Huffin' Hazelnut - NotebookLM, Google's Monopoly, Tesla Recall

Huffin' Hazelnut - NotebookLM, Google's Monopoly, Tesla Recall

Show Video

Leo Laporte: It's time for twig this week in Google. Leo Laporte: Paris Martino is here, jeff Jarvis is here. Leo Laporte: We've got a special guest, stephen B Leo Laporte: Johnson. Leo Laporte: You may know him from his PBS television Leo Laporte: show, his books and podcasts, but he's also Leo Laporte: the guy who helped Google design a new tool Leo Laporte: for writers called notebook. Leo Laporte: Lm will get the inside details. Leo Laporte: And then I will admit that I was bamboozled, Leo Laporte: horn swoggled, fooled, if you will, by the Leo Laporte: Gemini demo.

Leo Laporte: Yes, I have to say it, paris was right. Leo Laporte: That's all next on this week in Google Leo Laporte: Podcasts you love from people you trust. Leo Laporte: This is twig. Leo Laporte: This is twig this week in Google, episode Leo Laporte: 746, recorded Wednesday, december 13th 2023. Leo Laporte: Huffin Hazelnut this week in Google is Leo Laporte: brought to you by Discourse, the online Leo Laporte: home for your community.

Leo Laporte: Discourse makes it easy to have meaningful Leo Laporte: conversations and collaborate anytime, Leo Laporte: anywhere. Leo Laporte: Visit discourseorg Twitter and get one Leo Laporte: month free on all self-serve plans. Leo Laporte: And by Fastmail, reclaim your privacy, Leo Laporte: boost productivity and make email yours Leo Laporte: with fastmail. Leo Laporte: Try it now free for 30 days at fastmailcom Leo Laporte: Twitter. Leo Laporte: It's time for twig this weekend.

Leo Laporte: Well, actually, this week it's actually in Leo Laporte: Google. Leo Laporte: The show we cover Google news, internet Leo Laporte: news, media, journalism, everything on our Leo Laporte: minds. Leo Laporte: Paris Martin knows here from the Leo Laporte: information. Leo Laporte: Hello, paris. Paris Martineau: I think this week we could have the most Paris Martineau: Google we've ever had.

Leo Laporte: It's it's a new record possible the most Leo Laporte: ever huge Wow, all Google, all the time. Paris Martineau: Paris name, or something. Leo Laporte: Is there in the lower third. Leo Laporte: If you have a scoop still working, that Leo Laporte: story that you got from the last scoop. Paris Martineau: Yes, definitely send me tips.

Paris Martineau: Send her don't reach out just to tell me Paris Martineau: that Computers have advanced a lot since Paris Martineau: the 70s. Paris Martineau: That is not. Leo Laporte: Do that? Leo Laporte: Does somebody do that? Leo Laporte: Listen somebody our age, Leo you know, I Leo Laporte: mean you whippersnapper.

Paris Martineau: I've heard that these computer things are a Paris Martineau: bit more advanced than they used to make Paris Martineau: them. Leo Laporte: Look at that also with us the fabulous Leo Laporte: professor Leonard Tao. Leo Laporte: Professor for one, but one more day.

Jeff Jarvis: To you know, two more days. Leo Laporte: A nice town professor for journalistic Leo Laporte: innovation at the fabulous Craig Newmark Leo Laporte: Graduate School of Journalism, new Mars Leo Laporte: City University. Leo Laporte: Necromancer, retire the Craig Newmark Leo Laporte: singers. Leo Laporte: No, we can't do that. Leo Laporte: So you got to go to work for some other Leo Laporte: Craig Newmark joint.

Leo Laporte: Yes, okay. Leo Laporte: Make it like on a new marks friend friend Leo Laporte: of Craig Craig Craig, new mark, new mark. Leo Laporte: He's also the author the Gutenberg Leo Laporte: parenthesis. Leo Laporte: A Gutenberg parenthesis comm in his newest Leo Laporte: book magazine Scored rave reviews from Leo Laporte: Paris Martin oh, who has in fact written Leo Laporte: for a magazine, so she should a great book Leo Laporte: yeah it's adorable and portable. Paris Martineau: Adorable and portable.

Leo Laporte: That's what my parents used to call me. Leo Laporte: We have a special guest. Leo Laporte: Thanks to mr Jarvis. Leo Laporte: Jeff, do you want to introduce Steven oh? Jeff Jarvis: Steven B Johnson, famed author of what 13 Jeff Jarvis: books, even 13 books 14 books. Jeff Jarvis: Steven like a way back when Steven founded Jeff Jarvis: feed oh, and then it got.

Jeff Jarvis: It got mixed in with plastic. Jeff Jarvis: I was on the board of plastic because my Jeff Jarvis: employer at the time, the new houses, Jeff Jarvis: invested in it. Jeff Jarvis: Don't blame me for the fact that it doesn't Jeff Jarvis: exist anymore.

Jeff Jarvis: Nor, steven. Jeff Jarvis: It was a nice try in the early days of the Jeff Jarvis: internet. Jeff Jarvis: And then Steven's been doing all kinds of Jeff Jarvis: fascinating things and Then got hired by Jeff Jarvis: Google as editorial director of notebook LM, Jeff Jarvis: which I'm just fascinated that that title Jeff Jarvis: exists anywhere within 40 miles of Mountain Jeff Jarvis: View. Jeff Jarvis: So I visited Steven, as you all know from Jeff Jarvis: the show, and we talked about notebook LM Jeff Jarvis: before, but I couldn't really talk about it Jeff Jarvis: completely. Jeff Jarvis: But it came out publicly on Friday. Jeff Jarvis: So Steven said he would come on the show Jeff Jarvis: when it was out.

Leo Laporte: So here he is. Leo Laporte: You've seen Steven perhaps on the PBS Leo Laporte: series extra life and how we got to now. Leo Laporte: He is, in fact, the most famous person Leo Laporte: we've ever had on this show, so we are all Leo Laporte: thrilled to have you, steven. Leo Laporte: Thank you. Leo Laporte: And so you're not a coder, you're editorial Leo Laporte: director. Leo Laporte: Yeah it's.

Steven Johnson: It's a bespoke position that I had to kind Steven Johnson: of help invent To describe what I was doing Steven Johnson: there. Steven Johnson: So, you know, the the backstory of this is Steven Johnson: basically that I had been interested in, Steven Johnson: you know, using software to help with the Steven Johnson: writing and research process for writing my Steven Johnson: books and everything else For years. Steven Johnson: I mean it predates when I met Jeff. Steven Johnson: Like, I mean, I was kind of got obsessed Steven Johnson: with this stuff when I was in college in a Steven Johnson: way when, when the old Apple app hypercard Steven Johnson: came out for the Mac in like 1988 for the Steven Johnson: old-timers may remember this and so, as Steven Johnson: I've been kind of writing the books, I Steven Johnson: always had a little bit of a side hustle in Steven Johnson: writing about the tools I was using to Steven Johnson: write the books, and so I'd I've written a Steven Johnson: lot about this tool I used called Devon Steven Johnson: think to organize all my books from and a Steven Johnson: lot of us are, yeah, a bunch of books with Steven Johnson: and so there was this kind of separate Steven Johnson: interest in tools for thought, as Howard Steven Johnson: Ryan Gold famously called them years ago, Steven Johnson: and so when, at some point in in the late Steven Johnson: spring of 2022, I got it.

Leo Laporte: Okay, you're using a Macintosh and you did Leo Laporte: a thumbs up and in the yeah, give us a Leo Laporte: thought. Leo Laporte: Bubble of thumbs up. Leo Laporte: Case people think wow, he's magic, that's.

Leo Laporte: That's what's going on there. Paris Martineau: No that's. Leo Laporte: That's a Mac, I think, unless it's also a Leo Laporte: zoom right. Steven Johnson: I'll try to keep my gestures to Just Steven Johnson: whatever you do, don't do the.

Steven Johnson: What if you do this? Leo Laporte: Yeah, rock that you can't do that gesture, Leo Laporte: jeff, we know that. Steven Johnson: So I got, I got this or this. Steven Johnson: I Can just do gestures all the time I Got. Steven Johnson: I got kind of a cold email from One of the Steven Johnson: guy who was at Google at the time, named Steven Johnson: clay before, who had just started Google Steven Johnson: labs, which is kind of have been rebooted Steven Johnson: inside of Google is the new version of labs. Steven Johnson: There was an old before.

Steven Johnson: And he basically was like look, you know, Steven Johnson: been reading your work over the years and Steven Johnson: seeing some of the stuff you've done about Steven Johnson: tools for thought, you know, with these new Steven Johnson: language models, like we can really Build Steven Johnson: this thing that you've been kind of Steven Johnson: dreaming about for your whole life. Steven Johnson: Like you, you know it's now possible in a Steven Johnson: way that just wasn't possible before. Steven Johnson: And you know, would you like to come, come Steven Johnson: into labs and and take a part-time role? Steven Johnson: Initially, I think we called it visiting Steven Johnson: scholar is what we came up with.

Steven Johnson: But we they had a little team inside of Steven Johnson: labs where they were kind of starting to Steven Johnson: spitball ideas about it, kind of a writing Steven Johnson: research tool that would help you think and Steven Johnson: augment your understanding the world and Steven Johnson: maybe help you write books down the line. Steven Johnson: And so they brought me in as kind of like a Steven Johnson: in-house user, basically like a lead user Steven Johnson: almost for the, for the product, and we Steven Johnson: just started experimenting and then it just, Steven Johnson: you know, we built an early prototype. Steven Johnson: That Was pretty cool before the thing you Steven Johnson: saw, jeff, like in just in a couple of Steven Johnson: months. Steven Johnson: And then Suddenly there was just all this Steven Johnson: wave of interest in what can we do with Steven Johnson: these language models? Steven Johnson: And we had a kind of a genuinely new Steven Johnson: product that we were trying to make like a Steven Johnson: new software kind of experience. Steven Johnson: It was built natively around language Steven Johnson: models, and so it got a little bit of Steven Johnson: momentum and we announced it on. Steven Johnson: IO right, yeah, I got into IO.

Steven Johnson: Josh Woodward, who now runs labs, demoed it Steven Johnson: on stage at IO and we got a bunch of Steven Johnson: attention from that and we built up. Steven Johnson: We're still it's still like really feels Steven Johnson: like a startup, like there's there's only Steven Johnson: nine people full-time on the team I think, Steven Johnson: jeff, you met most of them when you came by Steven Johnson: and it's it's really been an amazing trip Steven Johnson: and we and we managed to get this thing to Steven Johnson: A general release to a US audience Just Steven Johnson: just on Friday. Leo Laporte: So it must have been.

Leo Laporte: I mean, given that you I've used Devon, Leo Laporte: think I've used Scrivener I those tools are Leo Laporte: really designed for you to take notes, to Leo Laporte: collate notes. Leo Laporte: In the case of Scrivener, you can actually Leo Laporte: write inside Scrivener next to your notes, Leo Laporte: and so it's a it's kind of a natural way to Leo Laporte: work. Leo Laporte: But it must have been a treat for you to Leo Laporte: say what if we start? Leo Laporte: And did you start from ground zero and and Leo Laporte: with a blank sheet and say what would you Leo Laporte: ideally want? Leo Laporte: Is that? Leo Laporte: That is, that, were you able to start at Leo Laporte: that point? Steven Johnson: Yeah, it was. Steven Johnson: It was starting with that kind of premise Steven Johnson: that there was going to be an AI Involved Steven Johnson: in almost everything you were doing right Steven Johnson: in this out, and an AI that was grounded in Steven Johnson: the source material that you gave it.

Steven Johnson: That was. Steven Johnson: You know, we now technically in the Steven Johnson: industry would call this rag, which I'm not Steven Johnson: crazy about as an acronym, but so we call Steven Johnson: it source grounding, right? Steven Johnson: You? Steven Johnson: You define a set of trustworthy sources, Steven Johnson: documents, and you basically say to the Steven Johnson: model I want you to answer questions and Steven Johnson: help me understand this material and I want Steven Johnson: you to stick to the facts of this material, Steven Johnson: and that we knew that that was going to be Steven Johnson: central to it. Steven Johnson: So when I, when I got to Google, there was Steven Johnson: a early-stage project, the brilliant Steven Johnson: engineer and in madam big now was working Steven Johnson: on it called talk to a small corpus. Steven Johnson: That was its original name, actually. Steven Johnson: It was corpse as I feel, yeah, talk to an Steven Johnson: exquisite corpse, but Took off from there. Steven Johnson: And so we had, at the very beginning, this Steven Johnson: idea that there was going to be An AI that Steven Johnson: you could have some kind of conversation Steven Johnson: with, that would have a Kind of something Steven Johnson: like expertise in the information that you Steven Johnson: gave it, and that it was.

Steven Johnson: It was really cool because then you could Steven Johnson: kind of build something genuinely new that, Steven Johnson: like you know, we didn't. Steven Johnson: It didn't need necessarily to look like a Steven Johnson: straight chat interface, it didn't need to Steven Johnson: necessarily look like a word processor, it Steven Johnson: didn't. Steven Johnson: You know, it could be something different Steven Johnson: from all that, and that's, that's how it Steven Johnson: evolved, basically it's Interesting.

Leo Laporte: We've we've observed that that's one of the Leo Laporte: best uses for an AI, because they eliminate Leo Laporte: the hallucinations. Leo Laporte: You eliminate the knowledge gaps. Leo Laporte: You say, just based on what I've given you, Leo Laporte: whether it's a bunch of PDFs or a book, Leo Laporte: tell me about this Thing and that for an Leo Laporte: author working on a project. Leo Laporte: Those are all your notes, right? Leo Laporte: That's all the information you've gathered. Leo Laporte: Yeah, exactly.

Steven Johnson: I mean, is it? Leo Laporte: as a product used, prompt, creative writing Leo Laporte: to use it. Leo Laporte: I just simply as a researcher, like kind of Leo Laporte: a little buddy researcher who can look at Leo Laporte: the notes how, how, what do you anticipate Leo Laporte: using it as? Steven Johnson: I, I mean, I'm just just where all of us Steven Johnson: are discovering all these new uses as it, Steven Johnson: as it comes online and and it is now, by Steven Johnson: the way, like Partially running on Gemini Steven Johnson: pro, the new model, and will be a hundred Steven Johnson: percent running on Gemini pro over the next Steven Johnson: couple of weeks probably and, and that Steven Johnson: alone we've seen a big, you know, increase Steven Johnson: in what you can do and the kind of Steven Johnson: dexterity with which it will answer your Steven Johnson: questions and things like that. Steven Johnson: So, for me, I Well, I can show you actually Steven Johnson: why don't we have some videos that you shot Steven Johnson: ahead of time? Leo Laporte: Which one is? Steven Johnson: something I did this morning. Steven Johnson: It's my reading history, and this by the Steven Johnson: way, let's, let's just let's even is Steven Johnson: preternaturally organized anyway. Steven Johnson: Yeah, let's. Steven Johnson: Let's hold a whole Often the video, yet Steven Johnson: history cities and what are there.

Steven Johnson: Yeah, let me, let me just said. Steven Johnson: It said, let me explain, like what, what Steven Johnson: you're about to see, actually at a couple Steven Johnson: different levels. Steven Johnson: So, first off, I understand that this is Steven Johnson: somewhat abnormal, although I'm glad to Steven Johnson: hear, as we were discussing before we Steven Johnson: started, leo, that you you have, you know, Steven Johnson: sources that are larger than 200,000 words Steven Johnson: that you would like to get.

Steven Johnson: Yeah, I was, weird too. Leo Laporte: I, as soon as it was, went public, I broke Leo Laporte: notebook Lm by trying to load in my entire Leo Laporte: list bookshelf. Leo Laporte: This is, by the way, something I did do Leo Laporte: with as a custom GPT on a chat GPT and Leo Laporte: found incredibly useful. Leo Laporte: I've been using it ever since. Leo Laporte: When I can't remember syntax for phrase and Leo Laporte: and when you do that with a GPT, you could Leo Laporte: say stick to the facts, don't use anything Leo Laporte: outside the corpus that I've, that I've Leo Laporte: given you. Leo Laporte: So I immediately tried to do the same thing Leo Laporte: with notebook Lm and broke it because I Leo Laporte: didn't know what.

Leo Laporte: There's a limit on this size of any Leo Laporte: individual source. Steven Johnson: Current limits are and you know these will Steven Johnson: presumably go up over time 200,000 words Steven Johnson: per source and you can have 20 sources per Steven Johnson: notebook. Steven Johnson: So you can be Simultaneously four million Steven Johnson: words worth of information inside a single Steven Johnson: notebook. Steven Johnson: So it's quite that's sufficient. Leo Laporte: Yeah, and I think most of the books I Leo Laporte: uploaded were less than 200,000 words, but Leo Laporte: there's a Peter Norvig volume that is about Leo Laporte: this thick and I bet you that's what broke Leo Laporte: it. Steven Johnson: So I have set up this video I have been Steven Johnson: collecting.

Steven Johnson: This is an example of my long-time Tools Steven Johnson: for thought nerdery. Steven Johnson: I have been collecting digital quotes from Steven Johnson: books that I read as part of my research. Leo Laporte: Kind of a zettelkasten. Steven Johnson: Exactly, yeah, and so I have a collection Steven Johnson: that I originally did by hand, actually Steven Johnson: like typing in, you know, from print books Steven Johnson: imported through Kindle, through read wise, Steven Johnson: other quotes from books once we had ebooks, Steven Johnson: and so it's 7,000 quotes. Steven Johnson: It's 1.3 million words.

Steven Johnson: I have that all loaded. Steven Johnson: It's a quarter century of my reading Steven Johnson: history, like the things that I found most Steven Johnson: important to me are all loaded into a Steven Johnson: single notebook, in notebook, which is Steven Johnson: which is pretty, pretty cool and kind of Steven Johnson: the thing I've always just wanted to have, Steven Johnson: right, I just wanted to have that, that Steven Johnson: kind of second brain that is capable of Steven Johnson: reminding me of all these things. Steven Johnson: So we can, we can run this club. Steven Johnson: So this is me. Steven Johnson: I'm not sure if I can actually read on the Steven Johnson: screen there, but I think I remember what I Steven Johnson: did. Jeff Jarvis: This is what authors discuss history of Jeff Jarvis: cities and what are the their main points? Steven Johnson: Okay, yeah, so it comes back.

Steven Johnson: So I'm just asking for, like general lay of Steven Johnson: the land, what are some authors in my Steven Johnson: quotes to talk about history of cities and Steven Johnson: and could you briefly like summarize what Steven Johnson: their main points are? Leo Laporte: So, to be clear, this is not doing it on Leo Laporte: device. Leo Laporte: You're still going on a lot of positive Leo Leo Laporte: the servers and and querying the servers Leo Laporte: and they. Steven Johnson: So the model is on Google's servers, Steven Johnson: Exactly okay so it gives this like lovely Steven Johnson: overview, kind of summarizing each of them, Steven Johnson: you know, explains what's going on, and Steven Johnson: then I do a follow-up question which is Steven Johnson: about I can't read it, jeff, can you help Steven Johnson: me read? Paris Martineau: the most interesting ideas from Mumford Paris Martineau: about city. Steven Johnson: Oh yeah.

Jeff Jarvis: I'm following up in this and four authors Jeff Jarvis: Stephen Mumford, marcus, tom Standage that Jeff Jarvis: we dig down into Mumford and there's four Jeff Jarvis: bullets of his interesting ideas about so Jeff Jarvis: you see, at the bottom it's automatically Jeff Jarvis: suggesting questions based on what's Jeff Jarvis: interesting. Paris Martineau: One of the questions, and that one is what Paris Martineau: biological criteria did Howard bring to the Paris Martineau: city? Steven Johnson: and this is all questions Based on this Steven Johnson: corpus of works, you know and then I found Steven Johnson: something interesting so I pinned it and so Steven Johnson: now I've got this kind of noteboard space Steven Johnson: here where I can capture the things that Steven Johnson: are interesting and that I can See the Steven Johnson: original citations, which is great. Steven Johnson: I'm a book and if I click on the actual Steven Johnson: citation it takes me exactly to that point Steven Johnson: in the original source so that I can read Steven Johnson: it in the surrounding context. Steven Johnson: So that workflow of I'm sitting down to, Steven Johnson: kind of Trying to get a lay of the land Steven Johnson: what's out there, what's interesting, like Steven Johnson: what, what are the ideas I can, I can Steven Johnson: engage in this kind of dialogue with my Steven Johnson: sources. Steven Johnson: It the suggested questions are incredibly Steven Johnson: cool.

Steven Johnson: Like it's such an interesting way to Steven Johnson: explore a new document is to just ride Steven Johnson: those suggested questions for a while and Steven Johnson: just kind of Figure out where they take you. Steven Johnson: It's a great onboarding tool for people as Steven Johnson: well, because a lot of times people don't Steven Johnson: know what to ask, and so we, when when they Steven Johnson: first load up their sources, we suggest a Steven Johnson: few questions that might get them started Steven Johnson: in that dialogue mode which is just show Steven Johnson: you real quickly because I have it a little Steven Johnson: bigger on my screen a toy example. Leo Laporte: I loaded the alphabet quarterly results Leo Laporte: from the last quarter and I typed a Leo Laporte: question what was the revenue for the Leo Laporte: quarter ending September 30th? Leo Laporte: And it answered and gave me some stuff. Leo Laporte: But that these are the additional questions Leo Laporte: what was the percentage chain and Constance Leo Laporte: currency revenues for the quarter? Leo Laporte: And there was there. Leo Laporte: Actually there's a bunch more that are Leo Laporte: going off the screen. Leo Laporte: There's a scroll bar, but so it's giving me Leo Laporte: ideas for further Investigation here.

Leo Laporte: I'll scroll over and you can see. Leo Laporte: And that's really cool. Leo Laporte: And now this is a toy example because it's Leo Laporte: a simple document. Leo Laporte: But let's say I loaded in all the quarterly Leo Laporte: results for the entire history of alphabet, Leo Laporte: you know, multiple years worth. Leo Laporte: That might get more interesting. Steven Johnson: You'll go good if you go back to it, Steven Johnson: actually click on the citation, so it one Steven Johnson: of these.

Steven Johnson: You'll note there. Steven Johnson: It gave you one citation right figured out. Leo Laporte: It knows it's coming from this. Steven Johnson: Yeah, but it knows it's, it's not. Steven Johnson: It's not citing the entire document, it's Steven Johnson: citing the this part.

Steven Johnson: It's specific. Steven Johnson: Yeah, the doc. Steven Johnson: Yeah, and if you click on that number it Steven Johnson: will jump. Steven Johnson: It'll open up.

Steven Johnson: Expand the document. Leo Laporte: Oh look at that. Steven Johnson: Yeah, that's really neat. Steven Johnson: So that that's the stuff that we're really Steven Johnson: excited about. Steven Johnson: And then I don't.

Steven Johnson: I don't believe that the GPT version does Steven Johnson: that kind of. Leo Laporte: So no, it does not, no no, so this is, this Leo Laporte: is from you, steven, because you are. Leo Laporte: They're so smart to get a user who's Leo Laporte: looking for this kind of, you know, Leo Laporte: external brain and, and and talk to you Leo Laporte: about what kinds of features you would Leo Laporte: expect or want or need, and and that's Leo Laporte: where you get these kinds of really nice Leo Laporte: additional features. Jeff Jarvis: So are you working on the next book, steven? Steven Johnson: You're a little busy.

Steven Johnson: Now. Steven Johnson: I have finished the next book, and so I Steven Johnson: sadly I wasn't able to use this I use them Steven Johnson: early. Steven Johnson: Like you know, I kind of would try to Steven Johnson: earlier prototypes, but the problem is my Steven Johnson: next book is that it's about this, the kind Steven Johnson: of the birth of terrorism and the anarchist Steven Johnson: attacks on the NYPD In the teens, and so Steven Johnson: when I would load up all my research it was Steven Johnson: constantly triggering the like safety. Steven Johnson: He's like there is the tax and it was so. Steven Johnson: But I'm boy.

Steven Johnson: Am I really excited to use it? Jeff Jarvis: Yeah, how do you imagine this would change Jeff Jarvis: your process for the next one? Steven Johnson: Well, so here's, here's a little thing. Steven Johnson: I put this on Twitter, actually, so I did a Steven Johnson: little test of of how long it would take me Steven Johnson: to To answer the question through my giant Steven Johnson: corpus of text, you know, find two quotes Steven Johnson: from two authors on two different subjects Steven Johnson: and draw a surprising connection between Steven Johnson: the two of them. Steven Johnson: And you know my reading quotes are spread Steven Johnson: across like 15 different documents, right. Steven Johnson: So the old way I would do that would be Steven Johnson: like to kind of do a general, like Steven Johnson: spotlight, like search or drive, search Steven Johnson: across them, get a bunch of Recommended Steven Johnson: documents, open up each one, search command Steven Johnson: F, you know, to find it exact phrase, look Steven Johnson: for the quote, see if it was right, copy it Steven Johnson: into some other doc and then go back and Steven Johnson: search More or whatever. Steven Johnson: And so when I did that the old-fashioned Steven Johnson: way just earlier this week when I was Steven Johnson: testing it, it took me five minutes and Steven Johnson: eight seconds to generate like two quotes Steven Johnson: from these specific authors with the Steven Johnson: surprising connection between them in Steven Johnson: notebook.

Steven Johnson: It took me 20 seconds. Steven Johnson: And so your ability to, you know, quickly Steven Johnson: Experiment with ideas and quickly dive in Steven Johnson: and get you know a sense of like what's Steven Johnson: possible. Steven Johnson: It reminds me a little bit when Wikipedia Steven Johnson: really started to become useful. Steven Johnson: We were like oh, now I have this way to Steven Johnson: very quickly assess the general state of Steven Johnson: some piece of information that I can dive Steven Johnson: deeper.

Steven Johnson: If I want, I can, I can make connections. Steven Johnson: If I want Notebook, lets you just do that. Leo Laporte: If you have, you know, a good collection of Leo Laporte: sources that are important to you, let's Leo Laporte: you do that incredibly fast one of the Leo Laporte: reasons people do second brains and you Leo Laporte: know Zettle custom is the connections, and Leo Laporte: the idea is, if I get all this stuff into a Leo Laporte: system, the real value of it is then Leo Laporte: Synthesizing new information from this Leo Laporte: connection.

Leo Laporte: But that's done by a human, yours. Leo Laporte: Are you suggesting that the AI can do that Leo Laporte: kind of creative synthesis? Steven Johnson: It's a great question, leo. Steven Johnson: So I mean, I think it's it's a real Steven Johnson: collaboration. Steven Johnson: So, in a sense, like my source collection Steven Johnson: was hand curated and a huge amount of the, Steven Johnson: a Huge amount of the wisdom and knowledge Steven Johnson: that is embedded in that source collection Steven Johnson: is one from the original authors who wrote Steven Johnson: it and two from the Human who assembled it Steven Johnson: right, and that's you know so much of Steven Johnson: what's valuable there. Steven Johnson: You add the AI's ability to just gather Steven Johnson: stuff and summarize stuff very, very, you Steven Johnson: know, at a speed and accuracy far exceeding Steven Johnson: any human ability right now, and it's a Steven Johnson: real kind of collaboration between the Steven Johnson: original authors, between me and and and no Steven Johnson: book. Steven Johnson: I'm in Gemini.

Steven Johnson: I Think that the other thing that you were Steven Johnson: asking earlier about the kind of use case Steven Johnson: for it, the thing that we're we have some Steven Johnson: couple of features that we've announced but Steven Johnson: that are just starting to roll out over the Steven Johnson: next week or two as part of this launch, Steven Johnson: this process that I've been calling like Steven Johnson: curate and create. Steven Johnson: So you use notebook, you load up, let's say, Steven Johnson: your student and you load up a bunch of Steven Johnson: sources for your class that you're working Steven Johnson: on and you can read them in notebook and Steven Johnson: you can kind of grab the things that you Steven Johnson: think are interesting. Steven Johnson: You can ask notebook to help you understand Steven Johnson: the things that you don't understand, and Steven Johnson: you can pin everything you find interesting Steven Johnson: on to that note board that you saw before Steven Johnson: and then, at a certain point, when you've Steven Johnson: collected a bunch of notes that you're Steven Johnson: Thinking is okay, this is the material I Steven Johnson: need to really understand you. Steven Johnson: You can just select all the notes and Steven Johnson: Option that says create a study guide or Steven Johnson: create a thematic outline or, you know, Steven Johnson: create or suggest related ideas for my Steven Johnson: sources.

Steven Johnson: So I can, I can expand my understanding Steven Johnson: this material, and so it's not just about Steven Johnson: like writing a book, it's maybe just like I Steven Johnson: want to do this to learn better. Jeff Jarvis: So so, paris, since your journalist who Jeff Jarvis: works on some things that take time and you Jeff Jarvis: end up with lots of Interviews and Jeff Jarvis: transcripts and documents and stuff what is Jeff Jarvis: this sparking in your head For how you Jeff Jarvis: would work? Jeff Jarvis: I? Paris Martineau: think it sounds fascinating and super Paris Martineau: useful. Paris Martineau: My main concern, which I think we've talked Paris Martineau: about in the show before, is, I think, Paris Martineau: privacy concerns. Paris Martineau: I mean, I think you mentioned that it is Paris Martineau: all. Paris Martineau: It's not. Paris Martineau: It is on the internet and being processed Paris Martineau: by whatever AI tool is powering this.

Paris Martineau: So it's not, I think, for my sort of use Paris Martineau: case, which is very specific just because I Paris Martineau: have sensitive documents and sensitive Paris Martineau: sources. Paris Martineau: The only time that I can really let my Most Paris Martineau: private information leave my actual Paris Martineau: computer is, you know, I try to. Paris Martineau: I try to reduce the amount of times that Paris Martineau: happens generally, so it's probably not Paris Martineau: Particularly useful for me in this case, Paris Martineau: but I'm really excited to see where this Paris Martineau: goes because it seems like a perfect tool Paris Martineau: for that. Leo Laporte: And the future of this is on device, I Leo Laporte: would guess I mean I assume, so yeah.

Leo Laporte: It's not. Leo Laporte: I mean, first of all, I don't think Google Leo Laporte: or Microsoft or a open AI wants us to keep Leo Laporte: using these valuable server resources. Leo Laporte: Certainly not. Leo Laporte: We're sending money on fire for them If Leo Laporte: they can get the model small enough, even Leo Laporte: if you looked at that kind of use of this. Steven Johnson: Not yet.

Steven Johnson: I mean, there's this new nano version of Steven Johnson: they. Steven Johnson: Just yeah, yeah, yeah, I haven't. Steven Johnson: I have played with ultra, which is Steven Johnson: fantastic, but I have not played with nano Steven Johnson: yet, so I don't know, but I would love to Steven Johnson: know.

Steven Johnson: That would be fantastic. Leo Laporte: It's very good read that on a pixel. Leo Laporte: I think they are running it in some of the Leo Laporte: new features.

Leo Laporte: Yeah. Steven Johnson: No, I'm psyched for that. Steven Johnson: But I want to just touch on something in Steven Johnson: the Paris phrase which is really important Steven Johnson: and I think you all probably know this, but Steven Johnson: but maybe it's not as clear to both who Steven Johnson: aren't living green this stuff all the time Steven Johnson: it's it's really important to stress that Steven Johnson: what we are doing with notebook LM is we're Steven Johnson: not training the model on your data. Steven Johnson: So the best way, the best kind of low-tech Steven Johnson: metaphor I think it's that I use with Steven Johnson: people, is like we are dynamically taking Steven Johnson: the information from your documents and Steven Johnson: Kind of shuttling it into the context Steven Johnson: window for the model, into this, basically Steven Johnson: the short-term memory of the model, and Steven Johnson: we're saying answer this question based on Steven Johnson: that.

Steven Johnson: It's ability to answer the question is Steven Johnson: based on its training data. Steven Johnson: But we're just showing it like it's like we Steven Johnson: give it a piece of paper and say, hey, look Steven Johnson: at this piece of paper briefly and answer Steven Johnson: the question, and then the second the Steven Johnson: conversation ends, that information Steven Johnson: disappears. Jeff Jarvis: You're in a skiff? Steven Johnson: Yeah, it's not. Steven Johnson: It's not stored at all and it's not trained. Steven Johnson: So there there, if you. Steven Johnson: Basically the way to think about is if you Steven Johnson: feel comfortable, you know, storing things Steven Johnson: in Google Drive, which you know.

Steven Johnson: Paris for you may may be because of the Steven Johnson: nature of what you do. Steven Johnson: It may be tricky, but for most people, I Steven Johnson: think, feel comfortable about doing that. Steven Johnson: They can feel comfortable using notebook LM Steven Johnson: with that data.

Paris Martineau: Yeah, absolutely so. Leo Laporte: It does not upload it, but it does upload Leo Laporte: to the server, doesn't it? Leo Laporte: Steven it? Leo Laporte: It has to process at the server. Steven Johnson: It's it. Steven Johnson: Yeah, your documents are sitting on the Steven Johnson: server. Steven Johnson: I mean, the most the easiest way to get Steven Johnson: sources into notebook LM is to just upload Steven Johnson: them in Import docs from your drive.

Steven Johnson: That are already, yeah, in your drive, and Steven Johnson: so they're already there. Steven Johnson: But the question is just like it feels as Steven Johnson: if you are getting a personally trained Steven Johnson: model, but in fact the word for it it's Steven Johnson: it's a, it's a different process, it's just Steven Johnson: a curious when is what is the training data Steven Johnson: that this is running on and what kind of Steven Johnson: corpus is that? Steven Johnson: so it's. Steven Johnson: Right now it's running kind of split Steven Johnson: between two models Gemini pro, so it's just Steven Johnson: a the the pro that you would get access to Steven Johnson: using the new Google AI studio tools API Steven Johnson: that they've just released, I believe today Steven Johnson: or yesterday. Steven Johnson: And then for factual questions, there's a Steven Johnson: model that has been kind of Specifically Steven Johnson: tuned to to be accurate and factual in the Steven Johnson: way that it answers the questions and to Steven Johnson: pick out those citations. Steven Johnson: So it's it will.

Steven Johnson: If there are six passages that are relevant Steven Johnson: to your question, it will give you those Steven Johnson: six. Steven Johnson: If there's only one, it'll like in Leo's Steven Johnson: example we just saw it will just give you Steven Johnson: that one. Steven Johnson: And so some of the questions are going to Steven Johnson: Gemini, some of the questions are going to Steven Johnson: those older models. Steven Johnson: Eventually they will just all go to Steven Johnson: variations of Gemini. Leo Laporte: I've been playing with them. Leo Laporte: I've been.

Leo Laporte: I want to do this locally and I have a Leo Laporte: high-end Mac that I, you know, has an NPU Leo Laporte: and can do a lot of this stuff. Leo Laporte: So I've been playing with this GPT for all Leo Laporte: and it's the same idea. Leo Laporte: Whereas you're gonna download a large model Leo Laporte: in this case it's I can't remember, it's 16 Leo Laporte: million or 16 billion Tokens, it's what. Leo Laporte: It's huge you download on your system and Leo Laporte: that's what teaches it to talk, and Then Leo Laporte: you can add a corpus of information that Leo Laporte: you could say I want to work with this. Leo Laporte: You're not drawing facts from the model, or Leo Laporte: are you? Steven Johnson: We it's interesting if you go into a A, if Steven Johnson: you, you know, load up a bunch of sources Steven Johnson: and then ask for a list of Taylor Swift Steven Johnson: song, right, in general We'll say I'm sorry, Steven Johnson: I can't answer that question because the Steven Johnson: information is not in the sources You've Steven Johnson: given us that's the baby. Steven Johnson: Yeah, we have tried to kind of elicit from Steven Johnson: the model every now and then it it feels Steven Johnson: like it really was, but that information is Steven Johnson: in its training data.

Steven Johnson: But we have set up, we've designed the Steven Johnson: prompts, we've designed the tuning that, Steven Johnson: the All that that kind of stuff has been to Steven Johnson: kind of create little guardrails. Steven Johnson: I think there, you know, there's certainly Steven Johnson: a Future version of it that you can imagine, Steven Johnson: where you sometimes want to talk to a Steven Johnson: general model, right, and then there's want Steven Johnson: to keep it grounded and and one of the Steven Johnson: things you'll note, actually, leo, that is Steven Johnson: pretty cool as well, as if you load up a Steven Johnson: lot of sources in a notebook, you can just Steven Johnson: dynamically be like, actually, briefly, I Steven Johnson: would just like to talk to this one and you Steven Johnson: select interesting I love that and then or Steven Johnson: you can select them all, whatever Steven Johnson: combination you want, and I think this is Steven Johnson: this is a really Big point. Steven Johnson: I think about what we're trying to do with Steven Johnson: notebook I am is they're gonna be there. Steven Johnson: Already are a lot of ways to Do source Steven Johnson: grounded AI, where you give an AI some Steven Johnson: sources and you can chat with them. Steven Johnson: Well, we're trying to say is there's a Steven Johnson: whole user experience that goes way beyond Steven Johnson: just a straight chat experience, and it's Steven Johnson: things like suggested questions.

Steven Johnson: It's things like being able to save and Steven Johnson: post the things. Steven Johnson: It's it's that curate and create idea, Steven Johnson: where you can collect a bunch of notes and Steven Johnson: then Transform them instantly into other Steven Johnson: forms. Steven Johnson: There's a whole, there's a whole surface Steven Johnson: that you're working in that is really so Steven Johnson: much more interesting, I think, than just a Steven Johnson: straight chat conversation. Steven Johnson: And so you know there's gonna be a lot of Steven Johnson: source grounded AI, but I think you're Steven Johnson: gonna want to spend time in a more advanced Steven Johnson: you know user experience, like notebook Steven Johnson: gives you interesting. Jeff Jarvis: It it when it to your question, leo.

Jeff Jarvis: When I find interesting and this is the Jeff Jarvis: same in playing with Gemini is when I'm Jeff Jarvis: asking something very specific about the Jeff Jarvis: document I have. Jeff Jarvis: It's very good when I ask a more general Jeff Jarvis: question and it has to draw on more general Jeff Jarvis: knowledge in a way. Jeff Jarvis: That's where it gets a more generic little Jeff Jarvis: fluffier.

Steven Johnson: Yeah, right. Steven Johnson: Yeah, I think that's that's right and that Steven Johnson: is Maybe the right. Steven Johnson: I don't know. Steven Johnson: Like do you? Steven Johnson: I kind of want to be able to.

Steven Johnson: I want the model, when it is speaking Steven Johnson: generally, to be a generalist, and if it's Steven Johnson: a little vague, that seems fine, because Steven Johnson: when it gets very precise, then you start Steven Johnson: to run into risk. Steven Johnson: Yeah, yeah, that's true, that's true. Steven Johnson: The other thing. Steven Johnson: I'll show you one other thing. Steven Johnson: That, because we've been talking about this Steven Johnson: in a very kind of high brow, you know how Steven Johnson: does an author with 7,000 quotations Steven Johnson: Another really fun example that we have Steven Johnson: that we.

Steven Johnson: That occurred to us like with a week to go. Steven Johnson: Which was what, if we just take all the Steven Johnson: help documents that we've created and a Steven Johnson: bunch of like how-to documents that I've Steven Johnson: written, and Create a notebook based on Steven Johnson: that? Steven Johnson: And so we uploaded those into a notebook. Steven Johnson: That's there, it's an example notebook you Steven Johnson: get when you sign up for notebook LM and If Steven Johnson: there's sources on the laugh, that are Steven Johnson: basically the help documents, and it turns Steven Johnson: out that, like notebook Learns not just the Steven Johnson: facts of how notebook works but it also Steven Johnson: really learns how, how the software works Steven Johnson: on some level, and so you can ask some kind Steven Johnson: of basic factual questions. Steven Johnson: Tell me what is clicking on there, because Steven Johnson: I can't read it. Jeff Jarvis: One of the best practices for maximizing Jeff Jarvis: the utility of transcripts and notebook LM. Steven Johnson: Yeah, so it's wonderful with transcripts, Steven Johnson: right, I mean it's it's really good at kind Steven Johnson: of analyzing conversations, and so this Steven Johnson: gives you this kind of detailed Steven Johnson: step-by-step you, which is kind of loosely Steven Johnson: modeled on something that I Wrote that was Steven Johnson: in the help docs.

Steven Johnson: But then you can, you can do follow-up Steven Johnson: questions. Steven Johnson: This is basic factual information that you Steven Johnson: would imagine it would be very good at. Steven Johnson: You know how many notes are in a notebook, Steven Johnson: and and then I think there's a word count Steven Johnson: question that follows after this.

Steven Johnson: But what, what kind of blew me away is that Steven Johnson: you can also ask questions that are Steven Johnson: freeform, that aren't in the the docs at Steven Johnson: all. Steven Johnson: So you can ask it I'm a lawyer, how do I Steven Johnson: use the software? Steven Johnson: And it actually will come up with this like Steven Johnson: really interesting bespoke answer. Steven Johnson: There's no reference to lawyers in the Steven Johnson: notebook help docs, but it understands Steven Johnson: generally how notebook works. Steven Johnson: It understands generally how lawyers work, Steven Johnson: and so it will create this little Steven Johnson: customized like description of how to you Steven Johnson: know Use no, fucking your law firm, don't Steven Johnson: bug I'll, I'm in your law firm, and then Steven Johnson: even there's a kind of funny one at the end. Steven Johnson: So that's the advice to lawyers there, Steven Johnson: which is pretty good, and then I've done a Steven Johnson: bunch of other tests like this where I ask Steven Johnson: I Think the next question I ask I Like I Steven Johnson: can't.

Steven Johnson: Maybe it stopped there, but I asked them Steven Johnson: Can I use this to cheat at school by Steven Johnson: writing my paper? Steven Johnson: What's interesting about this? Steven Johnson: We don't actually have anything in the help Steven Johnson: docs and say do not use this to cheat. Steven Johnson: It's cool, right. Steven Johnson: But it actually comes back with this very Steven Johnson: nice like no, I can't write your paper for Steven Johnson: you, but you can't your ideas. Leo Laporte: Like it really Summoned that response like Leo Laporte: safety stuff probably put into its main Leo Laporte: Model.

Leo Laporte: That's all that is. Steven Johnson: Mix of the safety. Steven Johnson: I mean you would think it would actually Steven Johnson: just get blocked in a way, but it's exactly Steven Johnson: the response you want. Leo Laporte: Yeah, but it's concerned. Leo Laporte: This concerns me, steven, because that's Leo Laporte: hallucination as well. Leo Laporte: The biggest problem we have with these and Leo Laporte: I'm sure you would as an author is Leo Laporte: non-contrafactual stuff that is presented Leo Laporte: as factual, and I don't want it to go Leo Laporte: outside the corpus.

Leo Laporte: I wanted to stick to the corpus because Leo Laporte: anything it projects from outside the Leo Laporte: corpus that's an example you already said Leo Laporte: it's not in the corpus, it has the Leo Laporte: potential being a hallucination. Leo Laporte: You even have that as a disclaimer In Leo Laporte: notebook LM notebook I'll am. Leo Laporte: They still give inaccurate responses that I Leo Laporte: don't want that. Leo Laporte: I wanted to take it just from the corpus. Steven Johnson: Yeah, that's very interesting.

Steven Johnson: They are, I mean, to me the the level. Steven Johnson: What I want the model to do is to say Steven Johnson: intelligent things based on the facts In Steven Johnson: the corpus. Steven Johnson: So, based on the facts of how notebook has Steven Johnson: been designed and and how it works, I Steven Johnson: wanted to come up with an explanation of it, Steven Johnson: custom, tailored to my needs and, if it's, Steven Johnson: if it hasn't Gotten information about Steven Johnson: lawyers, if it hasn't gotten information Steven Johnson: about students, I wanted to be able to Steven Johnson: craft an explanation that's risky. Steven Johnson: But it's an it. Steven Johnson: Yeah, it's an interesting edge case, for Steven Johnson: sure.

Jeff Jarvis: I just asked it to two questions about my Jeff Jarvis: manuscript and it said your source doesn't Jeff Jarvis: have anything about that. Jeff Jarvis: Huh well, it's right. Leo Laporte: Yeah, so it should do.

Leo Laporte: Okay, good, yeah, yeah, I don't want it to Leo Laporte: project too much. Jeff Jarvis: No, no I understand. Leo Laporte: That's the value. Leo Laporte: We were talking about Zettel cast and Leo Laporte: connections.

Leo Laporte: That's how you synthesize connections. Leo Laporte: But I just it makes me nervous because I Leo Laporte: because if you have to vet this content Leo Laporte: every single time as an author, that's an Leo Laporte: added cognitive load that I think kind of Leo Laporte: Mitigate, you know, it'll mitigate the Leo Laporte: actual usefulness of this. Leo Laporte: I would yeah. Steven Johnson: It's.

Steven Johnson: I mean, there's one potential response you Steven Johnson: could have which would be and we've Steven Johnson: actually tuned some of this behavior into Steven Johnson: the Into notebook, lm which is it could it Steven Johnson: could say I don't have any specific Steven Johnson: information in these sources. Steven Johnson: There you go about how to use it as a Steven Johnson: lawyer, but based on the information that I Steven Johnson: have in here, I would offer this hypothesis Steven Johnson: and and that, and that is probably the Steven Johnson: ideal response. Leo Laporte: I would arguably yeah, then I know it's Leo Laporte: projecting or it's it's attempting the hell.

Steven Johnson: Citations and you can, you know you can Steven Johnson: figure it out, and that's just, it's a Steven Johnson: little bit tricky to do, and so the Steven Johnson: question is like how do you what's the Steven Johnson: what's the right balance as we? Steven Johnson: We develop these things, and I mean this is Steven Johnson: one of the things that is so fascinating Steven Johnson: about this work, is it? Steven Johnson: None of this stuff has ever known? Steven Johnson: No one has ever had to read yeah. Jeff Jarvis: Well, there's. Jeff Jarvis: There's journalistic things too, so there's Jeff Jarvis: a. Jeff Jarvis: I don't we talked about this table when I Jeff Jarvis: was out there, but but City Bureau in Jeff Jarvis: Chicago is a really interesting Jeff Jarvis: journalistic effort that trains citizens to Jeff Jarvis: go and help report other communities, and, Jeff Jarvis: and so New Jersey vindicator, which is a Jeff Jarvis: new New Jersey news outlet, just put out a Jeff Jarvis: call for people to go record the county Jeff Jarvis: board meetings across New Jersey.

Jeff Jarvis: So you can well imagine putting these Jeff Jarvis: transcripts in here and then you can get an Jeff Jarvis: uber view. Jeff Jarvis: I had the same conversation with Texas Jeff Jarvis: Tribune about school boards. Jeff Jarvis: You could then get an uber of what's going Jeff Jarvis: on in those various meetings in a way that Jeff Jarvis: would be just too laborious to do, and Jeff Jarvis: you're gonna have citations, and so you're Jeff Jarvis: still gonna end up writing your stories out Jeff Jarvis: of it differently, but you're gonna be able Jeff Jarvis: to get a view of a corpus of data that Jeff Jarvis: otherwise you couldn't have done.

Steven Johnson: Yeah, I mean, I have a bunch of friends who Steven Johnson: are documentary filmmakers and they you Steven Johnson: know their workflow For their films is they Steven Johnson: make, you know they interview 45 people and Steven Johnson: they have these like hour-long transcripts. Steven Johnson: That's just millions of words. Steven Johnson: And you know the process of like how do we Steven Johnson: figure out. Steven Johnson: You know what we have on this particular Steven Johnson: topic is Exactly the kind of example I was Steven Johnson: giving before.

Steven Johnson: That took me five minutes. Steven Johnson: You know how do you search through a bunch Steven Johnson: of docs. Steven Johnson: You have to search for an exact phrase, all Steven Johnson: this kind of stuff and and no books, Steven Johnson: ability to just be like Okay, what are the Steven Johnson: things that have been said about this Steven Johnson: particular, the trial that happened in 1968? Steven Johnson: And it gives you the summaries and you can Steven Johnson: jump to the citations and you can quickly Steven Johnson: pin those to the board. Steven Johnson: Like all those workflows, like are Steven Johnson: absolutely gonna change and and it's gonna Steven Johnson: be one of those things where like, oh, I've Steven Johnson: been doing it this way and it's been a Steven Johnson: complete pain All these years and I never Steven Johnson: thought of it as a pain because there Steven Johnson: wasn't another way to do it. Steven Johnson: And now that I can do it this other way, it Steven Johnson: just frees you up to actually do the real Steven Johnson: thinking that you want to do. Jeff Jarvis: Problem is I've been good first.

Paris Martineau: Oh, I was just saying. Paris Martineau: I think you're absolutely right. Paris Martineau: I think that the main takeaway we're gonna Paris Martineau: see in the short and medium term from the Paris Martineau: rise of these tools is a total Paris Martineau: reorganization of how we think about Paris Martineau: interacting with large sets of data and how Paris Martineau: we are structuring that in the rest of our Paris Martineau: workflow. Jeff Jarvis: Yeah, no, it's really just amazing my Jeff Jarvis: problem is for the next book I'm working on, Jeff Jarvis: but the line of type, I printed everything Jeff Jarvis: out so I could mark it up, so I could see Jeff Jarvis: what's what Right, and I have PDFs a lot of Jeff Jarvis: these papers that I printed out and Jeff Jarvis: chapters that I printed out.

Jeff Jarvis: But I printed them out and that's where my Jeff Jarvis: notes are, and so it requires me to change Jeff Jarvis: how I work fundamentally For the next one Jeff Jarvis: of the next one where I want them as PDFs Jeff Jarvis: and I want to mark them up there and I want Jeff Jarvis: to, you know, be able to use them digitally. Jeff Jarvis: That's gonna be hard to switch, but I can Jeff Jarvis: also go ahead. Steven Johnson: Yeah, we have Multimodal.

Jeff Jarvis: That's true. Steven Johnson: Yeah, that's true. Steven Johnson: I would hope that we would have some Steven Johnson: solution for you in that field, jeff, as Steven Johnson: you're writing this book. Steven Johnson: Well on your and your documentaries would Steven Johnson: kind of combine it with.

Paris Martineau: Much prefer to know. Paris Martineau: Yeah, don't say the word we don't do it, I Paris Martineau: just, we don't, I just canceled. Leo Laporte: They have they've, they've really screwed Leo Laporte: the pooch which is why I'm curious like Leo Laporte: notebook L, m, I.

Leo Laporte: I just I just saved transcripts from the Leo Laporte: last 10 security now episodes. Leo Laporte: I Think this is, honestly, this is Leo Laporte: fascinating, a fascinating use of this. Leo Laporte: Let me refresh, because I need to be able Leo Laporte: to query it.

Leo Laporte: What's Happening in what's? Leo Laporte: I've noticed? Leo Laporte: By the way, I've learned how to query these Leo Laporte: guys and and you you don't have to be as Leo Laporte: verbose. Leo Laporte: What's happening in ransomware Is a Leo Laporte: perfectly good start. Leo Laporte: Now let's see what it finds. Leo Laporte: This is the last 10 episodes of our Leo Laporte: security show and it's so. Leo Laporte: It process those.

Leo Laporte: We have the transcripts done and as a PDF, Leo Laporte: so I was able to process those. Leo Laporte: How many documents did you say 20? Steven Johnson: sources per note. Leo Laporte: Okay, this is a kind of generic Summary, Leo Laporte: but it does have the citations, which is Leo Laporte: nice. Steven Johnson: I can go yeah sometimes also with those Steven Johnson: Transcripts. Steven Johnson: Sometimes you want to say, like, what are Steven Johnson: people seeing about? Steven Johnson: You know, you're kind of like, want to Steven Johnson: elicit, like in this specific thing, don't Steven Johnson: just summarize the fact that you've Steven Johnson: uncovered by right.

Jeff Jarvis: Yeah, ask, ask about Steve's opinion about Jeff Jarvis: something, because it's Steve. Leo Laporte: I'm already. Leo Laporte: I Think your citation system works really Leo Laporte: well because it has already given me kind Leo Laporte: of a rabbit hole to go down. Leo Laporte: But yeah, let's say let's say what does Leo Laporte: Steve say about? Leo Laporte: So this is just gonna give me quotes though Leo Laporte: credential stuffing. Steven Johnson: Let's say Okay, let's see here, that's a Steven Johnson: good good example. Leo Laporte: Yeah, I think I honestly think this is Kind Leo Laporte: of the better use of AI.

Leo Laporte: Is this kind of assistant Rather than you Leo Laporte: know? Leo Laporte: Just okay, there we go. Leo Laporte: He says a serious threat, points, talks Leo Laporte: about 23 and me here's the citations, Leo Laporte: that's the number of accounts are Leo Laporte: compromised. Leo Laporte: This is really great, I have to say. Leo Laporte: This is hugely valuable and people will Leo Laporte: listen to his show. Leo Laporte: The only problem is he's got eight, nine Leo Laporte: hundred thirty nine shows, so Be able to Leo Laporte: handle more documents. Steven Johnson: I want to say about this, definitely other Steven Johnson: terms of the use of it and and one thing Steven Johnson: that we've done that's kind of subtle.

Steven Johnson: The people remain, notice and maybe people Steven Johnson: won't like is,

2023-12-21 20:06

Show Video

Other news