if you want to use llms to write better with sentences brainstorming note taking all that jazz Steph Johnson Our Guest today has probably thought more about this stuff than anyone on planet Earth over the last I don't know 30 40 years stepen Johnson has written 13 books now he's using these tools all the time and he's working with Google to build their
notebook LM product so in this piece we talk a lot about structure like how do you structure your writing why is it so important how toes structure reced and into the background but why it's so important for writers to get right the dangers of over editing how to tell a story how to think about flow and all that jazz I think that Steven is one of the most prolific non-fiction writers
over the last 20 years or so and this is a master class in The Cutting Edge of AI how you can use that to improve your own writing at the level of the sentence the paragraph outlines and your overall piece so let's get into it here's my conversation with Steven [Music] Johnson man it is so cool to be doing this I'm so delighted to be here it is
so cool because along I was just saying to you in the elevator coming up at the St reges along a certain Vector of writing and thinking about tools for thought and how technology can amplify the writing process I think you've thought about this maybe more than anyone alive today maybe to a fault like maybe too much I've thought about it but um you know I'm really delighted
that there's an actual venue where I can talk about these things so thank you for creating this well you have been writing with technology and thinking about how to do this for 20 30 40 years years yeah and now you've had this opportunity to work with the team at Google on building notebook LM and thinking about how we can use technology and AI to amplify
thought what is the number one thing that you have taught the googlers you know I think the the folks that Google it's a Google Labs that were were building this notebook LM thing and and I think that I was originally brought in because one I'm a a writer and a writer who's written a lot about tools for thought as you said um but I think the the surprising thing that actually
surprised me in a way was something that I learned along with the rest of the team is that the the primary skill in getting a large language model to do something is command of clear persuasive Pros it's not a it's not it doesn't really matter whether you can program in C++ when you sit down to like write a prompt to talk to the model what you
really want to be able to do is like get the model to do something using convincing clear declarative sentences and so it turned out to be that like while I have almost no programming background there's a lot in the way that notebook Works where I I actually wrote The Prompt because we had a writer on the team right and so there was a quality of um kind of a surprising
importance of Pros style in in actually making the the application work properly which I think is just I mean nobody saw this coming in a way in the technology world that that writing skills would be so important to to getting machines to to function properly now and tell me what were some of the things that were really important to you as you began to
build this tool like the thing that stands out to me is the High status of citations and sources yeah the the sources are saved over time so what you can do is you can basically say I want you to pull from these canonical references of sources and then also it's very good about citing its sources in a way that say a tool like GPT isn't yeah well one of the
things that you know I'm sure we'll talk about is that I've had this long history of like keeping digital copies of all the things that I read so that they were in in the old days so that they were just searchable right it was just so if I had a digital copy of quote from books that i' read it research it was just a thousand times easier to like find that
quote six months later when I was you know working on the the book or the article or whatever that I'd done the research for what language models and AI have kind of changed all this is that it's no longer just about like searching with the kind of keyword string the way you know we used to do in in the old days um you can actually ask open-ended questions you can ask for summaries you
can ask for kind of highlights all these different things and if you design the system the right way so that you're not just talking to a language model but you're talking to a language model that effectively has read all of your notes or all of your quotations or all of your documents what we call sources as you said um suddenly you can do things that
go so far beyond search I mean one of my favorite things to do is um you can ask um you can ask something like in my reading notes what are the most surprising facts about dolphins and so it like on some level the the models actually are able to kind of understand you know in quotes what it means for information to be surprising right there
you're not just saying like find me all the things that mention Dolphins but actually find me the most interesting things about dolphins in my notes um and so we it was it was really animated by this idea that if you have a body of of core knowledge that is your kind of core source material and you attach that to a state-of-the-art language model the
whole workflow of kind of writing and thinking and researching is just going to be revolutionized um in one sense it's just going to be like 10 times faster like your ability to get to the Insight or the quote that you're looking for or to get a page of summaries of quotes that are relevant to the thing you're working on in that particular moment that's going to be much faster
but then you know we can talk about this in more detail like you also have this possibility of like really collaborating and kind of going to the AI and saying hey I just wrote this paragraph what did I miss like what else is in my reading history that might be relevant to the paragraph that I just wrote and for the first time in the history of computers the computer can
answer that question yeah I almost feel validated because many years ago I had lunch with a guy named Brian norgard who was really instrumental in building the UI ux Tinder of all things and I think he was a director of product there and he's a brilliant product guy and we sat down and we spoke about this idea that I had for this writing tool where what you
would have was on the left side of the screen you would be writing and on the right side of the screen at the top would be your notes and your notes would change and then there'd be like a public set of sources that would change and I deeply thought about building this product I was looking through my notes for it today and some of the even the
drawings that I did for it I did some hand drawings and I was like you kind of built this thing yeah you asked uh kind of what I brought to the Google team for this project like in a weird way I I do feel like I've been preparing to to make this tool for for almost 40 years like I've been thinking about this really since I was in since I was in college so you know 36
years or something like that and so you know kind of day one on the job everybody's like well what should we build and I was like well I've got 36 years worth of like ideas here like let's go you know and so I think we were able to kind of like we built it pretty quick with a small team in part because there was this stored up um body of of
thinking that I had well you were talking about surprise and one of my favorite things that's been said on the show comes from Derek Thompson right at the Atlantic and he says that interestingness equals surprise times novelty and I want to talk about your new book The Infernal machine and how you think about these elements of surprise these surprising stories as you
begin to weave together this tapestry of vignettes what are you reading for to find these surprises and then thinking through what actually makes it into a final book yeah that's a great question yes surprise is so important and such an important part of learning right you like you pay attention to things that surprise you like the the brain is constantly like
making predictions about what is supposed to come next which is by the way how a language model works as well which is interesting and that happens on a reading experience right you're like okay I'm reading this sentence and I kind of have a vague sense of where this is going and when you're bored you're bored because you know what's coming next and when you're
leaning forward and engaging it's because you're surprised by what's coming next and when you're really surprised by what came next you're confused right so so there's you have to hit this like sweet spot of I'm I don't quite know what's happening but I'm intrigued but I'm not completely lost and that's and that's the challenge of it in a lot of my books infernal machine
the new one um is a great example of this but um this is true and kind of half of them I would say uh a lot of the surprise comes from connecting different uh fields or different narrative arcs um uh different disciplines in unusual ways um and so you know in in infernal machine it's the story that weaves together kind of the birth of like
forensic science and the early detectives on the NYPD and the anarchists this incredible period of anarchist bombings that were happening in New York City really the worst period of terrorism in the history of the United States right around the turn of the century about 120 years ago um and the kind of birth of the FBI and Jay Edgar Hoover and so you have like all
these different threads like the invention of dynamite which makes terrorism possible um so it's a story that involves kind of science chemistry forensic science Anarchist political philosophy like you know the FBI and when when it starts you're kind of like I'm in Russia and I'm in London and I'm in Paris and I'm in you know New York City and all these different characters
who seem completely unrelated and slowly over time they kind of converge in the streets of New York and and for this kind of epic battle that unfolds um and so part of the surprise is just like the reader being like how is this all going to come together like how is this author going to get these 12 disperate people like in the same room on some level or
in the same plot and the Challen for that is you have to put enough kind of signals into the pros to convince the reader that you are going to pull it off because the risk is always when when people were reading in front the machine and in um early manuscript form the thing I kept asking is like in the first third do you feel it all lost and do you
feel like why are these people all in the same book or do you trust that it's going to converge and and in general I think people felt that but that was the thing I I spent a lot of time trying to figure out is whether it would you know you'd feel that confidence that it was going to come together right you were talking about how that's what llms
are doing they're guessing and I have to commend you I so you wrote that piece in New York Times magazine took you like seven eight months on GPT and I got to say that was my Quake moment for GPT I remember exactly exactly where I was sitting it's a 10,000w piece and I remember the feeling like you know what I wish I had I wish I had a heart rate monitor for how my
heart rate changed to that piece yeah but I got to tell you the highlight of that piece is the way you end the first paragraph and lead into the second paragraph you did such a masterful job I have to commend you can you talk about how you did that yeah so what you did yeah so I had that idea early on so there are two funny things about that
piece so first is um I originally I originally pitched it this is hilarious in the fall of 2021 and I went to my wonderful editor uh Bill wasik who's a great writer and his own right um and I said everybody's already written you know plenty of things about gpt3 everybody knows it's really interesting um but I want to write a different piece because it turns
out open AI has this really weird or organizational structure governance structure cuz it's like half a nonprofit and half a for-profit and no body knows about this and I think that would be an interesting piece which turned out to be very interesting because that's what caused the whole crisis uh a few months ago between the nonprofit group and the
for-profit side um so they the times was like great you know fantastic write that piece and so then I got access to gpt3 in I think it was like October of 2021 and I sat down with it the first night and just started experimenting and I was like oh man there is so much more to be said about this like this is so much bigger than I realized and so I went back to them and said I'll mention
the governance piece but I'm just going to write about what it means that computers have mastered language like that was the kind of framing of the piece um and so then early on I had this idea um given that large language models are just predicting the next word in a sentence um based on the previous words in the context um and given that the the
surprise and way I think everybody in the field feels that it was a surprise that that simple Act of kind of training and honing the the model's ability to to predict word sequences would turn out to be so powerful and so flexible and and creative in many ways um that that simple action would would generate so many possibilities um that's in some sense part of what the piece is about
and so it occurred to me that I could open the piece p and write a paragraph um and then omit the last word and so that the the reader would um and and the line is something like you know you're reading this magazine article but then something very strange happens it appears that the writer has deliberately omitted the final word of the
first and so then there's this Blank Spot and then I kind of step back and and say but you filled it in actually you predicted you you heard the word paragraph there um and just think about the kind of intelligence you have to build in your own brain to be able to predict that next word and all the kind of assumptions that go into the conventions of magazines and sentence
structure and the meaning of the words you've just read that enable you to make that prediction and that becomes the launching pad for the for the piece so I came up with this structure talk about you know the kind of structure of an article I came up with this idea I was in a hotel I I I just interviewed all the open AI people I was in a hotel room
in San Francisco and I was like sitting there like back in the hotel room drinking a glass of wine and kind of jotting some notes and I came up with this idea it's a reference by the way to the Italo calino novel if on a winter sign a traveler which I'm sure 1% of the readers actually got that too but it it's uh for people who are really into meta fiction from the 1960s it had added
meaning so I wrot I so I basically wrote that first paragraph and I was I was like a month away from handing in a first draft of this piece I had a ton I had not written any but it was 10,000 words so so I had a lot to write um but I sent it to Bill my editor at the times and I said the subject editor was are you going to let me get away with
this and he was like yeah I might actually I mean I'd like to see the rest of the piece but we might do it um and you know and then we ran it well was very self-indulgent and beautiful and actually like almost like moves with tears like it was so exquisitely done and I just it wasn't cheesy and you did it in a way where you said what you needed to say but more importantly you
showed what you needed to show oh that's very nice you thank you yeah I was very proud of that what are you going for when you write those magazine pieces because you did the Bitcoin one about four years before yeah yeah the the I've done basically four of these pieces for the magazine that have been very long and it's just an Incredible Gift to to
give a writer to give them that much length in in a magazine I mean it's just it's a lot of words um and and it it actually has been a it's a format that I think we'll see more of because it's a really lovely length um in that it lets you really explore a topic in detail um tell a complex story with multiple threads but you're not forced to fill out a 60,000 word book with that
information and there're just a bunch of stories or arguments that you know really shouldn't be condensed down to a 1500w block blog post but really shouldn't be expanded up into a um into a 60,000 or 100,000 word book um and on the Internet it's easier to you know you can publish those things that way or you can do multiple you know additions of it
substack uh email or something like that that tells a story that's that length but the magazine lets me do that and and they've been really a lot of fun to do and how how do you think about structured differently between 1,500 word article a 10,000w essay and a 100,000 word book yeah the biggest difference is between the the short format essay obviously in that
point you really you're limited to what you can do there's just a certain amount of um Headroom you just don't have so you have to pretty much make a point um and you can try and figure you know you're try to figure out like what's the what's the best opening paragraph here and you know you're going to tell one story maybe in the course of the blog
post or the the newsletter or the short article but you're not able to weave together things um in the same level but once you get to 10,000 words or 100,000 words I don't know they're possib I think of them quite similarly like for instance the the piece that I did about this guy Thomas midley Jr um The chist Who uh created both leaded gasoline and the CFCs that caused the hole in the
ozone layer one guy it was called the man who broke the world he dies he he gets paralyzed with polio and um builds in a lab because he's a kind of inveterate tinkerer and inventor he builds a Contraption to get himself out of bed with a series of pulleys and stuff like that and then he ends up um getting strangled by it and dies at the hands of the machine that he built now
oh wow it's believed that he actually used it to commit suicide but the story was that he accidentally died that way so so here's this guy who invented all these things that had all these devastating consequences and then he invents something for himself and it ends up being an instrument in his own death right so quite the metaphor quite
the metaphor right so that that was a great example of like the structure of a piece like that we um uh bill and I wrote this with with Bill wck as my editor um we're like it it it should be like um Sunset Boulevard there there's a dead body floating in the pool at the in the first shot and then it's like we're going to back up and find out how that
body got there and so like midley dies in the first paragraph in that piece um and then but you know you could have made a different choice you could have been like and the ultimate irony at the end of this whole story is you know that Mally died and the other thing about structure in in a piece like that um is did you aren't quite as bound to chronology in the way that you maybe are
with a non-fiction you know history book like the book kind of wants to flow in a linear fashion from from kind of start to finish with with flashbacks occasionally um but with that piece we you know we contemplated a lot of different structures where we're like maybe we you know maybe after he dies then we're like contemplating the landscape of like the Carnage that he
Unleashed on the world and the ozone hole and leted gasoline and all that stuff and then we come back to like understand how we invented it and what he was thinking at the time and I don't think you have an expectation that you're going to go on a strict timeline and in a piece like that whereas in a history book there's a bit more of an
expectation you do that tell me about these relationships that you have with your editors you've mentioned a few editors already yeah well another great editor story is um so I wrote a book um maybe my favorite book of all of them except for the new one which is is the new one always the favorite book it is a little bit because you're not sick of it
yet you know um but I wrote this book called enemy of all mankind which uh came out a couple years ago it's it's basically a story about a Pirate Henry ly who pulled off the arguably the largest heist in the history of crime in 1695 and and the story of how I got to that book we should talk about but um one of the things that was challenging about that book is that it had again you
know this multi-threaded structure where it was telling the story of ever himself and how he kind of assembles his crew he mutinies a a ship to take it over but it's also trying to set this context because it's right as the East India company is kind of taking off um the uh richest man in the world aranga the grand mugle of India is a major figure
so you have to kind of like explain the kind of birth of corporate capitalism and in public companies and the East India Company you have to understand why India was so wealthy at that point the Calico trade that was generating all this wealth that and so there's just a lot of scene setting to do and the first chapter is kind of like bounced around
you know in all these different directions and the main protagonist or you know kind of anti-hero of it Henry ever the pirate would go kind of unmentioned in the first draft he would go unmentioned for like 15 pages at a time and when my uh editor Courtney young at Riverhead who I worked with on a lot of books is fantastic um when she read the first draft she was like it we
can't ever lose sight of every for too long like that he's the one hook that the the reader has he knows that there's this pirate and or she knows that there's this pirate and um if you leave the pirate for too long they're going to be like where's the pirate I the Pirates you know they're gonna be lost and so she she actually said that she she built
you know they call it a crazy wall which is like the scene from you know the movies where somebody's you know it's like The Usual Suspects or something like that they've got all the suspects up on the wall and they're trying to see the connections between them up so she built basically a kind of crazy wall for the beginning the first third of enemy
of all mankind where she had like these are all the pieces and like how can we organize this so that one it goes a little faster it's a little shorter and we never lose sight of Henry ever for more than six pages or something like that we had some kind of goal like that and it's like having somebody who um just like a second pair of eyes who's looking at something that early and
thinking about the structure side of it um because later on you know when you share a book in manuscript form generally people are not coming back to you with comments like you should really reorganize the whole thing so that it it's told in Reverse chronological order something like that like no like that's not the kind of note you get you get
like hey I really like this I want to hear more about that character and I thought this was a little clunky in this part or I disagree with the argument but the the the Deep structure that's kind of baked into the book most readers it's like a law of the universe of the book I mean it really is like designing a house right so you're early in the house and you have the early
schematics and someone's like hey you put the garage all the way on the east side it needs to be all the way on the west side because otherwise they're going to walk in through the bedroom before they get to the living room and that's going to be a weird situation but if you've built the house and it's all there and you say something like that you're like yeah could you have please
tell me that told me this earlier once the house is built you need to be talking about the wallpaper you need to be talking about the furniture you need to be talking about what hangs on the walls so the actual content of what is worth discussing changes over time the so there two really crazy stories about structure in in in my career I think um
and the the the first one is uh my book The Ghost map which is actually the despite the fact that it's about chera is actually the best selling of all my books um I think it part because it has this kind of detective story structure it's it's uh following the this terrible outbreak of color in London in the 1850s and this guy Jon Snow who's not
the Jon Snow from Game of Thrones uh who basically solves the mystery of where colera is coming from which is a major medical advance in the period um and you know it's a major kind of breakthrough both in terms of Public Health but also the history of cities and all that kind of stuff and so it's effectively there's a killer on the loose in Victorian
London and there's a detective trying to track it down the Killer is a is a bacterium not Jack the Ripper but it's the same architecture right so so that that part of it was pretty clear like it's going to be there's going to be a kind of detective story structure to this but I also wanted to Riff on all these different ideas I wanted to Riff on the history of cities and the
relationship between microbes and you know the metropolis and I wanted to Riff on maps because there's a map in this in information design that's at the center of the story and I wanted to capture like how crazy London was at that period and all stuff so I had all these themes that I wanted to explore and at some point in the early stage of kind of
sketching out that book I'd done a lot of research but I hadn't really started writing um I realized that I could do it um where a structure where like each chapter would be a day there was an actual chronology to it so it takes place over about two weeks of this outbreak and so chapter is like you know um Sunday August 30th 1854 and there's a title whatever but each chapter is a day
and I realized that the given the events that happen on each day there was kind of a logical um opportunity to Riff on one of the themes that I wanted to cover that day and so I figured out oh wait day one I can do the whole thing about like scavengers and waste recycling all stuff in London because I'm introduced to the scene and I can start there and then day
two I can do you know the um birth of Public Health because it's starting here in day three because of the events that are happening I can do that and it's to this day like one of the things I'm most proud of that I've ever done in any of my books and um that book was reviewed extensively got wonderful reviews no one has ever mentioned that and I think that
it's one of these things like when structure works you don't notice it exactly you know and so it's but it's like design you feel it yeah well we were just walking out of the elevator here when we came up and I said ah it's a Norman door and what that is is the door pushes but there's a handle so you think it pulls like right when you push it should be
flat because that implicitly communicates to you that you push and then when you pull there should be the handle and I've gone in and out of that door like 25 times since coming to the hotel every time no I'm increasingly angry about it you know it's not it's it's not like I'm going to ever make peace with this and that's just like structure that structure is invisibly
telling you what to expect what to think and like you said when it's done right it just recedes into the background but this is the thing that writers know how important structure really is because once you find the structure a lot of things just fall into place totally so the other structure story that um is interesting which I think has a slightly
different lesson which is more or less the importance of structure which is the same lesson which is um I've had several cases where an enemy of all mankind the pirate book is is the most dramatic example of this where the the original idea for the book was the structure and then the content came later so with anime I had this idea of like it would
be really cool to write a history book that begins with an event that is ideally very short and compressed in time but something dramatic happens you know maybe ideally it takes place over like minutes not you know days and you start with that event in the preface basically and then you go back 300 years and you tell the story of all the forces that had to converge for that event to
happen the you know macro technological scientific forces the you know the individual lives the economic forces whatever it is and then the middle of the book converges back on that thing and then the second half of the book is like all the downstream consequences of it right I was that would be a really beautiful structure for book but I was like I don't
know what to put in that structure I don't have the content for it I just know that that architecture would be interesting and so in my I'm sure we'll talk about this in my spark file somewhere you know many years ago there's a note like it would be really cool to have a book that's structured kind of like an hourglass where there's a thing in the middle and you but and I
kind of described that structure and then I sat on that idea for literally years like I think it was maybe like three or four years it was in the back of my mind I wrote some other things and I was like I'd love to find the right thing to put in that structure and then at some point I thought oh what if the event is a crime which isn't the most you know
Shocking True Crime is obviously like a relatively lucrative field um but I was like the nice thing about crimes is that they're they often are fast like they happen in instance in you know a matter of seconds or minutes or something like that and so what if there's a crime that had a lot of forces leading into it and a lot of Downstream consequences and so
that narrowed it a little bit and so I could start basically I could start googling like I could start like famous crimes in history kind of like just exploring the world of like reading books about the history of crime and then I think maybe like a year later I stumbled across the story about Henry e and I was like that could fit and there's a note you know that's in my
files it's just like whoa what about this guy and there's like a quote from Wikipedia about him I'm like that could be kind of good that might fit that structure and then I dug around a little bit more and I was like okay that's it yeah I think that what's has shown up as a theme repeatedly is all this work that you do for many years without even
realizing that that work is going to have a payoff in the end so you're talking about that there where you had this idea for this this structure of a piece I've actually never even heard of someone who said I have an idea for the structure but I don't even know the content that needs to fit into that and then that's also where we started this
conversation where you said look I have been thinking about these things for 36 years and only now it feel like everybody's talking about it and it seems to me that the core thing there is this spark file where you throw ideas in and then things begin to grow organically from there it's like the seeds that you sew and then the fruits just sort of show up over time you're
like oh well but you were just sort of tending to that Garden without even realizing it yeah that so the spark file comes out of um the book I wrote where good ideas come from which the opening chapter which is called the slow hunch one of the key things I was trying to do in that book which is really about how Innovation Works and creativity Works um
was to say like the whole dominant metaphor that we have for creativity is like ridiculously over indexed on Eureka moments and light bulb moments and epiphanies and like I the Apple fell from the tree and I have a theory of gravity like everybody loves to tell the story like that but actually when you go back and look at like really interesting ideas whether they're creative writerly
ideas or scientific ideas they almost always have this very long incubation period where there's like a there's a sense of possibility there's a hunch that that there's something interesting but you're not really sure what to do with it you're like hey I've got an idea for a structure of a book but I don't know what the topic is you know like I
don't even know literally what it takes place or who the characters are and they just stay in that hun State and they kind of evolve and connect to other ideas a little bit but they often stay there for long periods of time before they really germinate to use the seed metaphor you using and so that if you if you believe in that theory which I do obviously um that means you have to be
like writing all those hunches down because those are and and and revisiting them because those are like the hunch is the most fleeting of ideas because it's not attached to anything it doesn't have any actionable you know utility in the moment it's just like I think there's something there and and by the way like um we talked earlier about surprise like
that surprise is kind of a like I just try to capture that surprise me even if I don't understand them um and kind of like lean into those things be like okay that was weird I wasn't expecting that and I don't even know what that means but because I don't know what it means like I should store that away because maybe I will eventually understand it
you know whereas I think a lot of times people are like well I would like to just focus on the things that I understand one of the things I just want to pop in here because one of the things that's important is that you're not overintellectualizing this you're not saying oh it needs to to work out in this particular way you're just like huh I just felt a ping of surprise I'm going
to trust that and I'm going to write it down I'm not I don't need to justify every surprise yeah and and this is um you know this is something I know you've talked to Thiago Forte who you know a power of mine and and someone we've worked with a little bit with notebook LM and like he's great on this too which is like capturing things before you know
where they go like the the idea this is why I'm kind of opposed to even tagging things um because I feel like you just want to capture things um without spending any time you want like frictionless capture as possible and any time you spend being like oh this idea connects to these seven other ideas and I'm going to build the tags and the link
system to like make sure that they're connected like to me that gets in the way of just reading and thinking and um and but but you do need to capture it and so the the way that I did that for for many years is that I would create what I called a spark file um which could have been called a hunch file if I'd been like consistent with my branding I suppose but anyway I called
it a spark file and it was just like it was initially a word doc then it became a Google doc um where I just put everything that occurred to me that was interesting at all and it would be like random quotes from things and random story ideas and um and you know oh my gosh there was his Pirate Henry ever you know maybe there's something there um
and it was I don't know it was longer than it was longer than any of my books I think it would ended up being like 120,000 words or something like that it was a big long file and the key thing that I tried to do uh was to revisit it every I don't know six months or so and really just kind of read through it um because the thing about the hunches is that you know the hunch you had in
2016 that kind of didn't make sense but seemed interesting in 2024 like maybe you have the missing piece that turns that into an actual like story idea or a startup idea or whatever it is you're you're doing for your work um and so it gives like it basically like gives you the opportunity to kind of like brainstorm with your past selves in a
weird way like oh I had this idea eight years ago and now I'm a different person and I see it in a new light and I can now finally kind of complete the thought in this way um so that that has been very valuable I've kind of switched over and I'm now trying to kind of do that in side of notebook LM which is kind of interesting and trying to figure out how
to design notebook LM to kind of encourage that workflow because now you know we can get into this more but now instead of just me rereading the hunches and in the spark file and making the connections on my own now I have this AI that can also make connections and that just was never really possible before um the thought that my computer might come
to me and be like Stephen you just wrote this paragraph but you've forgotten this thing you wrote a while back um that actually might be relevant to you like that possibility is really electrifying for me yeah have you've heard that there's a great speech that Johnny I gave at Steve Jobs memorial service and I was the chief designer at Apple and so close to Steve Jobs and he said
the thing that Steve Jobs understood better than anybody else is that that young ideas are fragile yeah yeah I've seen that and the young idea has to be treated and I don't know if he said this or or if it's how I imagine it that it's like a little Dove and it's like this little baby dove and you have to hold it in a very tender way and you have to
nurture it and if anything's too harsh the dove will just die and you have to sort of nurture it and let it grow and then only later can the dove fly but early on with those hunches there's a certain way like a tenderness a care that you have to give that idea because those ideas can be squashed so fast so easily like I it's almost like it's not so much that they need to be you can
almost like ignore some of the hunches like as long as you capture them and don't just like criticize them out of the gate um as long as you just give them some time um I think that that's that that's really the best approach with that kind of thing and this is a theme that shows up over and over and over again since you wrote that 20 five piece in New York
Times called tools for thought how can we improve amplify our Thinking by using these Technologies so yeah I first got the little taste of that in I was I was a sophomore in college was 1987 and apple came out with a program called hypercard that most people have forgotten now um this is before the web had even been kind of designed at that
point burner's Lee was still two years away from like kind of first version of that I was not connected to the internet in any way I didn't really get on the internet till I went to grad school um but hypercard was a kind of preview of the web it was a it had a kind of hypertext architecture you had these cards where you could put information on
them and you could basically make links between the Cards and it was kind of a note-taking tool it was kind of an open kind of human natural language programming language kind of tool and it never really took off but everybody who got got into it like I did you know Chang the course of their lives and I got obsessed with like oh I could build a tool to help me with my research for
my classes and I'm going to take my notes in this tool and I'm going to optimize it for the way that I like to work and I'm going to have this kind of research patol assistant here and so it had a little programming language and so I kind of this like little kind of taught myself out a program and and I built this program called curriculum that's what I called it and it was the
joke about it is that I like got so obsessed with building a tool to help me with my classes that I stopped going to my classes cuz I was like I just want to build this tool you know and I used it for a little bit but it it's one of these things where you get a taste of something that is possible and like the technology is not ready yet um but you
you get this little glimpse of what might be possible and it just sticks in the back of your mind and you spend the rest of your life chasing it and so that's kind of the journey I've been on I started using in the early 2000s I started using this program called Devon think or Devon think I don't even know how it's pronounced um it anticipated
some of the stuff that was in that's now in large language models in the sense that it could detect semantic relationships between blocks of text and so even if the two paragraphs in question didn't have any words in common if one was about you know kind of game theory and one was about um tit fortat behavior in bats that are AB linked to each other it was amazingly able at the
time to say hey these things are closer to each other in kind of the semantic space than these other paragraphs right and so what I did was I started um laboriously typing up all the quotes from books that I read I would sometimes have research assistants do it or I would do it I would try OCR but it was terrible but I would basically try and get these digital copies of all the
quotes from the books that I was using as research or articles or whatever and I would put them into think and then I could find things very quickly which is great and you know helpful like what was that quote from the Jane Jacobs book oh there it is okay great I could just paste it in um and and I would always put the page numbers there so I had
citations which was you know another Obsession like making sure you know the prominence of these ideas and getting that right um but there were the these hints of like maybe there's a higher level kind of collaboration here and the example that I talked about in this in this tius magazine um it was a tias book review piece um Tools For Thought um
when I was writing ghost map I had uh a theme about waste recycling and you know just how a city deals with basically human excrement which is like the major problem when suddenly you've grown to like two million people in the space of 50 years or whatever and so I had a big thing about like waste removal and Recycling and things like that and I
did a search in Devon think on that and it brought back this seemingly unrelated quote about how our skeletons are basically an evolutionary um kind of adaptation of uh calcium which is a waste product of cell metabolism so we were cells were expelling all this calcium and then Evolution was like hey actually I could use that maybe we could build a backbone
with that um and so seemingly unconnected ideas but I was like oh actually this is cool I can do a whole riff that goes from like the way the cities recycle the way Evolution recycles all kind of stuff and that became like seven pages of ghost map but when I stepped back I was like who had that idea like was it me or was it the software and on some level it was kind
of both of us like I described it I think in that piece as like a duet like it felt like there were two different kinds of intelligence kind of working together in that and a big part of it that is is crucial and it's crucial to what we're trying to do with notebook Alam as well is I curated all those quotes right so there's kind of like my human
intelligence reading all those books deciding that these are the most interesting and surprising and you know useful facts or passages and all those quotes and then there's the software's ability to maybe detect connections and and certainly scan the entire body of work faster than I could and then it surfaces a bunch of connections potential connections and then we get
back to my human intelligence at the other end of the chain saying like oh that's a keeper and so we're doing different things at different stages but like that was a new kind of creative workflow for me that I got a sense of in that well it's funny because you were talking earlier about these glimpses of the future where you see that something could exist but it
doesn't yet exist and you just have this this this dream and the dream that I have that I really want to manifest in the world is I want to be able to sit down at lunch with you and what I want to be able to do is basically have some sort of hologram type thing that is basically this third-party Source notes that is researching what we're talking about so
you and I are riffing and here we go so I have gone on a walk in Central Park every day for like the last four or five days and the architecture is fantastic so there's there's been a few buildings and now I'm starting to think okay what why is the architecture in Central Park so good because it's so good that something's going on okay so we started
talking about that and then you're like hey well what's an example and then there's that place like if you sort of walk up from uh call it like 6th Avenue and you sort of walk up and there's that area with the fountain then you walk through this beautiful place but I don't know what it's called but I know that it's really beautiful on top but now as we're having this conversation I could
basically like pull open the Hologram and you and I could look at it now we could begin to rotate it with our hands and then you and I could be looking at this image that's right there together and I know that the technology is basically there now with apple Vision Pro we're going to be able to see something like that and it's like that glimpse into the future and it's the
same sort of thing where over time as well the the tool the technology it becomes symbiotic with who you are so you begin to know it it begins to know you and you begin to work work together and it leads into that question well who discovered what and it's almost like when someone has one of those imaginary friends when when they're a kid it's
it's it's you're just in that duet together yeah I feel this way a little bit with my kids because they're so fast at typing on their phones huh that that will be sitting at the dinner table we don't you know have phones out at the dinner table but like some fact will come up and we'll be like well I don't know actually if that you know during
Kennedy's Administration if that's what they pass this is what we talk about at dinner and uh and and they'll just be like uh yes it was like they can just get to the fact so fast on their phones that I feel like they already have a kind of a version of that instant kind of knowledge bank that's there but um yeah it's going to be it's going to be really great and I
think yeah the part of the question is with the with the AI part of this all is to what extent does your kind of language model co-create collaborator um really get to know you over time right like these are your obsessions um well this is something that stands out with notebook LM is how much State there is so most llm tools you have the interaction and there's a
state within intera within the interaction So within that chat file on chat gbt it knows where you are what the conversation is like what you've built in is that you can save individual parts and then over time it retains a much deeper understanding of the individual who it's working with and that seems to have been a very deliberate design decision for you the the I think one of
the most powerful things about opag gullan which we're still one we're learning how to explain it to people and two we're still rolling out some of the features that really do this but um in a way it's we tried to redesign the whole kind of no ticking research tool um app application all the conventions of it um from the ground up knowing that you have a language model
there instead of like adding a language model into an existing UI we try to like think about how you would do it differently with that as a kind of a given and and part of what the interface is design to do is to give you really easy ways to focus the AI on different things and so the first thing you do is like you load up your sources that are
you know let's say it's reading uh for your class so you've got a bunch of teex books that you would load in there or something like that or it's your research notes for the book that you're writing so then the model is like kind of grounded in in that information so you are effectively saying hey large language model um I want you to be focused on this text and anything I ask
you I want you to answer based on this text and I want you to show citations I want to see the work I want to be able to jump immediately to the quote that you're using to answer this question um but then you can take some notes and you could be reading through you know your textbook and you could get to like a complicated passage and you can just
select that passage and say hey now I want you to explain this passage so I don't want you to pay any attention to anything else I just want you to look at this paragraph and and help me make sense of it then you take a bunch of notes and then you can select all those notes and you say now I want you to focus just on the notes that I've selected and I want you to turn those
into a draft of a blog post or I want you to create a study guide for me so that I can learn this material more and so what you're do now technically you could do this in chat GPT um you could go into your textbook copy the text switch Tabs go in there paste it in right please help me understand that get the understanding of that take that back
to your notebook and paste it in there but the the kind of cognitive load of switching but like the switching cost of going back and forth between all those different applications and tabs and things like that is just it's like really bad for your thinking and your creativity you're not in a flow State when you're doing it you're just kind of
like oh now I have to copy from here and paste in here and I hit return and this so notebook is is is designed to kind of like do that all in one unified kind of place um and I find it incredibly powerful but it's a different way of working and so like I I we joke the office like I no one knows how to use notebook LM you know better than I do like I have spent so much time with it
um so the the trick is now like explaining it to people and like getting people to understand this slightly different workflow tell me about the tagline then to go from information to Insight what is that process like and why is that this very Central oneliner yeah I so one of the things that I did the other day which was just amazing that you can do this um I have a
notebook inside of notebook LM where I just write all my random thoughts my kind of hunches about um the the software itself and what we're doing and so it's everything from I start drafting a little blog post about it or I have a kind of crazy idea for a new feature or notes to myself and it's just a bunch of random things and so the so the other
day I just selected the whole noteboard for like I don't know 45 different notes I think it's like 7,000 words worth of stuff selected the whole thing and I just said what is the most interesting idea in this collection of notes and it came back with I thought actually a a good one um so one of the things that notebook does is when you ask a question based on your source es
we suggest follow-up questions based on what you've asked and what your sources contain and so if you're reading that American history textbook and you ask about the Colombian Exchange it will say like oh you know what about asking about small pox and the impact on you know indigenous communities whatever it give you three or four and so you can
explore a complex new document by following these kind of chains of questions and it you it's a really interesting way to kind of get your bearings and something and I was describing it to myself in this note as that it was like conversational hypertext and there there actually aren't links right they're not blue words that you're clicking on you you're
just like I ask this one question and I ask a follow-up question I ask a follow-up question and that is a really powerful way to learn but it wasn't possible before unless you you know were lucky enough to like talk to the author of the book or like a teacher that was really well-versed in the book that you could ask a question of but you couldn't
do it in any other form and now now we can't well what I always say is that the llms are really good at three things they're good at diversity they're good at accessibility and they're good at speed so you say well you could do that if you could talk to the author and I'm not here to dispute that yes if you want to learn a subject you're trying to
learn about Robert Moses and Robert Caro was sitting right here I'm not disputing that he would be your guy but but but what llms are so good for is you can get the diversity through the interpretations of different people so I could get Robert Moses and I could get six other New York historians I could create a debate between a few of them accessibility I mean this Tool's
free and then the speed the instantaneity of it who knows how how would we even get Robert Koo here right now I wouldn't even know how to begin doing that I can do all three of those things instantly and when I think about where llm technology is so good get along those three vectors diversity accessibility speed Cario has got to be like 20 blocks from us right now
bet I uh we could probably find him but but yes it's a great point the speed thing is a really interesting one and you know what it reminds me of um it reminds me of something that people really I think underappreciated about Wikipedia when it when Wikipedia kind of got big enough and and reliable enough to to use um which is that Wikipedia and to a lesser
extent the web as well you know kind of post go Google web um uh it enabled you to do very quick explorations of ideas and for me this was an unbelievable godsend because I'm constantly you know jumping into fields that I have no actual expertise or training in and you know kind of be like what's the deal with that you know like what's the deal
with the history of air conditioning and oh that's interesting maybe I should read more about that um and so I think that that speed thing is really incredible and and that's part of what you know we're seeing again on a on a new level like your ability to um in s like kind of create a like dossier or kind of a briefing document on your notes like that you might do or research
system might do um that is very tedious work right like you interview somebody and you want to like what are the highlights and Main themes of this interview let's turn it into some kind of structure that's hours and hours of work and you can literally just like select that Source in Notebook LM and say like create a briefing doc with key
quotes organized by themes and it's like it's it's not always perfect but like the first draft of it is is crazy so that's winding all the way back to your question by like getting from information to Insight like it's like there's a bunch of raw data here I'm trying to turn it into like usable insights um it's it's just an extraordinary platform for that well to
your point about information to Insight one way to think about information is that it's you have a lot of you have a a lot of needle but you also have a lot a lot of hay stack Insight is about just the needle and one of the things that I find LM to be very good at is give me the on liner give me the one sentence and what I like to do is I just like to hit refresh refresh ref fresh
just to see how is it trying to compress and distill this idea not so I can necessarily just take what it gives me but so all of a sudden it's like a cubis painting where you can see the same thing from this angle from that angle little bit from above a little bit from below over and over again and it sort of gives me it's sort of like when you're
watching those NBA games or a football game and the camera sort of swings around from one lens to the other woo and I like looking at ideas from 360 angles like that you know one thing that um we haven't rolled this out yet but I'm really excited about this it's something I've always wanted as a writer um we haven't talked a lot about like writing on the level of sentences let's
get into it my skills skewed towards the stuff we've talked about the architecture the kind of um storytelling uh those kind of bigger things and I'm I find that I have to work hard to make my sentences interesting my default setting is to make this as intelligible as possible on the level of the sentence um so I think of like the sentence are very
clear the overall structure of the book and the argument of the book The Story the book is telling is actually very adventurous and maybe experimental and but this but you you're rarely laboring on the level of the sentence to understand what's going on um and I think that is good that has helped me in many ways but I sometimes find particularly with books like enemy of
all mankind or or infernal machine which I want to feel somewhat novelistic um that I have to wor a little bit to make the sentences more just aesthetically interesting and it also hurts me that I'm not a particularly visual person like my visual memory is very weak compared to my semantic memory my linguistic memory like I can remember I can think of a
paragraph when I'm on the subway and come home and like transcribe it like word for word almost um but I'll meet someone and my wife will be like what did they look like and I'll be like I don't remember I don't they were a man as far as I could tell wow you know I just don't have much of it so do you think that that's a trade-off I well yeah is it is it actually a trade-off
like to is it um you're so good at this that maybe your action is really bad I don't know whether and I don't know whether it's just because I focused on that and and that other part of my brain is ateed or if I came into the world with less visual memories probably probably some combination of that um but I used to joke that I would like never
be able to be a novelist cuz I would like introduce the characters and be like he had a nose and two ears just like could you give us a little work so but but on the level of sentence too like I sometimes find that um you just I think everybody falls prey to this you get stuck in a certain way a certain Rhythm certain you know um pacing of the
sentences and or sometimes you just write it out and you're like this is just a little bit boring but I can't think of another way to phrase this um and so we built this thing we have we haven't turned on yet but but it's pretty cool that we um I was calling paragraph thesaurus and it's basically like a thesaurus for the entire paragraph love it so I was like take
this whole paragraph and just rewrite it but keep the meaning the same but just like create different metaphors create different things whatever and because and and it's not so much that I would necessarily use it directly but it's just a way of like getting out of your routines well I think the core thing here is that the brain is much better at
contending with something that already exists than going from zero to one so totally what you're doing there isn't asking the tool to make you a statue you're asking the tool to give you different kinds of marble and then you can take those marble those different kinds and then turn it into the statue and what is happening right now is that the computers are getting really good at
going from nothing to a decently good something and a lot of the shallow critiques are I'm not getting a good something so two two stories first off what I I remember coming home and telling my wife about this paragraph The Source idea and being all excited about it and she's like um that is a plagiarism tool like that is like take this paragraph and rewrite it so that
it's no longer plag and I was like you're right so there's all I should say like there are issues with that that you know you have to figure out how you point should we just all be citing notebook LM G at the bottom I mean I almost want a disclosure on my site this I have pulled from notbook LM chat GPT every time you should just assume that
at this point yeah I have some ideas about how particularly in the classroom you you could you could deal with that it's a version of that um but uh but the other thing that is funny about the um paragraph the source which gets to the craziness of the language model so I was testing it on um uh Gemini Ultra which is like the the kind of like the Topline Google uh language
model that hasn't really been released I think maybe Bard Advance now uses it and we're we're starting to use it at notebook and uh so I was using another program to kind of test this thing i' written The Prompt I was basically like take this kind of somewhat flat paragraph and basically like make it more interesting and exciting whatever
and so I gave it uh a very Bland factual paragraph about the climate of Hawaii and I was like okay transform it with new metaphors and just make it more interesting and it came back with the most crazy paragraph that was like the climate of Hawaii is like a symphony that has been conducted by Mother Nature and the many motifs you so totally
useless like version and so all I typed in response was dude that is a little over the top and the model said uh you're right I'm sorry how about this version this is a little bit less successive and it was perfect and I was what I thought was just so crazy is that I could type dude that is a little over the top and that that was legitimate instruction to give a computer and it
would it would give you the the answer you were looking for it just was just crazy tell me more about the sentences what do you do to refined sentences refined words and one of the things that I've sensed you can tell me if I'm wrong but it seems like you're fairly skeptical of over editing that you're like I don't want to go crazy on my editing one key advice for writers
who are going to be published somewhere um and are lucky enough to have an editor is just trust the edit like don't you don't have to get it right because there is going to be an edit and if you have a good editor like things will get fixed and if you spend too much time in that first draft trying to get it perfect it's just is a waste of your
time like you you want to kind of charge ahead and get it done um but the the other maybe less um typical advice and kind of strategy that I veloped over the years is um that's specifically to do with the book but I think it's true to some extent of a longer article um is that you were really battling your own familiarity with the text because a book is like a song that you've heard too
many times if you read it 20 times it loses its appeal right and so uh I learned early on in my career that um my natural inclination was to sit down every day and reread the chapter I'd been writing um and then start writing which one was a waste of time took you know might take 20 minutes to read up to wherever I was in the chapter but two it meant that by
the time I finished the chapter I had read it 20 times or how many days it took me to to read it and so it had just all the pleasure and joy and surprise of reading the the my words was just wiped out I just knew it all by heart and everything seemed really obvious everything seemed really predictable so much of writing is projecting yourself
into your reader experience and mind and being like are you confused are you excited are you energized by this are you on the edge of your seat are you bored like is this working and um if you yourself have read your book a 100 times and you sit down to like do the first past edit of it and you read it basically 100 times cuz you've been reading the chapters as you go you can't
get into that mental state because you're so sick of it so you can't read it like a like a new reader basically would and so the technique that I kind of developed to to deal with that is to really try actively not to reread as I'm writing the chapters so I really sit down I do the hway thing of like leave yourself a little nugget so I often like the last couple of words will be like
and then do this and then do that that like little instructions to myself about where to go next so I have something to go on and I just write and then I finish the chapter and that I will print out and kind of like read or read on the iPad um and do an one pass edit on it and then I try and put it to bed and not reread it and so that means that like
for the longer books like infernal is like almost 100,000 words so it took me a long time to write um by the time I get to the end of it and really sit down to read the whole thing there's whole sections that I forgot you know and and every now and then I've had cases where like not only have I forgotten whole sections but I've actually like duplicated
sections like oh I forgot I already did that riff and now I'm doing it again here because I've totally forgot it but it gives me the sense of and it's also I find that um changing the place where you read and the format is really important um so like I will I used to print it out now I do it on the iPad so I can capture the edits as a PDF and go
back and look at them um but uh but changing the type face and where I'm reading it so I kind of I'm like I I make it look like a book and and so I switch over and I read in my chair not at the desk tell me about word choice or sentences are there any sentences that you feel like have really been some of your favorite sentences that you've had to develop and develop over time but you
feel like through that work you ended up with something you really proud of ghost map I had this idea from the beginning that I wanted to literally just immerse the reader in like literally in the of London that was literally like drowning in human excrement and was killing people and you had this kind of like scavenger class of of folks who had
who were playing all these kind of impromptu roles at recycling waste um some of them involving literally like human waste some of them involving just trash was out there and so there's this whole like typology of these different kind of waste removing um uh roles in the city that were not organized by the state at all they were just you know kind of organized by people being
incredibly poor um and so I was like I just want the book to just plunge into that world and there were two parts of that like the first line of that book is like you know something like it is it is the summer of 1854 in London is the city of scavengers is like the opening line and then there's just like long litany with all their names they have all these
like crazy kind of like Victorian London kind of Cockney like names and so it was the combination of like this very short declarative like Punchy sentence and then this like crazy Carnival of just terrible uh you know mid Victorian squalor and urban living um with all these like crazy categories and names stuff like that and so that like that was one of those things where like I I
had the I I had the idea for it um probably like months before I actually sat down to write it like I knew exactly what it was going to feel like um one of my favorite things to do that I've been playing around with a lot is really long sentences like 120 130 words and having communicating to the reader that that sentence is going to be a moment
so maybe there's something really serious and the writing is clear words lots of anglo-saxon words that are short they're direct and we're getting a boom boom boom boom to the writing and I'm over time beginning to expand into more romantic language French inspired pros and then adding a little bit more energy and then I sort of lead into this Crescendo of a super
long sentence and when you sit down and you say how is this like a 130w sentence how is it going to be and how am I going to make the reader implicitly intuitively understand that this is going to be a moment how do I do that through the construction of what I'm writing I don't yet have answers but it's been something that I've been
playing around with and having a lot of fun with it it's a really great idea so I have a couple of thoughts on that one is I'm kind of a recovering long sentence writer like I'm in recovery cuz I cuz I I I went so I was a semi otics major in in college and basically kind of studied like critical theory postracialism whatever you want to call it der FCO and all these kind of folks
who were very much in Vogue in the late 80s um and I went to grad school in English literature and so all my heroes like wrote These incredibly Baroque sentences that had a million parentheses in the middle to qualify the word words like when I you know I want to put that word under eraser even as I talk about the existence of language but as an aside
before I even get to the main point of the sentence I just want to say this other thing here and so like that was my like natural kind of mode and I and I had to like unlearn some of that to become the writer that that I became and I do think that I sometimes have overcompensated for that um and I I think that that you know a glorious long
sentence can be an incredible Tool uh to use as well but the other thing I mean I was thinking as you were describing it that like those setpiece sentences were like a big tracking shot in a movie I gu like okay now we're gonna have this crazy tracking job where we're going to go into the restaurant and we're going to follow them around whatever like I can't believe it's all
one shot I can't believe they're doing this you know this was hard to pull off you know and uh and I I have there are a couple of points um infernal machine the new one is like the most cinematic of all the books I would say and there's a um one of the challenges was which is like an architectural Challenge and then a challenge on the level of paragraph um
so it's kind of fun to think about how these things come together um I wanted to write a preference for the book that would like set up that there was going to be a kind of Thriller like structure with like literally like ticking clock bombs and police force and investigators and undercover operations that was coming in this book you're reading
because the next thing you're going get for the first story of the book is all these different threads starting in Russia and these Russian anarchists and these French police officers and you know you weren't really going to know where it was coming so I wanted to I wanted to plant a bomb in the first in the preface and so there is a bomb that
goes off in the um basement of the NYPD in 19 summer of 1915 and so I was trying to figure out like how I could create that kind of tension and I finally ended up ending the preface has this whole riff where it's like you know if there was any place that embodied the kind of clash between the new science of investigation and the the terrorist movements of the day it was probably the
the NYPD and the identification Bureau at the NYPD and you know it was it was this night in 1915 it's bustling with energy because there's just been this cases happened or whatever and then I was like I really want to convey the camera like pulling out of that room and going down the stairs and seeing a briefcase suspiciously left in the doorway and the sound of like a clock
ticking inside of the briefcase nice and so I had to kind of like I I so the the final paragraph is kind of like but it's not just the activity upstairs that would tell you this is interesting like you'd have to really understand what was going on you'd have to walk out and walk down the stairs and kind of recreated that shot without actually saying like
imagine this is a movie you know um so like that kind of that's kind of one of those wonderful things where you're like it's you you're trying to do something very Visual and and cinematic in a way on the level of like five or six sentences but the work that it's doing is all about supporting the AR the entire architecture of the book because you're like I have to make this little
moment so memorable that they tolerate all the crazy Journey that they're going to be on before they get back to it well you're talking about writing for video you've made a PBS documentary so how is that different for writing for the for text we did two of them actually how we got to now and exra life we also did um how we got to now was interesting
because we did a there book versions of each of those um how we got to now I really wrote it as a book as we were making the show so we there were some places where we did a script the the chapters of that book correspond to the episodes there were six episodes six six chapters and so in some cases we wrote a script first and then I adapted it into a chapter and
sometimes I wrote a chapter first um and so by far it was the most collaborative um and if you wanted me to submit a book for kind of best writing prize I wouldn't submit that I think because if there's any way to be lazy you will always choose to be lazy in you're writing like if somewhere if if like the problem I had with that book was like if
the script been written there was a bunch of language describing the invention of air conditioning say that was you know in the script and I'm sitting here being like all right now I have to describe the invention of air conditioning and I'll be like well that paragraph is right there and I have the rights to it like I can you know and so and and TV language is is different from
my style book language stuff like that I think you can feel that a little bit in um and for instance Wonderland the book that came after that was going to be a TV show and then kind of second season of how we got ton now and we interestingly we couldn't get funding for it which is annoying PBS wanted to do it but we couldn't get funding so it
never became a TV show but I wrote the book first and this the style of it is very different um but I did like it was a reminder of how um solitary if if you're writing books like generally it's a like it's one of the most solitary jobs until the book comes out and then it's a very very public job for like 3 weeks if you're lucky and everybody's like
critiquing your work and and you're out talking to people about it stuff like that and then you go back to this very like private World um but the shows were fun and and and working on Notebook LM is fun because it's it's so much more collaborative so it's a nice change of pace to just have other people yeah if you well I guess this did happen you walk into a college classroom and you
say all right this is going to be the Steven Johnson writing class how do you structure the curriculum what do students need to know that you can uniquely teach I mean I think it would be I think there's some little hacks that some of which we've talked about just about like the workflow the the way you edit the way you read um you know but I think the
thing that that I have thought about um and kind of push the envelope as as much as most folks is is using the digital tools um as a as a research and and collaborative kind of thinking mode whether whether it's notebook LM or Devon thinker scrier which I have used for many years and and loved um and just that idea that you know the the software has become vastly more helpful
in coming up with ideas and organizing your ideas and making new connections um that you know it's not just anymore about just being able to search and find the thing you want but actually is can be part of the creative process in a really kind of profound way um I think that's probably what that's why people would be attending the class I think
we'd be to get those kinds of insights um you know another thing that I wrote about a little bit with my my usus scrier um that's related to the the hunch idea but slightly different which is this kind of idea of clustering of ideas so um and and and notebook LM can do this as well I did it kind of manually in script ner um so for instance with with Wonderland which is the book about
the history of the play history of play and delight and amusement and how it drove a lot of surprising Innovations um I you know it's a very open-ended topic right I knew I was going to write a little bit about music I knew I was going to write about like early autom like robots and things like that and I knew I was going to write about spices
because that was interesting but you know it could be anything right so I would just like collect Snippets of things um and I kind of read very widely and I would just kind of throw the quotes into Scribner and Scribner has this very elegant mode for kind of like you can have these little Atomic units that can be quotes for little notes and things like that and then you can kind
of just drag them into folders and so I would kind of like be kind of looking at all the snippits and be like you know seems like there's an interesting little cluster here about the connection between Robotics and music that I hadn't really seen before but like there's something there like I've noticed there like eight things there and so I would
kind of like create a little mini folder for robotics music and I would just drag a bunch of the Clips in there and then you would kind of see like over time like oh that pile like you have these little piles of kind of like fragmentary ideas like piles of hunches basically to keep the metaphors Inta and eventually you kind of look down you be like oh I
got like three or four big piles here like this one's really developed this one's really developed this one's really developed and at that point one of the cool things about the interface of scrib Nar is you can take all those little Snippets and just turn them into a single document very easily um so we we kind of borrow that notebook has a feature that is combined to single notes
so you can grab a bunch of notes and put them into one note so you can then kind of treat them like a document and that process is kind of like once you get a big enough kind of pile of interesting ideas that seem to be clustering around a certain idea you're kind of like all right that's a chapter yes you know and but you have to go through that process
of like you know I'm sure it's different if you're like writing a like very traditional biography or something like that where like you really the structure is really like anchored in events that are known that you have a timeline before you start but I think for most books like the that structure is really open-ended at the beginning so you got to figure out like
what are like what are the little clusters of ideas that fit together the word that's coming to mind for me is emergence yeah yeah yeah yeah no it's a very very bottom up kind of way very bottom up which was in yes you're alluding to my second book convergence yeah and and like um in a way the like the trickiest book like that for me was
um where good ideas come from which was ultimately organized in these kind of patterns of innovative environment so this it talked about slow hunches liquid networks adjacent possible Serendipity error one or two other um and it took me months and months and months to figure out that structure um like I had a million stories and a million kind of lessons but I didn't I
just didn't know how to Cluster them and I had a in kind of building on emergence in a way um I had this initial idea that it would be kind of across scales and so there would be a chapter about like how your how Innovation and creativity works on the level of brain science and then it would be like your workspace and then it would be like the office and then
would be like this city and then it would be like the media platform or whatever and like each chapter would kind of like You' go up the chain like that and I like tried to write that I think I did a few drafts of chapters like that and it it got too it you got stuck at each level sure so you're like in the brain chapter you were just like reading Neuroscience
stuff and then you got to the cities and it was like it was Jan Jacobs you know it was like and the whole fun of that book was like I'm going to take you from here and there I'm going to take you all you know connect all these different ideas all these places but each chapter got two siloed um and so I finally um and this is actually when I
started using Scribner scrier had a kind of like note card interface I didn't actually write that book in scrier but I but I organized it with the note cards I wrote it I think in word I think maybe the last book I wrote in word um but you could kind of drag around these note cards in Scribner and so that was the way I was like okay I can see these
different patterns now I can figure this out um but that took it was really really hard and it you know it's in some ways is the hardest for me it's the hardest part of it um although I think it's one of the things that I do best um but it's it just takes a lot of thinking I asked about the class that you would teach and something I want to try with our rap
passage students is an exercise where you take a paragraph and then what you do to develop your voice is we have eight different images of architectural Styles artso modernist and then brutalist and you rewrite the paragraph based on the architectur AR sty I love that would that be fun yeah that's a great idea so I that's even more clever than something
I used to do consciously so when I was in college and I was you know kind of writing in this literary Theory mode um I would deliberately write papers in the style of like i' be like okay this paper is written in the style of Michelle Fuko and then I'd be like this paper is written in the style of gu spak or whatever and and I would kind of like try and and you know it ended up like um
you you eventually build your own voice out of some weird mix of all those different styles I would say imitate than innovate yeah you you start off you're imitating other people it's sort of like I love it when you go to the met and somebody has the easel and the can and they're they're sketching the the statue or the painting in front of them I was at the where was I I think it
might have been the British museum I think I know it was in London I think it was the British museum or the National Gallery and there was this guy and he's sketching this beautiful Greek statue with this movement and this flow and I went up to him and I'm like what do you what are you trying to get at he's like well I'm trying to sketch that but I'm
trying to exaggerate the motion so so in the Statue there's no wisps but in my drawings I can sort of add these lines and I can create this movement on with my pen that doesn't actually exist there I'm trying to communicate that and I was like yes yes yes that is exactly how an artist should be thinking it is the imitation try to copy it and bring it
into your own field of view but then to add something and not just copy but to really be generative there too I you know and that actually continued on I mean I remember um I was still kind of coming out of the academic Pros style so I started writing kind of Journalism in like 94 and 95 right as I was doing stuff on the web for the first time and um Gladwell
started publishing in the New Yorker right around that point yeah and I remember just being struck by the he would he would often um end sections with with very short Punchy little you know five-word sentences and like I don't think I'd written a five-word sentence in like 15 years at that point like I just been like all my sentences were like 100
words long and I was like oh that's so nice like you can write longer and then but you get to the end of that section you're like boom there's just this little button at the end and it's done and so I remember thinking that I remember um before I wrote ghost map I read uh Devil in the White City by Eric Larson and which was kind of like I was reading it somewhat to prep myself to
write it was the first book that I'd written in kind of historical narrative mode and so just like how he which which also had a kind of Thriller detective structure but was also trying to recreate a a city in time Chicago in that case not London but so it had a lot of commonality so it was just like studying like the way that he did that and I went um back to lon's book
actually The Splendid in the vile that he just came out with because um infernal machine has a lot of dialogue because there's so many um uh autobiographies that have been written Emma Goldman most famously the anarchist um and then there's so many contemporaneous news reports where seemingly reporters are like in the events as they happening it's very weird
and so I had I didn't have to invent any dialogue which I would never do I just had whole scenes where like conversations are on the record like they're they're part of the historical record and so I could recreate what people said to each other and I never really had that before a little bit of that an enemy but not very much um and so larsson's book splended in the vile
does the same thing cuz he had all these diary entries um during London during the Blitz and so I was like okay now I'm I'm gonna there's another novelistic feeling to this book here that I've never really done so I want to look at another non-fiction writer who's done it in that way so would you learn I just I mean I I don't I don't I'm not even sure
if it's in some of those things it's not even conscious you know you're just kind of like you just get a feel for like how it's going to work and you just kind of then you just remember that tone yeah I remember I had a conversation with my grandpa before he died when he was we had piece of paper we throw them in the trash I was like I was looking at the
trash can I was like okay I want to think about this he's like stop thinking stop thinking you're actually way better when you just throw the paper in the trash can and I was like Wow Grandpa you're right I could just sort of look throw look throw and there's about the Consciousness like a lot of especially with these sorts of things the body just
knows but before the mind even figures out what's going on and I think that's the yeah the style thing is very interesting because it often is something that you just you you you get in you slowly develop a Groove for a certain kind of style that you get by just reading other people and reading yourself and you slowly kind of get to it but it's often not you know the the
um architecture stuff that that we' spent so much time talking about is is very conscious you're very much aware like I am trying to figure out what is the structure for this book and what should the first chapter be like it's you know but like what is the vibe of this sentence or the way that I write is is is less um deliberate I think
sometimes um and and that is one of the places where I think again the the AI story has been so fascinating is and you really see this with the visual models like Dolly um but it's true also the language models which is the the ability of these machines to capture style like that you can say I want a picture of like Pokemon in the style of rembrand and it really like seems to get the
style of rembrand in this way that you would have thought that was yeah okay you can learn to identify a Pokemon versus you know a person but that you could capture like the the the the style of a a painter um in a in a really clever way um just by like predicting the next pixel which is basically all they do um is astonishing I want to close with one
final question what do you think is the core idea that your work is orbiting around because I see two parallel tracks and I've been thinking about this for the past few hours now of you have this narrative non-fiction drama track True Crime London disease and then you have this fervor for creativity and Innovation and connecting ideas what is
the synthesis of those two things I I I actually can't find it right now I think there are two I I do think you could divide my work in books into two different strains you there's a kind of Storytelling mode historical storytelling mode and then there's the largely history of innovation kind of how does innovation work that's mostly what the the other books are about in
one way or another um and a little bit of kind of Science in in there um but I think the the common theme you know in all the history books I was describing this in a in a substack email that I sent out yesterday um there's a whole riff in in infernal machine about so much of it is about dynamite and about bombs going off and so naturally I was
like well I have to understand what made that wave of dynamite bombs terrorist actions possible and so thus I have to understand how Dynamite came in the world which led me back to Alfred Nobel and his life which is extraordinary and got me thinking about like the state of chemistry at that particular moment in time and like why Nobel was able to like
invent this thing and why it basically like ended up getting coopted by all these political actors which was not part of his vision at all which is unintended consequences and technology is a big part of my obsession in all my books and so there's always this attempt in the story books and the history books to say here's what happened and I'm
going to tell you what happened in the most interesting suspenseful way possible so that you want to keep reading but I really want to think about why it happened and like what were the deeper forces not just the Arc of this person's life and the random collisions of their life but like what were the underlying forces that really you know when you see Paradigm shifts or
revolutions political revolutions technological revolutions scientific revolutions like what is why did that Revolution happen at that moment like why did why did terrorism get invented in this particular period of time why did the modern surveillance State get Ed in response to terrorism in that particular point in time those are big questions in
infernal machine and so those questions can only be answered um across different disciplines and across different scales time scales right um so to understand the anarchist bombers you have to understand the chemistry of dynamite um and you have to understand the economics of dynamite and you have to understand the political philosophy of anarchism and its relationship to
violence and you have to understand these particular actors and their particular world and you have to be able to go back in time in that case you know 60 years to Nobel in other cases in other points I go back 300 400 years to tell the story of how India became the richest country in the world at that particular moment in time when Henry ever was a pirate um so I think it's
thinking across disciplines and thinking across scales um that's a common theme in the idea books and it's a common theme in the history books um and the other words that come to mind is underlying Force as you said I really want to call attention to that cuz that when you said those words I was like yes the underlying forces shows up over and
over yeah and what's nice about it is it's an intellectual idea that when you do it right also carries that same weight of surprise that we were talking about at the beginning like you're surprised by oh I didn't think that event was going to happen and you're surprised by I never really thought about the fact that like a an invention with like nitroglycerin ended up like
influencing like the birth of the FBI that like I didn't think about that connection um and and so when you can do that kind of level of surprise where you have a page Turner surprise and you have like deep forces surprise at the same time like that's that's the sweet spot that I go for boom boom that was fun really fun really enjoyed it I could do that all day