Intelligent Machines 807 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. I'm Leo Laporte, jeff Jarvis is here, paris Martineau, our guests Ethan Sutton and Marie de Lourdes Zoyo created this little thing I wear on my wrist the Bee Computer. They'll explain how a personal agent on your wrist can change everything, plus all the AI news coming up next on Intelligent Machines Podcasts you love From people you trust. This is TWIT. This is Intelligent Machines, episode 807, recorded February 19th 2025. My life in tokens. It's time for Intelligent Machines, the show where we cover artificial intelligence, robotics and even those little devices you have sprinkled around your house that just seem to turn on the thermostat at random hours of the day or night. Joining me right now Paris Martineau from the Information. Hi, paris, hello hello Welcome back from your rock concert.

Yes, thank you for filling in for me while I was gone. I really appreciate it. Also here, Jeff Jarvis. He is an emeritus professor of journalistic innovation at the Craig Newmark Graduate School of Journalism. I'm trying to get past the song.

0:01:19 - Paris Martineau
You tried to speed through it. We're not going to let you do it.

0:01:22 - Leo Laporte
I thought if I said it fast enough, enough benito wouldn't push the button. Uh, and soon a prof. Well, now, I guess a professor at mount claire state university, and well, suny stony brook yep yep, oh, the montclair thing fell through. He's a fellow, no fellow.

Oh, you're a fella, okay fella, he's a fellow professor if it's at montclair, new jersey, it's probably a fella fella fella. Hey, we have a great guest on the show today. You've seen me many, many times. If it's at Montclair in New Jersey, it's probably a fella Fella Fella. Hey, we have a great guest on the show today. You've seen me many, many times talking about this silly little bracelet. I don't have to wear it. It's actually holding the B module which I'll pop out of the bracelet. You can wear it, whoops.

0:02:01 - Jeff Jarvis
That's it. That's it.

0:02:02 - Leo Laporte
Break the guest's big gadget before they even go, yeah, before you even introduced them I threw it in the air so you can wear it around your neck as a pendant. You can wear it, uh, as a clip-on. I like to wear it as a bracelet. It's always with me. Plus, it fools people because it looks like it's a fitbit or something sort of fitness thing. What is it really? It is a set of microphones very good microphones attached to a device that then streams to my phone and then sends it from my phone to an AI for analysis. Everything that I do say happens in my day gets sent to this AI for any summary from your rock trip Did.

0:02:43 - Jeff Jarvis
Oh yeah, the conversations. Yes, yes, Today, yeah.

0:02:45 - Leo Laporte
I have it all.

0:02:46 - Jeff Jarvis
Yes, yes, it's amazing. I had a good day today.

0:02:48 - Leo Laporte
The only thing I have to do it has a button you can push to turn it off, to have it stop recording, and I keep forgetting to do that during TV shows, so occasionally I'll say things like good luck with your rehearsals for Richard III. Yeah, well, we have an update on that we can talk about. Let me introduce our guests. It's great We've got the founders of Bcomputer. They had their coming out party in January at CES. Maria, I'm going to do your name. I think correctly, delordes Zoyo.

0:03:19 - Maria de Lourdes Zollo
So it's Maria Delordes Zoyo is my surname. Yeah, South America.

0:03:25 - Leo Laporte
So your first name is Maria de Lourdes. Yeah, and then Zoyo is your surname. Okay, You're from South. You're not from Spain, you're from South America.

0:03:33 - Maria de Lourdes Zollo
Born in Caracas, grew up in Italy. I'm a mix and I really appreciate the pronunciation because I think, like not basically no people understand that it's Zoyo.

0:03:51 - Leo Laporte
I love it. And I saw that you went to school in italy and I thought, yeah, yeah, that's definitely a spanish name.

0:03:53 - Ethan Sutin
So that's great. And with, uh, maria, her co-founder, ethan sutton. And am I saying sutton, right, uh, sutton.

0:03:56 - Leo Laporte
Either way, you guys met. Uh, what your previous startup was? Squad, right, that's right. Which then was purchased by Twitter and became Spaces.

0:04:13 - Ethan Sutin
Yes, so you're the founders of Spaces. Well, you know, I don't want to take full credit for Spaces. Esther Crawford also. Right, yeah, yeah, yeah, but we joined and launched it on Android. You know, interestingly, we worked on a speech-to-text for the captioning of that. You know, interestingly, worked on a speech to text for the captioning of that.

0:04:27 - Leo Laporte
So you both have a little bit of experience in AI, but tell me how you started be be dot computers the website, by the way.

0:04:36 - Ethan Sutin
Yeah, I will. I will keep it really brief, but there's there's some relevance to our last company actually with with B, because we actually started in 2016 the company we were all together at with esther that we sold to twitter. We actually started it as a personal ai company. So, like, me and esther were always kind of really obsessed about what what it would be like if we could have a computer, a machine, have like a really good model and understanding of of ourselves, of of humans, like personality, beliefs, attitudes and thoughts, and like what value would come from that. We actually started that and and launched it, but it was too early because, um, it was before transformers existed, before the transformers paper, before large language models and, um, I don't know if you remember, there was like a brief, very short-lived chatbot boom in 2016. Yeah, with tay microsoft's tay, the racist chatbot.

0:05:33 - Leo Laporte
Yeah, yeah, she rest in peace, yeah yeah, we all thought chatbots would be the big thing and it turned out they were terrible.

0:05:40 - Ethan Sutin
Yes, yes, and we were, we were excited about it and so, like we were one of the terrible chatbots at that time, oh dear, but like we were too early, but it, you know, I think that obviously that all changed with with llms, uh, so, anyway, we pivoted and we went through um like combinator and we, we built squad and we worked at twitter, but then, you know, now it's kind of we. You know, if you would have told us in 2016 the kind of natural language processing that's possible now, I don't think anybody working in the field would have really believed that we would have gotten so far be about the hot technology of 2016 google and it's.

0:06:30 - Leo Laporte
It is now about the hot technology of 2025, which is obviously uh ai just to flatten what I know jeff's gonna say, esther crawford was the woman who slept in her sleeping bag at twitter after elon took over. But you guys did. You were you under the? You didn't stay for Elon, right no, no, we left before.

0:06:46 - Ethan Sutin
Not related to Elon, but left before. So it did not get to experience any of the Elon, but that was my question yes, I knew, I knew where Jeff was going, I just had to know.

0:06:55 - Jeff Jarvis
I had to know yeah, yeah.

0:06:57 - Ethan Sutin
So so no, no, uh, no, no Elon stories and no sleeping in the office stories.

0:07:03 - Paris Martineau
I was about to say B isn't keeping you guys sleeping in sleeping bags in the office.

0:07:09 - Maria de Lourdes Zollo
We're for sure sleep deprived.

0:07:13 - Paris Martineau
Hopefully just a normal amount where you can be sleep deprived and sleep in a bed.

0:07:17 - Leo Laporte
Let me show you. I have my iPhone on the screen. This, by the way, only works with iOS currently.

0:07:22 - Ethan Sutin
Well, Android is actually in the Play Store now under preview. So is actually in the Play Store now under preview, so it's in beta testing. We're still telling people that it's like the full support's on iOS, but the Android's full featured and as soon as we. You know Android is a little trickier to iron out all the bugs with the device fragmentation and everything, but within about a week we'll probably do the full release.

0:07:45 - Maria de Lourdes Zollo
Yeah, I like that. You just did the announcement.

0:07:47 - Leo Laporte
Congratulations.

0:07:49 - Ethan Sutin
Thank you for holding that announcement for us it was in preview beta, but you can install it on Android now and it works Okay, good.

0:07:56 - Leo Laporte
Amazingly, it's. Currently this is the kind of Explorer's edition $50 and there's no subscription. Now that's obviously going to change. In fact, I'm here to lobby you for that to change, because I want to make sure that you get a regular income for this and I want to make sure we have access to the best AI models. Let me just, before we go any further, show people what I'm getting from this thing.

Here's my Bee app. It does a few things. It records all the time, so it's recording right now, it's just so. It also has previous days summarizes, the daily and, by the way, the summaries are fun because it's not only. It has a certain personality, which we'll talk about in a bit, but and I I stipulate this is not for everyone. I mean, there are probably a lot of people who say you have a microphone on at all times. What do the people around you think about that?

It gives me a summary of the day, event highlights, notice it's got a map. They even offer uh didn't today, but often offer links associated, uh, with the things you were talking about. It also keeps track of things that I'm talking about that I might have a to-do around, so it suggests these to-dos. This is from today, and if I think, oh yeah, actually I do want to do that, I can press plus to add it to my actual to-do list, otherwise I can delete it. I can then go to my to-do list, otherwise I can delete it. I can then go to my to-do list and you see, it's been recording. I have 34 to-do's, but these are all legit things I want to do. I haven't typed any of this. This is all from listening, transcribing, analyzing and then uh presenting, do you ever instruct me and say, don't forget oh yes, can.

We'll get to that in a second. I'll also show you. It is already 477 facts about me, the facts are often trivial.

Oh, and I get to approve them. All right, so let me review some. I'm associating with people discussing topics like MacBook Pros and Bitcoin. Yes, I've been receiving significant residual income from Megaphone and Libsyn. Yes, I don't know how I knew that I enjoy dark roast coffee with a preference for rich chocolate and bitter flavor. Okay, so this is one thing we'll talk about. That was Lisa. Now I can discard it, but it's actually not wrong. It was my wife, so sometimes it makes mistakes. That's true. I have an interest in AI topics, would you say. That's true.

0:10:25 - Speaker 2
Yeah, an interest in ai topics, would you say that's true, yeah I would say I like live events and shows in new york.

0:10:30 - Leo Laporte
That's oh. By the way, that's because I was watching saturday night live so we have an update.

Yeah, you're gonna tell me about that pet cat named sammy. True, sammy leo's cat is black fur coloring with a white nose and distinctive markings that represents a Hitler mustache. How did you know? Yes, so I can go through all these. I won't go through all 49. But it is over time and I think this is very smart built up this fairly long list of 485 things it knows about me. Trivial little facts, but certainly things that could inform an AI to give me better results. So that's just a quick tour of the interface. Let's just see what it thinks of the current conversation. I am particularly focusing on the integration of AI into the daily.

Yeah, so it doesn't yet know that I'm talking to its parents, but it'll get that probably. So a couple of questions right off the bat, and then I'll let you tell us, tell us more about this. Where is this going? Where's my recording going?

0:11:33 - Ethan Sutin
We don't store it, so that there's no audio recording. Well, where's it going? It's going from the device through Bluetooth, low energy, to your iPhone okay, the iPhone to the cloud and transcribed in real time, and then the audio is is not. That's the end of its life.

0:11:52 - Leo Laporte
So that's the privacy angle, at least so far. And by the way, that does answer a question I had about two-party recording in California. I have to ask somebody before I can record a conversation with them.

0:12:04 - Ethan Sutin
California is a two-party consent state and depends on the expectation of privacy by the other party. So in public you wouldn't. In private, yes.

0:12:17 - Leo Laporte
So we're a little bit skirting the law here. Rewind with its pin won't record unless you get an explicit. By the way, I don't have the pin yet I was going to say does that actually work?

0:12:28 - Paris Martineau
I'd be curious to see.

0:12:30 - Leo Laporte
They say they ask for explicit approval and then we'll record, and then the plod doesn't record unless you press a button, presuming the plod is presuming that you will ask before you record. This doesn't do any of that and frankly I think it's better because of it.

0:12:52 - Ethan Sutin
But it might be a little extra legal. Well, I think there's um. You know, it depends obviously on the jurisdiction, but you know also, I think, a little untested around. Yeah, so is. Is the is the summary and we target the summary to focus on the user. Um, through the speaker identification. Like I, I don't think if you went through a lot of your memories.

0:13:09 - Leo Laporte
Yeah, I don't have a transcript of things other well, I sort of do of things other people have said you don't really have any personal information.

0:13:16 - Ethan Sutin
I mean, I think this is like the key challenge we have to navigate is to kind of establish what the right norm is. I think that you would probably agree that it's gonna be the future to have like AI understanding us, and so the expectations are currently the way. They are now thinking about like, oh, somebody is recording me and listening to me, but if the expectation was, my AI is just learning about what I did purposely, not learning about anybody else, right? People will become more more attuned to that.

0:13:54 - Leo Laporte
So where is you don't reveal on the website or in your terms of service. You don't reveal the LLMs you use. Do you want to tell us now?

0:14:08 - Ethan Sutin
a llms you used. Do you want to tell us? Now there's a variety of llms we use, so, um the all, all of the pipelines you know have different, um kind of requirements. So summary doesn't require, like, necessarily the largest model, but like, for example, you know you want a very large model. So you know we use self-hosted models and then we do use commercial ones and you know, under the agreement of no retention, no training, so that's for to be very clear, no training on customer data at all from us or any commercial model we use. Okay, and so I can say, you know, from a provider.

0:14:56 - Leo Laporte
You know that you probably like. How are you going to pay for this?

0:15:00 - Maria de Lourdes Zollo
This has got to be expensive. Yeah well, we will introduce soon the subscription Good, I want you to. It will be $12 per month. It will introduce soon the subscription Good, I want you to, so it will be $12 per month. It will be a freemium model, so the basic features, as for example, memory transcripts, all of that will be free for the users. But the integrations, as, for example, you already have some of them right now for free, but Gmail, google Calendar how actually to do should be the digital side of you. That will be under the subscription model in the future and we are already testing them. We are testing agents and those as well. They will be under the subscription model.

0:15:40 - Ethan Sutin
And we focus a lot on.

You know, when we introduced the subscription to having a tier where people can kind of experience like the base, you know, of memory building, of understanding and learning, and you know you think it's very expensive, but with the right optimization, because it's like, oh, it's all time, it's really around voice activity, so like it can vary how much inference is needed depending on how much you talk, how much voice activity, and then with the premium you will and you kind of alluded to this have the more accurate models, the larger models, the models that take more GPU to run, and then you can still and we can talk about this later we have, we have a version we're testing where everything runs on your iPhone.

So this has both, you know, for people who are very concerned about control of the data, but also you know there's no inference cost to us if everything is running on the, on the, on the iPhone, um, and that won't be as accurate um, obviously, as a large model that you couldn't run on the phone, but actually for a lot of use cases it's still very useful so, um, it sounds like you got a lot of secret sauce in the AIs you're using.

0:17:09 - Leo Laporte
I don't want to press you. I mean, is the transcription done locally on the phone? Is it done in the cloud?

0:17:16 - Ethan Sutin
The transcription is currently done in the cloud. Okay.

0:17:20 - Leo Laporte
Your cloud or somebody else's cloud?

0:17:23 - Ethan Sutin
Our cloud, okay, well-. Oh, interesting, our, not our data center, um, but, but you do have your own local models?

yes, interesting um, we don't um, we, like I said, we do use some commercial models in some of the pipelines. But on the edge, on the mobile phone now there is a model running. It's very small model for um, but it is a neural network um for around detecting voice activity. We have a fully working version. So every part of it, there's five different models that are needed to do it on the phone and in that situation you could. It works on airplane mode. Oh really, the data never leaves your phone. So there's this activity, there's the speech to text, there's the voice embeddings for speaker identification, there's text embeddings for retrieval and then the llm for the summary and assistant. So all of those can work on the phone. Um, obviously they're. They're in like the three billion parameter range for the llm.

Yeah, they're too small, those models really I think they're good, good, like they're not as good, but like the summary, like we, we field test it and like it's still, you can still kind of experience the, the value of like the summaries. I think for the assistant it's a little rough. Yeah, you know, um, it's not so smart, but for things like summarization and like even generating insights and to-dos it's. It's surprisingly how good things are getting. You know, we have, we have Gemma, we have fee, we have like really good small models now, um, but like you can see the trajectory going where, yeah, where, like um, those models will be as good as some of the larger models now, but but the, you know, the best models will probably always remain on the cloud because once, once those get so good that ones on the we could run on on larger machines will be even better than they are now let me show you folks the settings uh page now.

0:19:25 - Leo Laporte
Uh on on this. If you show my screen, veneto, so I'm going to go into uh settings. Show my screen, veneto. You have my screen. Um, am I doing this?

0:19:38 - Paris Martineau
I'm seeing your screen you're seeing it. Okay, well, I'm seeing your screen on the camera thing, but it's not yeah, I'm sorry'm sorry, I'm having a little trouble here.

0:19:46 - Leo Laporte
Vinita's having some trouble switching. I just want to show you the choices you have as a user. Some of the things in here are really kind of interesting. So here's my settings. You train it on your voice. Obviously you can connect accounts. I actually created special Google Gmail contact and calendar accounts that extract from my real Gmail and my real mail so that I get rid of some of the stuff that I don't want you to worry yourself about. But you could. You could connect it if you again, I know I know not everybody would do this, but I like it. You give it location. I'm giving it everything. Location reminders, notifications, photos, contacts, all the health information from Apple Health. This is kind of fun. You have a choice of you can name your assistant and choice of some voices. Oh, you've added quite a few more. Holy moly, that's pretty new, yeah.

0:20:46 - Paris Martineau
Do you guys have a favorite voice?

0:20:49 - Maria de Lourdes Zollo
Yeah, what voice do you have? Yeah, the users love the voice.

0:20:51 - Leo Laporte
Yeah because you're going to talk to it. This is one of the things I think is the most fun is their button on the thing and you can choose what happens. For what is it? A double tap, single tap mutes it. Is that right? And the double tap picks one of these actions.

0:21:08 - Ethan Sutin
Yeah, so it's like a multi-purpose button, so the interface on the device is kind of intentionally pretty minimal.

0:21:15 - Leo Laporte
Which it should be.

0:21:15 - Ethan Sutin
That's perfect yeah, yeah, yeah. So single button is to toggle mute. Double press is like a customizable action button. Look at this.

0:21:24 - Leo Laporte
I have it on fact checker, so it's listening. You and I are talking, jeff, you say something preposterous. If I double tap this, my phone will then say what I've never used it. I've never dared, jeff, you ignorant slut. That's wrong Interesting anecdote. Is that about you or is it just a random anecdote?

0:21:48 - Ethan Sutin
um, it should be topical to what? Okay process.

0:21:52 - Leo Laporte
Current conversation, which is probably the one most people would pick, which means this conversation. I want a full summary of uh, relationship advice yeah, yeah, that's interesting.

0:22:04 - Ethan Sutin
And then roast I did try the roast and it was mean yeah, we actually, we need to remove that if somebody chooses roast, they deserve what they get yeah, I, you know this is like um, that one uses a local uncensored model. I think it's not it's pretty.

0:22:23 - Leo Laporte
It's like grok.

0:22:24 - Ethan Sutin
It's pretty but yeah, this is like um kind of an under underdeveloped area that we, we are you know we are testing around that um, but that's what I like.

0:22:33 - Leo Laporte
This is a. This is the early version, right yeah, oh, yes, yeah so so you can also, um, you do get transcripts and you can assign names. So, as I've been, as I've been talking to people, I've assigned names to them so that there are some names. But it's not. There are better models for this kind. Like claude is very good at this kind of thing, I I don't know what models you're using or whether that's even important to you, um, except that, as I mentioned, occasionally it confuses yeah, that's a.

0:23:01 - Ethan Sutin
it is very important to us, and so this is actually one of our primary focus is improving, but I understand this is an Explorer edition.

0:23:09 - Leo Laporte
This is like I only paid 50 bucks for this thing and, by the way, Maria, five bucks a month.

0:23:15 - Paris Martineau
No 12. 12? Yeah, well, that's still.

0:23:19 - Leo Laporte
I'm paying 20 for perplexity paying 20 for chat, gbt 20 for claw and everybody else is 20.

0:23:25 - Jeff Jarvis
That's a good deal complexity paying 20 for chat, gpt, 20 for claw and everybody else is 20. That's a good deal. What use cases? I can imagine many, but what are the A, most prevalent, but then B, any surprising use cases you're seeing.

0:23:36 - Maria de Lourdes Zollo
Yeah, so some of them. We are super excited about them. Students, students, in particular, students with ADHD, that they use. Bee to having the transcripts of their lectures, understanding what areas they should be focused on. Study as well. The neurodivergent use case.

0:24:00 - Ethan Sutin
Yeah, that really surprised us. A lot of people have wrote to us around, not just ADHD, but also people who've had brain injury or a stroke. A stroke, yeah. So we don't really even market at all to that but, I think there was just people looking for tools, but the really amazing thing is they wrote in they're getting benefit from it, which is not anything we had really um planned for in your discord.

0:24:29 - Leo Laporte
I've talked to several people who yeah yeah know who say this is great. There's a blind user who loves it. Um, it's really. Uh, that's a. It is absolutely an accessibility yes device and I'm an old guy.

0:24:44 - Ethan Sutin
We really want to kind of it is absolutely an accessibility device and I'm an old guy.

0:24:48 - Leo Laporte
We really want to kind of expand like, like really dig into that and uh, yeah, but ultimately the to me, and so this is a very this is like a very futuristic use of ai, but ultimately to me that's we're all disabled in various ways and as I get older, my memory fades and things. Yeah, I think this is a daily journal, if nothing else.

0:25:10 - Ethan Sutin
The reflection of like self understanding, reflective part of you know. A lot of users kind of overlap with that. I think that applies to people generally, like you know one of the other kind of um use cases that emerge for like people who are salespeople or talk a lot for their job and they want feedback. But not only do they, they've given us feedback on that, but then they're like, oh, make this insight use case more. They have feature requests for that, and so it's kind of crossing between work and reflection.

0:25:44 - Maria de Lourdes Zollo
Yes, yeah, that's, I think, is one of the things that I found. They have feature requests for that, and so it's kind of crossing between work and reflection. Yeah, that's, I think it's one of the things that I found like the most interesting is that some people are coming for professional but after they're asking parental advices or you know, like almost like the holistic coach aspect of the product, how can I be better on something? How can I have more personal insights about myself, patterns on my relationship? Um, things like that.

0:26:09 - Leo Laporte
yeah, there are people in the discord who are asking. So you can. You can talk to the assistant and say, um, you can ask it for things. Uh, there's one guy says I want all your answers to sound like it's bill burr. I don't know how, how permanent that is, but that's. He says it's very funny, um, but there's some people who say, hey, listen to this next conversation I'm having with my wife and let me know if I'm doing something that is, you know, non-productive or you know or whatever, if I'm being mean or um, and I guess that's the relationship advice button too.

0:26:44 - Ethan Sutin
Yeah, yeah, yeah.

So I mean you can, you know in the assistant, you know it will. You know, I kind of measure my life in in tokens. Now, in terms of tokens I produce, so like 250 000 tokens is like my informational output per day, um, but like that builds up over time and so so, like the assistant can actually, you know, traverse in my case you know 50 million tokens. So if I ask it to kind of analyze my, my behavior and trends, it can look historically and then you know kind of coordinate that with things that happened in my email and calendar and then kind of try and connect the dots. And I think this is this is something that's really useful, you know, and we'll only get better, like the kind of the more signals we can understand and the better we can kind of allow the assistant to to think about it, to iterate across and and pull things that when it discovers something, that then it can fetch relevant pieces and then continue on until it has something that's kind of substantial to the. That's kind of surprising to me.

0:27:50 - Leo Laporte
That's the point. Right Is the more you give, it's like rag. The more you give the AI of your life, the more insights you can ask, the more information you can get from it yeah uh, I, I think, is, I think this is now. There's a privacy concern and people want to make sure. But just to reiterate, you're not saving. What do you what? What do you have? What data do you save?

0:28:17 - Ethan Sutin
well, we don't save any audio so that that's processed in real time and and there's never any audio saved. To perform the functionality of the assistant, we have to have the conversation history to be able to then perform the AI on it, and we do a lot to make sure that, um, it's only ever decrypted at the point at the time the ai needs to to run inference.

0:28:51 - Leo Laporte
and so I mean, I think you, you probably are aware, like the apple intelligence, you know they have a hybrid yeah, and it's pretty stupid, to be honest with you, because I think they're unwilling to breach people's privacy to some degree. You gotta you you kind of know stuff to to understand stuff, yeah. So I mean, I think they have a hybrid. I know what you were. I'm sorry, I went off on a tangent. Yeah, they have a local model and they also will go to their particular servers in the cloud and then even further if. If they can't answer it, they'll refer you to chat, gpt, but then they'll warn you. You know we're about to send you somewhere where we don't control what happens yeah so is.

0:29:29 - Ethan Sutin
That's roughly the model you're using yeah, roughly, I mean we, we, we, you know to provide the service. You know, we, we use the models in the cloud, but we, we take every safeguard to make sure that the data is encrypted up until the point and that it's basically unreadable up until the point where it needs to be decrypted to AI. And we're always trying to find, you know, to get to like, for example, to get the Google Scopes, we had to go well, we voluntarily went through a third-party security audit. Oh, that's good. Yeah, we take it very seriously and like, we're committed to doing it every six months is like a best practice. And you know we come from a information security background too. You know we worked at twitter payment. We, we worked on like the first, uh, twitter blue, if you remember that wow, blessed soul, bless it tony, tony hale.

0:30:21 - Jeff Jarvis
we love Tony Hale, we love Tony Hale. What about the discoverability in subpoenaed? Where is it vulnerable?

0:30:31 - Leo Laporte
Oh yeah, Paris is a reporter. You wouldn't want to wear this to an interview with a source, would you?

0:30:38 - Paris Martineau
I mean, yeah, I think that's certainly a question. Does the data exist in some place that, hypothetically, a recording of your entire life could be subpoenaed?

0:30:49 - Ethan Sutin
So this is a great question. When you say recording, well, there's no recording, so we wouldn't be able to comply with any requests asking for audio because it doesn't exist. Right, the metadata and the summaries, yes, yes, we are vulnerable to that now, um, because you know we would fight, you know any court order but the because we have to decrypt it to um be able to run the ai on it, um, we cannot do full end-to-end encryption, um, because then, unless you're running on local model mode, so our plan around this is is kind of multi-step, so like we're going to be releasing very soon the fully local mode so you can turn it on and then anything, that's all the transcription, all the ai is, it's all happening on your device, so the data is on your device. So, um, that would not be vulnerable because we couldn't comply with anything, because there's no data that's possible to comply with, um, that you know. Again, they're not as accurate models, but they they still provide like pretty good results and they'll only get better over time as what you can run on the edge improves.

And then there's like the second option which we're working on now, which is where and not to get too technical, but like if we have the encryption key. Well, you guys understand. But like, if you have the encryption keys that you control on your phone and then the data is encrypted where you would like ephemerally grant access to run some ai, like when you're using the assistant, you know you would give us a uh decryption token that has a certain uh 12 hour, one hour life and then so we wouldn't be able to decrypt the data unless you explicitly allow it and that's time limited. And then, like, if we don't have your permission at that point and you can delete the key, revoke the key at any time, then there's no way to comply.

0:32:58 - Leo Laporte
Yeah, I think there. Look, I don't. I no longer care about privacy, so I'm a perfect guinea pig for this and I'm not worried about it. But obviously, before it became a real consumer product, people would want to know that that was the case and there's going to have to be a certain amount of trust. They're going to say oh yeah, marie and Ethan, I trust them to do the right thing with my data.

0:33:20 - Paris Martineau
How do you guys think, just generally, broadly, about the idea of us transitioning from a world where I don't know you can say something or have a conversation with someone and the only two people that know are you two, to a world where this is all increasingly recorded or searchable in some way. I mean, obviously this is a quandary bigger than your company, but I assume you've given this quite a bit of thought, so I'm curious as to your thoughts on it.

0:33:47 - Ethan Sutin
Yeah, I mean this is, I think, core to what we want to navigate and like, I mean, I think that we really believe that the value of what you can do with AI, when it has this context about you, is going to be so undeniable that that, you know, not just us, but you know it will be kind of normalized and so there will be some expectation changes. But I think that, like, core is like, if you want, like to have, like you know, an off-record conversation, like obviously that's, that's always should be um, you can press that button and it's off the record.

0:34:25 - Leo Laporte
Yeah, um you know.

0:34:26 - Ethan Sutin
But we do want people to to be responsible and and you know. But I think that, like, if you think about the future and google's promoting always a on ai and like it meta like with the ray-bans, um, you know, I think increasingly, you know, it will become kind of normalized. Like you know, maybe when Ring came out, you have a doorbell camera like that was kind of quite controversial at the time but like it just became kind of more normalized because there's a lot of value to be able to like, have some ability to, yeah, see who's at your door. But, like the, the key thing that we need to do is, like a, like you said, build trust, make sure we can do it in a privacy centered way. But also, you know, I think the value of having personal AI is just going to be more or less universal, and I mean the trade-offs there are real but they can be like maximizing the benefit and minimizing any downsides.

0:35:24 - Maria de Lourdes Zollo
Yeah, I also believe that we are in a transitional time, so right now it's a bit challenging and we are a startup, so there are many things that we need to take care about, that we should care about, but I think it's just a transition and will be the norm, and that's what makes me excited, like one day, actually, what AI can do for me, now that knows me so well about the, the actions, how actually, you know, I can be free in a way.

0:35:50 - Jeff Jarvis
That raises another question, then, of of kind of the portability of this Is. Is there a? Is there a, an ecosystem that exists, where, if someone comes up, do you envision an ecosystem where someone can create a specific application on top of what you have? I wanna create a romance advisor, or I wanna create a job advisor, or I wanna do a negotiation coach. Can you imagine that this is either portable, that I could take some kind of concatenation of what data is collected somewhere else, or that someone can plug an application in on top of what you have? Is that, is that in the future.

0:36:34 - Ethan Sutin
You think, yeah, absolutely. Well, we have a developer API currently.

0:36:38 - Leo Laporte
Yeah, when does that cause? I heard about that.

0:36:40 - Ethan Sutin
Now I'm thinking that's really cool. Developerbcomputer.

0:36:43 - Leo Laporte
Okay.

0:36:44 - Ethan Sutin
And so I mean I thinking that's really developerbcom computer okay, and so I mean I think that's definitely um a priority for us because, like both in terms of data portability yes, uh, you can, you can export data, but also like what, what you know, what kind of use cases that you know are beyond what we would build, like can be empowered by this right about like you know, the future, and like where there's different ways to exchange signals.

Maybe there will be some open standard even that will allow, like even more facilitated kind of um, uh, um exchange between ai systems where it's like you can authorize this ai system to have this you know explicit part of my context and there'd be some standard where they can just talk to each other. And I think, like one example you know this is all unreleased yet from Apple, but like they talk about the semantic engine on the iPhone where, like Siri can query apps about context. So you could see, like you know, siri could query and be like I want to know about this part of Ethan's life and then, if you would authorized it, then B can return like context from in real life that Siri wouldn't have.

0:37:54 - Jeff Jarvis
What about the agentic strategy? You mentioned at the beginning that you're starting to play with that.

0:37:59 - Ethan Sutin
Yeah, I mean, I think that that's like our frontier and like a key thing we're excited about, because, like one of the most valuable things you know that we could do is actually like proactively help and take action.

So, like you know, for example, a simple one is like if there's a to-do that that we could help with, the b could help with, like there should just be, you know, you could just say, yes, go do it as all of your preferences. It doesn't need to ask you every little thing. Um, you know you could just say, yes, go do it, yeah, as all of your preferences. It doesn't need to ask you every little thing. Um, you know, especially at the beginning, there'd be the appropriate amount of human in the loop. You know so it, you know, has your your kind of sign off it. But, like you know, for an example, like you know, a to-do is referencing. Like you know, oh, you need to follow up with this person from something that that be learned. You know we just have a button just like do you want me to draft the follow up for you and send it, and it can do that. But you know, larger, like multi, more complex, multi step kind of things that can be doing for you.

0:38:58 - Leo Laporte
Maria, can you talk a little bit about the roadmap, what I can look forward to something that I wanted to be to to mention as well.

0:39:05 - Maria de Lourdes Zollo
like speaking about agents, one of the things that I envision for the future is how agents can speak with each other, meaning like how, actually, personally, I can speak with my personal AI and how they can take actions on our behalf. We did some experiments where Ethan's agent spoke with my agents and they chose for us a nice French restaurant nearby, based on our locations, and actually it was quite good. That's wild.

Basically like the both agents that you know. They knew our preferences in terms of food, they knew our location, they knew when we were free, based on our calendar, and they figured that out for us. And that's where I want to go understanding how actually you know the user should not do things, but it's more like your personal ai that does those actions for you, and I feel that will make us even more connected as humans I wish I'd had this my whole life.

0:40:00 - Leo Laporte
It would be really fun to go back and look at different days in my life. Does this it saves? It saves it all forever. Does it roll off at some point or no?

0:40:09 - Ethan Sutin
no, so it's, it's. You know, everything is there unless you call there, unless you explicitly remove a memory so this is an incredible journal yeah, indeed, like the user.

0:40:21 - Maria de Lourdes Zollo
The other thing that amazed me is the fact that some users were sad or upset if they don't bring the bee with them, because after a while they get used to it. Yeah, you get up you know, like they're like, oh my god, like I love my day, or you know, I won't like to understand what happened, but now I cannot, because there is a point of self-reflections that b helps you to do so.

0:40:42 - Ethan Sutin
Yeah and the journal part, like one of the fun things that like at the uh around christmas when the spotify rap came out, like I just asked b like make a life wrapped, you know, in the spotify style for my life and it was like oh, you visited three, you know, had stats like three countries here were your key achievements.

0:40:59 - Leo Laporte
Yes, I feel like, in a way, I'm making an investment in the future of ai, like I'm now storing as much as I can, knowing that sometime this will be a treasure trove. Maybe it's a little trivial right now it's actually already quite useful but I can only imagine what it will become down the road.

0:41:17 - Maria de Lourdes Zollo
Yeah, and future, future. There is one of the things that sometimes we wonder around the vision of the product. So right now it's just audio, but for future there might be a vision, a camera component in the product as well a camera of some kind.

0:41:34 - Leo Laporte
Yeah, oh man, I really want you guys to uh succeed. I think this is thank you fantastic he's your best agent.

0:41:43 - Jeff Jarvis
I am, I is enthusiastic.

0:41:45 - Leo Laporte
I love it. Yeah, I talk about it all the time. Right now in the app, I can request an agent upgrade. Is that just a nonsense setting?

0:41:54 - Ethan Sutin
No, no, we can enable it for your account. We want to be very careful with the agents because we don't want any harm at all, you know so assume that I'm the person who you can harm all you want.

0:42:13 - Leo Laporte
Okay, and I, I am. I am well. No, I'm just all in on the potential for this and I've got nothing to lose at this point wait, I'm confused with the agent upgrade.

0:42:22 - Paris Martineau
What does that?

0:42:23 - Leo Laporte
Yeah, what do I get? Well, yeah, we, we the double O number.

0:42:29 - Ethan Sutin
The we, we have all right. So we have agents that run in multiple execution environments. So just to give a brief kind of outline of how it works, so you know the, the LM can do kind of you know tool calling. So like for things you already have, there's kind of already a little bit of agent. You can ask the assistant to send an email to like look up some information. It can send an email. So that's kind of one way where we have API access to take actions. We will use that. But, as you know, like we can't integrate with everything, a lot of things don't have public APIs. So we have two other agent execution environments the one that you saw in the request screen.

So you know, basically you get a lot of stuff done, probably on your phone. Most people's digital life it revolves around the phone. So, like the concept is, if you, if you get a lot done on your phone and we're trying to give everybody a personal AI what if we gave their personal AI, their own phone right that it could do things for you and had all the same apps that you do? And so we actually have an emulated cloud phone for users who are enrolled in the agent and it can use its cloud phone to perform actions for you and, um, that gives it, like through computer use, the ability to use any app that exists and it can do, you know, multiple steps. Uh, you know, we can try and give you a demo. We don't have the phone, unfortunately. I think we can't share the phone screen, but I'd love to give you a demo or follow up. We can enable it on your account.

0:44:07 - Leo Laporte
Just turn it on my account and I'll end up demoing it on the next show, probably, most likely by accident. So if you go now to Bcomputer, you've sold out how many? Well, you probably don't want to talk about numbers, but how many?

0:44:23 - Ethan Sutin
We have the next um shipment arriving um in on um soon yeah, 24th, it's the 24th it's it's shipping tomorrow yeah, and we'll arrive to to then be shipping to customers on the 24th so, uh, and again, I don't think it's for everybody.

0:44:43 - Leo Laporte
But so, uh, and again, I don't think it's for everybody, but it's only 50 if you want to play with it, um and currently there is go ahead, maria.

0:44:51 - Maria de Lourdes Zollo
Yeah, one of the things that we are seeing is that, um, some of our customers they're buying them, but they're buying beef, or you know, after they try it for family or friends.

0:45:00 - Leo Laporte
Yeah, I need it. I'm gonna need a new one pretty soon. I can tell I'm delaminating it already because I never take it off for Christ's sakes, but it's.

But the data is on my account, not this device. So, yeah, yeah, and when you come up with one that I could put in my ear or on my glasses that has a camera, I'll be getting that one too. I think it's. I think the idea is brilliant and the execution is surprisingly good, considering how early days we are with all of this. And now that I've talked to you, I think one of the things that wasn't apparent to me and I read through all the material I could on this is the complexity and the depth that you guys have put into the software. The back end of this is actually kind of interesting. It's very impressive. A lot of thought has gone into this. Um, I wish you the best thank you so much.

0:45:49 - Ethan Sutin
Thank you for the good work. Talk to us.

0:45:52 - Leo Laporte
Maria delordes Zoyo, it's great to have you, ethan Sutton, yes, to have you, founders of B dot computer at B dot computer on the web. Um, yeah, I'm. Can you turn on my? Uh, give me all the things yeah, I. Want it. I want it to be smart.

0:46:11 - Paris Martineau
I don't care if it's dangerous yeah, yeah, I like the roast this interview so that he could get an upgraded feature.

0:46:17 - Leo Laporte
That was the whole point, frankly. Yeah, it's just to get upgraded. No, I is. I was very impressed by this and I really have thought for a long time that the future of AI is personal AI One of the many, many futures, but personal AI has such an interesting implication in our lives. Yeah, thank you Maria, thank you Ethan, really appreciate it.

0:46:35 - Maria de Lourdes Zollo
Bye, take care.

0:46:38 - Leo Laporte
You're watching Intelligent Machines Getting more intelligent all the time. Our show today brought to you. We'll get to the ai news in just a bit, uh and uh, we've got some fun stuff for the back of the book. Uh, as well, our episode brought to you by zscaler this week, the leader in cloud security. Uh, as you probably know if you've been following this and certainly if you listen to security now, it's obvious that enterprises have been spending billions of dollars on perimeter defenses, firewalls. Then, of course, you have to get VPNs so that employees can use your network and your apps. Has this made everything safe? No, breaches continue to rise. In fact, ransomware is up last year, 18% year over year, $75 million record payout in 2024.

Clearly, what we're doing isn't working. These traditional security tools expand your attack surface with public-facing IP addresses that are exploited by bad actors. And, of course, the bad actors are getting worse because they're using AI now to craft their attacks, uh, which means they can operate faster and faster. And you know, the assumption on this is the really it's almost a bad mindset because you think I've got the perimeter defenses. So the assumption is anybody inside our network works for us or is supposed to be there, right, but we know now that's not the case. The breaches are going on like all the time. What do you do? Well, if you don't have, you know, some sort of protection, you're allowing them lateral movement anywhere in your network. They can find stuff before they even trigger the ransomware. They they look through your emails, they look for incriminating emails, they exfiltrate customer information, they blackmail you. With that. We're going to reveal everything. All because they have full access once they're inside. And, of course, part of the problem with this perimeter defense is the bad guys. When they start exfiltrating, they encrypt their outbound traffic and these VPNs and firewalls cannot really analyze this encrypted traffic at scale. It just allows all that data out, this encrypted traffic scale. It just allows all that data out.

Hackers, bottom line, are exploiting traditional security infrastructure using AI to outpace your defenses. It's time to rethink your security. You can't let these bad actors win. That's why you need Zscaler. Zero trust plus AI those are two concepts that you really need. Zero trust hides your attack surface. Your apps, your ips are invisible. There's no lateral movement. Users are only allowed to connect to specific apps. There's no assumption that well, you're in the network, you're good. No, you have to have permission to do anything and and it continuously verifies every request based on identity and context. Zscaler simplifies security management using AI powered automation, and it uses AI to analyze over half a trillion daily transactions. Looking most of those are fine, but looking for those suspect transactions that are going to become threats over time, it's really remarkable. Hackers can't attack what they can't see. Protect your organization with Zscaler Zero Trust Place AI.

You can learn more at zscalercom slash security. That's Z-S-C-A-L-E-R. Z-s-c-a-l-e-r. Dot com slash security. We thank them so much for supporting the show. We thank you for supporting the show by using that address zscalercom security. So there's an api. They seem like they have pretty good data security, at least intentions. Do you think you'd buy one of these things?

0:50:18 - Jeff Jarvis
paris, paris, you were the most dubious. I would say oh, you're muted.

0:50:26 - Paris Martineau
Even b can't hear you I took I, I muted it so that I could eat a biscuit, and then I forgot um, now, is that a biscuit biscuit or is that a cookie calling itself a biscuit?

0:50:35 - Jeff Jarvis
it's a biscuit biscuit, but last week you said spanner, you're trying to act, all oh, is she getting anglo on?

0:50:41 - Paris Martineau
okay, she is no, no, get off the lift Jen. I was reading something written by an Australian, and he said Spanner. To be clear, my answer to the bee question one brief aside though this is a biscuit, but they call it a scone at the bakery I get, so it's all kind of all a problem. My bee answer is it's not personally a technology or a tool for me, but I'm happy, you're happy.

0:51:07 - Leo Laporte
You seem very happy about it. What is it about it that turns you off?

0:51:12 - Paris Martineau
I'm not certain that there's a. I don't see a use case for it.

And I'm not certain I was thinking during that conversation I was like, all right, how would I use this if I wanted to try it out? During that conversation I was like, all right, how would I use this if I wanted to try it out? I'm like I, because I think the way to get some sort of use out of it, like you're describing, like they were describing, would lie in connecting it to everything, as you demonstrated connecting it to your reminders or calendar invite, having it in your apps that it can scoop up all that data and make intelligent-ish inferences from it or at least records of it.

But given my line of work, I'm just not certain that that's tenable in any way Like I use, I would have to somehow contain it to personal devices versus work, and in a way that I'm just not certain that it would. But even outside of that, the only compelling use case that I I mean, I think that it reminds me of that Black Mirror episode that I'm going to poorly describe because I haven't seen it.

Because I haven't seen it, but I've heard someone describe it. So it reminds me of that Black Mirror episode where it's something like there's a technology that can record all of your memories or every fabric of waking life, and it utterly destroys all of the characters, lives and sanity. Because I think that there is and maybe this is an outdated uh notion, but I think that there is something beautiful and messy about humanity and human existence to where I don't have a record of every embarrassing thing I've ever done, nor do all of my friends or enemies, and I'm just not sure that humanity is built to have that suddenly, even though that seems to be where we're going yeah, that's an interesting.

0:52:59 - Leo Laporte
Yeah, that's an interesting point that it's probably good to forget a lot of this or have an imperfect memory how about you, jeff um, it's a gadget.

0:53:09 - Jeff Jarvis
I'm like you I have the same disease you have leo. It's a gadget. I want to buy it. Right, it's a connected gadget that can do things that I can't even see. I'm fascinated, um, as I think it through. I mean, if I were teaching full-time, I could really see the use of it for student or teacher value you know, anthony niel is saying I don't talk much because he works at home.

0:53:29 - Leo Laporte
If you didn't have any audio, it wouldn't be that useful, I agree.

0:53:32 - Jeff Jarvis
Right, that's the thing. I'm sitting here writing my book.

0:53:36 - Leo Laporte
It's somewhat less useful for me for that reason, because the only person I'm ever having any conversations with is my wife. I don't go outside, but for so? Uh? But for a teacher, yeah.

0:53:44 - Jeff Jarvis
a student, yeah but some teachers get quite freaked at the idea of students recording. Really, yeah, why no?

0:53:53 - Paris Martineau
this is actually a huge problem right now it has been for years with the and I'm not going to go into details with the turbulent political climate people are experiencing all over the world, you you could imagine a situation where and I've heard anecdotally from professors over the past 10 years being like I'm worried that I'm going to make some offhanded comment in one of my classes, a student is going to record that, leak it to some media outlet and then I'll lose my job. I just think there's a lot of strange consequences that are opened up where suddenly everyone is recording everyone all the time, and that's not even considering the national security implications, like yeah, I recognize I'm in a somewhat privileged position where I'm not at risk, I'm not employed by anybody.

0:54:38 - Leo Laporte
I'm a white man, I'm older white man. I'm not I'm not really, you know, in a high-risk environment. I'm in a privileged environment.

0:54:46 - Jeff Jarvis
The other interesting thing about it is so professors some of them, not all of them so I'm very open and I put my stuff up online and I'm fine with that. But some have our copyright hawks of their own sort and they don't want their stuff recorded for that reason. And then there's an interesting thing about being caught in. So I had a minister once back when I was a decent human being who actually went to a church and he never let us put I was on the board of the church, believe it or not and he never let us put his sermons online. And he said he thought someone would steal them. And I thought, well, frankly, sir, they're not very much worth stealing. But then, as I thought about it, he I realized I'm pretty sure he was projecting that he was borrowing sermons.

0:55:35 - Leo Laporte
Yeah, and he didn't want to discover. If you think about it, writing a new sermon for some is quite a challenge. There are ai tools for writing sermons now. There are now. Father robert told me about this. Yeah, so I understand, and you don't want to write a sermon and have somebody else use it, I guess?

0:55:50 - Jeff Jarvis
And I can see. You know, if I were, let's say, an executive and I'm dealing in legal matters, personnel matters, that kind of stuff, or a dean, and I'm dealing with similar kinds of things, I can imagine my organization say don't you dare, and so it'll be interesting, no-transcript. Would this be useful for her?

0:56:14 - Leo Laporte
Yeah, she's Alzheimer's. Yeah, problem is right now you have to be fairly technical to use it, so I don't.

0:56:21 - Jeff Jarvis
But if the place were using it?

0:56:23 - Leo Laporte
Yeah, maybe Right. I mean, certainly that's an application and as I get older, my memory deteriorates. I think there's a lot to be said for that. I think that it's interesting. I think this is a larger conversation which, uh, society isn't crazy about these tools. The people who are, I think, may have an advantage. Do you think, in the in the long run, over people who, uh, in what way? Um more insight, more information.

0:56:53 - Paris Martineau
How does your phone knowing that your cat has a hitler mustache more information. How does your phone knowing that your cat has a hitler mustache provide you an?

0:56:59 - Leo Laporte
advantage. Well, if at any point uh there is a surgical uh solution for that, uh they may let me know.

0:57:07 - Paris Martineau
I don't know, no you're right, a lot, a lot of, has a half a hitler mustache and I'm gonna leave it on her face.

0:57:13 - Leo Laporte
The worst, the worst thing about this is you realize 99 of your conversations in life are completely trivial and are absolutely well, I can't so I was thinking about this, but my wife is very private.

0:57:24 - Jeff Jarvis
Uh, I could think her thinking are you knocking futz? No absolute way, though. I can imagine this use case, leo, in which she says give me that thing, I'm going to play back.

0:57:33 - Paris Martineau
I did, too, tell you about this, right, right I was going to say my immediate thoughts is this would be useful in arguments, and I'm not certain that it would actually be useful, would it?

0:57:43 - Leo Laporte
be good. It would be useful, but it'd be good is another entirely, I think you know, keeping track of agreement.

One of the things I think is useful is keeping track of the things I've said I'm going to do or I've agreed to do or want to do as a to-do list. That's pretty valuable. I've tried to do that in my own life and it's hard. I, you know, I remember that, you, you know to do that. The fact that it picks these up, but but the disadvantage is it can't read my mind, it doesn't know unless I say it out loud.

0:58:10 - Paris Martineau
So you know, hey, neural link is uh coming just around the corner for you.

0:58:16 - Leo Laporte
Just around the corner, scarlett Johansson urges the government to limit AI after a fake video of her opposing Kanye goes viral.

0:58:27 - Jeff Jarvis
I never saw the video itself. It was a bunch of celebrities, right.

0:58:31 - Paris Martineau
It's a bipartisan issue that enormously affects the immediate future of technology is very funny to say in response to an AI video of yourself going viral.

0:58:42 - Leo Laporte
Yeah, it's a little. I mean, look, I'm a big fan of Scarlett Johansson.

0:58:48 - Jeff Jarvis
She's the most powerful woman in Hollywood. She was being used by OpenAI, let's not forget. I think she's a bit bruised in that her voice was being used there.

0:58:53 - Leo Laporte
But this is an example of kind of knee-jerk, uh anti-ai movement, of which there will be a lot, I'm sure. Um, I don't think the government should limit ai because there was a fake viral video of her. No, that seems like a lot.

0:59:07 - Jeff Jarvis
It's like a big ask well, and I think there's a first amendment issue there too. Uh, yeah if I want to, if I If I wanted to hire actors to do exactly what that video was, I have a right to do that. Just because I use the machine to do it, I can't.

0:59:25 - Paris Martineau
Use the machine trained on her voice.

0:59:27 - Leo Laporte
Yeah, she has a right of publicity, publicity that you not use her image, and she could pursue it along those lines. That's fine. For her then to say no, government's got to ban. Ai is another. No, no, no, no, that's a bridge too far. Um, yeah, all right, just thought I'd bring that up.

0:59:47 - Paris Martineau
So, jeff, would you get a, b, and what would you use it for?

0:59:53 - Jeff Jarvis
you see, I had for a brief time the disease that Leo has, which was Kickstarter disease.

1:00:02 - Leo Laporte
I had to stop that.

1:00:03 - Jeff Jarvis
Yeah, he had a bad case. I had a light case, so I would buy stuff because I'm fascinated with it. You know, I have a Newton back there in my closet.

1:00:14 - Leo Laporte
What? Oh, I had several, but I got rid of them when I closed the studio.

1:00:18 - Paris Martineau
I got rid of them Of the Isaac variety? No, not.

1:00:21 - Jeff Jarvis
Isaac Newton.

1:00:22 - Leo Laporte
Oh, she doesn't know what a.

1:00:23 - Jeff Jarvis
Newton is oh my God, oh, she's so young. What's a?

1:00:26 - Paris Martineau
Newton guys. Before the Palm Pilot.

1:00:28 - Leo Laporte
Apple released a handheld device that you could write on.

1:00:33 - Jeff Jarvis
It was a very certain kind of writing you had to do.

1:00:36 - Leo Laporte
It was a flop. You don't know about the Newton. That's interesting.

1:00:42 - Jeff Jarvis
Hold on. We got a showroom for Doonesbury.

1:00:44 - Paris Martineau
Newton, you do have to show me Doonesbury Newton. Doonesbury Newton.

1:00:48 - Leo Laporte
So Doonesbury did a famous comic commenting on the very poor handwriting recognition of the Newton here's one.

1:00:57 - Jeff Jarvis
I'll put it up in the in the chat in the discord, the discord.

1:01:02 - Leo Laporte
Okay, let me go to the discord so that. But then this is from the. This looks like simpsons. So so bart is writing beat up martin and it says eat up, eat up martha. But you know, this was a long time ago, it wasn't. It was before the internet, it was before cell phones. So there were. This is not unusual that you have inventions that that eventually will become a big deal, but that, yeah, technology is way too early it was just too early, that's all right, so so so I would.

1:01:35 - Jeff Jarvis
I would love to have gadgets. What was your first pda?

1:01:38 - Leo Laporte
first paris, I gotta know, not public display of affection. What was your first? Handheld the answer was it the iphone?

1:01:49 - Paris Martineau
smartphone of some sort. Yeah, I can't recall it. I it might have been an and no, it might have been an early ip. I don't know. I'd have to do some research but it was a phone probably it was a phone with all screen, no keyboard, but I remember my mother having a Zune, my mom had a Zune.

1:02:15 - Speaker 3
Jeff, we're so old that wasn't a PDA Zune, was just a music player, was a yeah or no?

1:02:22 - Paris Martineau
what am I thinking? There was like a little switchy guy.

1:02:25 - Leo Laporte
I'll think it was probably a palm pilot or a cleo, a sidekick, a sidekick, you think she had a sidekick? That's pretty fancy. Oh, sidekick, mom had a zune she did have a a zoom

1:02:38 - Jeff Jarvis
leo, we were at south by southwest uh at the uh the dig event and microsoft was. It was a sponsor and they tried to give away a zoom and nobody wanted. Nobody would take it.

1:02:47 - Paris Martineau
What I'm thinking of is the verizon chocolate phone.

1:02:51 - Leo Laporte
She also had one of those and I don't know why that lies in my head prior to all of this, there was a whole bunch of different. They called them personal digital assistants. The newton started it, but they were all flaw. I mean, the palm pilot was pretty successful, um, but because they didn't have a phone in them. It wasn't until you could get the internet on them and the phone in them that they really became useful. And now this is our primary computing device, isn't it? Yeah, like this is, oh yeah, a lot of people don't even have a computer anymore. This is, they have a phone that's psychotic, but I understand well, you need one because you're a writer.

But if you didn't have to write?

1:03:29 - Paris Martineau
but it feels right. Sometimes you need to do like. Sometimes I feel like if I'm making a large purchase, it needs to be on a computer rather than a phone. It feels too trivial to do it on the small screen.

1:03:40 - Leo Laporte
When you bought your tv, did you do that on?

1:03:42 - Paris Martineau
the. I did yes because I you know I had you like that, by the way.

1:03:46 - Leo Laporte
Are you happy with it?

1:03:47 - Paris Martineau
oh, we talked about a little bit last week. It's fantastic, it's so. It's like jaw. You were right that I should have gotten the 65 inch.

1:03:56 - Leo Laporte
I did that on your recommendation and it is like yeah, you don't have to go to the movies ever again I don't, it's so big it's an lg oled oh no, she followed your instructions I got.

1:04:07 - Paris Martineau
I got a samsung s90 qd oled um. You got the best screen dots it's not going back from that, it's really. It's also so thin. It is like I was shocked when I took it out of the box. I had to ask my upstairs neighbor to help me lift it up there because I foolishly thought I could put it on my thing myself. No, not allowed to do that, uh, but the thing that I've really liked the most is I got a Blu-ray player as well. Oh good.

1:04:38 - Leo Laporte
And it's been delightful. What's the first movie that you watched?

1:04:42 - Paris Martineau
Wild at Heart, which was the movie that started us all.

1:04:44 - Leo Laporte
That's what you wanted. Yes.

1:04:45 - Paris Martineau
But now I've been watching all of Twin Peaks on Blu-ray and I really love it because I started watching Twin Peaks while I was waiting for the disc to arrive on whatever streaming service is available. That was arrive on whatever streaming service is available. That was fine. But watching it on blu-ray is totally different because they have the log lady intros which I guess aired when, uh, twin peaks was airing.

1:05:04 - Leo Laporte
But now I get to watch those and they're great this is a woman that carried around, as I remember, a log yes, but she at the very beginning of she's just kind of a side character in twin peaks.

1:05:15 - Paris Martineau
But something I didn't realize is when they were airing them on TV originally. They have these little intros that play before the show starts, where the log lady is sitting at a table holding her log and does a direct-to-camera monologue kind of about nothing but also about strange things that kind of tangentially relate to the episode. And it's beautiful.

1:05:37 - Leo Laporte
Would really recommend the log lady intros and in 4K 4k, there's nothing like it there's nothing like it well, I'm glad you like it. I think you did the right thing you gotta right.

1:05:44 - Paris Martineau
Before I move on, I have one question for you guys, which is I have a strange compulsion which has been brought up by this conversation. We just had to get old technology and frame it or like put it on my wall in some way, maybe in a shadow box, or just hold. I thought about this with the you know corded phones before. If I was to do this the uh, pda, what, what, what you know early 2000s phone, should I do what looks interesting, or is no well.

1:06:11 - Jeff Jarvis
Trio a trio is wonderful thing. Let me see if I get mine oh, he's gonna go get his trio.

1:06:17 - Leo Laporte
Ladies and gentlemen, see, this is where getting rid of the studio was such a wonderful thing for me because I really got rid of a lot of this stuff I have. I kept very little of it you're gonna come out oh is that beautiful?

1:06:31 - Paris Martineau
see, that's really beautiful.

1:06:33 - Leo Laporte
Actually, it really want to have it dismantled and framed, or you want to have it just framed as is Framed as is or shadow box. Yeah, I think you want that Because you know that's one thing people do now is they disassemble these things and have it all arranged nicely in a shadow box. I have a box downstairs.

1:06:49 - Jeff Jarvis
Next week I can bring it upstairs. I have a Newton. There was another device that you clipped on your belt. It was this really dorky thing and it dealt with a vertical blanking interval and got a constant feed of information. It would pick up something so you'd get. It was like a beeper, but just for your stock prices or news headlines. It was a way you were going to get news.

1:07:12 - Paris Martineau
And do you have that in a box?

1:07:14 - Jeff Jarvis
I have that downstairs, yeah.

1:07:16 - Paris Martineau
You got to bring that next week. I want to see that.

1:07:20 - Jeff Jarvis
I have an Intel Now that Intel's about to get broken up and die. I had an Intel, worked on an early, early tablet Just about the time Bluetooth came out. It wasn't Bluetooth, it was an Intel proprietary. It never came to market. I had the prototype because I worked on it and the idea was you would sit on your couch and it would replicate your laptop. There were no laptops. It would replicate your desktop screen and your den behind you so that way you could surf on your couch. I have a news pyramid. Did you ever have that one, leo?

1:07:55 - Leo Laporte
No, so it also worked on the vertical blanking. The vertical, oh it lit up a little bit.

1:07:59 - Jeff Jarvis
And no, this is different. It was vertical blanking interval and it would. So that was a constant stream of information and and you programmed it to just pick up what you wanted it's a news pyramid, but it's not the pyramid structure used in news.

1:08:14 - Leo Laporte
It's a different no it's used exactly it's a physical pyramid that has an antenna and it turns out the public broadcasting stations all over the country. Uh, would they had in the? In the broadcasting technology there was a little extra space that you could send data, but very, very slowly out over their antenna. So there were devices made that would pick up this data and display it, and that's what he had and it's called a new talk about a gadget news pyramid.

1:08:46 - Jeff Jarvis
Yeah, I got that downstairs. Uh, there's also.

1:08:48 - Paris Martineau
Have it google glass, oh yeah you can't even find this on google.

1:08:51 - Jeff Jarvis
I just did a couple you want me to run downstairs now and get the box?

1:08:55 - Paris Martineau
I mean, you don't have to.

1:08:56 - Leo Laporte
No, no, no, let me show you, though this is what we were talking about the disassembled, um, you know, kind of blown up. This is an iphone 6. These are on etsy, so, um, it's kind of cool. A lot of people do these. A lot of our people in our audience do make their own. They'll, they'll, disassemble something and frame it. Yeah, I think it's kind of an interesting way to show old technology. I don't? You might as well do that with your news pyramid, because public broadcasting stopped sending data in the vertical blank a long time ago, did I?

1:09:26 - Jeff Jarvis
ever tell you about my use of public broadcasting with the india journal no history.

So the first task I did when I got to advanced publications in jersey city is, as the as stevehouse, my boss said we were the shadow of the Statue of Liberty, and all these waves of immigrants come in and the time Italians and English, I mean Irish and German, and so when we were there working there was Indian and Pakistani, and so we wanted to serve them. And one, one story from all of India and the newspaper every week wasn't gonna do much. So we started the India Journal. This is 1990. This is pre-web, just pre-web 1994.

And so we found a whole bunch of old computers and put them out in the neighborhoods and people there would send in reports and would come into a bulletin board we had. And then every day in what we called the black hole of Calcutta, which was below the press room in the building Sri Ram Krishnaswami, would turn out a two-page news bulletin about both local and diasporal news from India and Pakistan, in English, yes, and it went out over the vertical blanking interval into the NBC signal, down into a TV antenna over the Shivam Deli on Newark Avenue, down into a kitchen microwave cart that I put together, and in there was a laptop and a big printer with a big red button. We worked with MIT Media Lab and they assisted had to have a big red button and you hit it and out would come the day's India Journal in the midst of the Indian neighborhood. It was killed. You want to guess why? You want to guess who killed this project? Who, the Teamsters? Oh, because it was seen as an alternate delivery of the newspaper.

Oh man, the vertical blanking interval was a handy way to transmit a constant flow of information and was used. It was there, it was that. Was that not the basis of teletext?

1:11:23 - Leo Laporte
and video text. It was. That was teletext. Yeah, yeah, yeah, teletext over VBI. But it wasn't even 300 bought. It was pretty slow, yeah, yeah, so, oh, this is such a rat hole.

1:11:36 - Paris Martineau
What a beautiful world. Oh, this is such a rat hole. What a beautiful world.

1:11:40 - Leo Laporte
So, weirdly, one of my first computer programs back in the 90s was a demon dialer. I had a bulletin board system, and so you had to dial into it, right? I only had two lines A demon dialer.

Well, it was very hard to get in. There were only two phone lines right. So I wrote a mac program that would run on the vertical blank interval, because there is a moment in time on all crts when the the cathode ray tube right, there's a gun that's firing what is I don't even remember now some sort of electrons at the cathode ray tube to light it up. It goes all the way up and then there's a moment in time when it repositions that it's not doing anything. So the computer is kind of waiting for the gun to go back down to start drawing the next screen. So that gives you a very small millisecond time, length of time that the computer's just sitting there and I.

What I wrote was one of the first multi. I think it might be the first multitasking program for the macintosh. Uh, during that vertical blank interval it would stuff dial commands into your modem. It took quite a few of them, but by the end of it it would fill it up and then you could dial the modem and if it got a busy, signal would start over, but if it didn't, it would beep a lot a loud I think you might have advanced past.

1:12:58 - Jeff Jarvis
Uh, we didn't go back far enough for paris. Here the old billet bulletin boards. Paris, you weren't bulletin boards, I got I've got you know bbss.

1:13:06 - Paris Martineau
Okay, I know bulletin boards. You know everything else you just mentioned I was trying to hold in my head. I wish that there was a book that I could read, that I would understand all of them. Let me share this fascinating to me yamming crt yolks.

1:13:27 - Leo Laporte
Do you know that in the old days they didn't have flat, thin screens like your tv.

They had a cathode ray tube I had and they this is the little thing that's firing at the, the gun that's firing to all to uh align these is called yamming them and it's very difficult. So it turns out. We've only recently learned how they did automated deflection yoke yamming. Here's a video from Thompson. This is a video from Thompson Electronics on how we aligned the yokes in a mass way in the factory. The article starts when I worked in a TV repair shop. The first thing they told you is never touch the yoke, for two reasons. One, it could electrocute you. But two, once it's out of alignment, it's almost impossible to get back in alignment. So here's Bill from Thompson Consumer Electronics. He's talking about how they came up with an automated way to do the yamming. I don't know when this is recorded, but look at that screensaver.

1:14:22 - Paris Martineau
Look at that screensaver so good. They managed to not have a refresh line on that screen.

1:14:31 - Leo Laporte
Oh yeah, well, you have to have a special camera that you could adjust the shutter speed. So just as we've done here, so that you don't get uh, because you know you notice I you don't have this refresh problem with uh lcds, so my lcds don't, but this mac is a crt. In fact was it you, benito, who got this working no, it was russell or what is that fun?

1:14:53 - Paris Martineau
is it a screensaver on that? Or yeah, it's flying. That was the big.

1:14:57 - Jeff Jarvis
That was the funny. Everybody had the same sense of humor. At the same time, we had a universal sense of humor and posters flying was funny to everyone.

1:15:05 - Leo Laporte
Everyone, we always. Anyway, this Mac, which is Jammer B gave me before he left, is the original Mac. It has a cathode ray tube and cathode ray tubes, because of the vbi, will have a line going across them because they, unless you can get the frame rate of the camera to be exactly the frame rate of the tube and that's what russell or somebody did maybe it was burke, I think burke did. It was burke or russell he figured out how to get this frame rate of my very fancy sony fx30 camera to adapt so that we didn't have we used to have a line going up and down on that. So he fixed it, which is pretty cool, okay, so let's go back to ai I was just about to put up a picture from my memory lane, but go ahead.

Oh, go ahead, go ahead. No, do it, do it, do it.

1:15:48 - Paris Martineau
I don't want to stop you, I'm just I'm. I'm writing down all these terms that I can uh educate myself later, guys.

1:15:55 - Leo Laporte
I had never heard of CRT yamming before. No, I haven't either. That was brand new.

1:16:00 - Paris Martineau
I was going to say okay, good, because I still don't fully understand it.

1:16:04 - Leo Laporte
You want to jump into the future? Microsoft has announced a new quantum chip, the Majorana One, chip that they say is going to transform quantum computing. Now we've been talking a lot about quantum computing. Wait, is this the thing that they say is going to transform quantum computing? Now, we've been talking a lot about quantum computing.

1:16:19 - Paris Martineau
Wait, is this the thing that they're saying? They've created a new form of matter.

1:16:25 - Leo Laporte
Yeah well, it's not a new form, because it's been around since matter began Well yeah, that was going to be my point.

1:16:32 - Paris Martineau
You can't really create a new one. Well, you could, but it's not that kind of new frame of matter.

1:16:35 - Leo Laporte
It's a very subatomic particle. Really create a new one? Well, you could, but it's not that kind of new frame of matter. It's a. It's a. It's a very subatomic particle that was discovered in 1937 by ettore majorania, an italian, who then disappeared mysteriously, by the way, at the age of 31. Um, anyway, the uh. This is a breakthrough in quantum computing. Is it gonna? Is it gonna make suddenly quantum computing happening? No, no, but it is a way of getting more uh qubits onto a single chip and it's it's not electrical.

1:17:07 - Jeff Jarvis
Can you explain that to me? I heard something about that it's not using electrons.

1:17:11 - Leo Laporte
It's using electrons, okay, it's using this majorana particle. Uh, that was hypothesized about in 1937, but, uh, I think cern probably discovered. Uh, maybe not cern. Uh, anyway, some, some super particle accelerator discovered it. Uh, and this is, microsoft has created what they call the world's first topo conductor, a new type of material that can not only observe but can control majorana particles to create more reliable qubits.

1:17:43 - Jeff Jarvis
I never thought in my life I'd say that sentence well, leo, if we want to go even further into the future line 117 oh boy story I put up just before we went on yes a uh for the first time. According to wired, a team at university of oxford has succeeded in getting two quantum processors to connect to each other and transmit the same information using particle entanglement. Again a sentence I never, yes, so they weren't actually connected explain that for someone.

1:18:15 - Paris Martineau
This is the weird.

1:18:16 - Leo Laporte
This is the weird thing about quantum so you know, traditionally, in order to get from point a to point b, you have to cover every point in between yes right. Well, it turns out quantum particles somehow manage to get from point a to point b without doing that no, it's not, it's not a role, it's not movement, it's too. They're not moving.

1:18:37 - Speaker 3
No, it's just communication are connected to each other they're connected without any connection right so they're just they're entangled faster than the speed of light in theory right right, right, right, well, right.

1:18:49 - Jeff Jarvis
One will change a state and the other one will change too, but the reason I say travel is because one thinks of data traveling between those two but it was right to correct you, because that's the old way to think yeah, that's the past.

1:19:01 - Leo Laporte
Yeah, uh, in the experiment, light particles remained in the same place, but because they were entangled, the computers could see the uh information and work in parallel, even though they were not next to each other. Well, they're next to each other, but not in the same spot, and there's no entangled qubits. Is that what it is? Yeah, it sounds like it pretty much.

1:19:24 - Jeff Jarvis
It's entangled. Yeah, it's quantum entanglement, why a pair of linked particles, even at a distance, can share the same state and therefore transmit the same information. If one changes, the other instantly reflects it. Instantly, in our view, because it's faster than light reflects it.

1:19:42 - Leo Laporte
And that's what's interesting, because you're implying that there's some sort of communication. When we're communicating, we are traveling as electrons down a pipe, and so there's traveling involved and that takes time. It's the speed of light, but it still takes time. This takes no time because it's simultaneous, and it's a puzzlement, because how does that how the greatest minds have been?

1:20:08 - Speaker 3
trying to figure out how to use this to make an ansible yeah, basically an ansible is a sci-fi concept of instant trends instant traffic faster than light communication yeah, um, yeah, that's different this is not to say it would be communication, not travel not travel.

1:20:26 - Leo Laporte
I'm sorry, I keep bringing the word travel into it. Yes, this is what wired.

1:20:28 - Jeff Jarvis
Wired makes for just for you. Leo wired said this. This is not to be confused with the conventional idea of teleportation, which involves a hypothetical immediate exchange of matter in space. In the experiment, the light particles remained in the same place, but entanglement allowed the computers to see each other's information, as you said.

1:20:46 - Leo Laporte
Yes, so big breakthroughs in quantum computing. In the last, google also has a quantum chip Willow that solved a benchmark that they say solved it in five minutes. That said, a conventional computer would have taken 10 quadrillion years to solve. We live in interesting times, I have to say we do. Coming up on Intelligent Machines, we've got Ray Kurzweil. The guy who coined the term, wrote the book the Age of Intelligent, intelligent machines. His new book uh is coming out soon. It is um frankly. He says we will merge with ai. I can't wait to talk to ray kurzweil. That'll be in a couple of weeks. Uh. We've also uh communicated with stephen wolfram and he has a very good book on how ai works. He's been very good actually writing about ai uh. So stephen wolfram will be joining us as soon as we can work out a time that works for both of us.

1:21:40 - Jeff Jarvis
So we've got some good guests some other people in the proverbial pipeline smarter people than I who can explain this.

1:21:48 - Paris Martineau
You know, maybe we should get somebody on quantum computing, on it would be good, because that confuses the hell out of me I'm literally looking up quantum entanglement as uh defined in the outer wilds game, because I remember when I played that I kind of understood it briefly but I asked, uh, one of the ais I think it was deep seek to explain quantum, quantum entanglement to a dog, and it said something like well, imagine there were a million tennis balls and you could catch them all at the same time.

1:22:22 - Leo Laporte
You know what? Maybe I'm not as smart as a dog. Anyway, uh, let's see what else. Uh, what else?

1:22:28 - Jeff Jarvis
in the ai news there's lots of stuff um I I enjoyed some information reporting in line 25 oh, I wanted to talk about that. This is really well done, I think by. Okay, who was it?

1:22:42 - Leo Laporte
erin woo erin woo she's talking about? Uh google, uh, google's ai efforts marred by turf disputes, and there's a picture of demise Hassibis. Sir Demis Hassibis, who's the CEO of Google's Deep Mind Google, she writes, scored a rare hit with AI-powered Notebook LM, which, of course, we talked about. We had Richard Gingras Was it, richard on?

1:23:06 - Jeff Jarvis
No, no, no, we have Stephen Johnson on that. Stephen Johnson, that's right. Who we'll get back? Yeah, let's get him back.

1:23:12 - Leo Laporte
People love it.

1:23:18 - Jeff Jarvis
That was love it. That was the one that not only summarizes your documents.

1:23:19 - Leo Laporte
I think everyone that I've talked to has used notebook lm says this is cool. It was one of the um ai breakthroughs or ai products that made a lot of people say, oh, maybe there is something here. Yeah, uh, particularly the feature that would let you make a podcast out of your you know your documents. Uh, after the product's release, sundar Pichai praised the small dedicated team that built that notebook LM, but the path to launch wasn't quite so smooth as he made it sound, writes Aaron Wu in the information, with an exclusive before notebook LM launch, staffers working on it in the company's labsashed with their counterparts at workspace, which runs gmail docs.

According to four people with knowledge of the situation she's got some sources workspace employees argued notebook lm will conflict with our plans for existing apps like google docs. They even tried to shut it down. This was a big problem at microsoft back in the day, where the office team would say, no, you can't do that, or the windows you can't do that, it's going to hurt our business. In the end, leaders of both teams stuck with notebook lm and it launched as a standalone website, although it is now a service in workspace.

1:24:31 - Jeff Jarvis
But I think if workspace had won and said, no, no, we're going to own this, they would have concoctvated. However, you say that down to features that fit within workspace right, whereas notebook lm was important on its own. And hiring stephen johnson he's editorial director of workspace that I'm sure that title exists nowhere else in google. The team allowed it to exist to serve writers like stephen allowed it to exist to serve writers like steven.

1:25:00 - Leo Laporte
Some of the tensions flow from how google is structured. Demisa sibis runs google deep mind. Thomas curian runs google cloud. Differing priorities between the two units have caused some tensions. Top managers at google cloud debating how to improve their unit's relationship with deep mind. Um, yeah, I mean, look, it's pretty clear that, by the way, gemini 2.0 is pretty good, right, in fact. Uh, let me just see the ai arena score. Uh, which has got you know it's. Uh, this is. This is a people uh test and vote on which ais are better, and it's kind of the most um, useful way of what's. Am I at the wrong spot? Maybe I'm at the wrong spot. It's the, it's trans. Look, it's striking my mouth that's the right.

That's not the tracker I was thinking arena baby that's not the ai arena I'm looking for I like it.

Uh, what did I do wrong? Where did I go wrong? Anyway, here we go. Um, it ranks. It's actually really useful because it ranks the ais and it's changing constantly. But gemini, for a while, was on top. I haven't checked it recently. I think I'm still on the wrong site. No, that's the wrong place. Anyway, I think Gemini is pretty good. I think DeepMind got a lot of people's attention, you know. Did you guys watch the video I sent you?

1:26:37 - Jeff Jarvis
I got a third of the way through it, leo. It is very informative, yeah. However, I've never had a best, better way to go to sleep. The um, the um. The affect is everything. Well, we took out your kidney and then we're going to take out your heart simple things like multiplication addition. I found it gripping it is, but it's exactly this voice for the entire.

1:27:05 - Leo Laporte
So this is andre carpati, who is the one of the founders of openai, and he's currently working in an ai education. Uh, how?

1:27:13 - Jeff Jarvis
long is it leo?

1:27:13 - Leo Laporte
three and a hours, but it's everything you need to know about how modern models are created, including this breakthrough in reinforcement learning. This is actually the relevant portion where they're talking about AlphaGo, which is the blue line here. Using reinforcement learning actually got better than the red line, supervised learning, which means they got better than the best humans. This is the breakthrough. See the problem with uh the current common models using supervised learning never get better than a human because they're basically learning from what humans have done. They're remixing human intelligence. But this new thing that deep seek is doing is reinforcement learning is actually potentially giving us where ai is, the ability to exceed human capabilities and it did it within go very early on. Any event, uh worth highly recommended. It's a deep dive into llms like chat gbt, if you're interested. It's the first thing I found. That really makes sense to me. It's a primer.

1:28:16 - Jeff Jarvis
It's a very useful yeah, it is good, but but oh, the voice just a soft-spoken man, and that's all. That's all he wants no, no, no, it's fine.

1:28:26 - Leo Laporte
I'm not criticizing, it's not criticism it's just, we're never going to get him on now no, no, no, he's brilliant.

1:28:33 - Jeff Jarvis
He's explaining everything I've been trying to get him on yeah, that's what I'm saying.

1:28:37 - Paris Martineau
He explains everything so calmly that it's comforting, and I I was trying to remind jeff of what he normally watches to go to sleep, which is diners, uh, drive-ins and dives, as I learned last week what guy fieri puts you to sleep, he falls asleep to guy fieri regularly and the thing that I've been thinking about since you said that, jeff, this is just what goes on in my head as I'm going on.

My day is one day. I was walking around and I was like that means jeff's wife must fall asleep to guy fieri um.

1:29:07 - Jeff Jarvis
We fall asleep at different times wow oh, oh so

1:29:11 - Speaker 2
he's already asleep, so here's the llm arena uh, that I was looking for.

1:29:17 - Leo Laporte
Grokk 3, which has just come out we're going to talk about that in a second is currently ranked one, but look, gemini Flash is number two, gemini 2.0 Pro number three. Then ChatGPT 4.0 DeepSeek.

1:29:31 - Paris Martineau
How is this being graded? What are the Arena scores?

1:29:34 - Leo Laporte
Yeah, it's just such a game. You have one against another, humans are doing it and then giving them the score, uh, it makes it a little volatile. You see the number of votes. For some reason the gemini stuff was getting a lot of votes all of a sudden because it was very low and then it suddenly went up, and so I think there's I think it's a little bit of gaming going on, but it is the closest thing. It's hard to judge. You know, I use them all and you know I just find some are better at some things than others.

1:30:02 - Speaker 3
Um, I'm really happy with perplexity, which gives me a good variety of models yeah, I was saying that recently I think, huh, we should have a leo score for like I should, I should have leo arena would, that's too much trouble rank all of us against one another yeah, right it's

1:30:20 - Jeff Jarvis
uh, it's a hard thing to do a child.

1:30:22 - Leo Laporte
I know that. All right, we're going to take a little break. You're watching, uh, intelligent the age of intelligent machines has come. We are uh, this is the brand new rebranded this week in google I am we call it intelligent machines, focusing on, I think, the most exciting uh area and technology ever more than the invention of the computer, the microprocessor or even the internet.

Uh, we're talking about ai, but not just ai robots, the smart little widgets and gadgets you have all around your house that seem to you know have a mind of their own. Uh, paris martineau and jeff jarvis and I and we're so glad you are here have you tried grok three yet?

I have not so this is uh elon musk's uh moonshot. Spent a lot, a lot of money to build what I think most people consider the most powerful AI uh computer ever oh, absolutely the hardware, the hardware hardware um, he, uh, he has X is built with Elon's money a data center in Memphis containing, ready for this, 200 000 nvidia gpus to train grok 3.

That's musk claim. Now I don't know if musk making it up or not, but it was 10 times the computing power of the process of the predecessor, grok 2. And this is you know. This comes back to that article we talked about some time ago, the Bitter Lesson. Remember this. You're not paying attention at all, are you?

1:32:06 - Jeff Jarvis
No, what is the Bitter?

1:32:07 - Leo Laporte
Lesson Rich Sutton, who's an AI professor in.

Canada wrote this six years ago. In summary, the lesson is it's not how the model is made, it's the amount of computing you can assign to it, that there's a direct correlation, he says. One thing that should be learned from the bitter lesson is the great power of general purpose methods, of methods that continue to scale with increased computation, even as the available computation becomes very great. In other words, the lms get better and better and better without modification if you throw more hardware at it. Two methods that seem to scale arbitrarily in this way are search and learning. He also says, by the way, this is an important uh insight.

The second general point will be learned from the bitter lesson is the actual contents of minds. Human minds are tremendously, irredeemably complex. We should try stop trying to find simple ways to think about the contents of minds. That's not. What we're trying to build is another mind. We want AI agents that can learn and discover like we can, and that's the. To me, this is the whole key, and and the breakthrough that's starting to happen is getting machines to think of things that aren't just remixed human content. Does that make sense?

1:33:25 - Paris Martineau
Well, they're not thinking. No, they're not thinking.

1:33:29 - Leo Laporte
No, and that's again. You know we talk about AGI and why it's a terrible term. No, but they could come up with new novel ideas, like new proteins, or they can inspire.

1:33:40 - Jeff Jarvis
Google just started a new assistant for scientists, not to do the work for them, but to be kind of on the side and saying think about this too, think about that too. I think I think that, yes, it can. It can do that, um. But I think this whole war, you know, I think we're going to commodify. Ben evans said of grok congratulations to grok on spending a lot of money to be yet another company on the leaderboard for a commodity technology with no defensibility that we know of. And I think that this Well, that's Me too-ing. Wait a second, this me too-ing.

Of these models as they stand, I don't think no one's leaping ahead. They're a little faster for a hell of a lot of money. And I'm more interested, more excited about A applications, like we had the first half of the show, but B, about working on different visions for how these things work. And I think that's where Jan LeCun, we talked about two weeks ago, talks about teaching them. Reality is the next step. There's next steps, but the fact that Musk is spending this fortune to say, oh, I, oh, I'm a bazillion fraction of a second faster than that model, I'm more powerful, doesn't excite me it's not about speed.

Well, whatever the measurement, is it's it's about?

1:34:52 - Leo Laporte
well, I'll tell you what the measurement is it's about coming up with insights that are beyond the human oh, you're getting near agi sand sand time you don't think that's possible, sand I. I raise move 37 here's a move and go that deep that alpha go did after. Uh, by the way, the way alpha go learned was the current hot thing in ai, which is reinforcement learning. It learned and if you would watch the video, you would know this.

1:35:19 - Jeff Jarvis
I won't finish it.

1:35:20 - Leo Laporte
I just need to have a little-. It learned by playing billions of games against itself Instead of, in other words, ingesting all the existing human knowledge about the game it taught itself-. But it's a new way to learn. It's a new way to learn yes.

Well, this is reinforcement learning and by doing that, exceeded the human capability. And this is move 37, a move that the experts at the time, the master Go players, said that's a terrible move. Everybody knows that's a terrible move. Why did it do that? That's a mistake and it turned out it was a huge insight and it played a non-human move. And that's what we want AI to do is come up with things that aren't remixes of existing content. That's what most ai is right now, but I don't think that comes out of necessarily more power, more power well, that's what the bitter lesson says.

It does, it does. It also comes out of reinforcement learning, deep seek.

1:36:15 - Jeff Jarvis
I think there's other methods in other ways. I think we're an american macho. You gotta watch this video. So deep seek is exactly. There's other methods in other ways. I think we're an American macho, you got to watch this video.

1:36:20 - Leo Laporte
So deep seek is exactly the same as everything else, right up to the very end. You know it does this, builds the models, does the tokenization, uh, does it, does the, uh, the fine tuning. But at the very end, they're doing this new step which open a. I immediately said, oh, we have that too. We knew about. Thatseq was the first public model to do it publicly and, more importantly, I think, do it openly so that people could see what it was doing.

It used reinforcement learning, it actually used a punishment and reward system and by doing that and that's also what AlphaGo did in effect, by doing that can exceed, can go beyond what is essentially just rehashing existing human content. It's as if you had a textbook, a physics textbook. The first stage of fine-tuning would be reading the content, adding it to your token base, right, your LLM model. The second stage would be all right. Now here's the problems in the book. Solve those, and that's, that's the, the um. What is it called? It's, uh, supervised fine-tuning, I think where it. But it never goes beyond what a human's already done, right, because it's just learning the solutions to those problems. This most recent step, and if you get to the end of the three and a half hour video. He talks about this.

1:37:37 - Jeff Jarvis
This is what deep seek did is this reinforcement learning, and that's where the capability can go beyond what's in the book. Things like like protein folding. We're beyond what a human can do.

1:37:52 - Leo Laporte
It's raising the ceiling, raising the floor and that's really amazing, the ceiling and we could. That's when it gets interesting. Yeah, we're there for the many, many functions, just not this ag ibs yeah, by the way, karpathy said I was given early access, uh, to grok3. Um, we should mention he's a former head of ai at tesla. I was given early access to grok3 earlier today around him.

Yes, yeah making me think one of the first I think one of the first few who could run a quick vibe check, making me think one of the first I think one of the first few who could run a quick vibe check. He says first, grok three clearly has an around a state-of-the-art thinking model and did great out of the box on my settlers of katan question. And he gives this question. Few models get this right reliably. The open ai thinking models, the ones you pay 200 bucks a month for, get it too, but all of DeepSeek R1, gemini 2, flash and Claude do not.

However, grok did not solve my emoji mystery question where I gave a smiling face with an attached message hidden inside Unicode variation selectors, even when I give a strong hint on how to decode it in the form of Rust code. The most progress I've seen is from DeepSeek seek r1, which once partially decoded the message. So that's a negative. And then there's a question mark. It solved a few tech tacto boards I gave it with a nice chain of thought. So I asked it to generate three tricky tic-tac-toe boards which it failed on, but then so did o1 pro can I take you to line 112 and gary marcus's uh contrarian as his usual view of grok yeah and I'm trying to get gary on the show because I think he'd be great to have on anyway.

Just well, just to finish before we move on. Okay, um, karpathy's summary is as far as a quick vibe check over two hours this morning, grok three plus thinking feels somewhere around state-of-the-art territory of open ai's strongest models and slightly better than deep seek r1 and gemini 2.0 flash, which is quite incredible considering the team started from scratch a year ago. That's the other thing that all of this compute power gives them is. You know, open ai has been working on this for how long? Five, five years.

1:39:59 - Jeff Jarvis
In less than a year they can get to state of the art. Call me skeptical of claims, but we'll see about how they're doing it.

1:40:07 - Leo Laporte
Yeah, I think we know how Grok's doing it. R1 is a little bit more of a question mark because it's kind of behind the iron curtain what?

1:40:16 - Jeff Jarvis
do you call the Chinese curtain I forget Anyway, all right, what line again, One 12. If you go to a Gary Marcus, who's a wonderful curmudgeon but an Intel, but you know one who knows his stuff, he starts.

1:40:33 - Leo Laporte
I DM him and, if you would as well, we should get them on the show. Yeah, I've, I've, dm'd him and if you would as well we should get him on the show.

1:40:41 - Jeff Jarvis
Yeah, I've, I've I've connected with him. Okay, my davos thing, yeah, um. So uh, he starts off um, I don't know if we want to go into this or not, but uh, well, elon musk, uh, praising how it criticizes the information, but we can skip that before we do, though, let me just point out nobody has been able to reproduce this, and when I give grok 3 the same question, it says the information is very reliable.

1:41:03 - Leo Laporte
It gives me a great summary of this, so this is probably, I think, made up. It's fake, or it's him doing it. It's fake. Go on for it.

1:41:08 - Jeff Jarvis
That's what I joke go further down where, um uh, he gives its own test, grok. My favorite is draw a picture of five different basic geometric shapes and label each one. Okay, the triangle is a turgle, the circle is, it is queer. The square is a rectangle, the circle is a rectangle and the hecticon hexagon is a pentacle now these other ones are what open ai or no, it's all him he's asking, it's four no, no, oh, it's four different efforts, so in some cases it's correct.

1:41:43 - Leo Laporte
Uh, no, no, none of them are 100 correct, no?

1:41:47 - Jeff Jarvis
hey, go down, draw me a calendar of this month circling today's date. The calendars look quite nice, but the circling is blindfolded. Uh uh, pin the tail on the donkey right and this is, you know, agi 8 around the corner. Folks, I think we need voices like like Gary here to be ballast. All right, I'm gonna bring back a voice from a little bit longer ago.

1:42:13 - Leo Laporte
I'm gonna bring you a. You know who? Richard feinman is um great physicist long long gone by the way uh, he was a physicist at, uh, caltech, I think. Brilliant guy long gone nobel prize winner, I believe very, very, very ted, very ted. Talky asked in this is nine, but he was in Oppenheimer.

1:42:37 - Paris Martineau
Oh, then I do know of him.

1:42:38 - Leo Laporte
He was asked in 1985, do you ever think there will be a machine that will think like human beings and be more intelligent than human beings? This is. I think this is about four or five minutes. I'm going to play it, though, and I'll stop it before it's all the way through. You should go watch this.

1:42:55 - Jeff Jarvis
As long as it's not three and a half hours leo no, but listen fine.

1:42:58 - Leo Laporte
First of all, feinman is very compelling. He's also.

1:43:00 - Speaker 3
I actually saw this at what, like when we start first started talking about ai, I I saw this and this kind of formed the basis of my philosophy of ai ah, benito says so, let's listen.

1:43:13 - Richard Feynman
Let's listen to feinman's answer first of all, they think like human beings. I would say no, and I'll explain in a minute why I say no. And second, that they'd be more intelligent than human beings is a question. Intelligence is to be defined. If you were to ask me, are they better chess players than any human being possibly can be? Yes, I get you Someday.

1:43:35 - Leo Laporte
Someday.

1:43:36 - Richard Feynman
They're better chess players than most human beings right now. Yes, I get you someday. Someday it did happen. There's better chess players than most human beings right now. One of the things, by the way, that we always do is we want the darn machine to be better than anybody, not just better than us. If we find a machine that can play chess better than us, it doesn't impress us much, we keep saying. And what happens when it comes up against the masters? We imagine that we human beings are equivalent to the masters in everything. Right, the machine has to be better than a person in everything that the best person does at the best level. Okay, but it's hard on the machine.

But with regard to the question of whether to make it to think like a machine, my opinion is based on the following idea that we try to make these things work as efficiently as we can with the materials that we have. Materials are different than nerves and so on. If we would like to make something that runs rapidly over the ground and we could watch a cheetah running, we could try to make a machine that runs like a cheetah, but it's easier to make a machine with wheels, with fast wheels, or something that flies just above the ground, in the air. When we make a bird, the airplanes don't fly like a bird. They fly, but they don't fly like a bird, okay. So they don't flap the wings exactly. They have in front another kind of a gadget that goes around. Or the more modern airplane has a tube that you heat the air and squirt it out the back, a jet propulsion, a jet engine has an internal rotating fans and so on, and it uses gasoline. It's different, right?

So there's no question that the later machines are not going to think like people think. In that sense, With regard to intelligence, I think it's exactly the same way. For example, they're not going to do arithmetic the same way as we do arithmetic, but they'll do it better. Let's take very elementary mathematics, arithmetic. They do arithmetic better than anybody, much faster and differently, but it's fundamentally the same because in the end, the numbers are equivalent, right? So that's a good example, example. We're never going to change how they do arithmetic to make it more like humans.

1:45:44 - Leo Laporte
That would be going backwards so you're looking for a machine that can circle a date in the calendar because you have that as a model of human intelligence but I could tell you right now I could. I mean emacs could make me a calendar with the right date on it 40 years ago, right when richard feinman was doing that talk in 1985.

1:46:08 - Jeff Jarvis
but that's, that's a bad, that's a, that's a, a phony test, it's a paper tiger. I, I listen I constantly argued that two things one, agi is bs. Two, that, uh, to try to make something human is the wrong scale, that's a species hubris and that's not the goal. And I certainly argue that the machine can do tons of things that we cannot do today. But I think that, but this is. I'm always going to fight this AGI BS, I'm going to fight it from. And the people who push AGI are the same ones who believe in the philosophies of Tess Creel. I'm doubly dubious of them. As a result, let's talk about it in terms of its actual capabilities, what we push it to do, what we can't do, fine.

1:46:54 - Leo Laporte
But don't be the guy that says, oh, humans will never fly because we can't flap our wings fast enough. It's not AGI, isn't, and you're right. Maybe some people who are talking about AGI are thinking in those terms, but that's not really what's going to happen.

1:47:08 - Jeff Jarvis
Yeah, they want to replace human beings and all that, and that causes all kinds of fears, and that's the exact wrong way to talk about this. We're actually agreeing here, because I'm arguing that that's limiting to talk about the machine.

1:47:23 - Paris Martineau
If we limit it to human capabilities, that that the meaningless, sorry paris you were starting I was just saying. I agree with you and I think also, when we're talking about any new technology, but especially a technology that is in the peak of its hype cycle, it uh would behoove all of us to I don't know keep our conversations grounded and firmly planted in reality, because ultimately that's more interesting and, I think, useful and it will result in, I don't know, more productive and interesting conversations.

1:47:52 - Jeff Jarvis
Darrell Bock. Where do you want our feet? Paris, Firmly in the sand. Paris Huggins.

1:47:56 - Paris Martineau
Ideally in the air, please, darrell.

1:47:58 - Leo Laporte
Bock. More importantly, how do you pronounce cheetah?

1:48:02 - Jeff Jarvis
that's, that is the best part of it that was really good.

1:48:05 - Paris Martineau
Yeah, that video, you think he but you know what?

1:48:08 - Leo Laporte
it's richard feinman. So I'm gonna guess that's the real way to pronounce cheetah. Let's find out. I don't know how to, how to find that out. Well, you go to you go.

You just go to google and say cheetah, oh, I don't do google he only perplexity is now I just let me see if perplexity if I say how the word cheetah is pronounced cheetah let me do, I'm doing she rhymes with c. Ta sounds like ta and table, so I don't know why he said sheeta, but I'm guessing, knowing it was feinman, that that's the african pronunciation wait, wait, wait.

1:48:45 - Jeff Jarvis
I'm doing the british. Now hold on a second here. Cheetah, no, cheetah.

1:48:49 - Leo Laporte
Still still heartsy well, what's the new jersey pronunciation? Because that's where feinman was from. Run was from run. What the hell is that? Get the hell out of here.

1:49:08 - Speaker 2
Uh, all right, uh did, I was smart I did not buy the humane pin.

1:49:10 - Paris Martineau
I admit I have. I did really think about asking the b people about their thoughts on the humane pin.

1:49:15 - Leo Laporte
I bought many dumb ai devices. I uh, I'm wearing some right now. I even have those glasses with the rubber nose but I did not buy a human AI pin because I saw the video and I thought that is not going to. That's not and you know what. You were right. It's a $700 brick of e-waste. Jason keebler writing at 404 media only 10 months after it's released the humane ai pin, some would say killed by marquez brownlee, who said it was the worst product he's ever reviewed he killed itself um, it was killed.

Before that, it already had dozens of reviews saying it was truly terrible yeah, the company sold its ip and the software to hewlett packard and says we're turning it off. On the 28th, which is nine days away, the server will be turned off every. Nearly every function of the ai pin will stop working on well, it'll still check your battery yeah, I want to show you the faQ because it says some functions will continue. What would those be?

You could tell what the battery level is. But the humane servers are gone. Center access will be retired. If you bought in the last 90 days, you're going to get a refund. The rest of you? Hey, congratulations. You now know. Will my AI pin work with another cellular provider after the shutdown?

1:50:42 - Jeff Jarvis
no, nope, it's not a real, can I?

1:50:44 - Leo Laporte
port my number to another device. No, can I use my ai pin for offline features? Yes, after february 28th 2025, ai pin will still allow for offline features like battery level, but will not include any function that requires cloud connectivity, like voice interactions, ai responses and center access couldn't even think of a second thing after battery level.

1:51:07 - Speaker 3
Yeah, it was just battery level, it was just battery level etc which I think is very funny I feel kind of bad for them, I mean.

1:51:20 - Jeff Jarvis
I saw somebody on Twitter, on one of the socials, said yeah, well, don't dance on their grave. And somebody came back and said they danced on the grave of the iPhone, of the phone. Yeah, they thought the phone was gone.

1:51:30 - Paris Martineau
They raised how many millions of dollars? Yeah, they're fine.

1:51:35 - Leo Laporte
Do you think they made money Like they went home with money, the founders certainly got paid for their time what was the juicer thing?

1:51:45 - Jeff Jarvis
um juicero, yeah, just this is.

1:51:48 - Leo Laporte
This is juicero. It was pretty apparent and actually the story of the company it was that they pivoted. They really were just trying to come up with a projector screen and then they ai became hot and they so they added ai on top of it. But never.

1:52:01 - Jeff Jarvis
So now they're gonna. Now they're gonna add things into meetings for hp you're watching intelligent machines.

1:52:06 - Leo Laporte
Jeff jarvis, paris martineau, I'm leo laporte. We'll have more in a moment. Just did you see the the piece about, uh, these ai assistance at meetings, and how terrible it is. Let me see if I can.

1:52:20 - Paris Martineau
I don't know what piece you're talking about, but I will say it's truly the bane of my existence when I get in a Specifically like a company Zoom meeting and someone has their little Otter AI thing there. It's like we're not allowed to record our company meetings, guys.

1:52:34 - Leo Laporte
Oh you're not.

1:52:35 - Paris Martineau
No it's been a whole thing.

1:52:37 - Leo Laporte
They make announcements, there's a whole bloomberg's weekend essay from Chris stoker Walker. Please stop inviting AI note takers to meetings. Using artificial intelligence to summarize meetings raises questions about etiquette, privacy and the purpose of the meeting. In the first place, it sounds like uh, the information some of them aren't just note takers.

1:52:58 - Jeff Jarvis
But a wonderful former student of mine named joe md, who's at montclair state uh, he plays with all the tools, so his will will tell you who's speaking the most in the meeting. I think that's good. Tone is in the meeting. I think that's great. All kinds of things added in. So we did it one time with with the, the, the dean of the school. He was like all nervous, like like, oh no, I'm talking too much.

1:53:20 - Leo Laporte
Well, if I were a woman at meetings? I would definitely run that thing and say look, how many words you've got.

1:53:26 - Jeff Jarvis
Or podcasts right.

1:53:27 - Paris Martineau
Paris. We could use it here, certainly.

1:53:35 - Leo Laporte
And again he raises the issues, or actually one of the interview subjects raises the issues of hallucinations, which I think is not as much of a problem as it used to be, especially when you're working off a corpus of knowledge like a transcript of a media.

1:53:49 - Jeff Jarvis
Well, this goes to the piece, if I may. No, I don't know.

1:53:52 - Paris Martineau
Especially with depending on what transcription service you're using or what model hallucinations can definitely.

1:54:00 - Ethan Sutin
I've noticed this with.

1:54:01 - Paris Martineau
MacWhisper, because you can pick from dozens of different models and there are some that it has very specific hallucination bugs.

1:54:12 - Leo Laporte
Margaret Mitchell, the co-author of the famous Stochastic Pirates piece. I want to get her and or Timnit Gebru on. Let's make a note of that. They would be great.

1:54:22 - Jeff Jarvis
Margaret, I think, has a brand new book out, so that can be more reason to.

1:54:25 - Leo Laporte
She's chief ethics scientist at Hugging Face. We asked the CTO of Hugging Face to join us and he said he only does in-person calls. Only ones where you can hug his face I said I think that means no, but I I don't know unless you want to say no I think it's a creative way to say no.

Anyway you. She says. Using an ai note taker is much more like recording a meeting than taking your own notes. Quote some of the etiquette around recording applies. For instance, anyone using an ai note taker should let others in the meeting know well, of course, and give them the right to veto the ai's attendance. Zoom has it built in? You can push a button on a zoom call. In fact, do we do that? Could I do that on this call? Yeah, ai companion request access for ai companions. Ask the host. I'm asking the host right now. Are you the host, benito? Yeah, it's me. It's asking me, do we?

need an ai companion in here well, I would, we could at least okay. So, uh, we just turn it on, so I'm not going to push the catch me up button. Oh, look at the questions you can ask. Was my name mentioned? Are there any action items? So it's really good for somebody who's sleeping through the meeting. What?

1:55:47 - Paris Martineau
topics have been discussed. It really is for someone who wasn't paying attention.

1:55:50 - Leo Laporte
Yeah, so I'm going to leave that on, and at the end of the show we can say, hey, did they mention me?

1:55:56 - Jeff Jarvis
By the way, the book just for the video, for the record. This, uh, the third of three authors, uh, the first one listed for the stochastic parents paper is emily bender from university of washington. She has a new book out called the ai con how to fight big tech's hype and create the future we want. So I think she doesn't? She's not weighed down by sand. I would say a Paris.

1:56:18 - Paris Martineau
Yeah, sandless.

1:56:22 - Leo Laporte
I'm going to defend myself. You know cause? The first guest, zach Cass, was very pro, was an accelerationist and I think what we have to do, and what we plan to do, is get all points of view on this. Yeah, he really was. I will. I will paint myself as an accelerationist. I know I am. I don't know, paris, you're kind of in the middle. You're saying jeff are you? Are you a negative? Are you?

1:56:43 - Paris Martineau
are you more like emily bender and it's all a big hype I would say my natural position towards all things is skepticism yeah that is the nature of my job that's.

1:56:56 - Leo Laporte
You know, you can't go wrong. You might have a miserable life, but you can't go wrong. Oh, I definitely will. That's where the nihilism comes in.

1:57:04 - Paris Martineau
But such is the curse I've been tasked with.

1:57:08 - Leo Laporte
And Jeff, you're a little bit more positive, aren't you, I think?

1:57:11 - Jeff Jarvis
I'm positive, but people think that I'm all because I wrote a book about Google and such, that I'm all because I wrote a book about Google and such that I'm all accelerations, I'm not.

Google certainly thinks you are when they try to pay you for things, right, they don't pay me. But well, that's about my stupid industry. That's about newspaper publishers being jerks. No, I'm skeptical, and I think just on today's show I'm skeptical and I think just on today's show I've been arguing more than you about how AGI is BS and how the boys are dangerous and how we have to be careful.

1:57:50 - Leo Laporte
So no, I'm no accelerationist and I'm damn the torpedoes full speed ahead. I can't wait until our AI overlords are in charge.

1:58:03 - Jeff Jarvis
I'm an optimist. I think the UK, in its current discussions about regulation of AI, has been surprisingly in my view I didn't expect this at all, I think enlightened, where their starting point on training is that reading text online is a right of the machine to learn good good whoa I didn't expect that kathy gallus has argued for that the right to read and after, uh, the paris ai summit.

I think what we've seen is a shift in the attitude in europe. Is that, well, being the biggest regulator in the world hasn't gotten this very far exactly, and so I think they're more open to things, but they're still going to be, by nature, cautious yeah, I think that's okay.

1:58:42 - Leo Laporte
There's a mix. We need that. I think there's a realization, though, uh, when it comes to national uh authorities, that if if you don't do it, somebody else might, and so we are in a arms race of ai, and so to be too cautious or too safety focused might be a mistake from a national priority point of view. Mitchell goes on to say, talking about the AI assistance as we seek to have more automation in traditionally human tasks, we cede our control power and privacy to AI systems and the companies that own them. That's a good point, I think. A critical thing that ai note-taking lacks, which human note-takers have, are the non-verbal cues from the speakers. A sarcastic note, a joke rather, said in a meeting, could become an action point when cranked to the compute power of an ai system.

1:59:33 - Paris Martineau
That's the kind of hallucination that probably could go to hr yeah, I do think it's worth noting that, again, having a record of stuff like this and having, in some cases, perhaps a sarcastic joke transcribed literally and taken literally and putting your meeting notes at work could be a problem for a lot of people uh.

1:59:55 - Leo Laporte
The piece concludes. This is ch Chris Stokel Walker talking. If workers trade in-person engagement for AI readbacks and colleagues curb their words and ideas for fear of being exposed by bots, what's left If the humans step back? All that remains is a series of data points and more AI slop, polluting our lives.

2:00:19 - Jeff Jarvis
That's a little harsh. I think that's trying to find too much meaning in it. I think Paris's attitude is right. I don't want a record of this. This is company stuff. It's a little rude. Stop. It's more that etiquette level than it is the oh my God, it's ruining humanity level. So I'd counter that.

2:00:44 - Paris Martineau
You've got to stretch it to make an essay, baby.

2:00:46 - Jeff Jarvis
Exactly, exactly.

2:00:49 - Leo Laporte
I'd counter that with the piece from Ars Technica, researchers have made a significant breakthrough in 2025, using ai to design a multi-step enzyme capable of breaking down certain plastics, potentially revolutionizing efforts to combat plastic pollution. I think that's interesting. This was published in science a significant advancement in ai driven enzyme design. Now that's. To me, this is an example of going beyond what humans have been able to come up with, to something to come, something new yeah, it's not cool.

2:01:23 - Paris Martineau
I will also note, though you are reading perplexity's summary of this article, you're not reading the actual article yeah, I don't want to give uh technica too much.

2:01:34 - Leo Laporte
Uh, all right, I'll read the source. Actually, Ars Technica was just to be honest and fair. Ars Technica was just summarizing this scientific piece from a bio RxIV preprints computational design of serine hydrolysis.

2:01:56 - Jeff Jarvis
That's a preprint server.

2:01:57 - Leo Laporte
Yeah, yeah, it's a preprint server. Yeah, yeah, it's a preprint server. So this is a site. I mean, they were just interesting, but so I'm just continuing to synthesize the information and say look, it looks like. Maybe not, but it looks like an ai may have come up with a new enzyme that we otherwise didn't know about. That's, that's a good use. Yeah, nicola, the electric autonomous truck company has filed for bankruptcy. There's one of them bye-bye oh, you know, we could have a ai graveyard segment we could just things that didn't make it.

2:02:38 - Paris Martineau
You remember after it could be the sort of stuff that washes up on the beach would you guys leave off?

2:02:44 - Leo Laporte
I'm gonna get this guy on the show and then you can decide no it wasn't so late for him all of the uh, you know your little journey into being a radical I, I, I've, I have, like a good ai, synthesized all of the different data points, including those that you and jeff have proposed, and have come up with my theory of the future is the sandman a real person?

2:03:11 - Paris Martineau
leo, yes, he is okay, you're sure you sure Is he tall, I imagine him being tall, I imagine you looking down at the sand and there was only one set of footprints and that's because he's carrying you.

2:03:29 - Leo Laporte
Is that what happened? Wouldn't that be good? Oh, that was good. That was Jesus. I think that did that right. I walked with Jesus.

2:03:37 - Paris Martineau
Was that who you were walking with?

2:03:39 - Leo Laporte
no yeah. I, jesus, google Lens has in Chrome for iOS that you know they didn't put that circle to search thing in, but this is kind of like that you can.

2:03:52 - Jeff Jarvis
You could put the change log in front of this one if you want, it's the Google change, or Google's more deeply integrating lens visual search in the Chrome.

2:04:02 - Leo Laporte
Weirdly, though, for iOS, I guess, cause you already have it in Android with circle to search, so you don't really need that. Uh, so if you're using Chrome on your iPhone or iPad, you can now search screen with Google lens as you're browsing the web or watching a video. You don't have to do a screenshot. There's a new shortcut in the three-dot overflow menu and it will then give you an AI result based on there. It is search this screen based on what it's seeing. Based on what it's seeing, google credits advanced AI models as allowing lens to go much further and provide information on the contents of more novel or unique images. This is really the successor to search right, like image search Is it I mean image search.

2:04:52 - Paris Martineau
they've had image search for a while, for a long time.

2:04:55 - Jeff Jarvis
Is this better? This is better, isn't it? Well, because it does more than search. You can query about what. It's a way of saying. I remember many years ago the first time I went to MIT Media Lab and it was a whole big deal. One of their first demonstrations was a big screen and some shapes on it and the instruction set of saying put that there, so it understood the antecedent. What is that? What are you pointing to? That's what the circle really means is you can. You can say I'm talking about this and then you can ask a question about it. That's just another ui, but it's more than search, I think youtube has added ai to youtube.

2:05:33 - Leo Laporte
Right, there's now a an ai button. Does everybody got to have youtube? Everybody an ai button. Does everybody got to have youtube? Everybody got to have youtube uh, I think that's cool, I guess I mean it's google. Let's see what google. This is a little inception moment. Can I do uh ai on our show um, there's a, there's a.

2:05:52 - Jeff Jarvis
Uh, there's a button that shows you interesting maybe because, it's live, it can't I can't do that, maybe yeah but go to last week's show. There's that, there's that, there's that I can do it there. Oh okay, and oh okay, that's interesting don't go to last week's show, go to a few weeks ago why shouldn't I go to last week's show?

2:06:14 - Leo Laporte
because you don't want to? What are you talking about? Things you don't want to talk about what are you saying, intelligent machines? Let's go to, uh, the last episode, which was 806. All right. Well, actually we got episode one, which is 805. Yeah, that's all right. So so here it is now. Where's the ai? Is it on this?

2:06:38 - Jeff Jarvis
no, maybe it's in the uh, the mobile app. I don't know that. That video will show you where the button is I know I shouldn't have watched the uh youtube video. Look, there's an intelligent machine store.

2:06:49 - Leo Laporte
I didn't know that you can shop the intelligent machine store that's fun all right, maybe it's down there maybe it's down there.

2:07:00 - Jeff Jarvis
Okay, so on my phone. I put up Karpathy's video and now I can ask about the video.

2:07:06 - Leo Laporte
Oh, that's cool. So instead of watching a three and a half, hour video.

2:07:10 - Jeff Jarvis
So if you go to your phone there's an ask button, you can do that. Nice, please tell me what the point of this video is. Please tell me what the point of this video is. I Please tell me what the point of this video is. I'm not being a smartass. I think it's really brilliant.

2:07:22 - Paris Martineau
I like the loud voice you use to talk Machine.

2:07:27 - Leo Laporte
Hello machine. I think this is a brilliant life hack A ring that adds an extra finger to your hand so that any picture of you will appear to be ai generated that's great.

2:07:45 - Paris Martineau
People have been saying you should wear that if you're going to do crimes if you're going to do crimes, wear the finger so we're going to get past that position

2:07:55 - Jeff Jarvis
so uh, the, the, the youtube. Uh, I'll turn down the video audio here. The YouTube comes back pretty quickly. It says the video is all about explaining large language LLMs like chat, gpt, exclamation point. Why the exclamation point? I don't know. It goes through the steps involved in building an LLM. This is the cool part. Here are some of the highlights of the video 12 seconds in the goal of the video. 34 seconds in You'll learn about these models One hour seven seconds in the goal of the video 34 seconds and you'll learn about these models one hour seven. Oh wait, is that one minute? Yeah, I guess one minute. So it only lasted two minutes 32 seconds and it came up.

Oh well well, it tried it explains common crawl, but I think if you went through this, that would be a very useful thing to come up with its own I mean again, my question is are these hallucinations, though?

2:08:44 - Paris Martineau
I mean I'd be curious if you have it, go through a whole three-hour video. There's probably going to be some things come up that they didn't actually say well, they're much like.

2:08:51 - Leo Laporte
There's two things. There's not no this. Watch the video.

2:08:55 - Jeff Jarvis
You'll understand what a hallucination is and why it's not going to happen there well, there's also randomness that you are you gonna get the same answer twice in a row. No, no, can I go to the?

2:09:03 - Leo Laporte
story I want to go to on line 110 yeah you'll like it I think the overton safety and um hallucinations has shifted which line 109 on probabilism.

2:09:14 - Paris Martineau
It's a real problem I had one of these services I had gemini summarize an article two weeks ago on the show. It provided a very succinct, well-done summary with all the stuff I wanted in it and then I went to go check any of the details it gave me and like half of them were made up.

2:09:32 - Jeff Jarvis
Oh well, that's not good, that's still the case. That's still the case? Yeah, absolutely.

2:09:34 - Paris Martineau
And that's Gemini, which I assume is powering this.

2:09:37 - Jeff Jarvis
And if you do it the second time, you're going to get a different answer. Yeah. But the issue here is that we expect computers to be deterministic, but like a spreadsheet, you put data in-.

2:09:48 - Leo Laporte
No, it's stochastic.

2:09:49 - Jeff Jarvis
Yeah, you put a question in and then you get an answer and what this argues it's a really neat piece that then came that was deterministic to deterministic to deterministic. Then came deterministic to probabilistic to deterministic Things like targeted ads. You put in a specific question. It did some stuff and you don't really know what, and it came back with an answer and it may be the right answer, maybe the wrong. There was no right or wrong answer Then. Where we are now is probabilistic to probabilistic to probabilistic, where it's all kind of just close enough for jazz and generative AI and it's going to change every time and what we want to get to is probabilistic. I can ask a vague question. To probabilistic it can do its thinking in all kinds of ways no, it doesn't think. But to deterministic, where it's going to give me an answer that I can rely on. We are not there yet. And paired with this is a story from our friend Brett Taylor, who line 110, who we ought to have on as well. Brett, I think, is now the chair of OpenAI, right?

2:10:57 - Leo Laporte
Yeah, I think so. I think that sounds right.

2:11:00 - Jeff Jarvis
And he's CEO of Sierra and he was at Salesforce and was at Meta. So what he's saying here is that artificial intelligence agents yeah, they're not quite right, yeah, they're going to get stuff wrong, but kind of lump it, don't wait for them to be 100%, because you're never going to get there. Accept that it is imperfect. He says, rather than say will AI do something wrong, say when does it do something wrong? What are the operational mitigations that we put in place to deal with it? So what I hear Paris say is that's not acceptable, that's too much effort and too much risk. And Leo is saying yeah, accelerate, let's see what it can do. It can do amazing things. So where's your tolerance for, whether you call it hallucination, whether you call it randomness, whether you call it just being wrong, you?

2:11:45 - Leo Laporte
should be. I think it's fair to say you should be able to trust a summary of a document to be accurate to the document. I'm very shocked that there was hallucination, to be honest with you, because it's supposed to be working directly from the document.

2:11:58 - Paris Martineau
I typed in to Gemini hey, this anecdote you gave me, where'd you get it from? And it said oh, I made it up.

And it gave a little description of it. We talked about this in the show and gave a little description of it. We talked about this in the show, and I think this is just.

My constant frustration with these tools is, yeah, I think they could be really useful, but if I am, with some frequency, encountering problems like this, then I can't really trust them to do tasks that I need them to do, which would be stuff like, hey, an article that I meant to read that I want to talk about in the show. Can you give me a high level summary of it, because I've only read half of it, and point out any interesting anecdotes? It can't do that, so I'm not going to use it for that. And then it's like where does the buck stop? If I have to be, because I mean, then my next logical step in thinking about this goes to. Well, if I want to use something like this for the purposes I want to use it, I have to go through and check everything it gives me before repeating that to other people, and that's just not particularly useful. I'll just read the article myself.

2:13:00 - Leo Laporte
Yeah, no, I would agree with you. It's not been my experience that when working with an AI on documents that I've given the AI, that it hallucinates on those documents. So that's troubling, to be sure. I haven't seen that. But I still do you still see, oh yeah okay, again, I keep on saying this.

2:13:22 - Jeff Jarvis
I'll say it again. The other issue to me is the randomness. Ai has built in randomness, so you don't get the same answer twice. That alone is an issue when it comes to reliability and credibility.

2:13:34 - Paris Martineau
And the thing that I think that makes this particularly difficult is the way that these tools work is they're trying to produce something that sounds right. They're trying to produce something that is right, but it doesn't know really the difference between sounds right and is right, and so, in the case of that article, it gave me exactly what I was looking for Anecdote about. I think it was that article we were talking about about tech companies removing tampons and period products from men's restrooms and kind of a squabble, and it gave me an anecdote about uh oh, I remember this between a manager and an underling about they'd set a comma, a pithy comment like oh well, if you need pads, so much bring your own.

And I'm just like right, what a strange little detail to make up and insert in it.

2:14:25 - Leo Laporte
But it's the sort of thing that if my natural inclination was not to check, everything yeah, I would have just assumed that was in the article and if you have to do that then it's not that useful. I completely agree with you.

2:14:37 - Jeff Jarvis
I understand if you're using it for, like in the google example for a scientist, if it's, if it's to your side and it says have you thought of this? Have you tried that? Why don't you think of this's okay, because it's up to the human then to decide what to pursue. And if it does inspire some new path, cool, they're going to do it in their scientific method and and and um, they'll have confidence. But if you use it to replace your task because it's better than us, Let me ask did you just?

2:15:06 - Leo Laporte
give it a URL, or did you paste the article and ask it for that?

2:15:11 - Paris Martineau
I gave it a URL, but then I tried again with pasting the article.

2:15:15 - Leo Laporte
And it wasn't better with pasting.

2:15:17 - Paris Martineau
I mean both times. It gave me something that wasn't right.

2:15:21 - Richard Feynman
Okay.

2:15:23 - Leo Laporte
I feel like that's a flaw in Gemini, to be honest with you.

2:15:30 - Speaker 3
It might be paid versus free version, because if you don't, you're not seeing that you're paying for all those services. I'm not sure if paris is paying for those services yeah, I'm not paying for them yeah, I pay for everything but I mean, I think that's also kind of.

2:15:40 - Paris Martineau
So that's kind of a problem, because the average well, I want better I want the best, ai yeah yeah, but the average person is going.

If they have any exposure to these tools, it's going to be for a while as a free user. And especially if kind of the end game for these products is a use case like yours, leo, where you're not using search engines like Google, you're not going to read the Ars Technica write-up or the preprint, you're just trusting these services for all your information, then we get to a kind of dangerous place. If they're very adept at making things that are not quite right but sound very right, which is kind of their whole thing. I will say, though and I've said this before on the show my favorite use case for AI remains that if I ever can't think of a synonym or a specific word, chat gpt can find it every single time, and it's fantastic yeah, because it has basically a giant thesaurus built in, but that's really what it is thesaurus, but then also I can just describe it in the most, in the shoddiest way possible like it's like this word, but not.

But. The vibes are slightly more positive and it'll, it'll get right there.

2:16:51 - Leo Laporte
It's great I'll have to, uh, I'll have to look at the synopses that I'm getting and see if I can find issues. Um, maybe it's because I don't use gemini, but, um, I mean, I don't have this problem, I don't think, with the perplexity, but, uh, maybe I do and I just I don't know.

2:17:09 - Paris Martineau
I think it could also be related to. I mean, if I'm thinking of that specific use case, I asked for you know a summary, I asked for you know an overview of, like any main points, arguments or counter arguments, and then like an overview of any interesting anecdotes. And he gave me all of those.

2:17:27 - Leo Laporte
So he was trying to help you yeah.

2:17:29 - Jeff Jarvis
He was interesting anecdotes gave me all, so I was trying to help you? Yeah, he was trying to provide everything I asked for, and it doesn't know how to say sorry, no, there's nothing interesting yeah, what would please?

2:17:37 - Paris Martineau
you I mean, that would please me if they're just like yeah, not there, okay, I move on uh, let's take a break and get the pics of the week.

2:17:48 - Leo Laporte
As we wrap things up on this issue of intelligent machines, we don't yet have a guest for next week, right, benito? We? Uh, still working on we will fill that hole.

2:17:59 - Jeff Jarvis
We have some good uh quite a few good ones, so yeah, so we will get benito. If you need me to go to some of the people my name's on, let me know. Okay, thanks.

2:18:10 - Leo Laporte
I'd like to get Dave Marcus on Gary Marcus, gary Marcus.

2:18:18 - Jeff Jarvis
That would be great, and I think that. Who did I just mention the author, co-author, margaret Mitchell? No, margaret Mitchell. Well, that's one. Yes, anyway, we've got some ideas.

2:18:27 - Leo Laporte
We've got some ideas. We've got some ideas and if you're an AI wizard, let us know. Email us and we'd love to hear from you, oh, emily Bender, yeah, yeah, yeah, I put her on there.

2:18:38 - Paris Martineau
Daniel Oberhaus, a former reporter at Vice and Wired, just has a book out on. Ai and AI-powered therapists.

2:18:48 - Leo Laporte
Oh, and we'll, in just a moment, check with Zoom to see what it's learned from this show. You're watching Intelligent Machines with Jeff Jarvis and Paris Martineau. All right, let me see. Let me see, was my name mentioned. Let's see. It's analyzing now. And yes, leo Laporte's name was mentioned multiple times throughout the meeting transcript. Leo laporte appears to be one of the main participants in the discussion contributing to various topics including ai technology, recent developments and personal views on ai enhancement. Are there any action items?

2:19:27 - Ethan Sutin
analyzing this is zoom's ai companion leo researcher, reach out to emily bender to invite her as a guest, investigate potential hallucinations and ai summaries from complexity.

2:19:38 - Leo Laporte
Benito continue working on security. Yes, for next week's show, jeff, assist benito and reaching out to potential guests if needed. Very nicely done, very nicely done. Uh, were there any? Uh, let's see, catch me up, I don't want to. Well, all right, too late, I clicked were there any disagreements?

oh, that's good, let's see. Were there any disagreements during the meeting? Paris expressed skepticism about the reliability of ai generated summaries. Well, good yep. Citing experiences with hallucinations in gemini's outputs. Leo was surprised by this, as it hadn't been his experience with other ai tools. Jeff acknowledged seeing similar issues with ai generated content. The group also had differing views on the pace of ai advancement, with leo positioning himself as more of an accelerationist, while paris maintained a more skeptical stance. See this this is great. How could you not love that?

of course, if you were in love I think the action items is the most useful. I think that's great. Uh, let's get our picks of the week. I've got one. Can I start?

2:20:58 - Ethan Sutin
I don't usually do a pick of the week.

2:21:01 - Leo Laporte
I want to give you guys all the time in the world for this, but it happens. This is an old friend of ours. Ginaani, who was the original host of this show, has just posted something at ginatrapaniorg called my Life in Weeks. This is apparently something people do nowadays.

2:21:19 - Jeff Jarvis
It's just the kind of beautiful thing that she would make.

2:21:21 - Leo Laporte
Yeah, and she actually. The software is available. She talks about where she got the idea. She adapted the code from Buster Benson, so what it does. In fact, if you go to the Buster Benson, you enter in, you can do this, and you oh well, she's actually doing a page. Let me see if I can find the. Yeah, here it is. You can enter it in. There are pages where you can enter in your birth date and it will make the chart and then it's up to you to fill it in.

So I was born. And then you see, the first few years, nothing happened. Then there's kindergarten, the tiny tot schoolhouse where Miss Mary and Miss Ninny were my teachers. Reagan was inaugurated. Elementary school First computer.

What was Gina's first computer? Ah, he bought her an IBM PC Junior. That's manyina's first computer. Ah, he bought her an ibm pc junior. That's many people's first computer. Anyway, it goes on and on. Saw my first web page. Graduated college elevators, first professional ig. I teach up at the elevator company where my dad and brothers worked. I didn't know gina was an elevator engineer. No, might have asked her about that. Uh, here's her founding of lifehacker. She started lifehacker. I do believe we are in here. The first podcast, april 1st 2009, launched this week in google with leo and jeff, later coasted and co-hosted all about android with jason, in beta with kevin purdy, and the post light podcast catalyst started. Think up, she, uh she talks about getting married, having her child. I think this is really cool. That is my life in weeks and, of course, uh it. The last entry is made this february 5th, and then there's a lot more weeks left it's great you could fill.

Do you think you would do this?

2:23:11 - Jeff Jarvis
I'm not that organized. I started to do it. Oh yeah.

2:23:17 - Paris Martineau
Yeah, that would require a lot of effort on my part. B would be perfect for this, though. Yeah. By the way too many personal things about your life in there.

2:23:28 - Leo Laporte
On her 100th birthday she has a to-do item hug her partner and her daughter at the beach. That's a good goal for your 100th birthday. That's very in the year 2075. Very cool, paris, your pick of the week I've got two very different picks.

2:23:47 - Paris Martineau
The week one is a delightful long, long, long read published by the new york times that is a look into the murdoch's succession drama that has got more twists and turns than the show succession and it does have a literal succession the tv show side plot in it. It is such a fascinating it honestly flies by. It's really I don't know. I just really enjoyed it. Would recommend that if you're looking for something to do, and then I will also write while you're on that.

2:24:17 - Jeff Jarvis
While you're on that, may I say for a second the companion piece to this, the bookmark which I just put it there is the Atlantic McKay Coppins interviews James Murdoch.

2:24:28 - Leo Laporte
James is on the outs right because lachlan is going to run the empire well, no, no, no, I know there's a court.

2:24:36 - Jeff Jarvis
There's a court case going on well the new york times piece, which is really wonderful. We should give credit to the reporters there because it was a great report the new york times piece.

2:24:43 - Paris Martineau
They somehow this is a sealed court case. They somehow got all of the documents 3 000 discovery 3 000, which basically shows all of the dirty laundry discussions between all of the Murdochs during this entire period. I read it straight through Same I, literally. I thought I was going to take it and read it on the train I sat there.

2:25:02 - Jeff Jarvis
No, no, no, no, no. Jim Rutenberg and Jonathan Mahler credit to you, it's a really good report. This, jim Rutenberg and Jonathan Mahler credit to you, it's a really good report. This is the best of the New York Times, so is it not?

2:25:11 - Leo Laporte
the case that Lachlan beat out James for the Empire.

2:25:15 - Paris Martineau
Well, no, lachlan and his daddy tried to get all of them.

2:25:19 - Jeff Jarvis
They lost, changed the trust.

2:25:21 - Paris Martineau
They tried to change the trust to get out. Basically everyone, including James, that is not Lachlan they lost.

2:25:31 - Leo Laporte
I didn't know they lost. I know this was ongoing and that they had sealed it. Uh, we did not know they lost the. The theory was well, you can change the trust if everybody benefits from the change, and murdoch's contention was well, having lachlan run fox will be much better than having j James runs Fox, because he shares my political points.

2:25:51 - Paris Martineau
One of the small things in this that I particularly loved was all of the project and code names they use, which are just like the plan to kick his three children out of the trust that they are entitled to and give it to his other son is called Project Family Harmony.

2:26:13 - Jeff Jarvis
If I may, let me give you this from the Atlantic story, Because this is just. You gotta read both of them. So in the deposition, Rupert's attorney asked James a series of withering questions. Have you ever done anything successful on your own? Why were you too busy to say happy birthday to your father when he turned 90? Oh geez.

2:26:32 - Leo Laporte
This is straight out of succession. Yeah.

2:26:35 - Jeff Jarvis
Does it strike you that in your account, everything that goes wrong is always somebody else's fault? At one point, the attorney referred to James and his sisters as white privileged, multibillionaire trust fund babies. And Lachlan's not Exactly Well. Rupert is fun babies, james did, and lachlan's not exactly well. Rupert is uh, james is best to concentrate, but he couldn't help stealing glances at his father. Rupert sat slouched and silent throughout the deposition, staring inscrutably at his younger son. Every so often, though, he would pick up his phone and type. Finally, james realized why quote he was texting the lawyer the questions to ask oh jeez, how effing twisted is that? Oh, the two stories together.

2:27:15 - Paris Martineau
Oh, that's what I'm doing after I.

2:27:17 - Jeff Jarvis
Oh yeah.

2:27:18 - Paris Martineau
Oh yeah.

2:27:19 - Leo Laporte
Can I just have the AI and analyze it and give?

2:27:21 - Paris Martineau
you something. You want every delicious detail no you want every word of this we love watching. You want to feel it, you want to see the text go by.

2:27:30 - Leo Laporte
They've got a great graphic with it. It's fun. Yeah, we loved watching succession and I knew that succession was at least partly based on the murder it in turn, as it plays a role here.

2:27:38 - Paris Martineau
Hey, interesting you know there's a big happening in succession that, um, I guess there's also succession spoilers in this article. Oh, you're kidding.

2:27:48 - Ethan Sutin
Yeah, yeah.

2:27:49 - Paris Martineau
Because a big happening in the last season of Succession inspires real world events and drama in the lives of the Murdoch.

2:27:58 - Jeff Jarvis
They all deny watching Succession, but it had an impact.

2:28:02 - Leo Laporte
Apparently, they heard about it.

2:28:03 - Paris Martineau
That's what my other pick of the week is very different. It's a recipe from the new york times for roasted chicken thighs, hot honey and lime, and it's changed my I don't know. I've been eating it all week.

2:28:15 - Leo Laporte
It's really I live, I live on, I live on chicken thighs I don't know how I've been sleeping on chicken thighs now but now I'm gonna be living on them. They're anything to do with. The chicken thigh is gonna be good, yes I'm gonna eat chicken thighs.

2:28:27 - Paris Martineau
After this, it's gonna be great oh yeah, way better than breasts anyways, yeah, breasts are dry and boring. They're white, terrible yeah, and this is super easy you could pot it in. You pop it in the little oven. It probably will be done before you finish reading the murdoch article.

2:28:43 - Jeff Jarvis
So you know, enjoy all right, so it's my turn oh, I'm making this.

2:28:48 - Leo Laporte
This sounds so good. It's really good, okay, and very simple, by the way it's very simple, uh, healthy, reasonable calories, you know thank you for that.

2:29:04 - Jeff Jarvis
I'm putting that in my recipe box can you send it to me, since I don't subscribe to the food?

2:29:09 - Leo Laporte
Oh yeah, oh yeah. You know what I do. I have a program called Paprika that will go to any webpage and Scoop it up. Scoop it up and put it in my recipe box.

2:29:21 - Jeff Jarvis
Oh, I see oh.

2:29:23 - Leo Laporte
So yeah, I'll show you, watch this. So I just copied the URLl for the recipe and I'm going into paprika into the browser, paste in the url and then I say, yeah, yeah, yeah, no, no one wants me to log in, all right, well, I'll log in later. So but I would download that and it makes a beautiful little recipe card out of it with the picture and everything.

2:29:46 - Paris Martineau
So I put your link in the Discord. Anybody can get the gift link in there.

2:29:51 - Leo Laporte
Oh, you did a gift link, aren't you sweet. Thank you, aren't you kind, jeff Jarvis.

2:29:58 - Jeff Jarvis
You have a pick. For us, our game through time has been trying to get Leo to spend money on the show.

2:30:04 - Leo Laporte
It's not a hard game, is it?

2:30:07 - Jeff Jarvis
It's not hard. He didn't get the humane pin. He didn. It's not hard, he didn't get the humane pin he didn't get that.

2:30:13 - Speaker 2
Did you try to get me to get that? If you go to line 144.

2:30:15 - Jeff Jarvis
No prepared to translate the video on line 144 okay, okay, so first yeah, this is so complicated.

2:30:22 - Leo Laporte
The sound starts low, but okay, there's the video, and do I have to do something? Uh, with it's in german that's so over 50 years.

2:30:31 - Jeff Jarvis
the Thermomix has been the heartbeat of the kitchen. Millions of people love it every day to do their own cooking, 1971. Looks like a blender.

2:30:41 - Leo Laporte
Looks like a blender with a base.

2:30:44 - Jeff Jarvis
Oh, now it's looking like the Thermomix 2004. Oh, that one looks fun. That's the one I have the music comes up.

2:30:53 - Leo Laporte
Oh, that's the one I have, sorry 2019.

2:30:55 - Jeff Jarvis
Yeah, Is there a new one this year? The heartbeat gets even faster.

2:31:01 - Leo Laporte
Oh no, please tell me no. Oh, this is very Sherman here. They need some.

2:31:12 - Jeff Jarvis
Sherman, now Like never before. Oh my god, the music. This is the interstellar. Service like never before. The only bad thing that Thermomix, besides the fact that it's $1500 Is there's a monthly subscription, intuitive like never before. Behind Rook, I forget what that is Like before. Inspired, Inspiring as never before. There it is.

2:31:42 - Leo Laporte
Leo, I don't need a new one but oh, it does look better.

2:31:48 - Jeff Jarvis
Look at that tablet. Oh, it is Leo. Oh, it is better Look the peppers fly Experience like never before.

2:31:55 - Leo Laporte
Look at the kiwis fly.

2:31:56 - Jeff Jarvis
Look at the raspberries fly Raspberries, the onions, not together.

2:31:59 - Leo Laporte
So really it's a blender with an even pot Quieter than ever before.

2:32:04 - Jeff Jarvis
Jeff, I just want to watch.

2:32:06 - Paris Martineau
you translate German videos all day More powerful than ever before. Einfach mehr. So, wie wie zuvor.

2:32:15 - Leo Laporte
Jawohl, dann sich was und wie so einzeugt? Kein Wobbix, kein Wobbix.

2:32:24 - Jeff Jarvis
Kein Wobbix.

2:32:25 - Leo Laporte
Oh, it does look prettier, so wie wie zuvor. Oh, it is a lot prettier than TM7.

2:32:29 - Jeff Jarvis
Kein.

2:32:29 - Leo Laporte
Wobbix TM7. Ja, oh, it does look prettier, oh, it is a lot prettier. Yeah, damn Zeven. I didn't know, you spoke German. Oh yeah, that's his thing, man.

2:32:38 - Paris Martineau
I mean makes sense given all that I know about it. What is?

2:32:42 - Jeff Jarvis
a Thermomix oh it's Leo's little.

2:32:47 - Leo Laporte
You know what Stacy got me to get one.

2:32:49 - Paris Martineau
Hey, you could get this new one and then give your old Thermomix to me.

2:32:54 - Leo Laporte
That's what Stacey wanted, that's a good idea, but she's gone now, so you get it. That's a very good idea.

I don't mind that so it's a food processor kind of, but really what it is, it's a blender with a heated thing and then it's got inside. It has a scale, so what it does. And then it's got this Android recipe tablet. So you pick a recipe and then it says, okay, put you know 600 grams of potatoes in and it tells you how you know how many grams you've been you've put in. And then do this and do that. It heats it, heats it and cooks it. It makes an incredible risotto, incredible mashed potatoes, incredible soup.

Yeah, I've never made bread in it. I guess you could. I mean, yeah, I guess you could, interesting. It's a ridiculous kitchen appliance, but I tell you what most of my kitchen appliances have been relegated from the counters because lisa likes kind of a clean lines, that one's on the counter because we use it a lot.

2:33:58 - Jeff Jarvis
So the I found this because dite online did a story. Uh, the headline is the redeemer. The new thermomix steams all the world problems.

2:34:08 - Paris Martineau
I love the drama.

2:34:10 - Jeff Jarvis
I'm reading the Google translation for ease right now. Since life is often an unfortunate affair, mankind has decided to cook something nice for itself from time to time. Over the centuries, the machinery in the kitchens of this delightful planet has grown quite enthusiastically, even if people hardly have time to cook.

2:34:26 - Leo Laporte
You can make yogurt with it. I mean, there's a lot you can do with it here. Here is an example of a poor housewife, house flower, who is doing just all the work and then she disappears you don't need the housewife anymore. You've got yogurt, making yogurt with it that's cool, that'd be good.

2:34:50 - Jeff Jarvis
We're on the food. Uh, I haven't. I couldn't listen to this because I discovered it while we're on the show. But the next line, 145, it looks funny. I read, I read it the new australia. Australian lamb commercials are renowned for being funny, for the okay here we go Australian lamb. You want to start back at the beginning, of course.

2:35:11 - Leo Laporte
This is for 2025.

2:35:13 - Jeff Jarvis
Is the sound on?

2:35:14 - Leo Laporte
No, nine. All right, now everybody get ready. We're going to do this.

2:35:19 - Richard Feynman
What do you reckon?

2:35:24 - Paris Martineau
I don't know, it's a dog, arguably.

2:35:26 - Leo Laporte
Oh my God, that's the funniest thing I've seen all week the comment section. Who would do?

2:35:32 - Jeff Jarvis
that to a dog that looks like a dog I drew in grade two, Gina Smith this is literally you. It's literally me.

2:35:38 - Ethan Sutin
Everyone calm down, it's fake.

2:35:42 - Leo Laporte
So it's like if the whole state were real life is the comment section. This film was made with 100% real this comment section is getting toxic.

2:35:49 - Speaker 2
Made was made 100 real comment section toxic do you think this is happening everywhere?

2:35:57 - Leo Laporte
madness we're killing the environment we're saving the environment they only work when there's enough wind.

2:36:04 - Maria de Lourdes Zollo
I made five thousand dollars in a week working from home with this one simple trick. I love renewable. Co2 is perhaps the near lowest in Earth's history.

2:36:13 - Jeff Jarvis
You're right. Next thing, you'll tell me the Earth is flat.

2:36:17 - Paris Martineau
See Flat Earth, another in a spaceship that could be AI. No, that's.

2:36:22 - Jeff Jarvis
AI, it's achievable. Naturally, I have a body similar to that.

2:36:27 - Richard Feynman
Looking good, Koshy.

2:36:34 - Paris Martineau
I made $5,000 in a week. That's the way you make the perfect cuppa Milk before tea.

2:36:37 - Leo Laporte
This is a crime, Same as with cereals. Milk first. She only has a handful of commenters by the way, it's not real.

2:36:41 - Richard Feynman
Oh, we're fighting over tea now.

2:36:43 - Maria de Lourdes Zollo
Could be worse. How.

2:36:51 - Jeff Jarvis
She broke dancing, that's the famous Australian break dance If Prince could tell you this it'd be a lethal dose. Ooh, another keyboard warrior, gina Smith literally not you. So not me. It's fake. What is everyone hating it's fake, she sucks you suck.

2:37:08 - Paris Martineau
Booked and reported.

2:37:09 - Speaker 2
Booked and reported, Booked and reported World tour please. It's fake Everyone relax, it's not real.

2:37:17 - Paris Martineau
It's all democracy. What's next?

2:37:19 - Leo Laporte
What does this have to do with lamb? You'll get there.

2:37:22 - Paris Martineau
How long of an ad is this?

2:37:25 - Jeff Jarvis
It's one of those ones that's so special they do one a year. There's got to be something that unites us. You know when this is coming.

2:37:32 - Leo Laporte
Yeah, he's got a little lamb on the babby.

2:37:37 - Paris Martineau
Oh, that does look pretty good.

2:37:38 - Jeff Jarvis
It does look good, doesn't it? Lamb chop on the babby as if a torch to civilization. Here comes one honest man with a lamb chop A bite. Thumbs up.

2:37:55 - Speaker 2
Come on, everyone, get out of the comments and into the cutlets.

2:37:59 - Paris Martineau
Oh, that actually was worth it that one line did sit.

2:38:01 - Leo Laporte
Get out of the comments and into the cutlets.

2:38:03 - Speaker 3
I have such a ditsy daddy. I've been a big old gritty guy. Now this is real.

2:38:08 - Speaker 2
I'm Gina Smith. It's literally you. Oh my God, it's me, Gina Smith.

2:38:12 - Paris Martineau
Wait, why is everyone being so nice? I think everyone's a lot nicer in real life than we are online.

2:38:17 - Leo Laporte
I guess the anonymousness of the internet promotes disagreements and the platforms that we use. Yeah, we get it, just ate some lamb.

2:38:25 - Jeff Jarvis
That's the best part, isn't that great? That's the best part, isn't that great.

2:38:32 - Leo Laporte
That's the best part. I hope people post nice comments about this ad. And he looks straight at you, straight at you. He's still looking at me. Stop looking at me. I'm not going to post, stop. I told you Stop, no, don't, no, go away. I thank you for joining us for this bizarre edition of intelligent machines. Special thanks to our guests from bcomputer. Great fun having maria, lourdes, de lourdes, zoyo and uh, ethan, uh, oh god, I'm never gonna get their name.

2:39:05 - Jeff Jarvis
It's on the rundown. It's on the rundown, it's right there.

2:39:08 - Leo Laporte
Right, sutton, he likes sutton, sutton, sutton, sutton, ethan sutton and maria de loret, zoyo. Uh, from b computer. Really fun having them on. I can't wait to hear what the bee has to say about it. Uh, thanks to all of you for watching. A special thanks, of course, to paris martineau, weekend reporter for the informationcom. You must subscribe. You'll find her on signal if you've got a tip, martineau.01. Thank you.

2:39:34 - Paris Martineau
Now you can go off and enjoy your biscuit I've already consumed it oh I will be enjoying my chicken thighs though did you make some more?

2:39:43 - Leo Laporte
I have I made some.

2:39:45 - Paris Martineau
I made some last night, been nibbling all day.

2:39:47 - Leo Laporte
It's a delight uh, thanks to jeff jarvis, emeritus professor of journalistic innovation at the craig, at the craig newmark school, graduate school at the city university now at montclair state university and suny stony brook we have a new president as of today, who is really interesting.

2:40:12 - Jeff Jarvis
Oh good, she's now the Dean of Engineering at Princeton, but she's in the Marconi Hall of Fame for the work she's done on Wi-Fi and telecommunication.

2:40:23 - Leo Laporte
And she's now the president of SUNY Stony Brook. Yes, how interesting we should get her on the show. Thank you, jeff. That's a way for me to suck up. Yeah, I like that, yeah, oh, she's great. She's great. Jeff is the author of the Web. We Weave Gutenberg, parenthesis and many other fine volumes available all at jeffjarviscom. But they're real books. It's not like self-published.

2:40:48 - Jeff Jarvis
He actually has a publisher and you get it from the publisher, it just has links to it, and, and this is now out in paperback, and I just found out that I'm going to be able to do an audiobook version of the magazine. Oh good, that'll be really good that'll be really good, because it's in your voice person. Yeah, it's about your story yeah, excellent, I'm happy thank you all for joining us.

2:41:09 - Leo Laporte
We do Intelligent Machines Wednesday afternoon 2 pm Pacific, 5 pm Eastern, 2200 UTC. You can watch live if you want on YouTube, twitch, tiktok, xcom, linkedin, facebook, kik but you don't need to watch live. The live version will have all of the unexpurgated AI insights from my phone. The edited version will all be blurred out Apparently.

I revealed far too much. I didn't realize at the time, so that's why you might want to watch live. Of course, it's easy to watch after the fact. We have copies of the show at our website, twittv. There is an Intell machines uh channel um on youtube so you can share clips if you wish. Uh, and, of course, the best way to listen or watch is to get the audio or the video in your favorite podcast player. Subscribe and you'll get it automatically. Uh, just in time for th morning. Thank you everybody for being here. Thank you, paris, thank you Jeff. Have a great week. We'll see you next time on Intelligent Machines.