This Week in Startups - AI-Powered "Tools for Thought": Exploring NotebookLM with Steven Berlin Johnson | E1869

Episode Date: December 20, 2023

This Week in Startups is brought to you by… DevSquad. Most dev agencies only offer developers. Why? Because product management is hard. Get an entire product team for the cost of one US developer plus 10% off at http://devsquad.com/twist LinkedIn Marketing. To redeem a $100 LinkedIn ad credit and launch your first campaign, go to linkedin.com/thisweekinstartups Fitbod. Tired of doing the same workouts at the gym? Fitbod will build you personalized workouts that help you progress with every set. Get 25% off your subscription or try out the app for FREE when you sign up now at fitbod.me/TWIST * Today’s show: Steven Berlin Johnson joins Jason to demo Google’s NotebookLM, a new AI-powered research assistant he helped create at Google Labs. They dive into Steven’s journey to Google Labs (1:28), NotebookLM's unique features (14:09), the rights of authors in the digital age (40:22), and more! * Timestamps: (0:00) Author Steven Berlin Johnson joins Jason (1:28) Early experiences with the internet and Google’s Project Starline (8:01) DevSquad - Get an entire product team for the cost of one US developer plus 10% off at http://devsquad.com/twist (9:00) The significance of document organization, the concept and features of "sources" in NotebookLM, and the impact on writing (14:09) Steven demos Google NotebookLM (28:06) LinkedIn Marketing - Get a $100 LinkedIn ad credit at https://linkedin.com/thisweekinstartups (29:36) Steven’s NotebookLM demo continued (33:47) Jason showcases how he utilizes NotebookLM for TWiST's Business Breakdowns segment (38:56) Fitbod - Get 25% off at https://fitbod.me/twist (40:22) The rights of authors in the digital age (52:01) Superintelligence and the trajectory of language models * Links: https://stevenberlinjohnson.com/writing-at-the-speed-of-thought-21dfb7f689e4https://stevenberlinjohnson.com/good-ideas-the-four-minute-version-7e7856e69621https://www.wired.com/story/googles-notebooklm-ai-ultimate-writing-assistant/ * Follow Steven: https://twitter.com/stevenbjohnson Check out Steven’s website: https://stevenberlinjohnson.com/ Check out Steven’s podcast: https://podcasts.apple.com/us/podcast/american-innovations/id1370092284 • Follow Jason: X: https://twitter.com/jason Instagram: https://www.instagram.com/jason LinkedIn: https://www.linkedin.com/in/jasoncalacanis * Great 2023 interviews: Steve Huffman, Brian Chesky, Aaron Levie, Sophia Amoruso, Reid Hoffman, Frank Slootman, Billy McFarland * Check out Jason’s suite of newsletters: https://substack.com/@calacanis * Follow TWiST: Substack: https://twistartups.substack.com Twitter: https://twitter.com/TWiStartups YouTube: https://www.youtube.com/thisweekin * Subscribe to the Founder University Podcast: https://www.founder.university/podcast

So I wrote this book called The Ghost Map. And so I would often start off just to figure out where Bard was today. I would be like, hey, let's talk about Stephen Johnson's book, The Ghost Map. And so one time I did it and Bard came back, it was like, oh, I would love to talk about that. That's a thrilling medical mystery set in 1854s London about the Dr. John Snow in his investigation. And it's a tale that weaves together a number of different themes, blah, blah, blah. And I finally got to the end of it. I was like, oh, well, thank you very much.

I'm actually the author of that book. And that said, oh, I am so excited to meet you. Mr. Johnson. If I had any idea, I'm so sorry I didn't recognize you. And I was like, that's fine. There's no way you could have recognized you.

It's just one of those moments where I was like, what is even going on? This weekend startups is brought to you by Dev Squad. Most dev agencies only offer developers. Why? Because product management is hard. Get an entire product team for the cost of one U.S. developer

plus 10% off at DevSquod.com. slash twist. LinkedIn marketing. To redeem a $100 $100 LinkedIn ad credit and launch your first campaign, go to LinkedIn.com slash this week in startups. And FitBod. Tired of doing the same workouts at the gym? FitBod will build you personalized workouts that help you progress with every set. Get 25% off your subscription or try out the app for free when you sign up now at

FitBod.me slash twist. All right, everybody, welcome back to this week. startups got a really special guest today. I have known Stephen Berlin Johnson for a long time. We met in the 90s, which was pretty much the best decade of the last century, I think. Maybe the 60s. Some people might make an argument. But when we were kids, we were running around New York City. It was called Silicon Alley back then.

And he was running a website called feed, a zine. And I had a zine. My zine was print. His was online. I was doing Silicon All the reporter. And we were all trying to figure out what would happen with the internet. Stephen went on to write 13 books. We're good ideas come from, actually.

I read that one really good. And I think he reads his own books on Audible, so you get to hear his voice. He's got a great podcast called American Innovations. And now he's working at Google. And apparently, Stephen, you're a developer. It's so nice to see you after many years. I know, it's been a really long time. It's been a really like, but you're looking good.

I'm very impressed. You too. You too. Life is good, right? I mean, but here we are. You and I started on the internet before the web existed. We were doing online services. We were hanging out with people making CD-ROMs, Voyager, Blender, whatever.

It was like a really interesting time. The internet happened. And somehow you wound up at Google building Notebook L.M. So I guess let's just start with that. What is Google Notebook LM and why are you building it? And why did you take a gig at Google? Yeah. So in a weird way, it's like a 40-year story.

because I had this long obsession with tools for thought. I mean, when we first met, I was thinking about this stuff. I really started when I was in college when HyperCard came out for the Mac in 1988, which was kind of proto, almost web-like thing that you could organize information. You create these little stacks of cards, and you could kind of link between the card. Yes, it had the first hyperlinks. It had the beginning of hypertext, although it was not a network thing initially. It wasn't really connected to the Internet at all.

I think maybe a later version and finally got wired up to the outer world. But I just had this glimpse of, oh, I could use software to help me think and create and have more interesting ideas and make connections and not just use it to kind of format my papers. You know, there was just a little hint to that.

I think a lot of us who get interested in technology gets this little taste of an idea at some formative point. And the tech isn't there yet, but you're like, I know someday I'll be able to do this. So that was always in the back of my mind. Some of the things that we did at feed, trying to experiment with different ways to use hypertext. That was one of our kind of calling cards in the early days, creating new ways of kind of connecting ideas and print through text. And then I was always, you know, I wrote a lot about the tools that I was using to write the books.

You know, I wrote about the tool Devon think that I used for a long time. I am a big Scribner fan. I've written about that. And in my book, Where Good Ideas Come From? I talk a lot about how you create. create environments that allow you to think more creatively. And so I talked about software in that context as well. So there was this kind of long history of this.

And then, of course, language models came out. And, you know, we had, you know, in the early days of, you know, kind of behind the scenes, Google had Palm and Lambda. And then, of course, GPD3 had come out. And then, and so in the spring of 2022, I got a kind of a cold email from a guy named Clay Bevoir who had started kind of rebooted labs at Google. And he had read a bunch of my books and had followed this kind of train of thought in this interest in tools for thought. And he reached out and he said, hey, do you?

I want to talk to you. And so we met with this is actually the crazy stories. We've met in Project Starline, which is Google's new kind of hologram technology. Oh, yes. This is like a holodeck. You go into a phone booth, I think, and the other person's in a phone booth in New York and I'm in San Francisco and it kind of projects a full 3D model of the person. You're not wearing glasses. You're just sitting there and so. What's it like? It was one of the, you know, I mean, this is, I say this, I said this before I joined Google. It was top five impressive technology demos I've ever seen in my life. It's, it's uncanny. And during the conversation, you. you know, Clay kind of persuaded me to come be a part. You said, we've got a team, just a small little team, but they're ready to kind of build something in this mode with language models.

Like we can finally build the thing you've been dreaming of your whole life. And why don't you come initially part-time to Google and just be in the room with us and help us create. This is a big part of the lab's ethos, was to bring people from the outside at the early stage of these products and help develop them. And I thought, you know, that's a pretty cool, interesting idea. You know, you don't get an opportunity like this very often. And then I got out of the hologram, the Project Starlight meeting, and I thought, he literally put me in a reality distortion field.

Yes. This is not a term of art. This is a reality. And just one more thing on Starline. So you sit and the other person is sitting across from like, and it feels like a table and there's depth to it. But it's some sort of a, is it a television screen or a projector? I mean, here's like an image of it. But how does it get the depth feeling?

It's tracking the location of both of your retinas. It's send the screen. It's just, it looks like a regular television screen, but the screen is sending a different set of pixels to your left eye and to your right eye. And so it only works currently, I believe, with just, you know, one person on one side at one point person or the other because you have to send those pixels directly to it. But it creates a very powerful thing.

And my favorite example of it is that I, when I was testing it in another time, I wanted to kind of lean forward to see if the illusion goes away if you lean forward. And so I leaned forward and I had this very strong visceral feeling of, oh, I'm invading the space of this other person. I'm like, all my breath, all this stuff. And then I was like, wait a second. This person is like, put the house away. You're like leaning in for a kiss or like bad breath. I mean, it's like all the things you're not supposed to do. And you can kind of get too close.

That's wild. Is it going to come out as a product ever? Where's it at? I shouldn't speak to you. Yeah, okay, right. Yeah, it's in labs. That's enough. It's in labs.

Yeah. All right, going from an idea sketched on the back of a napkin to a robust, stable product requires a wide range of skills. You can spend ages looking for a one in a million developer who can do it all, or you can quickly ramp up an entire product team to help you build and launch your product with our partner, Dev Squad. Def Squad provides an entire development team packed with top talent from Latin America. Your elite squad will include two to six full-stack developers, a technical product

manager, plus experts in product strategy, U.I and UX design, DevOps, and QA, all working together to make your SaaS product a success. You can wrap up an entire product team fast in your time zone and at rates 75% cheaper than a comparable U.S.-based team. And with DevS. squad, you pay month to month with no long-term contracts. Take the hassle out of assembling and managing a sprawling team of freelancers and work with a group that's ready to hit the ground running. Visit devsquad.com slash twist and get 10% off your engagement. That's devsquad.com slash twist. So you mentioned Scribner.

I remember when that came out. You're an author of many books. I am the author of one. But when you're writing a book, especially if you're writing a book in the veins of like yourself or, you know, Malcolm Gladwell, a lot of times my understanding is you're trying to take disparate ideas, stories. and kind of pull them together and hey, this has been your life's work.

Feed. It was a feed before RSS existed. And hyperlinks, you were trying to tell stories. So when you clicked the link, you went somewhere and that was an epiphany or an emotion or, you know, a change something about your perspective when you went to that destination. It was part of the fun of clicking on hyperlinks, which seems incredibly basic now,

but it was incredibly mind-blowing at the time. But the process of writing these books takes years. It takes trying to put all this information together in some way. And that's what you've built with notebooks. So maybe you could just show it to us. And I remember Scribner was like this too

because you had kind of like post-it notes kind of all over the place. And, you know, architecture. You do a table of contents. There's been software like the brain that Jermakowski was obsessed with. If you remember, where you would just kind of make little neurons

and still is, apparently. I don't even know the company exists, but I know Jerry's brain exists somewhere on the internet. It's incredible. Yeah. I mean, I think one of the, things that's, before we kind of dive into it in detail, like one of the things that's important, maybe it wasn't even fully clear to me when I started this, is that what I do with the books

is just a more exaggerated version of something that I think a lot of people do, anybody who works as kind of a knowledge worker in any discipline, which is that feeling where you're sitting down, you're trying to generate your ideas about something or kind of build the first draft of something. And the information you need to pull off. that job is scattered across multiple documents. And, you know, in my case, it might be scattered across like hundreds of documents, you know, but I just finished. It has, you know, literally I think something like 400 newspaper articles are part of the research

for it. But, you know, even if it's 10 documents, that's a lot of information. And, you know, we just are constantly in this state where we're like, oh, I've got 10 tabs open. I'm trying to figure out this thing. I'm trying to put together this blog post. I'm trying to put together this marketing plan. I'm trying to synthesize these ideas for a legal brief, whatever it is.

And the information is kind of scattered everywhere. And we've never really, you know, so many people talk about the experience of like you go through your tabs and you command F to try and find the thing you're looking for. Then you find it finally. And then you copy and paste it back into the other dock, which says in the other tab. And it's incredibly laborious work. Yes. And it completely, it's so disruptive of your flow state, like your creative state. You're not thinking.

you're just doing this kind of menial, you know, archival labor, trying to like find the thing you're looking for and not just kind of thinking and writing. And there was a reason for that. There was just there wasn't any software just wasn't smart enough to kind of find and summarize and make sense of the meaning of documents until now. And that's the key part, right? I mean, you could store documents. You can retrieve documents.

You could edit, cut and paste them. And search Google's, you know, Wheelhouse was incredible. and then, you know, organizing stuff, but it didn't understand the entire corpus. And whether you're doing something like, I don't know, you're doing M&A and you got a document library of, you know, all the documents to close this M&A deal or an investment,

or you're writing a book and you've got, you know, a hundred different articles and you're writing a biography of somebody and, you know, if you've got all the articles written about them, you can't keep it all in your head. That's just not how the human brain works, right? And you meet an interesting human, right? if you meet somebody super interesting, Jared Diamond or, I don't know, pick somebody who's got a lot of, a big brain who has synthesized a lot of stuff for 30, 40, 50 years.

When you talk to them and they put it together, that's kind of like the great interview, right? When we find a great interview and that's the one that blows everybody's brain, like, wow, this person is doing that in real time. The documents are their brain. Yeah, actually, I'm glad you put it that way because that's the thing I've been trying to explain to people a little bit, which is one of the things that the notebook L.M lets you do, and we can explain it more detail. in a second, but it gives you all these tools for basically having conversations with documents. And, you know, having an open-ended conversation with a book or a chapter of a book or a collection of quotes from books that you've assembled over the years is just something that was not possible before. Like, you could, if you were lucky enough to meet the author of a book, you could have

a, you know, if you met Jared Diamond, you could have a conversation with him. And that would be kind of like having a conversation with this book. If you could meet an expert who had, studied Jared Diamond or a tutor or a great teacher and maybe have a conversation. But now, like, you can just upload these documents and you can. Let's show people. We got to show people this. All right. So.

Sure screen. And I'm going to show you what I literally got so excited about this because I was, I'm working on a project. I can't wait to show it you and get your feedback. But it's, this is a very mind-blowing. I would show you. So fittingly, okay, here you. Let me share.

If you're listening in the audience and you want to go find this, just go to YouTube and type in this week in startups. You'll find it. And I'll try and describe it. Well, sports cast it. Yeah. Well, sports cast it. Okay, so I thought I'd start appropriately enough where I've loaded up like five chapters

from that book where good ideas come from as sources. And so when you begin in notebook, L.M, you know, one of the first things you do is you kind of define the documents that you want to work with in a particular notebook. So as we said, those could be, you know, a bunch of marketing documents. It could be research for a book. It could be legal briefs, whatever it is. It's relevant to the job you're working on. In each notebook right now, you can have 20 documents.

They can be 200,000 words each. They can be PDFs. They can be docs. We can be copied text. We're going to add a bunch of formats, as you can imagine. Those are your sources. Those are your sources. Those are listed on the left here with these little document previews.

Yeah. And that's an important word. The source has a very specific meaning here. And so once you load them into notebook LM, the AI is grounded in the information that is contained in those sources. So it takes about, I don't know, For loading these five chapters, I think it takes about like 12 seconds or something like that. And then at that point, it's like the AI.

It's like Notebook LM has become an instant expert in those five chapters, which is just astonishing to me that a computer can do this. And one of the things that we've done, we've spent a lot of time with inside a notebook LM is like setting up guardrails. So if you try to ask questions that are outside the boundary of the information that is containing those documents, generally, it's not always perfect, but generally notebook LM will decline to answer. It will say, I'm sorry, I can't answer that question. It's not contained in the sources. Okay. You can see how that's a great use case in the classroom, right? So you can see a teacher loads up a shared notebook with a bunch of documents that they,

that they're using for the syllabus for the class. And the student can use those, but they can't kind of go outside the boundaries of that. So I've loaded up here, you know, I think it's five chapters from where good ideas come from. And I actually just did a question before just to preload it. So there's a whole riff about. 9-11 and the kind of intelligence failures in 9-11 in that book. And so I ask a question, you know, this is not keyword searching, right? This is the age of language models.

You can ask these very sophisticated questions. So I ask what happened with the FBI and 9-11 and what was the significance of it? And it's going to answer that question based on not the general significance of it. It's going to answer it based on the significance that I, you know, kind of endowed it with in writing this particular book. So it's going to be based on the subtleties of the, the interpretation that I wrote in the original book. So this isn't going out to the open web and finding a New Yorker story or the Wikipedia and then hallucinating or a bunch of conspiracy theorists.

This is, you know, you've got a narrow corpus here to really return a tight summary. Yeah. And what it ends up doing is it greatly reduces the hallucination risk. I mean, you will still see, you know, occasional mistakes. Or sometimes it will just, it's more like it gets confused. sometimes if the information is a little bit confusing, but reduces it significantly. And you get these, I mean, this is just amazing. This is, by the way, this part of it, we're kind of rolling over to having everything be powered by Gemini Pro, which has been fantastic in our early testing of it.

And so this is an answer that Gemini Pro returned. It formats it in this really nice way. Like it has kind of bold text for the key terms. It gives you nice little bullet points. It does this, you know, really elegant kind of overview of, of, of, of, of, you know, the situation. It's nice, kind of presented in just kind of a nice way. We spent a lot of time trying to get that style right. Like, it's

very much like editorial product, you know? This would be the equivalent of if you had a great writer and you said, hey, summarize this very long story or a series of books, you know, in our magazine or zine, so that a casual user can, you know, within five or

10 minutes of reading, 400, 500 words, really understand it and not have to do a lot of work, a lot of heavy lifting. So it says here, hey, you got the Phoenix memo, you've got something that's overlooked. So you're doing headings, you're putting them in bullet points, you're making it easy for people to digest knowledge. And so somehow you've trained the model that this is an academic synthesizing world in which

you're working, yeah? Yeah. And so that's really useful. But then with every answer, we also give you citations. And so you can always, in a sense, check one, you can just double check that, you know, What did the model use to come up with this thing? So if I mouse over these citations, I can see these are the original quotes from my book that it used to piece this together. And sometimes you use that just a fact jack.

But sometimes that's actually what you want to do. You want to actually read the original text. And this is just a faster way to get to it. And so you can click on any of these. And it immediately opens the original source and takes you to the point in the original document. So you can actually like read it originally. So your ability to just kind of navigate through, I'm not going to, I don't have it on this computer, but I have on my main Google computer, I have a collection of quotes that I've been assembling like digital quotes from books that I read for the last 20 years. It's like 1.3 million words of quotes from like my reading history of the last 20 years. And I can literally just sit down and be like, hey, what are some interesting things about dolphins? Yeah, tell me about love. And it'll find all these things.

And then I can zip immediately to kind of the original passage and see it. And that's the speed with which you can move through information and have it summarized for you is just incredibly powerful. Now, the other thing that's, that's, you know, we were talking about having a conversation with a document. So one of the things we found when we started doing early testing in this this summer at a few schools around the country, a few colleges around the country, was that people didn't know really how to ask questions. You know, they would sit down, they were like, what do I say? You and I are trained as a journalist. And so we know we know how to ask questions, you know, we know how to kind of think in that, you know, dialogue kind of mode. But it's not something that's actually taught very, very often.

And so what we started to do is actually to use the model to constantly give you suggested questions based on the content of the source and based on what you've just talked about. And so you can see right below here, there's a set of suggested questions here, like, what was the central claim of the Phoenix memo and why did it fail to prevent the attacks? How did the automated case support system contribute to the failure of the FBI to connect the Phoenix memo? So that's a pretty good one. Let's see what happens. I'm going to click on this.

We're doing. We're now on live demo mode here. We could ask it, why didn't anybody pick up that the terrorists didn't want to learn how to land the planes? It looks like a little bit of a red flag. Like, sure you don't want to learn how to land a plane? Nope. Just how do I get it up there and keep it straight?

And this is such a good answer, man. I'm still astonished at this, that this is possible. So, yeah, it's talking about basically there was this automated case support system at the FBI, and it was just, it was designed. It was the opposite of notebook I love, actually, now that I'm thinking about it, this is a really good example because they had all this information and they were unable to connect the dots. So the software here has summarized all that stuff. It's figured out, you know, it gives me a bunch of different details from that story,

and then it gives me a nice little summary at the end. And then again, I can go look at the citations. Like that, that's the, the kind of the core, kind of very beginning of this project. But anything you find that is interesting. So you're engaged in this and you're exploring, you're asking questions, you're following, you're clicking on the follow-up questions, you're following citations. You find something interesting, you just pin it. And then you've got this kind of noteboard space. This looks great on a big screen, by the way.

It's incredible. Yeah, I have a wide screen. I was doing this up by an example. It's really nice. So then you have all these notes that you can go and refer to. So you're able to just constantly like grab things that are interesting. And what's coming like any day now, you're going to be able to select a note or a set of notes. And we're going to automatically give you a set of options when you select those notes.

And those options will be things like create a study guide or convert into a thematic outline or suggest related ideas from my sources. What sources am I missing? Finding new sources is the one I was looking for. That is not short term coming out yet, but that is definitely where we're headed. I mean, for sure, I mean, it's Google. Well, here, you know, your work is great. If it said, you know, hey, I found three other sources, would you like to add the actual 9-11 commission report, which is a giant document to this? would you like to add this Amazon series, a fictional series, to your, or the script from it?

You know, now all of a sudden, it's this research assistant on either side of you, answering the questions, and this is how documentary films are made, right? Or a book like yours or Malcolm's where you're trying to pull together themes that, you know, only a human previously could make these connections. And when you make those connections and they spark something in you as an office, fitter, we want to share those, right? And we want to have that spark happening in your brain, right?

Absolutely. And that's really the magic of being an author, in my mind, is when you can get into, when you can, and Stephen King wrote this, do you ever read on writing, his book, his autobiography? He did, yeah. It's incredible.

I mean, it's, I mean, there's many reasons it's incredible. Perhaps one of the best is that he wrote Kujo while doing copious amounts of cocaine with his nose bleeding and just, And that kind of tracks, right? Cougat's pretty... Yeah, having read that book, I think it's probably...

Yeah, it probably feels like it was powered by cocaine. But he said, you know, teleportation. What writing, great writing is, is you can basically teleport through time and space. Whatever you write, it just magically appears in the other person's brain. So if you were to describe your desk right now and the microphone, et cetera, and somebody read it 100 years from now, all of a sudden you would have manifested in their brain, this image. And that is just... you look at writing that way, that you're literally creating a vision in somebody else's brain,

you really understand the power of it and why it's important in the world. And this it just takes it to a whole other level that most people are not even thinking about right now. Well, that's a great point. And it reminds me of something, you know, Tiago Forte, the second brain, I wrote how to build a second brain. Really interesting guy. We've been talking to him about notebook because it's right up in his alley. And he has the thing about like taking notes and capturing things that are interesting to you and kind of storing them is this way of, you know, you're trying to do it to like send a message to some future version of yourself who's going to need this thing in five years. They don't know why, but like this piece

of information. And so, you know, I haven't kind of fully done that just the kind of cleaning up of my documents to do this. But, you know, by the end of the year, early next year, I'm going to have one notebook that's effectively like all the important things I've ever read. read and all the important things that I've ever written are going to be live in this one notebook. And my ability then to find that idea that I jotted down, you know, in 2007 that is so long gone, you know, to my actual physical brain's memory. But that, you know, the combination of this software and my curating all the information initially is going to, is going to enable me to find those things again.

And it really evokes, you know, like, where you always wind up in these discussions is consciousness, right? Like, what is our consciousness, but a collection of the things we've read, we've written, we've spoken, we've experienced. And so now we start looking at what is actually being created with these LLMs. And I think it's either explaining to us

how basic our brains are in some ways or how incredibly complex they are, because if you did have all of your writing, every instant message, every email. And then you start thinking, well, every song and movie I've consumed, every book I've consumed, everything I've written, you've now got almost your consciousness, you know, in a book here. And that gets really trippy because it's a perfect consciousness as opposed to ours,

which has bias or something. When we actually become conscious of something, we know this because in different scientific studies, if they play some music or, you know, give you a cent or take you to a location in your high school, all kinds of other memories come out. But this is like becoming perfect memory to point about, yeah. We're trying to, it was interesting. I think I think you saw that Stephen Levy piece that he wrote about it in Wired. And I think he had gone into it thinking it was just going to be a really good search engine for his own stuff.

And he was like, it turns out it has a little bit of its own opinion about things. Like it's steering you towards certain ideas. And, you know, we've spent a lot of time trying to figure out, like, what is the right balance here? because you want your kind of assistant to be smart and help you develop your ideas, but you also want it to be, you know, to play second fiddle to your ideas, your own thinking. One of the things that's interesting about it is, in general, we have tried to create a voice for the, for the AI that doesn't have a first person voice. So it's not trying to be your buddy, right?

It's just living you the information. It's not, we're like, we don't want it to pretend to be a person. It just should be somehow just an incredible service that finds the information you're looking for and summarizes it and doesn't. So every now and then it'll sneak in, but it won't generally say things like, sure, I can help you with that. Yeah, that kind of stuff. You got it, buddy. Business to business marketing is not an easy job. It's much different than business to consumer advertising.

Why? Well, the enterprise buying cycles are very long, and they're filled with decision makers. and those decision makers are going to kill your deal. If you can't get to them, that's why you need to check out LinkedIn ads. LinkedIn has amazingly, but not unexpectedly, past a billion users. This includes 180 million senior executives. There's also 10 million C-suite executives. Those are the CEOs, CFO, CTOs, the chief strategy officers, chief finance officers.

This means 18% of those users are the ones who are the decision makers. How do you get to them? You get to them through LinkedIn in a respectful business environment. They're ready to accept a business message, as opposed to, you know, another platform where they might be consuming cooking videos or podcasts or political discourse. No, LinkedIn is about business. You want to get people when they're in that cognitive mindset and they're willing to accept a business to business to business message. 79% of B2B content marketers said LinkedIn ads produces the best results for paid media. This is obvious. I can tell you this is true. When you think about business, you think about LinkedIn. It's just exactly what comes to mind. So here's your call to action. Make business to business marketing everything it can be and get $100 credit towards your next campaign by going to LinkedIn.com slash this week in startups to claim your credit. LinkedIn.com slash this week in startups. No spaces, no dashes. LinkedIn.com slash this week in startups for a hundred dollars in credit terms and conditions do apply. Let me show you one other thing and then I want to see your you've been doing with it. But I just want to make sure that it's clear that this also has, it doesn't always have to be such a highbrow.

use case. So this is something that occurred to us very late of the process where we were like getting ready to launch and we were like, wait, we've written all these help documents for using notebook LAM. What happens if you just load those up as sources into notebook? And it becomes an instant expert in how to use notebook, which, you know, it turns out to be a really just as a kind of onboarding guide, as a help desk. It's so powerful. So anybody, whatever kind of service you are or company you are, you can just load up the documents that describe how your system works, and then you can share notebooks with other people, and so then they can ask questions and engage. And one of the things I like, so you can ask these kind of prosaic questions, like,

how do you upload a PDF to notebook LM? It'll be good. But generally, this is, I'm a lawyer. How could I use notebook LM? So we don't have any language in there about like, lawyers using notebook LAM. All we just have is a description of the product.

So generally, this is generally this generates a pretty interesting answer. Like it kind of understands, yeah, this is great. So it understands how the software works and it understands generally how lawyers work. And so it's able to synthesize those two things

and come up with this actual kind of precise. And you can actually dive in deeper and say, actually can give me a step by step, you know, set of examples for how I could use it to draft legal documents or something like that. Right. You just go down the rabbit hole from there. Yeah.

It's really interesting, like if you were to think about instances where you need a large corpus of documents. And in fact, when we were talking about the Phoenix memo and the failings of that computer system, kind of alluded to the fact that, hey, you know, if you're trying to do a case, you're trying to solve some puzzle in the world, which is what law enforcement does or mathematicians, etc., you know, there are going to be these examples of the language models being able to, if given the right corpus of information and given the right prompts

and somebody actually reading the result and checking the work, we're going to solve a lot of mysteries. You know, like, just like these crazy people who get obsessed with, I don't know if you've seen these lunatics like on subredits and excitement, like, and they make documentaries about them. I think there was one like something about cats or something,

but there are these people who, I don't mess with cats, I think it's the name of the documentary, but there are these people who are like stay-at-home crime hunters, like detectives from their keyboards, and they find a case of a missing person or the, you know, somebody who was murdered and they are inspired by other podcasts about people being murdered to then go solve it,

and they become online salutes. And it's usually they don't find anything, and they just are twisting, you know, reality. Yeah, it creates... And sometimes they do figure stuff out. Well, you know, it reminds me of what you said at the beginning of the, the conversation about how mesmerizing it was to click on a link in 1994, right?

You're like, I clicked on this blue word and I was taken from one server to another's to a completely different document on the other side of the world and it just seemed mind-blowing. And it was just fun to, in those early days, just clicking around the web was kind of just fun as an adventure. And I think that there's something that is like that that I'm seeing again in a book, L.M, in part because of the suggested questions. I mean, that's just, that has that same feel of like you jump into a big archive of information. Yes.

You ask an opening question or you, you know, we'll actually recommend topics that you can explore right out of the gate. But, and then you just kind of ride those suggested questions for a while. And you get, it's a great way to kind of engage with, you know, kind of initially figure out what's in this text and discover new things. And I could imagine somebody who was going down a rabbit hole of some, you know, complex crime, you know, investigation and had whatever 30 PDFs of evidence there. That that could be particularly intoxicating. Let me show you what I did.

This is related to the pod. So, you know, I'm doing this pod for like 1800 episodes and like, sometimes these things have to remain interesting to the host, right? Like the art has to be interesting to the person producing it or else it's like, it just becomes a chore. And I was like,

you know, I just love movies. And I love business stories. So let's just do like this business breakdowns where we find, you know, a great movie about business. And then we kind of do like a bullet by, you know, either timeline kind of approach or the key lessons from a movie or from a success story.

And so your mind is already racing with the possibilities of movies to do because you're well read, I assume. So Ray Kroc, who created McDonald's, wrote a great book called The Founder. And he was an insane founder, like pretty sharp bell bowed and sane.

And I know about this because I'm a, big fan of Mark Knopfler, the lead singer of Dire Straits, and when he went into his solo career, he wrote a great song called Boom Like That. This is probably one of my favorite songs of his solo career. And he had read Ray Crocs, grinding it out biography, and wrote a song based on it. The guy who made the movie The Founder with Michael Keaton

had heard Mark Knopfler's song, found the source material of the book, and then convinced Michael Keaton to make a movie. The founder is an incredible movie that nobody to remember. Have you ever seen it? Never seen it. And I assume you haven't read Ray Crocs.

I have not. I'm grinding it out. Right. So this is like absurd shit, but I read it and it spoke to me. Now, I got to it. Mark Knopfler song. The movie, then I found out the movie was based on a book. Okay, here we go, right?

Because I went down the rabbit hole. So then I was like, okay, I'm doing this episode. It'll come out after this episode comes out where we're going to break down the story of Ray Kroc through the song by Mark Knopfler and the movie. and the book. Three interesting pieces of source material. So then I went out and I know, and I paid for all this stuff.

So anybody who wants to give me a hard time about copyright, I found what I assume is the public domain version of a PDF of the script of the founder. I found a public PDF of the book, grinding it out, which I have three different versions of I've paid for. So please do not sue me. If this is your book, I assume I'm legally allowed to do this.

And then I found a YouTube video. Now, the YouTube video, I thought I could just drop it in there since it's Google, but you guys are in a version 1.0, not yet. But I know there's transcripts, so I opened the transcript, and I very awkwardly had to, you know, drag and drop and cut that transcript in here. So I got my three. And then I started asking things, hey, what were the key moments, you know, in the history of McDonald's?

And then I said, well, what page is that in the screenplay, right? And so I haven't gotten two down things, but, you know, I also asked, What is Ray Kroc consider his the key to a success? Now, I don't know where this is coming in from citation-wise. But he said it's perseverance and determination, yada, yada like you would expect. And so I'm just started my adventure, but there's probably 20 other documents. I don't know to find the Harvard case study on this and the pen case study on it. And you have put the song lyrics in.

What are the, let's see, what are the suggested questions, actually? So what were the three decisions the McDonald brothers made regarding the design of their restaurants that made it stand out from other drive-ins. Oh, I know this. It didn't have seating, right? It was a counter, so I know it didn't have seating. The limited menu, that's right, and the streamlined process. And the unique building design is there.

The brother's restaurant was in a building with a red and white design was eye-catching. I guess they didn't. The restaurant was a standout from other drive-ins. Interesting. Yeah. So look at the citations, too. Um, so just coming from the script, you'll see. So when you roll over, yeah, you can see it's coming from the,

grinding it out, which is his book. Yeah. Yeah. Yeah. So I mean, I, I, you know what? I, I've only been in this for like a half hour. I was like, most of my, if you do a listening lab with me, like the hardest thing I had was getting information in. Um, and I tried to do this before I found out about notebook L.M. when I saw Stephen Levy's story on you and said, hey, I know, I know, I know you, Stephen. And I know you, Stephen. And I, I said, get them on the pot, I want to talk to him about this. I just couldn't figure out how to get stuff into it. That was my blocker. And then I was like, oh, wait a second.

And then I took the same PDF and I tried to put it in Claude. And Claude was like, yeah, that's way too big. And I was like, ah, God damn it. The clean thing right now, by the way, for folks, is either a simply formatted PDF where it's just kind of a straight text thing. If you have a PDF of a book, you know, you have all this complicated things where, like, it'll grab the, like, headers and the page numbers and footnotes and it'll be a little confused. Anything that's in doc format is going to be good.

But we're going to get better and better. And the other thing is that Gemini is natively multimodal. So there's all this incredible stuff that we're going to be able to do with images as well. I mean, right now it's all text. But we were just doing a bunch of experiments the other night with images. And it's astonishingly good. All right. You know, I've been on a health kick over the past year.

And you know, I care about data-driven solutions. and if you listen to this podcast, I bet you do too. So let me tell you about FitBod. This is a data-driven workout app that blends machine learning with exercise science. FitBod creates custom dynamic workouts, programs, based on your fitness goals, your experience, and most interestingly to me, the available equipment. Let's say you got a bunch of kettlebells, or let's say you're at some, you know, sparse gym at a hotel, or you're on vacation.

You've got nothing. Well, FitBod will maximize your fitness games by varying the intent. intensity and the volume between your sessions and leverage the equipment you have or don't have, as the case may be. You can customize the length of your workout, what muscles you want to target, and so much more. So let's say you want to get a 30-minute workout in. And I want to do chest, triceps, and amps. But I'm staying at an Airbnb. There's no equipment. FIPAA can create a perfectly optimized workout for me based on these parameters. And it will do it for you to check it out. It's amazing. The design of this app is extraordinary. I was able to invest in it. That's how impressed I was

with it, FitBod takes the guest work out of fitness. Just open the app and start making progress. You deserve it. Get 25% off your FitBod subscription or try out the app for free when you sign up now at FitBOD.m.m.m.m.m. slash TWIST for 25% off. Have you looked into the rights and you're an author? If I asked it, if I had purchased your book to allow me to talk to your book, as an author, you'd be cool with that, I think. Yeah. Well, I certainly would be. I would be as well.

I would want people to do that. Yeah. The other skills. You know, one of the things that we like to recommend people do is, you know, you can use, if you buy an ebook, um, on the Kindle or the playbook store, you know, you're allowed to, save quotations from those books. So as you read, you can save quotations and they're wonderful services like readwise, um, uh, that will allow you to, um, uh, that will allow you to,

export those quotes to a dock. And then you can just bring those and then you can bring those in. So you wouldn't have the whole book there, but actually sometimes you don't want the whole book. You want the passages, the most important ones for you. And then, so that's what basically I've been doing with my set of quotation. So that's a great way to do it. But I agree.

I think, you know, there's a logical place we could end up where if you buy an ebook, you could read it in an ebook reader or you could read it inside of notebook L.M. I would love that future. Yeah. I mean, I think we, I wonder how our publishers, or Harper's or whoever your publisher is, is like thinking about this. Because I feel like you could charge an extra 10 bucks

for a digital book to allow it to talk to it, you know, and to query it. Or it could just be like, it's almost like a new format. Like I wonder if Apple and Amazon must be thinking about this as a feature. You were thinking very much along the lines that I've been thinking. So I would hope that we would make some progress on that front next year. But the other thing that's really worth pointing out to people, and this is important whether you're an author or not, is in terms of privacy and security, we are not training the model on the information you upload in those sources, right? So the model has been pre-trained.

What we are doing, the easiest way to put it is, like, we're putting, if you know the AI language as you do, we're putting the information from your sources briefly into the context window of the model and asking you questions based on that. for people who don't know what that means, it means we're basically showing the model, we're giving your information to the model's short-term memory. And the second you end your conversation, it remembers nothing. And we do no training based on the information of the document. So that means you can use it with private documents, corporate documents, or a rights holder can feel confident that if somebody is taking some quotes from a book that they've read that they purchased, that information is not somehow getting into the training data for the models.

It's a key issue for, yeah, I think it's a key issue for authors. I was really flabbergasted by OpenAI's approach and you work at Google, so I'm not going to have you comment on. I'll just tell you my opinion that they just took open crawl and some other corpuses and trained their model on it. And they know full well that they're, that's just taking the open web and that the open web has all kinds of stuff that hasn't been cleared. And then they train the model on it. Now, it's six of my books were in there. Right. Now, as far as I'm concerned, they owe you a licensing fee.

And you know the books are in there because when you ask very specific questions, it's going to give you the answers. And in the earlier versions of chat GPT, you could ask it, is Stephen Berlin Johnson's books in here? And it would actually tell you, yes, I've got them right here. In your case, did they subsequently take them out? And what do you think of these, just broadly speaking, the rights of authors in terms of model training? Because it feels profoundly unfair to me. Yeah. I probably should be delicate about this.

Sure. Oh, right. Because of where you work. You know, but I do think it does feel like, one, as an author, I want people to be using language models with my work. I think that's a way that people are going to be exploring information. So I'm very comfortable with people. If they do it in a proper way, they're paid for the book, they should be able to interact with a model. And I do think that, yeah, if models are being trained on copy, you know, written,

material that there's, there needs to be something that's going on with the rights holders there, but I honestly, I haven't spent that much time thinking about that side of it because I've just been so immersed in the product side of it. I mean, in some cases, it's super obvious, right? Like, I was asking Chat GBT to make me like a Jedi night bulldog, and it was like, sure, here's a Jedi, and I was like,

okay, now make me Darth Vader as a bulldog. And it was like, sorry, I can't do that because of our content policy. And I was like, okay, I get it. You understand, like, a Jedi is a category, but not a character. So they must have taken the entire Disney corporations and said, for Dolly, let's not make images of Marvel characters. Let's not kick the number one rights holder who is the most litigious and thoughtful and, you know, about this. And so it declines to make that.

And so then I just said, I'll make a Sith Lord. And it literally made me Darth Raider as a bold. Okay, here we go. I talk about, you know, searching authors' work. So in the early days when Bard first came out internally, I was kind of, I was kind of. at home or the rest of my family was skiing. This was like a year ago. And I was just like, okay, this is our new model.

I got to test it. I figure out everything. And so one thing I would do, like, Bard's kind of intelligence in those days would go up and down as they were retraining and things like that. Variable. Variable. I have like a standard kind of question that I'm asked.

So I would often ask, so I wrote this book called the Ghost Map. And so I would often start off just to figure out where Bard was today. I would be like, hey, let's talk about Stephen Johnson's book, The Ghost Map. And so one time I did it and Bard came back. It was like, oh, I would love to talk about that. That's a thrilling. medical mystery set in 1854s London about the Dr. John Snow in his investigation and it's a tale that weaves together a number of different themes and I finally got to the end of it. I was like,

oh, well, thank you very much. I'm actually the author of that book. And said, oh, I am so excited to meet you, Mr. Johnson. If I had any idea, I'm so sorry, I didn't recognize you. And I was like, that's fine. There's no way you could have recognized me. Right. It's just one of those moments where I was like, what is even going on? So this feels to me like part of Google Docs eventually. How does a lab product, where does it go from here? Because it feels like a product.

And listen, I'm tip of the spear for you and use cases. So I'm wanting to pay $199 a year for it to get like the feature set or whatever. I pay $20 a month for it because I would use it. And my production team would use it for this podcast and things we do because we frequently have a guest. And I would have taken interviews with you. I would have said, hey, what are the most interesting things he's been asked in other interviews, right? And I would want to pull in podcasts, et cetera.

So for me, it's a great paid product. But how do you think about taking it from a laboratory experiment and productizing it? How does that work at Google or how do you think about it? I mean, genuinely, it's not a cop out to say we don't really know because this iteration of labs is a new one, right? It's a new labs now run by a wonderful guy named Josh Woodward, who, was also instrumental in and bringing me in. And we have kind of graduated up to a, you know, a public launch in the U.S. We're still billed as an experiment, although BART is still built as an experiment, too, I believe.

And we're trying to figure out, you know, what's working, what's not, what actually the path is, if people like it as much as we think people will like it, particularly as we expand it, we make it easier to bring in sources, make it easier to discover sources, all the things we talked about. I, you know, I don't know what becomes a bit.

It's in a nice spot, I think, where it does something different than what docs does. Yes. And or slides does, you know, we're for the, you know,

almost all of our features are like helping you think and understand. And there's almost no, like our formatting is like bold and italics. Like that's, you know, write a note, that's all you have, you know, enjoy it. So it's not at all about creating the final product at all, but it is a place where you can synthesize across lots of different docs. And so I think it, you know, it complements Google's existing offerings, whether it graduates up into some more elevated points, I don't know.

It's always a challenge with big companies. You can build these really amazing things. And then you have to figure out how they live post, you know, in a laboratory. It does feel to me like this also, based on your U.X. feels like if I put this on a giant screen in a conference room, you know, with the way the post-it notes are kind of designed, the notes and the material, we could all be sitting in a conference room working on a book together or working on a documentary series or the writers on the Simpsons could be doing a retrospective of the last, you know, 10 seasons. And wow, this could be

quite, you know, powerful to be on a giant whiteboard and moving it around like a minority report and asking you questions. Like, is it? This is it? seems like a really good brainstorming tool is I guess what I'm getting at in a whiteboard sense. The shared notebooks, we've just started to explore. Like, you can share a kind of read-only notebook where you can just ask questions, which is great for like the help desk kind of use case. Right. Don't screw with the source material. Yeah.

And then you can share one where you can write your own notes and do all this stuff. We don't have a lot of the technology is very basic right now. Like everybody's note seems to be authored by the same person. You know, we're just getting started. But yeah, I agree that that's really useful. I mean, I keep thinking about it is like, what am I? like are these drafts of things that are sitting on that note board? And the other thing that we're going to be able to do is like you can grab a bunch of notes and combine them into a single note.

And so there's that process. I think that people are going to have. I want to pin a bunch of different things. This is about to roll out. This isn't live yet. You didn't miss anything. So you can be in this mode where you're like, okay, I'm going to pin a bunch of ideas up there. And now I'm going to kind of consolidate them into a single note.

And then I'll use some of these tools to maybe turn it into an outline or convert it into whatever format I want. So I think there's a lot. a lot of stuff that's going to start to happen as people use that interface. It's really a new UI. It doesn't quite look like anything else that's out there. And that was kind of our thought. It's like there's an opportunity now to create, just like we needed to create a new thing called web browsers because this thing called hypertext and HTTP, it needed a new software

category. We think that language models are going to necessitate the same kind of interface revolution. So this is our first stab at it, which is just so, it's just so fun. Yeah, no, it's super mind-blown. Because if you think about it, like, there's the source material, there's the queries and the questions you've asked it. And then there is, well, what do I do with that afterwards? And how does the rest of the world interface with it?

And you're right. Some people might just want to ask questions like you. And it just becomes like, hey, we're talking to Shakespeare about all of his plays or we're just, you know, have every Simpsons episode here. And we're just looking for funny moments that have to do with donuts. but then it could also become a script. It could also become an outline. It could become project management.

So the output could be the LLM. You tell the LM, I want to make this into a podcast episode. That's one hour long with two hosts. And it's like, okay, I get an idea. That's going to be about this many words. And yeah, just tell us which 20 things are the most interesting that we should talk about, right? Make this into a script. Make this into, you know, a summary or something.

I guess what we wrap here, I'm curious, you know, knowing what you know, And I know you've studied all these different technology changes. I remember listening to your audio book. We were talking about the Mendici's and Glass. Which book was that? How we got to know. Yeah, how we got to know. Yeah, it's really good.

Yeah. And so you've seen these changes and your inflection points. When you look at this one, language model specifically, this ability to ask questions, and it is a recency effect here, obviously. Like, we're pretty enamored with us right now. We're enamored and confused. And yeah. But where do you think this one winds up?

Because it does feel like it's building and building and building from the open internet, broadband, unlimited storage, everything, just to kind of to this moment, right? And then also consumers and customers putting so much data into the internet. Like, when we look back on this, like, what equals this in terms of potential impact? What feels like? Yeah, there are kind of two questions, really in a sense, like, where does it end up? Is it really a question? Like, where does it end up in 20 years? is a huge question that I'm probably not qualified to answer.

But in terms of, you know, this existing technology as it is today, you know, if you imagine, you know, somewhat similar incremental improvements over the next three or four years that we've seen over the last two years, which have not been incremental, they've been more than incremental, then I think it has to be considered that, you know, for me, the single most important technological revolution of my lifetime. I mean, you know, I would say it will exceed. I would have said before that it was the personal computer and the graphic interface and the web and mobile were the, you know, kind of the biggest ones. And this seems like it is ultimately going to be more important. But one of the points that I tried to make when I've, and I wrote a Times Magazine piece about, mostly about GPD3, but about language models before I came to Google. And so this was like April of to the 2022.

And the point I was trying to make is like you can be agnostic to the question of whether language models are going to lead to, you know, artificial general intelligence or some, you know, or true understanding or consciousness or all these kinds of things. And I suspect language models as themselves will not lead to that kind of breakthrough. And still think that they are enormously significant. And that once, once the computer is able to. able to manipulate and summarize kind of meaning and make associations on a level of semantics and not just like find, you know, text, but actually be able to talk to you about like,

okay, I've taken your idea and I've summarized it so that a five-year-old can understand it or I've taken your idea and I've connected to these other ideas and I've made a little, you know, analogy here between these two different nights. Once you do that, there's just a whole host of things that no computer in the world could do three years ago that now, you know, anybody can do with, with the web connection and soon enough will be able to do, you know, on device on their phone. And that just unlocks so many doors of possibility that it doesn't really matter whether they ultimately become sentient or become true rivals to human intellect. They're just going to be enormously useful. And that's what we,

that's, that's what really, you know, when I got that call from labs, I was like, yes, this is a time to build this thing. Right. This is a great, you know, the opportunity is just so fantastic. And it also, and I,

and I want to use it myself, you know, so I'm animated by this like desire to build the thing I want to use. I think when you know, when you and I have been at this now for, gosh, 30 years.

Hundreds of years. It feels like it. But, you know, if you're, if you've been assessing technology for three or four decades, right, uh, since we're teenagers,

uh, looking at PCs and dial off services. You know, you, you, you kind of get a sense for like, yeah, this is a big one.

This one. I wonder how you think of, you know, putting aside, you know, crossing different valleys, et cetera, but just super intelligence, how do we even define what super intelligence is? I mean, smarter than any human you've ever met, more than any human who's ever lived, it's pretty clear that this is on that trajectory already. And it doesn't feel like it's very far off than being smarter than any human who's lived, right? you know, take it to like the simplest example of, um, that I gave in, in those demos, like the, the, how to use notebook LM, um, you know, example notebook we have. So how long would it take a human being to get enough expertise about how to use notebook LM, having never seen it before, so that they could explain how it could be used to a lawyer or to anybody else who came along and said, hey, I'm a, whatever, I'm a marketing director.

how can I use this product? Like, they would have to read, you know, if they came up with zero knowledge, they would have to read through the documentation. You know, they'd have to probably mess around with it a little bit. It would take them, I don't know.

Hundreds? An hour, maybe two hours, maybe 10 hours, like to understand, to be able to answer that question confidently and quickly to someone asking it. For notebook,

LM, that is a kind of intelligence, right, to understand a system, be able to improvise and answer based on kind of a novel, new input of like, I'm a lawyer,

I'm a marketing director, or whatever I am. And, you know, so a human that might take somewhere between an hour and 10 hours probably. And for notebook LM, it takes 10 seconds. That's how I want to digest those doctors. So I don't know what that is, but it's something. When you hear people say it's a parlor trick, language models, what's your take on that, just broadly speaking? Like, what are they missing?

And what are they getting right in some cases? Like, why does it feel like that sometimes to people? Yeah. I've been working on a piece a little bit of kind of notes to me. myself about this and I might write someday looking back on this whole experience. Like, part of it is, is that very fraught word understanding. Like when you say, you know, if you say notebook LM understands X because it's read these documents, it's an expert in X because it's read these documents, you know, it's processed

these documents. That causes a certain, you know, subset of people who think a lot about AI to really object. And they say, no, it does not understand it. It's just statistics. It's just a stochastic pair. It's just, you know, it's just predicting the next word. It doesn't understand anything. You know, on some level, that is kind of true.

And if by understanding you mean it is conscious of it or it is having an internal sentient experience of the knowledge, I 100% do not believe that, you know, Gemini or or ChatGPT have any interior mental life. Right. But it is doing this thing now that until two years ago, required human understanding to do. There was no way to get to that result unless you understood. And now I can do that thing.

And so the fact that we use kind of the language of understanding to a shorthand for that, I think is not inappropriate, but I also understand why it, you know, kind of rubs people, people the wrong life. Triggers people, I think. It's, and it's very hard to. Here's like, maybe my favorite interaction with Gemini so far. This wasn't a notebook I'm on, but I was testing it with, um, uh, AI Studio.

So one of the things we want to do is, you know, have these writing tools that will roll out in early 2024 where you can write something and you can select the text and you know, you can kind of ask the model to transform the text in various different ways. And so one thing I was trying to do was to say, here's some boring text and try and make this rhetorically like more interesting with some more metaphors. And so I gave Gemini a passage of description of the climate in Hawaii. And it was just very like scientific, but kind of dry, just a bunch of facts. And I said, make this, you know, metaphorically.

more interesting. And it returned this like completely overwrought thing. It was like, Hawaii's climate is like a symphony that has been conducted by Mother Nature. It's something, you know, it just was crazy. And so I, all I said to Gemini was, dude, that is a little over the top. Gemini's, Gemini responded, you're right, that was a little bit excessive. Here's a better version. I think this is better. And it was perfect. And I was like, the fact that it understands, dude, that is a little over the top. And completely comes back with the right response. It's just, I mean, that's just nuts. It feels to me, this is like, this is where I go to.

I think like if you do believe in simulation theory, like this is the final level of where, like, in the matrix, you kind of wake up from the matrix. Like, we're building this thing. And it's thinking like us. And I asked it like, yeah, dude, a little over the top, bro. And it's like, okay, got it. And then whichever of us kind of realized that what we just recreated was our own brains first,

kind of wins this video game that some sentient being created a billion years ago in some other dimension, which is, let's see if we can make life forms that figure out that they're building computers that are themselves. And then it's just like Prometheus or aliens or something like that where, you know, there are the engineers, but who made the engineers? And it's like, oh, it's just a, it's turtles all the way down. you may have unlocked it. I don't know. If the simulation comes to a halt,

but if this, the pod goes out. It could be. Yeah, exactly. We'll know who cracks. All right, brother. Listen,

everybody go check out. Just Google, Google notebook L. I'm, start playing with it. You can find Stephen Berlin Johnson everywhere.

He's on, I think you're still on Twitter X, yeah? You're somewhere over there. I am. Yeah, Stephen B. Johnson.

I've been tweeting. I'm going to say it. Yeah, do it. I'm tweeting a lot about notebook. I'm going to be sharing a lot of,

like, tips and things. about how to use it, which is, I've just stored up. Get in the arena. It's chaos. Exactly. In the arena, it's chaos over there. If you really want to get a great interaction going, just bring up Alex Jones and Tucker

Carson and freedom of speech. You'll get really great threats going. It's awesome. It's really chaos and it's full contact. All right, everybody. We'll see you next time on this week. It starts. Bye-bye.