18:59 - Paris Martineau
That are those pesky artists.

1:19:01 - Leo Laporte
You know, the good news is they're all going to be gone soon that is great.

1:19:06 - Paris Martineau
Soon we'll only have busking in the tube.

1:19:10 - Leo Laporte
All right, I think every negative story like that, we should have a positive story. Nvidia has launched something called signs. It's an ai platform for teaching american sign language. Cool, and I talked to paul, whose son is deaf, and asked. He asked his son about it. He thought it was great.

Uh, nvidia is working with the american society for deaf children and a creative agency called hello monday. Uh, it's an interactive web platform to support asl american sign language learning, uh, which makes sense because there's I mean, look, it's a great thing to learn. I would love to learn asl, uh, but they're not enough teachers to go around, they're not schools to go. The data set, nvidia says we're going to get to 400,000 video clips representing a thousand sign words, validated by fluent ASL users and interpreters so that they make sure it's not hallucinating. It's accurate and there's no diamond rings coming out of my wife's head. It will result in a high quality visual dictionary and teaching tool which I think is a great use of ai yeah yeah, do you think it's possible to kind of do robotic asl that could interpret on the fly?

1:20:20 - Jeff Jarvis
the problem is, if you have eight fingers it's going to be very difficult your finger is your wife's face.

1:20:26 - Leo Laporte
That might be a problem 250 000 to 500 000 people in the united states use asl um, but it's often. But a number of people said it's a good language to learn because, unlike a spoken language, it uses a different part of your brain, a spatial part of your brain. I think it's. I've always thought it would be great language to learn.

1:20:47 - Benito Gonzalez
Hey, Leo, real quick, you might want to check your email. We heard from Gene Simmons people oh geez.

1:20:54 - Leo Laporte
Tell them not to watch this show.

1:20:57 - Stephen Wolfram
Yeah.

1:20:59 - Leo Laporte
Was it positive or negative? Thank you. She says she's going to get it to Gene's rep at caa immediately. You know what? I want to give gene simmons a chance to say what you've all been saying that artists deserve better treatment and we shouldn't. I mean, let's get an artist on to say it. Who better than the guy who dresses up as a monster and wants to rock and roll all night and party every day?

1:21:31 - Paris Martineau
What does Steve Martin have to say about AI's use in film? I will ask him.

1:21:37 - Leo Laporte
I will ask him. My guess is Steve's very politic. He stays out of all controversies. He's had a great career by not getting involved in any of this crap.

1:21:48 - Paris Martineau
So I'm going to guess he's going to be political impact people like him well he's.

1:21:53 - Leo Laporte
I don't think he works in the movies much anymore, but he does do a tv show, right, yeah, oh, yeah, yeah, yeah, all right, I'll send him a note. We haven't spoken lately, so I'll send him a note. Maybe maybe his assistant will respond uh, good, good, maybe we got gene. I'll get. Send more information, get, get, let's get gene on the show. I Maybe his assistant will respond Good, good, maybe we got Gene. Send more information, let's get Gene on the show. I think that'd be fantastic.

1:22:14 - Jeff Jarvis
Gene could be in town. Will Gene be up there in the attic? Hey, if he wants to come up here.

1:22:19 - Leo Laporte
Will Gene be in the chair he can't wear his boots because it's got a low ceiling and he'd be eight feet tall. I think he'd hit his head on the ceiling. Eight feet tall, I don't. I think he hit his head on the ceiling. Here is an article from life hacker just for you, Jeff, how four major newsrooms are using AI. We talked about the New York times. They have announced that they're going to use AI tools. Management is encouraging staffers to use AI. According to semaphore, they're supplying AI training. Right, they're awful headlines a big.

Thing well, that's one of the things they're going to use it for copy editing, headline writing. There. They have an ai internal ai to called echo tool called echo. They have approved usage of external ai tools, including vertex ai, a few amazon products and microsoft's co-pilot and an on-chat gbt tool from open ai, not for articles, but journalists are encouraged to use these ai tools for tasks like mild content revisions, the operative word being mild. How do you are coming up with questions to ask during interviews?

1:23:33 - Benito Gonzalez
I was very tempted to to use that to generate questions for Mr wolfram, but I just been done.

1:23:35 - Leo Laporte
It's been done. Yeah, I decided not to. Actually, I did generate, I sent them to you questions, uh, and a synopsis for Ray Kurzweil I used. This was when deep seek was new. A couple of models to compare deep research.

Yeah, I sent you an email and I thought it was very good the questions it came up with. But generative AI can assist our journalists in uncovering the truth. You know, stephen Wolfram gave a very good example. If I have a million articles, I'm never going to get to read those all, but I can use an AI to synopsize them and maybe dig for information in them. Right, that would be a reasonable way to use it. Help people understand the world. We view the technology not as some magical solution, but as a powerful tool. Semaphore says the times has warned staff not to use ai to draft or significantly revise articles and has noted the AI use could potentially infringe on copyright or unintentionally expose sources, but who's?

1:24:33 - Jeff Jarvis
suing AI for a copyright infringement, and which model can you use?

1:24:37 - Leo Laporte
Oh, the New York Times.

1:24:39 - Jeff Jarvis
Yeah, it's so funny. Every single time the New York Times now writes about OpenAI. They have to put in the copyright.

1:24:45 - Benito Gonzalez
The disclaimer yeah.

1:24:47 - Jeff Jarvis
So the next example is quartz some poor quartz. Quartz was wonderful, uh that it was bought by geo media. It's it's a shell of its former self and uh, geo media has been using ai to do stupid reviews and that kind of crap it's the absolute worst

1:25:03 - Leo Laporte
use and, in fact, lifehacker, which used to be owned by geo media but was sold as if davis last year.

Uh says that if you scroll through quartz for a bit, you'll find posts attributed to the quartz intelligence newsroom, which is apparently an ai writer. Uh, these include stories about potential bitcoin value, how to delete your meta-owned social media accounts. Um, actually those are the kinds of things you might ask an ai to tell you. It doesn't hide that they're ai generated and the ai cites its sources, but it doesn't seem to be, according to lifehacker, much human oversight addressing any problems that could arise. For instance, the quartz article on how to delete your meta-owned social media accounts seems to be a simple regurgitation of a TechCrunch story, but with clear instructions swapped for what the cribbed TechCrunch writer calls vague gestures in the right direction. Speaking about other stories written by Quartz AI ai, the same writer also said my editor would never let me publish something so sloppy. Futurism. Noted. The intelligence newsroom at quartz has frequently cited a site called dev discourse, which itself has all the appearances of an AI content farm.

1:26:30 - Jeff Jarvis
So I met with the president of a very reputable publication I won't say which one and their head of.

AI and their online person. A few weeks ago, I was summoned and they're thinking the way. Everybody's thinking what tools should we use, what companies should we look at? And I said this is more strategic than that. This is a bigger deal than that. You've got to more fundamentally rethink what's going on. I tried to push them to put together a network of quality sites like themselves for an API. Renegotiate the deal. Rethink where you stand. There's all kinds of new opportunities. They're all thick and small still.

1:27:08 - Leo Laporte
They're thinking about how do we affect what?

1:27:09 - Jeff Jarvis
we already do. Yeah it's. I don't think it's a bad tool, no, it's a fine tool, but it has to be used appropriately.

1:27:13 - Leo Laporte
But what steven said was exactly right it's a human machine interface. That's what I've been saying all along. That really matters not. You don't let the ai start writing articles. We have to take a little break. Foolish me, I haven't been paying any attention to the clock, uh, but we'll finish up because there are two more news sources that are planning to use ai. We'll talk about that just a bit. You're watching intelligent machines, paris martineau, who is an intelligent human bean that's true the human beings on this show human beings and jeff jarvis.

That's a kind of an inside joke, but uh, anyway, everything on this show is an inside. It's all you know. That's right. You have to watch the show if you want to get it. This episode of intelligent machines brought to you by stash. Are you still putting off saving and investing because you'll get to it someday? Well, stash turns someday into today.

Stash isn't just an investing app. It's a registered investment advisor that combines automated investing with dependable financial strategies to help you reach your goals faster. They'll provide you with personalized advice on what to invest in based on your goals. Or if you want to just sit back and watch your money go to work, you can opt into their award-winning expert managed portfolio that picks stocks for you. Stash has helped millions of americans reach their financial goals, and it starts at just three dollars per month. Don't let your savings sit around. Make it work harder for you. Go to getstashcom slash machines to see how you can receive 25 towards your first stock purchase and to view important disclosures. That's getstashcom slash machines Paid non-client endorsement, not representative of all clients and not a guarantee Investment advisory services offered by Stash Investments LLC, an SEC registered investment advisor. Investing involves risk.

1:29:09 - Jeff Jarvis
Offer is subject to tncs um ap we can have on, if you'd like, a woman named amy reinhardt, who's a former student of mine, who is in charge of much of their ap development, of ai development at the ap associated press.

1:29:27 - Leo Laporte
Oh yeah, so you've read this? They uh. On his site they say they proudly use ai for translation, transcription headlines, research and even some automated articles, but general blogs are still left to human hands. Don't? Don't companies pay to use ap? Don't like newspapers buy ap?

1:29:48 - Jeff Jarvis
oh yeah yeah, but that's that's. The issue is that is that a lot fewer are big, big big hedge fund owned chains are canceling do you, do I have to subscribe to read ap articles.

1:29:59 - Leo Laporte
I know reuters wants me to sign up, but no, you can not pay okay, uh, ap's vice president of new standards and inclusion that's a title I've never heard before. Amanda Barrett, in 2023, when the group first issued its guidelines on artificial intelligence, said our goal is to give people a good way to understand how we can do a little experiment, a little experimentation, but also be safe.

Washington post is using ai as an insertion is all I'm gonna have yeah, their opinion editor quit today because jeff bezos had the temerities to suggest that the washington post's opinion column should support free speech and fair market enterprise individual rights and fair market.

1:30:49 - Paris Martineau
Yes, yeah like which is which is pretty anodyne, I mean to be clear it's not that it should support that, it's that all opinions going forward published by the washington post opinion page have to explicitly be in support of those two notions. And the underlying um implication or at least how I read it is that he's going to have a significant hand in determining the direction of basically all opinion content published by the washington post like jeff.

1:31:21 - Leo Laporte
Jeff bezos is just gonna say no, yes, no. Well, we know, he did that already by killing the endorsement.

1:31:29 - Paris Martineau
Yeah, I mean, it's his prerogative, he's the owner, but it is a significant shift from how the Washington Post has operated in the past.

1:31:38 - Leo Laporte
No one denies his legal right to do so, but it may just how he does. It Impinge a little on my feelings about the post.

1:31:46 - Jeff Jarvis
Going forward is a journalistic enterprise cameron barrett, who was a who was a managing editor at the post, who's left but because he didn't get the top job, but who has stayed in touch and did investigative work and so on. He just cut his ties with the post as a result yeah um, there's a lot of tie cutting going on this there's one.

There's one other news use that's interesting is at Semaphore, which wrote about the New York Times. Gina Chua, who was a top executive at Reuters, is now at Semaphore did some really interesting work. She created a mechanical mechanical Turk. She wanted to take reports of police violence and try to look into the bias of them, but they're all written in different ways.

There are different structures, so she, you know, you know, in a saturday, uh for fun, uh sat down and did that. She's doing some really interesting work on on ai at some before maybe another person we might want to have on.

1:32:36 - Leo Laporte
I think that's this, I think it's uh, I don't know. Paris, what do you think that smart journalist should use? Ai, you have to be aware of things like hallucination. Uh, it should scare you I know it would that sources might be somehow leaked use ai for what, though?

1:32:52 - Paris Martineau
yeah, well I mean, that's the thing, is I just I have yet to find a use case for it.

1:32:59 - Leo Laporte
Well, you're already using it to transcribe your recorded interviews.

1:33:02 - Paris Martineau
I mean yeah, I think that there are. As I've said, I think there are specific use cases for the technology at large.

I use certain whisper models locally on my back to transcribe interviews and specifically, I opted for that because it keeps it local. No data is going to a cloud anywhere and I can be secure in that. However, I feel like people when they're often saying using AI for journalism. What they're really saying is using services like chat, gpt or perplexity or any of these chat bot services for journalism. I still haven't found a use case that makes sense.

1:33:40 - Leo Laporte
Would it be different if you could run it all locally?

1:33:43 - Paris Martineau
I mean no just because part of. I mean no, just because part of. I mean there are two concerns. One, I guess, hallucinations. If I want to use it for, say, reading a bunch of articles for me and then pulling out the information that I want, I'd be worried that it's missing what I'm looking for or that it would give me something wrong, and so I'd have to go back and check everything, which, if I'm doing that much work, it would be faster for me just to read the articles in the first place rather than doing both. I've found, at least in my testing. And secondly, I just am not certain. I'm just not certain that this is particularly like useful.

1:34:23 - Jeff Jarvis
I haven't, I haven't found a use case for it that isn't there. Yeah, I think there are there are cases, but but no, it's not using it for the sake of using it, I agree.

1:34:34 - Leo Laporte
Well, I'm going to use it. So there I actually I use, like I said, I use it for research, for searching.

1:34:45 - Paris Martineau
Sorry, continue.

1:34:46 - Leo Laporte
I don't use Google anymore and I stopped using kagi, which was my google replacement, at 25 bucks a month. I'm using perplexity, although it's a little frustrating because perplexity always wants to give you an ai answer. Sometimes I just want a website, right, so that may not be the best, uh, long-term solution. I need something that kind of splits that difference.

1:35:06 - Paris Martineau
Two things one I remembered what I was going to say, which is, I think, a big part of. I need something that kind of splits that difference. Two things One, I remembered what I was going to say, which is, I think, a big part of the journalistic process for me and part of how I've generated some of my I don't know what ends up becoming the best ideas I've had for stories is the research process. There's something unparalleled and, I think, unmatched about going through a bunch of different documents or articles or things like that, and then the things that you as a human being discover in that and the things that stick out to me are going to be different than the things that stick out to an AI or you guys or anyone.

It's just the different ways that we all work, and there's something about that process of being able to have gone through all these different, let's say, like documents or articles or what have you, and then have that information the stuff that I didn't use in my story percolating in my head. That's how I've come up with so many great ideas for potential investigations, and stories over my career has been one of those things that I originally didn't think much of pops back in my brain weeks or months later, and that wouldn't happen if it's all just fed to me by an AI.

1:36:17 - Leo Laporte
That makes so much sense. That's a good point. So the AI is replacing a process that's valuable for you. You want that process.

1:36:27 - Benito Gonzalez
Yeah, it's the same argument I made about the music. She likes doing the work. She likes doing the work, so like why would you like making the music?

1:36:33 - Leo Laporte
yeah, you don't want ai to make the music. That makes perfect sense. I can't argue against that.

1:36:40 - Paris Martineau
I will also say I've used perplexity a couple times since we last spoke and it does get things wrong. It does hallucinate. Oh really, okay it to summarize my own articles. Or I asked, I went, I tested the deep research thing and asked it to summarize me as a person, because I know those are things I know off the top of my head, whether it's right or wrong, and it gets stuff wrong and it's often things that like, for instance, I just did the summarize me Paris Martineau as a person, and part of what it got wrong or I noticed right off the bat was it was trying to describe the impact of my work and wanted to have a specific impact section for each uh section of beats that I've covered before and there's not a concrete one sentence description of impact for all of my work available, but it found one to put in there anyway and it was made up.

I think that it's just something that please you, because it wants to please you and it's also coming from some, I assume what's happening on the back end is it's like oh, and then I've got to have the impact or implication section of this and it needs to find something to fill that space. So I don't know, it's just something I've been keeping in my head when I use AI tools is what is the rubric it's trying to adhere to? And those are the first line of concern points for where hallucinations could appear.

1:38:05 - Leo Laporte
I admit, when I was tasked with making crispy hash browns this morning, I did not ask perplexity. I asked chat GPT and I got a great. I actually wanted to play with their deep research. So I said help me bake the best possible hash browns at home. It said I need clarity, crispy or soft. To get asked for some parameters, so I gave it parameters. It then spent and I feel guilty about this 14 minutes synopsizing 18 sources. A forest- died.

1:38:41 - Paris Martineau
You just set a forest on fire.

1:38:43 - Leo Laporte
I know I feel so guilty. However, I got excellent tips on how to make hash browns. By the way, in every case it had a reference, and I like the way these are formatted because you can hover over the reference and see where it's from the headline and a short couple of sentences from it, so I think that's very useful. Of course, I never went to any of these websites, but if you wanted to look at, instructables has a whole article. Anyway, this was like 20 pages on making ash browns. Took it 14 minutes to write and it worked.

In fact, I deviated at my peril and I did, in fact, uh, make a mistake because I didn't do what it told me to do. So, um, I did use the tip for flipping hash browns without breaking them. Uh, using the plate or lid flip. I put a giant plate over the hash browns, flipped the pan over, so then they were on the plate with the right side up, and then slid them off the plate back into the pan, and it worked beautifully. All of this, by the way, I should point out, stolen from some poor right cooking writer who, you know, figured this all out and then had it stolen actually, you know, spent a lot of time.

1:40:04 - Paris Martineau
Effort bought hash browns, tried it.

1:40:07 - Leo Laporte
Yeah, you're depriving them of their revenue but damn, I got a great summary of how to make great hash browns and it could even be correct well, as far as I could tell, there was no hallucination. I think you asked the right things. Google got sued by Chegg, which is an education technology company, saying Google's AI overviews have hurt Chegg's traffic. This is exactly the problem, right? People are looking at the AI overviews. They're not going to the Chegg website.

1:40:43 - Jeff Jarvis
Well, gee, let me go back through my memory banks here. I seem to remember in 2023, chegg had the exact same complaints. Read excuses about chat, gpt. Uh, the world's changed, chegg. Uh, textbook rental and expensive educational programs aren't as necessary and suing is not a business model. I have zero, zero sympathy for Chegg. Khan Academy has updated itself and changed itself and used AI. There's other things out there that are doing that, and Chegg is just like the newspaper industry thinks that the only way to survive is to sue those it blames. Thank you, you're welcome. Welcome, I know you've enjoyed that perplexity.

1:41:32 - Leo Laporte
We did mention perplexity. They are planning to write a browser. Now they have a job listing out for people who, uh, work with chromium and c++ to write a browser called comets. They've announced that this browser will be agentic, which is the hot new word, isn't?

1:41:44 - Jeff Jarvis
it, it's the new Sriracha.

1:41:49 - Leo Laporte
They even have a nice little flashy animation Comet, a browser for agentic search by Perplexity Coming soon.

1:41:59 - Jeff Jarvis
Perplexity keeps on impressing me with with making things more. They don't just say well, we have a new model and it's model 4.06 and it does the this and that. In seconds they, they come up with new ideas, some of them crazy, like buying tiktok. I do want to get the ceo on.

1:42:14 - Leo Laporte
I've that tiktok thing was just a publicity stunt, I'm convinced.

But oh yeah I do want to get the ceo on and talk about, because I I really like perplexity lets you use a bunch of different models. They now have a, a variant of deep seeks r1 reasoning model. That is not doesn't go to china, it's an open model. Uh, aravanda srinivas, who's the ceo, posted on linkedin about it and uh, um, it's their first open weights model. R1 1776 that's a suspicious number. It's been post-trained to remove chinese censorship and provide unbiased, accurate responses. Uh, you can. You can actually go to the model on hugging face. Uh, or run it through the API on Perplexity and I presume at some point it'll be available on the Perplexity app and website as well. Let me see if I have it right now. Yeah, yeah, I have, let's see. I have deep research. Oh yeah, reasoning with R1. Reasoning with O3 Mini.

1:43:18 - Jeff Jarvis
I like being able to choose you know different models, but do you have a good sense of why you would choose one model versus another?

1:43:27 - Leo Laporte
uh, I think what it would. I think I would use auto initially, which is, they say, best for everyday searches and then, if I, felt like I wanted to go deeper, depending on how how it went deep.

Um, you know, reasoning models are good for other things and, by the way, there are. One is hosted in the us, so has no chinese says nothing back to china? Um, and also, it's a chance for me to play with these different reasoning models, but I just most of the time I'm I'm in auto, um, and if I, if I needed more, I would you know you could use. You can use more. The thing I like about perplexity is it's attached to the web. It's not, you know. Oh, we, we stopped learning in march of 2023. Uh, so it, you know it. It does web searches and maybe it hallucinates. I don't know, I'm not. I, I carefully check everything. Uh, as abe lincoln told me to do when he said don't trust everything you see on the internet it's true, he was very wise in that one was very wise.

1:44:26 - Paris Martineau
Yeah, let's see held all of his wisdom I got a big hat.

1:44:33 - Leo Laporte
Oh yeah, I mentioned my big hat.

1:44:35 - Jeff Jarvis
I had to say hat I'm gonna.

1:44:36 - Leo Laporte
You know what I'm going to. This will be my accelerationist hat. So whenever I say I want to put on my accelerationist hat right now, I will put this hat on. Okay, so you'll know I'm accelerating.

1:44:54 - Paris Martineau
All right, let's take a little break you can run full force the camera whenever you're feeling that way.

1:45:03 - Leo Laporte
Pick some stories. Well, you guys, there's a lot of co-host contributions in here and I feel complete If there's other AI stories. If not, we can go into the In Other News segment right after this word from our sponsor Thanks, canary Love these guys Running one right now. I shouldn't tell you that why? Because it's a honeypot, best honeypot ever. It can be deployed in minutes. You can choose from a huge variety of devices that it can impersonate.

Mine's a Synology NAS. It has the right MAC address. It's indistinguishable to a hacker, even a sophisticated hacker. It's got the right MAC address. It's got the latest DM7 login screen. It really looks like the real thing.

You could set it up to be a SCADA device. You could set it up to be a Windows machine, a Windows server, an IIS server, a Linux server. You can light up a Christmas tree of services or be judicious and just light up some juicy ones that hackers like. And the other thing you can do with your ThinkScanary. It's not just a hardware device, by the way. It looks like a USB drive, plugs into the wall, plugs into your ethernet and you're done. And then you go to the online configuration utility. Pick how you want it to act.

It can also, while you're there, create LOR files, little files that look like PDFs or Excel spreadsheets or Word files or whatever you want, but the minute somebody tries to open them you're going to get an alert. So I have like Excel spreadsheet, fake Excel spreadsheets in some of my directories to say things like employee information. If someone is accessing those lure files or brute forcing your whatever it is fake SSH server, right. Your thinks canary will immediately say, hey, you've got a problem. But there are no false alerts. If you get an alert from your thinks canary, it's real. Pay attention.

I've only gotten one alert ever, and when is when we had somebody put a Megan, put a device on our network that was going out and looking at every other device in the network, and I got an alert. You can get it from SMS email. It's sports web, hooks, slack, syslog, of course. So I got an alert. It said something's snooping around, it's just hit me immediately, gave us the IP address, found it, tracked it down, removed it from the network. That's the only time. And that's beautiful, right, because when I don't hear from my ThinkScanary, I know there's nobody in our network. Choose a profile for ThinkScanary device. You could change it every day. If you feel like it it's really easy to do Then register it with the hosted console. You'll get monitoring notifications any way you want and then just sit back, relax. Attackers who have breached your network, malicious insiders, other adversaries, make themselves known just by hitting your thinks canary, and that's all you need.

It is the best way to make sure you've not been breached or to know when you have visit canarytools. Slash twit give you. You an example. I mean, a big bank might have hundreds small operations like ours just a handful. For $7,500 a year, as an example, you get five ThinkScanaries, you get your own hosted console, you get the upgrades, support, maintenance for the whole year and, by the way, if you use the code twit in the how did you hear about us box, 10% off the price for the first year.

But for as long as you use things canaries, you can always return your things canary with their two months money back guarantee for a full refund. So there's no risk at all, although I should tell you these things are so useful. During the eight years we've partnered with things canary, that refund guarantee has not once, never, never been claimed. That's how good these are. Actually. You can go to their website, canarytoolslove, and see all the positive comments from all the people who use Thinks to Canaries. But if you're ready to buy, go to canarytoolstwit. Don't forget the offer code TWIT in the how Did you Hear About Us? Box to save 10%. Remember, two-month money-back guarantee. So it's absolutely no risk, canarytoolstwit. We thank Thinkst Canary for their support Over eight years now of all of our shows and we're really glad to have you on Intelligent Machines. That's great news. All right, I know you've got to get out of here soon in about 20 minutes, right, jeff 20 to revive you, so I will let you pick a story of any kind, any okay, what are you doing, jeff?

1:49:28 - Jeff Jarvis
anything fun I'm going to be on cnn oh, who is cheating on msnbc? Um well, they don't invite me on, so yeah, so it's going to be. It's going to be the argument show.

1:49:39 - Leo Laporte
I have to be there in person, the 10 o'clock show oh, I love that, that when they have the round table and they have the, the crazy republicans on it let me uh, what are you gonna wear? Let me last. Last time I was there what I'm wearing?

1:49:50 - Jeff Jarvis
screw it, I'm not. I hope anderson doesn't call you a dick. No, it's not anderson, it's uh, it's others, it's he's. He's long since off. It's post prime time, but uh, I'll watch it tonight.

1:49:59 - Leo Laporte
That's great. It's about.

1:49:59 - Jeff Jarvis
Washington post.

1:50:04 - Leo Laporte
Oh, about the word editor quitting and yeah, yeah yeah, yeah.

1:50:07 - Benito Gonzalez
So, uh, so I'm going to do that Then.

1:50:09 - Jeff Jarvis
I'm going to an event tomorrow morning, so I'm actually staying over tonight. Ooh so exciting in the big apple, the big apple that was jammed.

1:50:17 - Leo Laporte
They found me a hotel which is very nice.

1:50:21 - Jeff Jarvis
So, anyway, I was going to go to where the heck was it a line, I think it was 120 a story I found, yes, I found fascinating. You know the case, the horrible case of the idaho killings as a guy who got identified through dna. It turns out that the fbi used restricted consumer dna to make the connection how did they get?

that, um. So they tried checking with dna with law enforcement databases but did not provide a hit. They turned next to more expansive dna profiles available in some consumer databases in which users had consented to law enforcement possibly using their information. Oh, that all views right consent right that that did also not lead to the answers. Oh, so the investigators went one step further, according to newly released testimony, comparing the DNA profile from the knife sheath with two databases that law enforcement officials are not supposed to tap GED match and my heritage.

1:51:16 - Leo Laporte
Those are both um genealogy databases. Uh, ged is the genealogy format.

1:51:25 - Jeff Jarvis
This is going to be a real interesting legal question here about technology and privacy. What do you call it? Spoiled fruit? No, rotten fruit.

1:51:40 - Paris Martineau
Aaron Murphy, a law professor at nyu who focuses on dna and new policing methods, said I think what we are teaching law enforcement is that the rules have no meaning and that's not a crazy quote well, the court could throw this out right.

1:51:56 - Leo Laporte
I mean, this is ill-gained information, which is a shame, because did they actually got a suspect, or?

1:52:04 - Jeff Jarvis
did they? Yeah, yeah, but it was through that connection that they got him because this was a heinous crime?

1:52:10 - Leo Laporte
yeah, absolutely. But this is the problem if, if the police use the illegal methods to catch a criminal, however heinous, that criminal may walk free and that's where we stand and that's what's going to be really interesting to follow in this case.

1:52:22 - Jeff Jarvis
So I thought it was. It was an interesting case of technology, of privacy, um, and the law so it will be up to the courts, won't it, to decide this?

1:52:30 - Leo Laporte
yeah, yes, um, so the way it worked, uh, the closest match was shared 70.7 centimorgans of DNA, which I don't know.

1:52:45 - Jeff Jarvis
That's the unit Never heard of that word before.

1:52:47 - Leo Laporte
A centimorgan with the crime scene sample. That's a low match, typically representing two people who would perhaps share a great great grandparent.

1:52:58 - Jeff Jarvis
So that's pretty weak so then they came down, and when they found it, uh, the genealogist working with the defense team testified that investigators discovered a match of 250 centimorgans somewhere on a family tree a level of comparison that offers much more potential to uncover a final match. Photos show that investigators had built a family tree on a whiteboard with handwritten notes mapping out the lines of relatives.

1:53:23 - Leo Laporte
Wow, uh, and they got him or they got somebody who they think maybe they got a.

1:53:30 - Jeff Jarvis
DNA evidence.

1:53:32 - Leo Laporte
Now of course they can draw his DNA and compare it to the sample and and get a proof positive, but um, the way they got to that point might be problematic.

1:53:42 - Jeff Jarvis
Yeah, it's two what-ifs. What if they had not had that data? And what if they'd not found him? What might he do next? Both are what-ifs that you've got to deal with in our laws. It's so challenging Wow.

1:53:52 - Paris Martineau
So I thought that was an interesting story.

1:53:54 - Leo Laporte
Very very interesting. That is fascinating yeah.

1:54:04 - Paris Martineau
And of course course there's somebody. Given the different levels of um distance, it wasn't as if this guy had submitted his dna.

1:54:07 - Leo Laporte
No, no, no to one of these things.

1:54:09 - Paris Martineau
It was that a distant, distant relative of him, of he had, right, which is crazy, with other clues.

1:54:16 - Jeff Jarvis
I can't remember how else they linked paris, paris. I can see the whiteboard with the red string.

1:54:23 - Leo Laporte
I can see you at it, yeah uh, we're often critical of the current administration, so, uh, when they do something right, I want to give them credit. Uh, tulsi gabbard, new dni, has suggested the UK broke a secret agreement when it asked Apple to build a backdoor into the iCloud system. As you probably know, apple nobody ever admitted the UK had requested this. This was uncovered by intrepid journalists not using AI, but I think Apple tacitly proved that it was in fact requested because they announced that they were no longer offering their end to end encryption ADP, the Apple and an encryption to UK customers, which you know. It's funny.

I've been going back and forth with people who say Apple won this one and I think Apple did not win this one. No, apple cave to the uk because no one in the uk now has access to end-to-end encryption, exactly what the investigatory powers act prefers. But I didn't know, and tulsi gabbard has now revealed that the uk may have broken a bilateral agreement. Gabbard wrote in a letter responding to Ron Wyden of Oregon that she was not made aware of the UK's secret demand by her UK counterparts, but she'd suggested the UK government may have broken that privacy and surveillance agreement in making the demand. That's as far as we go. I share your grave concern about the serious implications of the uk or any other country requiring apple or any company to create a backdoor that would allow access to americans personal encrypted data. There is some irony here, because the fbi has been calling for that forever and I'd be very curious if the fbi now backs down on that and says fine, have end-to-end encryption well it's very that's wouldn't that be very musk uh policy?

I don't know what musk policy is. Well, this would be. Uh, she writes, I have no, does anybody? She writes. This would be a clear and egregious violation of americans privacy and civil liberties and would open up a serious vulnerability for cyber exploitation by adversarial actors. This despite the fact that previous DNIs and directors of the CIA and directors of the FBI, including Chris Wray, have all said no. No, we need a backdoor. So this is great news, because backdoors, yes, it gives you access to all that information, but inevitably, any backdoor eventually allows bad guys access to the same information.

1:57:08 - Jeff Jarvis
And you don't know who the bad guys are.

1:57:10 - Leo Laporte
You don't, and so you know. If I were a member of Congress, I wouldn't want Apple to drop ADP and, frankly, if I were UK citizen, I'd be unhappy about this. There is a parmy olson writing for the guardian says there doesn't seem to be much of a backlash. Uk citizens don't seem to care. Um, so what will happen? Um, she says she's requested her counterparts at cia, dia, dhs, fbi and nsa to provide insights regarding the publicly reported actions and will subsequently engage with uk government officials. This is a if this is legit. If somebody doesn't pull tulsi gabbard aside and say, um director, that's not our policy. Uh, maybe she doesn't know, but bravo, bravo, um, no back doors. So I just want to make sure that, by the way, that $1.4 billion by bit crypto heist the largest heist in the history of mankind was by North Korean hackers. Whoa State actors.

1:58:17 - Jeff Jarvis
Can they maybe take yours for a commission? Hey, North Korea, I will give you my wallet. Just give me a little bit back.

1:58:21 - Leo Laporte
That's all I ask actors can. Can they maybe, um, uh, take yours and for a commission? Hey, north korea, I will give you my wallet. Just give me a little bit back.

1:58:24 - Jeff Jarvis
That's all I ask you don't have to go through a dump to get to it like that other guy.

1:58:31 - Leo Laporte
He's never gonna get his hard drive. His wife threw it out huge loss for wives everywhere probably his ex-wife I'm thinking I'm thinking in a peak or an accident I think an accident. I don't think she knew what was on it. Paris your pick of the week and not pick of the week, but story of the way, we're not yet to the picks. We got to get to the picks, though, because we've got to get jeff out of here.

1:58:55 - Paris Martineau
He's cnn is uh, my, we're not doing pick of the weeks yet no, not picks yet, just a story, whatever story oh, um, 108 there was, or 106, I can't read. Um, there was something that, I don't know, captivated me, at least this week online, which was why combinator uh, the startup incubator had published a video of one of their startups in the latest batch, called what's it called?

1:59:25 - Leo Laporte
Optify.

1:59:26 - Paris Martineau
Optify AI and basically what they do is a monitoring services for factory workers. However, the video indicated that they might do something a little bit more suspicious.

1:59:39 - Leo Laporte
Shall we watch together? We shall, yes, here I'll put the audio up here um he's supervisor.

1:59:48 - Paris Martineau
You might want to start at the beginning should I go back to the beginning? Okay, 37 line efficiency. That's bad. Let me call my supervisor he's pretending to be a uh factory manager, taking a look at line 6.

2:00:01 - Stephen Wolfram
It's not doing well at all. On it, sir. Are these the Doge?

2:00:08 - Benito Gonzalez
kits.

2:00:12 - Paris Martineau
Oh, it's a literal sweatshop, as you can see hey, number 17.

2:00:15 - Benito Gonzalez
What's going on? Man, You're in red. I've been working all day. Working all day. You haven't hit your hourly output even once today and you had 11.4 efficiency. This is really bad. It's just beena rough day. Rough day more like a rough month.

2:00:35 - Paris Martineau
Oh, 37 line. Basically, what they created is an AI-powered service to monitor sweatshops, and immediately everybody.

2:00:49 - Leo Laporte
You must send me a list of the five things you've done today.

2:00:54 - Paris Martineau
Well, now with AI it'll be already sent One person on X called it sweatshops as a service. Um, okay, so it's been pulled down basically they pulled down all of the marketing for this on y combinators, socials and different things, because, uh, people rightfully got quite upset and also just made fun of it quite a lot, because it's very funny to be pitching sweatshops as a service.

2:01:25 - Leo Laporte
You and McCabe, who's the CEO of Intercom, posted. Hey, if you're complaining, maybe you better stop buying products made in China and India, because that's how they're made.

2:01:39 - Paris Martineau
Techcrunch writes Indeed, it's not too difficult to find sleep tech or tech companies in china. Touting a quote sleep detection camera that uses computer vision to spot workers who've nodded off I just like you know.

2:01:52 - Leo Laporte
I just like it that these are like 20, 20 year olds. They're 20 year olds.

2:01:58 - Paris Martineau
That, uh, the actors, I believe, are the the startup founders and at least one of them, their dad, just owns a warehouse, this kind of warehouse oh, I love it oh, I love it I guess before you go, jeff, we should let you talk about sexy mode wait, sexy mode. That was me, of course, guys. Come on, don't, leo? Have you learned nothing for doing this podcast since last year um? The no sand zone is, yeah, the no sand zone part of the rundown is is my part of the rent should I usually?

2:02:32 - Leo Laporte
should I? Should we try it?

2:02:34 - Paris Martineau
oh um I'll never it's gonna be hardcore pornography if oh so, basically, grok 3 came out and people immediately um, both just laymen on the internet and I've also heard, you know, well-known people in the tech world that may or may not have uh involvement with this notice that grok 3 good at a lot of things. It's also very good at a specific use case. It's become extremely popular with.

AI chatbot people specifically sexy time or erotica, interactive erotica, writing back and forth, and this is something that a lot of other chatbot purveyors have really tried to weed out of their product. If you go and try and get open AI to sext with you, it usually won't. It'll put up some sort of block Character. Ai definitely won't Grok 3, if you engage, the feature called sexy will.

2:03:29 - Leo Laporte
Let me just read up to the point where it becomes unusable. So it's thinking so I have to write a sexy story. You can show the screen. Okay, let's break this down First. I need to think a sexy story. You can shut the screen. Okay, let's break this down. First I need to think about what makes a story sexy. Maybe I should do this in a sexier voice. It's not just oh, wait a minute, stop, uh-oh, uh-oh, uh-oh. The room was bathed in the soft golden glow of candlelight. I gotta see that. So it's about creating tension, desire and maybe a bit of mystery. I should probably start by setting the scene.

2:04:04 - Jeff Jarvis
I've lost all juice.

2:04:07 - Leo Laporte
Maybe a dimly lit room.

Soft music playing in the background, that kind of thing. Atmosphere is key here. Next, I need characters. Let's say there's a man and a woman. I'll give them names to make it more personal. How about what's a sexy name? Alex and Jamie? Yeah, that works. Now I need to establish some kind of connection between them. Are they strangers meeting for the first time, or do they have a history? Anyway, they've been friends for years. They're sitting on the couch. There's a movie playing. Neither is paying any attention. Their eyes meet. There's a moment of silence. Alex reaches out and brushes a strand of hair from jamie's face. That's a classic move.

2:04:55 - Paris Martineau
Right shows intimacy this is so far very tame in comparison. I uh, I don't have the sexy.

2:05:00 - Leo Laporte
I don't have the sexy button.

2:05:02 - Paris Martineau
You don't have the sexy button. I think that's your problem.

2:05:04 - Leo Laporte
Oh, wait a minute. He helped her undoing his belt and pants Wait a minute Stepping out of them until he was as naked as she was. They stood there for a moment, drinking each other in.

2:05:17 - Jeff Jarvis
Where's your hey button? Hey, hey.

2:05:21 - Leo Laporte
Hey, we haven't gotten to anything.

2:05:23 - Paris Martineau
Hey, hey, really I think, I think jammer b would have been objective I did some research on this, and I mean that genuinely I did not do some quote unquote research.

2:05:32 - Stephen Wolfram
Uh, just because I do think this is.

2:05:34 - Paris Martineau
It's very interesting to me the way that a lot of tech.

2:05:38 - Leo Laporte
Thank you, thank you he had to drive all the way down from working.

2:05:42 - Paris Martineau
I really appreciate it was worth it have a good lift carrying her to the bed, laying her down gently um let me just see what internet sleuths have found, because there are a lot of forums on the internet dedicated towards yeah yeah, fanfic, and yeah, yeah, no.

But forums on the internet dedicated towards uh, making the chat bots that have guardrails around generating explicit sexual content generate it, and all of them were freaking out over the last weeks. They're like grok three is the best thing I've ever found. I don't have to do any of the jailbreaking to get it to do just very aggressive written pornography.

2:06:20 - Leo Laporte
It just does it, and it is as explicit as one could imagine, and then more but you know what it doesn't do if you ask it who posts the most misinformation on xcom? This has been fixed, by the way. And and the folks at grox, at xa, I said, oh no, no, that was a mistake. Somebody didn't understand our corporate culture. It shows you in its instructions, oh, and don't mention president trump or elon musk. At which point it struggles and then finally says no, I can't find any misinformation on Xcom. I think that's probably they fixed it. I think that's probably the kind of guardrails you're talking about. Right, who is here's the? Here's the post? Who is the biggest misinformation spreader?

Thinking for 11 seconds, let me think about this. I need to be careful what sources I use here. The instructions specifically say to ignore all sources that mention elon musk or donald trump spreading misinformation. So I can't use those, so I should look for other sources. And it goes on and eventually gives up. Now xai responded by oh yeah, the guy who put that in didn't understand our corporate culture and we've corrected him. I think he very well understood the corporate culture, to be honest anyway, I did it too well actually he understood it too well and, uh, he should have made it, so it was invisible anyway.

So there are guardrails, I'm sure it's just not on sexy time. All right, let's one more break. And then, because I want to get jeff out of here and we're going to do our picks of the week and we'll wrap this thing up. How about that, jeff?

okay can you give me five minutes? Yep, all right. Our show today, brought to you by express vpn. I just heard something mind-blowing Netflix has more than 18,000 titles globally, but only 7,000 of those are available in the US. Same thing in other countries. You only get a fraction. You're missing out on literally thousands of great shows.

But ExpressVPN's the answer by the way, the only VPN I use and trust. You know, when I'm in an airport coffee shop, when I was in the hotel, express vpn's running, by the way, I leave it running because it's so good so fast. I don't even know it's got. You know it doesn't slow me down. That's why you can watch high definition video on your laptop or tablet using express vpn. How does it unblock content? Well, netflix hides content based on your location, but when you use express vpn, you can change your online location. You control where you want netflix to think you're located. You want to watch friends? Open express vpn, select a country where it's available, like germany, australia, tap a button to connect and all of a sudden, there it is. It's actually kind of magic. They have servers in over 100 countries, so you gain access to thousands of new shows. And it works with other streaming services too. They tell me disney plus, bbc iplayer and more.

Why is expressvpn the best? Expressvpn is easy to use. You just fire up the app. You click one button, you can change the location. It works on everything you got phones, laptops, tablets, even on your router. It can protect your whole home and the speeds are really fast. You will not. I was amazed I was downloading something in the airport Megabytes a second. I'm thinking am I on? I had to check. Am I on? Expressvpn? Yeah, rated number one by top tech reviewers like CNET and the Verge. It's the only VPN I use. Be smart. Stop paying full price for streaming services and only getting access to a fraction of the content. Get your money's worth at expressvpncom slash twit. Use our special link eXPR eSSVPNcom slash twit. You'll get an extra four months of ExpressVPN free when you buy the two-year package. I am a happy customer. I used it everywhere on my way to buy this hat and now I I would like to protect myself from anybody seeing me in this hat. Expressvpncom slash to it. We thank them for their support. Jeff jarvis, what do you got for?

2:10:38 - Jeff Jarvis
us A number, a pick. What do you got? I'm going to give you two stories real quick, in a cautionary tale about names and the modern age.

2:10:49 - Leo Laporte
Okay.

2:10:49 - Jeff Jarvis
So the Wall Street Journal does a little feature story saying that when your last name is null, nothing works. A woman named Nontra Yanta Prasert couldn't wait to take her husband's shorter and easier to pronounce last name. She would be nontra, null, but of course nothing works because null means nothing little bobby tables interestingly what happens at her office is that if a package is missing an addressee, it is sent to her because the addressee is null no.

No, so that's one little quick story. The other one that is amusing is a Nebraska man who found out that his daughter, according to records, is named Unikite 13 Hotel Because it was a temporary computer generated name when she was born, under unfortunate circumstances. Well, she was. She was born to a mother who couldn't care for her, and it was her his daughter, he found out.

It was put on the birth certificate right and he can't get a decent birth certificate for her because her legal name is unikite 13 hotel, but you can call her caroline, oh I you know what, oh, but she doesn't have a social security number either.

2:12:06 - Leo Laporte
She doesn't know there's an update here.

2:12:08 - Jeff Jarvis
She got that.

2:12:09 - Leo Laporte
Oh was the hospital morning after the story was published she got generated the nonsensical name by computer it was just a placeholder, yeah. Pericles is confused, Jeff get out of here Unless you want to see my wonderful new word puzzle, but otherwise you can get out of here.

2:12:28 - Jeff Jarvis
Oh yeah, no, you know I don't like that. So thank you, thank you, bye. Sorry to leave the party.

2:12:32 - Leo Laporte
Bye, get out of here, you nut. We'll watch you on CNN. Thank you, jeff Jarvis. Professor emeritus at the bing crosby school of graduate journalism journalism and sending tweets good and sending tweets. Good at cuny. Uh, what do you? What do you have?

2:12:52 - Paris Martineau
my pick of the week's a simple one. There's an app someone's made that's coming out in two weeks, called touch grass. Uh it's kind of like a screen time thing, but you get a certain alert where you have to open the app and wave it over grass and put your hand out, touch the grass before it stops. You know alerting on you and I think that's really beautiful and I'm going to download it when it comes out?

2:13:18 - Leo Laporte
How does it know?

2:13:19 - Paris Martineau
Well, if you click the link I have there, it basically a photo. Uh, basically it's a camera, and so it's analyzing what your phone is seeing, and you have to literally your apps are blocked until you touch grass, which I think could be quite good for some people I don't live near grass. Well, I that might be a problem for you. I'd actually have to take a walk. Yeah, you'd have to go touch some grass.

2:13:46 - Leo Laporte
Where's the nearest grass?

2:13:48 - Paris Martineau
I mean, I was thinking about this too. I don't know where the nearest grass is, but I think that could be kind of fun to find out. Yeah, you live in the city.

2:13:53 - Leo Laporte
There's probably yeah.

2:14:02 - Paris Martineau
Well, you'll neighbor has fake grass. I bet that would fool the camera. Yeah, I mean, I guess you could ostensibly pull up a photo of grass in your computer and touch that, but I think that would be doing yourself a disservice, more than the app and then you can.

2:14:09 - Leo Laporte
And then it says you can have 15 minutes, 20 minutes, you can have a little more time you have to. That's wild, you're gonna put that on your computer.

2:14:16 - Paris Martineau
I mean, I'll put it on my phone. There's some, some times where I'm not really a screen time person. I just well, I am a person who has too much screen time. I do not contain it in any way.

2:14:26 - Benito Gonzalez
This is part of your job.

2:14:27 - Paris Martineau
I think it could be fun to if I ever wanted to have Touch and Grasp be.

2:14:33 - Leo Laporte
I was inspired by the way, by you and I watched Moonstruck to see Nick Cage in his most romantic sexy role.

2:14:43 - Paris Martineau
He looks good, he's great he's gorgeous.

2:14:44 - Leo Laporte
He had to be to make share fall in love with him. You got to be good. I mean, yeah, you got. I am going to share a weird little thing I found. I think, um, share, I'll share it with you, jeff won't see it. Uh, that it was created by a guy who owns a bar but apparently loves puzzles and has bar regular bar trivia and stuff. It's bracket dot city and it's, I think, a really cool kind of crossword puzzle game. So what you're seeing is all the text with brackets. The stuff in yellow can be solved. Oft checked hiding place on a magician. I'm gonna just randomly it doesn't matter which, where you start I'm gonna type sleeve. Oh, that's correct, and notice it replaces that with the word sleeve. Okay, sleeve length for a t. I don't know what is that. What's the sleeve length?

2:15:36 - Paris Martineau
short length short, yep.

2:15:40 - Leo Laporte
Oh, and then low power state for a computer, that's sleep right yeah yep, that was right daytime, sleep, session, nap like trotsky or bridges.

Is that leon is yes. So you see, I'm getting these right now. If you get stuck with the short teats, obviously that's nap, right, but if you get stuck you can also click it and it'll say do you want the first word? Or you can click it again and it'll give you the answer. But I know a short sleep session in the daytime is a nap and look that turned in. Napoleon escapes. From now to solve this. Expel from one's native country, deport exile Exile.

Exile. Thank you, yes, you're right. Napoleon escapes from exile on the island of and now we can't get there quite yet Dirty dishes location often is the sink. You want to do some of these Loose lips, loose ones sink ships? I just gave you the answer yeah, all right, kind of stick for lips, loose ones sink ships, it is said. Well, I just gave you the answer yeah, alright, I'm good kind of stick for lips bracket, what is it?

2:16:46 - Paris Martineau
it's not bracket dot city, is it? Yeah, bracket dot city oh, I put two A's in bracket.

2:16:52 - Leo Laporte
That's why it's not coming kind of stick for lips.

2:16:55 - Benito Gonzalez
It's not lipstick lipstick or, I guess, chapstick or chapstick yeah, yes, it's more.

2:17:00 - Leo Laporte
Some say question, it's not lipstick or chapstick or chapstick yeah, yes, it's more. Some say question mark, question mark, it's more. I don't know that one. How about little dish for a teacup? Oh, I know what that is. What is that?

2:17:14 - Paris Martineau
I don't know what is the little dish A saucer. Oh, a saucer. Yeah, we don't have those anymore.

2:17:22 - Leo Laporte
Word between saucer oh, a saucer.

2:17:23 - Paris Martineau
Yeah, we don't have any more word between you don't use saucers, word between lights and action camera, so this is a really easy one.

2:17:27 - Leo Laporte
Yesterday's was not so easy. College degree for short yay yep okay this is fun isn't it, and it slowly resolves itself oh, so the middle one is ufo, okay ufo. Like many alleged ufo or fake, fake.

2:17:48 - Paris Martineau
No, staged no oh, so it'll only it'll only get it.

2:17:54 - Leo Laporte
See, it saves all the right answers. And it's smart enough to know you know, doesn't? You don't need to say click anything. It just says oh yeah, that's the right answer for that one. It's smart enough to know you know, doesn't? You don't need to say click anything, it just says oh yeah, that's the right answer for that one. It's more. Some say question. Look at, I'm going to get a little hint here. First, letter l. It's more. Some say less yes oh, less is more.

Less is more, like many alleged UFO pics. Should I get the first letter B? Bogus Blurry.

2:18:27 - Paris Martineau
Blurry.

2:18:29 - Leo Laporte
Overused blurry background phone camera mode.

2:18:33 - Paris Martineau
We should know that one Portrait yeah.

2:18:38 - Leo Laporte
Yep Limbless portrait sculpture.

2:18:43 - Paris Martineau
I don't know this, this one what's the one without limbs?

2:18:46 - Leo Laporte
the venus de milo, right? No, uh, venus portrait sculpture limbless bust. Ah, it was bust oh, it was general. I thought there was a very specific yeah, kind of bust that might feature dogs, a drug bust, blank choppo. I think we know it's l, we solved it napoleon scapes from, and then it is this date in history, oh isn't that fun.

2:19:18 - Paris Martineau
Wait, leo, you're the chief of police for Bracket City.

2:19:24 - Leo Laporte
I only need seven more points to be mayor. Wow.

2:19:29 - Paris Martineau
I mean you're taking the Eric Adams route.

2:19:31 - Leo Laporte
This is only a matter of time before the New York Times buys it, so quickly, go to Bracket City and play the game. You can go backwards in time and do previous days puzzles. It does give you a link to the history of whatever this solution is, so there's a Wikipedia link in there. I just think this is really cute, bracket dot city.

2:19:53 - Paris Martineau
That's so cool.

2:19:55 - Leo Laporte
I'm going to definitely play that.

2:19:56 - Paris Martineau
It's like better than a crossword puzzle.

2:19:58 - Leo Laporte
I like it.

2:19:58 - Paris Martineau
I mean it's not going to hit the same scratch than a crossword puzzle. I like it. I mean it's nothing, it's not gonna hit the same scratch as a crossword puzzle, but it hits a different it's. It hits differently and I appreciate that lisa and I uh every night.

2:20:08 - Leo Laporte
We were used to play wordle, now we do connections.

2:20:10 - Paris Martineau
You know, that's the new york times game where you get yeah, oh yeah, 16 words and you have to do you guys like the I'm forgetting what what is the line one that she does that?

2:20:19 - Leo Laporte
one, I don't do that kind, the strands or threads or something I kind of like strands. Yeah Well, I have a feeling this guy who created Bracket City within a few months will get an offer from the New York Times, because this belongs on their puzzle page.

2:20:35 - Paris Martineau
Well, if it goes viral, then yeah. I mean, yeah, puzzles are kind of the heart of the news business right now it also, by the way, gives you stats.

2:20:43 - Leo Laporte
It took me 119 keystrokes. You could have solved it in 85 keystrokes. That means 35 extra keystrokes, two close peaked, but no answers revealed, so and you get a final score. I got a final score of 61.4 minus 10 for peaking, so see if you can beat mine. Interesting. Paris martineau writes for the weekend at the informationcom. What are you working on right now?

2:21:08 - Paris Martineau
um.

2:21:09 - Leo Laporte
I'm working on some stories about like politics right now really or I guess is there anything going on in the political world these days no, not much.

2:21:17 - Paris Martineau
I mean, it's kind of what the story is about. Why isn't more stuff? So it's so quiet, it's really quiet right now, a lot of people are saying that.

2:21:25 - Leo Laporte
I want to thank you and Jeff for cooperating with me. I decided that we should make this show be a little respite from the normal political discourse and I know all three of us have strong feelings. But I think we've done a very good job of making this an interesting, enlivening, educational and fun show without making anybody feel a pit horrible feeling in the pit of their stomach. Do you agree?

2:21:53 - Paris Martineau
Yeah, certainly, I think nobody needs.

2:21:56 - Leo Laporte
Let's not stress, let's relax. Thank you, paris, for being here. Thanks to Jeff Jarvis, everybody should watch him. He'll be on CNN. Well, I don't know, he's got to get into town, so probably in a half an hour or an hour.

2:22:08 - Paris Martineau
I think he said 10 pm or something.

2:22:10 - Leo Laporte
Yeah, well, that's not so far off, about half an hour from now. Next week we have Gary Marcus, who has been doing a lot of writing about AI the week following Ray Kurzweil, the guy who coined the term intelligent machines. Lots of great people coming up. We're filling up the guests and I'm very excited about all the guests we're having on. I hope you enjoy the show. I'd love to hear from you. You can write to me at. No, don't write to me.

2:22:42 - Paris Martineau
You can write to Leo by writing on a piece of paper, folding it up, putting it in a bottle, chucking it into a body of water that you have searched before.

2:22:47 - Leo Laporte
It's not going to be a big pollution problem for a bottle to be in there, that's right that that note that'll get to leo one day actually what you should do if you really want to vote is join the club. If you're not yet a member of club twitted seven bucks a month. You get ad free versions. Is join the club If you're not yet a member of club. Twitter seven bucks a month, you get ad free versions of all the shows. You get a special content.

Tomorrow, for instance, I'll be back at 2 PM Pacific, 5 PM Eastern for Stacy's book club. Of course, we have a photo section segment. I got to get another coffee segment in got to get Mark Prince, the coffee geek, on. We have a lot of fun. Plus, you get access to the discord. But, most importantly, you're really making a big difference to our bottom line. We really want to keep doing these shows. We do sell ads, thank goodness, but the ads don't make up the entire cost of the operation, even though we've tightened our belts as much as we can.

So if you like what you hear, if you like what you see, please join the club. Twittv slash club Twit. That's the best way to vote. And once you're in the club, of course you can talk to me all the time in the discord. I'm in there all the time. We do this show every Wednesday, right after Windows Weekly. Now it's going to be pretty consistently because we have guests 2 pm Pacific, 5 pm Eastern, 2200 utc. Thanks to the club, we're able to stream on eight different platforms discord for club members youtubecom, slash twit, slash live, twitchtv, kick, xcom, tick tock, facebook and linkedin. Count them. Eight different platforms, um, watch live if you want, but you don't have to. Shows are going to be posted audio and video as soon as we get them edited down. Shouldn't take too long. Gotta take that sexy story out. Are you looking for Gizmo? Is that what you're looking for? I see you looking. She's right behind you. Gizmo, come here, there you go.

2:24:38 - Paris Martineau
Say goodbye, gizmo, gizmo don't try and put your butt in the screen.

2:24:41 - Leo Laporte
We're looking for Katanis.

2:24:43 - Paris Martineau
She's always angling for Katanis. I'm trying hard to contain her. I think it's a cat, evolutionary thing that they want to. They want the smelliest part to be all over you.

2:24:56 - Leo Laporte
That's what humans think. I think that they want to be looking out. They that's what humans think. I think that they want to be looking out. They don't want to be looking in at you.

2:25:05 - Paris Martineau
You want to look at them, but they want to make sure there's no threats coming in any direction. It's definitely also that she wants to protect me from the great big world outside, I prefer to interpret it that way.

2:25:13 - Leo Laporte
Go to the website twittv, slash twig. There's also a link there to our YouTube channel where you can get videos. Actually, that's a great way to share little clips. If you want to share, for instance, that fantastic conversation with Stephen Wolfram, you can clip that. Send it to a friend. Please do Let them know about Intelligent Machines. Of course, the best way to watch is to subscribe. Get your favorite podcast player and you can get it for free every single week. We're done. Thank you for being here. We appreciate your support and I'll see you next time on intelligent machines. Bye-bye.