Ep. Ep 381: Life Advice from Legendary Writers

26:00He owns an office building. A nondescript office building in a nondescript part of Bing or Main. You go into this building, which they say looks like Dunder Mifflin, and it's like all this king memorabilia and things from his movies and his gothic Victoria. There's a really cool space where him and his staff are, but he couldn't put it in a cool place because otherwise people just start showing up. That's why we're above a restaurant because otherwise people knew where we were, Jesse. The problem that me and granny both have people are always trying to chase us down leo is always like i want to know about your latest book maybe we'll do a movie about it like leo man chill chill okay you gotta do a podcast about uh fixed schedule productivity i don't have time to talk about your movie ideas but i thought that was cool so i like that piece all right so what was king talking about here uh the question that he was answering is like what's your schedule as a writer Writing is hard.

26:57Waiting for the muse or being like, hey, I'm going to get some writing at some point today when I feel up for it. Like that's amateur hour. The writers who write a lot like King, and King, he writes a lot. He's definitely a high-volume writer. They do the same thing every day. Some days are better than others, but that's what adds up over a lifetime to a lot of books. Have your schedule. Put aside the time for things to matter. Do that time. Have a ritual around it. You can have a taxiway as he talks about it for your mind to get into the mode and then do the work. Well, you can imagine this is something I'm happy to generalize.

27:28Make regular time for the things that matter most in your life, whatever your equivalent is of working on book manuscripts. Don't wait for having a lot of free time or feeling inspired for it. Put it on your calendar. Have a ritual surrounding it and just do it on a regular basis, even if inconvenient, even if it doesn't always go well, and even if sometimes nothing of interest is produced. That is how you make progress on hard things that require a lot of time. If we're going to make this concrete, the particular tactic I want to talk about is autopilot scheduling. This is where you actually have the same time, same place, same day for certain activities, and they're on your calendar in repeat for weeks, weeks, weeks into the future.

28:04So that time is protected. Nothing gets scheduled there. If you went into the day and say, I really want to prioritize my big project, that day might already be fractured too much for you to make progress. But if 10 to 12 for the next six months on Tuesday is put aside for like I work from home and then I go to this library and that's where I work on whatever the thing is, the new skill you're learning or whatever, it doesn't get taken up because you've protected that when you're making meetings or appointments. That time hasn't been available and you have a routine. You do it week after week.

28:32You don't have to think about it. You don't have to burn mental energy or convince yourself. That actually gets done. You got to autopilot to schedule the stuff that you know you have to do or the stuff you know you really want to do on a regular basis. That is the absolute best way to make progress on it. So you follow something like Stephen King's lead. All right, we got one more idea here. For the fifth idea, I'm kind of pulling a fast one, so I apologize. I am going to turn away from the pantheon of quote-unquote legendary writers. and our final idea is interesting, but it's going to come from an idea who is much less impressive than the other four.

29:08And I'm talking about myself. So I actually have a quote from an interview I did a couple of years ago from which I will draw my final suggestion. I couldn't help myself. I don't know. I've done a lot of interviews. So I put one of mine in there as well. Before we get to that final piece of advice that comes from yours truly, however, we have to take a real quick break to hear from the sponsors that make this show possible. from athletic stuff like a full court pickup game swish to athletic stuff like a half mile stroll get those steps in head to sierra or sierra.com for the brands you want at the prices that let you do it all from athletic to athletic sierra's got it are you looking for a gift for your family that they will find meaningful let me tell you exactly what you should get an aura frame now these are digital picture frames to which you can easily upload photos you can do it straight from your phone with just a few clicks after you take the picture.

30:01And then the digital frame will rotate through the images that you have uploaded. So we bought one of these Aura frames for my parents and my wife's parents. They love them, right? It's become a habit for us. If we're doing something interesting, we upload a couple photos. And now my parents and my in-laws, they feel like they're kind of a part of what's going on. They know what's happening in our life on a sort of, you know, hey, you guys went on this interesting trip this week and we got to see it. We got to know what's going on. It really is a great gift and they love it as well. Here's a cool thing.

30:30If you're buying an aura frame as a gift, there's a feature where you can pre-upload a bunch of photos before it ships. So when someone opens it, there's a bunch of photos to show right away. Because otherwise, it's a little known fact, Jesse, the default photos in the aura, it's a lot of me giving like thumbs up and finger guns. A lot of like this and a lot of this. It's part of my sponsorship deal. You know when you're in-laws seeing a lot of like me doing finger guns so you can pre-upload the photos ahead of time. I wonder if there's something in our ad copy that says do not imply do not imply that Cal Newport finger guns has anything to do with this product.

31:05Here's the thing, you can't wrap togetherness but you can frame it. For a limited time, you can save on the perfect gift by visiting AuraFrames.com to get$35 off Aura's best-selling Carver Matt frames. These were named number one by Wirecutter. you also need to use the promo code deep questions at checkout to get that discount so that's a u r a frames.com promo code deep questions this deal is exclusive to listeners our listeners and these frames sell out fast so orders your now orders your order yours now to get it in time for the holidays support the show by mentioning us by mentioning us at checkout terms and conditions apply.

31:50I also want to talk about our friends at Caldera Lab. Here's the thing about being an aging guy. Not that Jesse and I are aging, but we're in our upper 20s now, so we have to start thinking about these type of things. When you're young, you don't think about your skin, but then one day you look up and you're like, yikes, I look like a grizzled old pirate captain. What happened? This is true, Jesse. I told you, I won't get into the details, but a professional movie shoot with a professional crew and professional makeup artist. Really interesting makeup artist, real pro. She's worked with like Tarantino.

32:20She worked on sinners, like whatever. She had to put makeup on me for so long before I could get on those cameras. I was like, Oh my God, I'm old. Like after a while, it was like, she was holding like a sandblaster and just going on. So here's the thing. We get older, our skin starts to age. This is why men need to know about Caldera labs, high performance skincare line, which is designed specifically for men. Their products include the good, which is an award-winning serum packed with 27 active botanicals and 3.4 million antioxidant units per drop. The eye serum, which helped reduce the appearance of tired eyes or dark circles and puffiness.

32:58And the base layer, a nutrient-rich moisturizer infused with plant stem cells and snow mushroom extract. That makeup artist told me, like, you got to moisturize every day. Like, you got to moisturize. So anyways, and I told her about Caldera Lab. I was like, I have exactly the product. All right, so anyways, this stuff works. In a consumer study, 100 % of men so their skin looks smoother and healthier. Skin care doesn't have to be complicated, but it should be good. Upgrade your routine with Caldera Lab to see the difference for yourself. Go to calderalab.com slash deep and use code deep at checkout for 20 % off your first order.

33:30All right, Jesse, let's get back to our deep dive. We're on to our final piece of writer advice that we're going to apply to our lives. This one comes, God help us, from me. I'm going to kind of cheat here, Jesse, because I also directly give life like advice about your life more generally than just writing. I'm going to give a quote here. This comes from an interview I did for the New York Times Magazine back in 2023. Speaking of photos, by the way, there's another collision of worlds because the photo for this New York Times Magazine was David Marchese's talk column. I've told you about it, but you had to go to New York.

34:07They had a fashion photographer do the photos. and he was so disappointed with me to have a writer come in like i'm like a gargoyle who you know can't control his face and he had just done a photo shoot with iggy pop who had like come in shirtless it was just like fascinating to look like and i'm like

34:25he's like look less terrifying light's bad anyways so this is a quote that is both about like writers but also about everyone's life outside of writing. So I'm jumping straight to the life advice with a double acting piece. I'm kind of cheating here, Jesse, but it's my own quote, so I didn't get away with it. All right, let me load this up here. Here's what I said. The critical mindset shift is understanding that even minor context shifts are productivity poison. When you looked at an email inbox for 15 seconds, you initiated a cascade of cognitive changes. So if you have to work on something that's cognitively demanding, the rules has to be zero context shifts during that period.

35:05Treat it like a dentist appointment. You can't check your email when you're having a cavity filled. You have to see it that way. All right, so that's kind of about writing. It's kind of about life outside of writing all at the same time. But here's the generalized point I'm making when doing something important to you. Professional or personal. So like having a conversation with someone on the phone, watching something meaningful, or you're like doing something professional. I'm coming up with a business strategy. I'm writing like a really important memo. You have to remember that the absolute poison that you need to avoid is context shifts.

35:42This is what we get wrong about productivity in the digital age. It's what we get wrong about efficiency in the digital age. We're used to productivity being about producing things faster. What's going to make me move faster? Because when I was building Model Ts, the way I got the productivity ratio, cars produced per input put in, the way I increased that ratio was moving faster. I had a faster, more efficient way of putting things together. So we think that's what productivity means. But in the digital age, that is not our problem. Our problem is not that we move too slow and that we need to find ways to move faster.

36:14The problem is, is in the digital age, the things we're creating, we're creating with our brain. And our brain is a really fickle thing to work with. If you change the context even briefly, you show it a brief distraction, you look at an email inbox for 15 seconds to check for a message, you just go to see what's going on on social media real quick. the brain goes haywire it initiates a context shift this is an expensive cognitive operation and before this shift is complete you try to come back to the thing you're doing so it tries to abort that shift but before you can fully lock into what you're doing you check something else it's a cognitive disaster your cognitive capacity is reduced you're exhausted the stuff you produce isn't you actually end up producing is not very good it would be like if in a factory we discovered like the way the human body worked was uh like let's say you're you're like moving a wrench like Let's say the way our human body happened to work is that if you turned your head too far, your arm strength disappeared for 10 minutes.

37:10You have a hard time holding a wrench. Let's just say that was true. We would keep that in mind when designing factories. Like, okay, let's be careful. Let's not build our factory in such a way that the workers keep having to turn their head because they're going to lose their arm strength and they're going to have to wait for a while until they can really speed up again and get the steering wheels back on. Well, this is what's happening, cognitively speaking, in knowledge work in a digital age. Context shifting takes all the proverbial arm strength out. Our ability to concentrate really dies down.

37:37But instead of protecting ourselves from this, we make it even worse. We've added even faster, more ready forms of context shifting and distraction. So that's what productivity is in the digital age. Avoiding context shifts, not trying to go faster. Avoiding context shifts. In fact, I've talked about this in multiple books of mine. I will take on significantly more cumbersome processes, slower, more structured processes that are a pain in the moment. I'll take on a huge amount of hardship if it saves me from context shifts. That's the productivity poison. So that's the advice I want to give here at a high level is you need to think about context shifting as the thing you want to avoid to be productive, not getting faster.

38:21Now, if we want to get more concrete, read Deep Work, read A World Without Email, read Slow Productivity. These books of mine get into it, but I'll give you two concrete ideas you can do just to sort of get more used to thinking about that. In the world of work, if you're working on something deeply, you should have a set rule for what happens in deep work blocks. How long do I want to go? And my rule is no context shifting. Start a timer. Put the timer on your phone. Let me tell you why. because now your phone is, yeah, it's in front of you, but it's full screen timer. That's the purpose is it has a timer on it.

39:03And the way you're going to run this process is that if you context shift, that means looking at anything that's not the work you're doing. So email, Slack, your text message on your phone, social media, YouTube. You have to start and restart that, stop the timer. Okay, here's how far I got. And then you have to restart it again. I only got seven minutes. I only got 15 minutes. So now your phone, which will be one of the biggest sources of context shifting, is pre-mocking you. You see that number on your, that's what's on your phone is seven minutes, like the timer's going. And so to look at something else on your phone, you have to swipe that time to the side.

39:41Like, you know exactly, oh my God, I only made it seven minutes. Like you have to confront what you're doing. So I think have your phone with a timer full screen. That's what you're timing. And be like, I want to get an hour, I want to do a half hour, whatever it is. and like, hey, if I contact shift, I'm going to restart. At home, the answer is really simple. You got to plug your phone in the kitchen. Just plug it in the charge. If you need it, go to the kitchen. It's not with you while you do other things. It's not with you while you're watching TV. It's not with you at the dinner table. It's not with you when you're reading, when you're doing other things.

40:07You actually have to do those things. There is no easy distraction to turn towards. You'll lose a lot of your taste for the contact shifting. This really has to do with your short-term motivation systems, what's happening in your brain. Don't worry about the neurological details. Keep your phone plugged in in the kitchen when you're at home. I can tell you what, though. It's going to make certain shows more boring, especially if you're watching Netflix. Netflix does more of this supposedly. Netflix is doing more of this engineering shows with stimuli stacking in mind. They assume that their viewers are going to be looking at their phones while they watch the show.

40:36So they have to telegraph much more clearly what's happening. So you can kind of listen out of one ear or be brought back up to speed of what's happening. My wife and I are watching the Claire Danes show, The Beast Within Me or something like that about an author and et cetera, et cetera. I feel like they're doing a lot of that in that show. They're like, oh my God, you just told me this. And that is bad for these reasons. And I worry because I worry that I'm now responsible for the thing that you did. You know, like there's a lot of this type of telegraphing because I think they're assuming. So you might find Netflix kind of boring, actually, if you do this, but it's worth doing.

41:09All right. Those are my five ideas. I think, Jesse, it's time for some takeaway music.

41:26All right. So when it comes to shaping a more intentional and meaningful life, there are all sorts of different sources of wisdom. Today, we looked at one source that's particularly close to my heart, which is writers. We took five pieces of reasonable advice about the craft of writing, and we transformed them into five more general ideas about cultivating a deeper life. and this resulted in five concrete suggestions for actually putting these ideas into action in your own life uh let's review these jesse can you bring this up on the screen here i have them here so our five pieces of advice were uh have a lifestyle vision and use an insight journal to help edit it practice diligence and deliberateness with seasonal projects make sure that you have a training plan use evidence-based planning when trying to make decisions in your life turn every page.

42:16Actually study the thing closely before you go after it. Use autopilot scheduling to make sure that you're making progress on a regular basis without having to expend any decision-making energy on the stuff that matters. And finally, minimize context shifts in almost anything you're doing important in your life, whether it's personal or professional, you need to not be trying to shift your context between different things. Do those five things. Those are five ideas that come out of the world of writing, but it will apply to the life of basically anyone who's listening to this show. and you will find some more depth.

42:46So whether you're writing a book or authoring a deeper existence, how you approach the task matters. And it was worth taking a moment to learn from those who have spent their time mastering this craft. All right, there we go, Jesse. I love the generalization part. Yeah, I don't know why. Maybe it's because I know writers, but I feel like writing advice often is pretty generalizable to life. The other field where this is true is athletics. Yeah, golf. Golf, like all lessons come from golf. But like in athletics, there is a lot of reason in golf and other places. There's often these lessons that carry out to other sorts of things.

43:19Do you think Alan, if he had known Robert Carroll's career trajectory and writing these like epic biographies, he would have said to every page. Yeah. Yeah. Because that was advice when he was writing for a magazine, right? Yeah, I know. I think like he created a Frankenstein's monster. Sounds like, all right. So I'm going to go away for 12 years to the Johnson library. Just like reading, you know, the menus from the White House chef menu. Yeah, I guess you can take these things to an extreme. I think I would be good at that type of lifestyle. It's like an Indiana Jones when they go into the basement.

43:53There's like never ending. Yes. And then in the very back. So here's the thing you don't know about that scene. In the very back of that giant warehouse where they're taking the Ark, you can see in the distance Robert Caro just opening up the boxes and. I'm turning every page. Hey, there's an Ark here. What's in there? Face melt. that'd be a twist the arc ends up killing and melting the face of Robert Carroll that would be a deep dive you know that's a matte painting that scene it's a warehouse just like a sound stage that's how they did all the effects back then in the 70s and 80s they painted that warehouse you paint it on a piece of glass and then you leave like a little bit like unpainted at the bottom and you put that later so you have a new film just someone pushing that cart on like an empty concrete floor and a soundstage and then you film it through that matte painting and then that gets on the film so now it looks like you're walking into these like this vast you know background and they're big these matte paintings are big like the size of our wall in here yeah uh old school special effects are pretty cool all right uh let's see what else we got going on we got a bunch of questions coming up so what's happening in our show we got questions from you about these type of issues etc like living a deeper life etc um and then and we have call we got a case study back by popular demand in the third act I'm going to react to some more comments so kind of get back get the word of the people and respond some I agree with some I don't some I think are funny and some they change my mind on so we'll get into all of that as well my good friend granny might stop by at some point to talk writing I'm going to like make that people think that's true that me and David Grant are just like really tight granny's going to stop by we're going to talk some Tufts lacrosse you know research, talk about researching and Tufts lacrosse.

45:39It's going to be a good time. All right. Enough of that nonsense. Let's get into some questions. All right. First question is from front of the show, David Dwayne. I'm currently reading Jim Lehrer and Tony Schwartz's book, The Power of Full Engagement, which encourage you to manage your energy with equal focus as you manage your time. What are your thoughts on managing energy versus time? I like that book a lot. I remember that book from the nineties, The Power of Full engagement. And that was the idea. We talk about time management, but not energy management. But if you're careful about increasing your energy and then matching your energy to the things you want to produce, you're going to get a lot more out of your business.

46:17So it's a really smart idea. These are the type of books I came up on in like the 90s when I was reading all these business books as a teenage entrepreneur. They really influenced the way I write today. I think it was probably for the good. It's a style of book that's not as, maybe it's not as popular as it once was. It really was big in the 80s or 90s. Books that are just like, hey, here's an idea. I mean, you see this so much in my writing. Here is an idea for organizing a part of your life or doing this type of work or organizing your tasks or trying to maximize what you get out of your day.

46:47I'm just going to walk through it. It's like a concrete idea. We get this wrong and we talk about it. We have steps to it and matrices. It was just like this sort of business pragmatic nonfiction was a really big thing because people are like, yeah, it's hard to be in business and I want to know how to do better. these books are a little bit less common now but it's definitely like my writing is basically a mixture of like Jim Lerner and Tony Schwartz style writing with like New Yorker writing and I just put those two things together so they were both very influential so I like the PowerPoll engagement I met Tony Schwartz actually back in I got it would have been like 2015 I did an event at Lincoln Center 99U conference and I was speaking at one of the stages at Lincoln Center about so good they can't ignore you and i remember it because like i was backstage was me and bernie brown and so we were like yeah that was that event well it was a cool event and i might be mixing events up i think that was an event i met gretchen rubin was there and we were hanging out and i met tony schwartz there interesting guy i think he co-wrote one of like donald trump's books like the art of the deal type books or whatever and then he like during the first donald trump election like came out against him and it was like a big controversy at the time or something so i i don't know tony schwartz interesting guy um i like powerful engagement i'm a big believer in energy management this is an idea maybe we should do a deep dive on this book at some point um it's an idea that really should come back it's a really smart one low energy you can't do much work so it matters so do important stuff when you have high energy less sports if you have low energy but also manage your energy itself with diet and exercise and sleep that point that the physical matters for the cognitive i think we accept it more now but we don't do much about it and so i think there's a good book i recommend it it's like again it's 90s style business books you didn't have to have stories not like an adam grant book where you're going to like let me have stories and ideas or whatever um it was just like here's something else that's important we worked with our clients on it here's like the five things we tell them to do like that's what these books were maybe they're still writing these books i don't know all right who do we have next next up is lauren i struggle I'll go with me time because of my family, work, and hobbies.

48:59When I sit at home, I feel antsy. What should I do? I think it's a bias towards action. I think it's absolutely fine. And if anything, it helps stave off anxiety for a lot of people. The sort of sit still, like let me just have an empty evening, go for a slow walk or take a bath. Like some people, it makes them really anxious. Activity makes them feel better. I think there's nothing wrong with that. and I think this is really a personality type I know people that have to constantly be working on projects and other people who really would like to minimize work altogether like they're like I want to just sit and enjoy the sun in the afternoon and listen to a ball game and like I don't want to do things and there's people that are in between so there's nothing wrong with the bias towards action as long as here's the caveat if all of that action is imposed from outside it's extrinsic it's inflexible and it's deadline driven then you're going to burn out and be stressed.

49:56Like that's just the people, like the people who have like the really hard jobs that have to work a hundred hours a week. They have a lot of action. They're also burnt out. What you need if you have an action bias is a lot of action that is autonomous and flexible. You're choosing to work on it because intrinsically it's interesting to you, but it's also flexible. Like if you don't do it today or you're feeling sick and you're like, I'm just going to cut this short, nothing happens. It's just you choosing to work on these projects and you make progress as you can. And you're not stressed out if something you can't make progress on for a while.

50:23that works really well for action oriented people intrinsic autonomous flexible action then fill your time with as much action as you want that's not going to stress you out because you don't have deadlines you're not falling behind you don't have a sense of like being forced to do something you don't want to do so i like it i wish i i do have an action bias but i don't have an extreme one i wish i did it does help my anxiety though when i try to stay busy with intrinsic flexible autonomous things. I think it's a pretty good this is sort of Arnold Bennett's How to Live on 24 Hours a Day. You should really be filling your time pretty intentionally.

50:57There's nothing really good. You recharge when you sleep. Otherwise, you're supposed to be doing things. It should be outside of work. Things you choose that are meaningful, flexible, and autonomous. That's basically his book. I think it's a good idea. Alright, who do we got next? Next up is Frank. How should someone think about Reddit, which blends meaningful information and attention traps? Yeah, I think Reddit, it falls into a category of multiple other online information sources in today's day-to-day. Where on the one hand, it is true that the information source has a lot of good information in it.

51:32That there's a lot of people who are contributing information or if it has really good creative output. Like Reddit will be an example of a place where there's a lot of really good information. Like people talking about stuff they know about. And on YouTube is another example. There's like a lot of really good creative output, right? There's junk on there, but you can also see really interesting, cool stuff as well. At the same time, it's also true of these services. They have attention traps. They can put you in addictive loops. They can give you that consistent purity signal, reward signal that overwhelms your short-term motivation system, and now you just turn to it for numbing or distraction.

52:09It's just if I just get lost in Reddit, there will be debates and arguments, and they're suggesting new things to look at. what about this argument over here? And you go read that. And next thing you know, you've been on it for hours. YouTube, of course, is the same way. You go down those auto recommendations and God knows where you show up for those types of information sources. I always say like, you need to use them. Like you would use a reference library and not like you would use your television. So if you use them as I am bored or uncomfortable or feeling strained and I'm looking for numbing or distraction, let me just go and see what I find.

52:44you're exposing yourself to the attention traps. If instead you say, I am looking for this particular thing, let me go see if I can find it on here. You're extracting the valuable information. So like, oh, I want to, I want to know more about, um, I'll give a concrete example where Reddit was useful to me. This was back when I was writing my book, Rolled Without Email. I was looking for a good concrete example about the arrival of email. What was it like when email arrived at companies in the 80s and 90s i mean i'd done a lot of archival research about the like business journalism about email and how is how it's being written about but i was like what was it like in the companies and i was just searching around reddit on it's just using google searches and i found like a great reddit thread where like an engineer from ibm was like let me tell you like someone asked that question he's like let me tell you i was involved and let me tell you the story and then i kind of tracked him down it was and it was really useful that's good reddit bad reddit It would be if I then was like, let me just stick around and now see people just like fighting about, you know, email blame wars and Usenet and people getting like whatever, like nonsense.

53:46That's like not really relevant to what I was trying to do. YouTube is the same way. If you want to look something up, it's like a great resource library. Also, if there is like a particular show you like. They say, yeah, I know that show comes out on Mondays and I'm going to watch the video on YouTube of it. That's actually that's fine as well. you're like look i'm a big fan of like mark rober science videos or i'm a maker and i like the wicked makers channel or something like this and they have a new video out i'll go and set it up and watch it on my tv like that's fine as well but if you're just sitting there on your ipad following the recommendation uh you're following the recommendations you end up in some sort of like weird corner or something like completely unusual or weird or mean is happening that's a bad use so that's the way i think about those type of dual use platforms for appointment viewing or looking things up.

54:34They're great for distraction or numbing. Danger Will Robinson. So don't use them for your distraction or numbing. Have higher quality things that you use for distraction or numbing and use those that have the attention traps just for looking things up or appointment viewing. I talked about that in my book, Digital Minimalism. There's a chapter on joining the attention resistance. That's one of the strategies of actively resisting the attention trap economy is having this particular uses for those platforms and in particular being very careful about what you allow to be used for distraction and numbing.

55:05All right. What else do we got here? Next up is Maya. Can you explain the difference between the Lincoln Protocol on reading specific books for a purpose and the more expansive view of reading for the sheer joy and mental workout of reading? Should I shift my own eclectic reading habit towards the Lincoln Protocol or some hybrid of the two? You should do a hybrid. These are two related but not completely the same type of purposes. the Lincoln protocol is to seek out books that are going to make your life better and that you maybe have to stretch to to understand or build up to be able to consume and that's like an important way to self-improve and it also makes your mind very sharp not all your reading should be you know trying to pour over Euclid like Lincoln did to try to learn geometry you should also have reading you do for fun and you do both like this I just like this it's elegiac it's it's fun it's an escape or it's just really interesting and also i'm working on something over here that is a it's hard and pointed towards a particular purpose do both those things you know what that ratio can change depending on the season of life and your energy if it's a busy period like i'm about to enter thriller december you read a lot of dumb stuff january you know some other periods or in the summer sometimes when like i have a lot of energy i'm maybe reading a lot more lincoln protocol type expansive so the ratio can change but they're two separate related but separate types of reading.

56:25I think it's important to have both. The big thing to avoid is having none because then you're letting the devices win. They are just going to get those tentacles into your brain. It's not good. All right. What do we got? Do you know any of the books that you're going to be reading for three or three or December? I'll read the Dan Brown book. You got me for sure. Some folks have been asking. They want to read along with you. I don't know yet. I'm going to read that one. I haven't really. It's secret of secrets. That's no one that came. Okay. So I'll read that. I got to see what the new crop of thrillers are that I should know about.

56:57And then I go back and look for classics. Like I like first books often. Oh, here's someone's been writing for 25 years and I haven't read the first book. I like first books. so i haven't got into it yet but maybe maybe i should try to announce try to remind me i will i'll remind like i'm working on this one or here's an interesting one i'm thinking about yeah thriller december all right do we have one more question yep next up is welby do you think that active podcast listening done with focus and reflection can ever rank high in the learning process or is it more often than not just another form of distractions disguised as productivity for the knowledge class well i don't think it's a pure distraction i don't think it's productivity i think gets entertainment.

57:36You know what I mean by that? It's like any other source of entertainment, watching a TV show or listening to a radio show. It's something that's interesting to you. And some of the stuff might be kind of dumb and some of it might be kind of smart, but it's not, it's not, I'm not being productive. I'm not trying to like master, if I really want to like learn a new skill that's important for my job, I should actually learn that skill and get direct to practice and do deliberate practice. So think of it as an entertainment, but it's a good source of entertainment because you can focus it and target it, right?

58:03But unlike TV or radio, you could find shows about exactly what you are interested in. Like my kids and I were getting excited about going to one of our local theme parks. There's, tell you what, podcasts where all they do is review theme parks. Oh, there's a great one. Jesse, I hadn't heard this format before. And I feel bad because I forgot what it was called. But it was a travel log format. so he would go to an amusement park and take audio and then he would edit it and narrate it and it would be like he would hear the audio in the background like and then i went to like the such and such ride the line was not that bad i talked to someone in the uh and then when i clicked in it to the and the turns let it out he's just kind of like narrating then i got this food and it was actually pretty good he's narrating his day with audio clips in it and it's like super interesting relaxing aspiration like man i just kind of want to be at the park with this guy the new yorker does similar stuff with their podcast i just listened to that one darkness in the darkness with a um she did the research of that crime yeah my wife was talking about this yeah i like that that's a cool but so here's what i thought was cool about that format is it's a great format but in like true crime you have to do a lot of work because you have to like actually have a crime that you're investigating this guy was just like traveling and but it ran into the travel channel like yeah it's like any show that's popular travel channel but like this on this very particular thing I happen to care about that moment.

59:25And there are shows like that for anything that you're interested in, and there's probably good shows on it as well. So it's like hyper-targeted cable TV. That's the way I think about podcasts. And you're like, oh, this is good. I have a good source of... I'm not watching a lot of TV shows, but I have podcasts. It's a higher quality source of entertainment. So it's not about productivity. It's not like mindless distraction either. So I'm a fan of it. No attention engineering either. You've got to win fans the old-fashioned way. Be interesting. Be useful. Be entertaining. Be somewhere where people want to keep coming back to.

59:55They have to just hear about it and subscribe and choose to stay. So there you go. be listen, more podcasts, especially ours. All right. We got, speaking of podcasts, we got reactions coming up where I'm going to actually read. God helped me read the comments. And before that we have a case study and a call as well. So a lot of good material still to come, but first we got to take another quick break to hear from our sponsors. Sierra has all the best active and outdoor brands for the super athletic stuff like running gear for cruising up the trail. Woo! And the super athlete-ish stuff like fishing gear for chilling by the creek.

1:00:29Nice cast. Fitness apparel to push for higher reps. You got this. And golf balls priced so you can afford to lose one. Or a few. Head to Sierra or Sierra.com for the brands you want at the prices that let you do it all. From athletic to athletic-ish, Sierra's got it. Look, if you run a small business, you know there's nothing small about it. every day there's a new decision to make and even the smallest decisions feel massive. So when you find the decision that's a no-brainer, you should take it. And when it comes to selling things, using Shopify is exactly one of those no-brainers. Shopify's point-of-sale system is a unified command center for your retail business.

1:01:07It brings together in-store and online operations across up to 1 ,000 locations. It has very impressive features like endless aisle or ship to customer or buy online, pick up in store. And with Shopify POS, you can get personalized experiences that will really help shoppers keep coming back. And they will based on a report from EY businesses on Shopify's POSC real results like a 22 % better total cost of ownership and benefits equivalent to an 8.9 % uplift in sales on average relative to the market set surveyed. So get all the big stuff for your small business, right? With Shopify, sign up for your$1 per month trial and start selling today at shopify.com slash deep.

1:01:48Go to shopify.com slash deep. That's shopify.com slash deep. Also want to talk about our friends at Vanta. Customer trust can make or break your business. And the more your business grows, the more complex your security and compliance tools and demands become. That's where Vanta comes in. Think of Vanta as your always-on AI-powered security expert who scales with you. Vanta automates compliance, continuously monitors your controls, and gives you a single source of truth for compliance and risk. This is really important stuff. Compliance and risk monitoring is one of those overlooked time taxes that can really weigh down businesses, especially during phases where all of their focus should just be on growing.

1:02:33Vanta makes all that so much easier. If you know what the term SOC 2 compliance means, then you really should be checking out Vanta. So whether you're a fast-growing startup like Cursor or an enterprise like Snowflake. Vanta fits easily into your existing workflow so you can keep growing a company that your customers can trust. Get started at vanta.com slash deep questions. That's V-A-N-T-A dot com slash deep questions. All right, Jesse, let's get back to the show. All right, so we're going to do a case study here. This is where people send in their descriptions or accounts of using the type of advice we talk about the show in their own life.

1:03:14It is also an excuse to play our often requested case study music.

1:03:30All right, now we can relax. Today's case study comes from Owen. Owen writes the following. In 2024, I received some hard-hitting feedback where I almost lost my job as a project manager. I could barely focus for 15 minutes at a time, and I wasn't making progress on my projects. I was behind schedule and way over budget. My problem was that I stayed up late every night scrolling social media. I came across your YouTube video, How to Reinvent Your Life in Four Months. I started with discipline in the fall of 2024. I prayed, set my bedtime alarm, and connected with friends and family every day that I tracked those core habits.

1:04:03I wrote out my lifestyle-centered career plan. I implemented multi-scale planning. I started a values document. These adjustments helped decrease my daily screen time from 9 hours to 1.5 hours per day. This past March, I decided to let these new systems run for a bit. I really like Cal's advice. Run the systems for a while, take note, and then start from the beginning again. So it's November 2025, and it's time for me to start back at discipline. I read Slow Productivity over the summer, and I started implementing the ideas. I got pushback on not opening emails immediately. I got pushback on muting my team's notifications.

1:04:34However, all this pushback came from colleagues at the same level. My bosses and managers didn't notice. In fact, they just commented on the improvement in my work. My systems and habits provide rigor and efficient structure to my professional endeavors. These systems allow me to use my time efficiently and lift the cognitive load. But most importantly, my systems continue work, confined work to designated times and locations. My work shutdown routine ensures that I can confidently put work aside. I have many pursuits outside of work that energize me on a weekly basis. Oh, and I love that case study.

1:05:07It both shows the problem of living a fully distracted life. it will just keep pulling more and more from you until like, oh, and you're up all night long scrolling social media. Like you're literally toiling for free in the attention factories of Mark Zuckerberg, right? So you can buy the other half of Kawhi while your main job is suffering. Like, look, you can barely focus. So the danger of living with uncontrolled distractions there in the story, but also we see some realities about fixing. In that video, Reinvent Your Life, one of the things I talked about is like, let's go in order. like discipline and organization uh i had a few things i said start with some simple habits in each just to signal to yourself that you take those seriously and then rotate through and keep working on those parts of your life there's sort of a rotation to that well look at the difference it made in owen it like it it slowly if you focus on the right areas start small but are continuously going through them and improving slowly over time real change can happen the other thing that happened here is you put into place the type of ideas i talked about slow productivity and his boss is like, wow, your work's better.

1:06:10Because ultimately that's what matters the people who are in charge of your job is how good is the work you're doing. To your colleagues, it might be, make my life easier by answering my email. Make my life easier by being on teams all the time. But for your bosses, for your customers, for your clients, the value of what you do matters. And taking your time, doing things well, obsessing over quality, spending more time on each thing you do, not doing too much things at once, those principles of slow productivity, it produces better work over time. It produces more work over time because you have less of the overhead tax eating away at your schedule.

1:06:41It's a great example of the type of things I talk about. The motivation for my work is the digital age is eating us alive, and we don't even realize why. We weren't ready for it. But there are ways, if we're systematic and careful, that we can push back against that erosion and still build successful lives that are meaningful and not overwhelmed. Again, I hate to harp on this, but it's why I hate this straw man that any talk about systems and organization or improvement is somehow like hustle culture or productivity culture, or it's coded as like online bros or something like that. Not about hustling, not about how do I produce more and crush it or whatever.

1:07:21It's about surviving our current digital moment that's eroding all the things that makes life meaningful. And we didn't even realize it's happening until it's too late. It's about thriving as a human, it's incredibly humanistic what I talk about in an age that's increasingly anti-human and digital so there we go oh and great case study all right do we have a call this week we do all right let's hear this like to get your opinion on where somebody like me should be focusing their efforts for the next five years um and when i say somebody like me i mean somebody who works in technology not necessarily cyber security um i'm doing all the stuff I can with optimizing my time and trying to develop a deep life.

1:08:06And a lot of the things you've covered, they've been transformational. But I just thought if anybody knows, it would be you guys. All right. So Jesse, he's asking about what he should focus on professionally. Yeah. Okay. I cut off some of the beginning of it because it was a little long. All right. So he's saying like, look, I don't know what. He's doing all the advice I talk about about your life and being intentional. But he's like, what should I be focusing on? Probably he's worried about AI if I had to guess. This is where, going back to the deep dive, evidence-based planning matters. You want to identify what are the skills that are in the same arena where you play that are producing value and are useful.

1:08:49Actually producing value and useful. Like if you do this thing, customers will pay for it. if you know how to do this thing, other companies are going to try to hire you away. I saw this person get hired away at double the salary because they have those skills. So things that you have evidence are valuable and are adjacent or in the same arena where you already play. Always be thinking about that and surveying that and be ready to make moves every like two to seven years. And that's like deliberate diligence to get there. Okay. Now I need the diligently and deliberately. So over time with focus and deliberately working on things that actually make me better, not the things I want to do move towards those, You probably should be doing that every two to five years, maybe every two to seven years in most tech industries.

1:09:29The important thing here is the evidence. You need evidence that this is producing money. Otherwise, you'll get lost in trends and then you'll end up, you know, writing all of these like PyTorch calls to LLMs to build these like simplistic agents that like everyone and their brother is building and they don't really do much. and there's no actual value proposition there and it doesn't help. Where what you really need to be learning was like, you know, some sort of new like virtualization technology that is being used on the new class of servers that are happening for like hot swap, you know, whatever, web apps.

1:10:03Like, oh, this is like really a better way of doing things and it's highly valued and because it's hard, not everyone knows how to do it. You got to constantly survey that arena. You don't want to change all the time. If you're changing every year to something new, you'll never catch up. But every two to seven years, maybe every two to five, you should be like, am I moving diligently and deliberately towards a new skill that's in the same arena that I play so I'm not learning something completely new and that I have evidence which is going to be people are willing to spend money to hire or pay me clients or people trying to hire me away do I have evidence like this is actually valuable all right good call let's go on to our final part recently we've been having a tradition here of me going back and responding to comments from recent episodes we're gonna continue doing that today we actually have them all on screen today as well if you're watching instead of listening.

1:10:47Our last episode was about LLMs and whether or not they could be conscious. I was responding to Brett Weinstein's discussion of AI consciousness on the Joe Rogan podcast and I said, look, I think there's something sort of somewhat pre-modern about this. I think he told a story to explain what he sees and now he's reacting to his own story, not to the reality of how these systems are actually architected. As with all things AI, this generated some responses. I don't know if the inboxes were like Jesse, but you know, generated a lot of comments. And so I figured I'd go through and let's respond to some.

1:11:19Some are really smart. Some are smart, but I disagree with some I agree with. And some I think are just funny. All right. Let me bring the first one up on the screen here from Brandon Harrell, 962, who says LLMs do have a goal. Understand the world well enough to predict the next token. All right. So that was in response. I was saying in that last piece, they're not LLMs don't have like goals. They're not trying to And Weinstein said something about LLMs are doing experiments to try to understand how different things affect the world and what things affect the world in ways they like versus not.

1:11:53It's sort of implying that they would have goals towards which they might manipulate us eventually. And I was saying, well, they don't have goals. It's what's trained. They're static. What Brandon is talking about is like, well, technically you could say they do have a goal, which is to predict the next token. That's true. But I don't think I would use the terminology goal. I would say LLMs are optimized in training such that once deployed, the tokens they generate tend to minimize loss on the token guessing game. That's what's really happening during training is we're adjusting weights in a way that reduces our average loss when we give it a bunch of token guessing examples from our testing set.

1:12:31And we kept adjusting these weights until we were happy with the average loss we were getting in the testing set. To me, that's different than talking about goals because goals to me require some notion of agency, right? You need like a dynamic system, which is evaluating possible actions based on competing priorities and some sort of notion of value judgments. A static LLM doesn't have that. It was optimized until it did well on a particular game. And then once you deploy it, it doesn't change again. So somehow that seems different to me than having a goal. So it's like saying, I guess saying like LLM, their goal is to token predict is like saying a lawnmower's goal is to cut the grass.

1:13:10Yes, they optimized in the development and training of the lawnmower to be good at cutting grass. But then once you deploy the lawnmower, it has no agency, right? It doesn't evaluate the world and make decisions and pursue and evaluate different options. It just runs. Nothing about it could ever change. It just runs the same way every time you turn it on. and the way it runs tends to be good at cutting grass. We'll see with an LLM after trained, it is static. And so once it's deployed in the world, it produces tokens. It does pretty well because we optimize it, do well at that. But without agency, I don't like the terminology of goals.

1:13:41All right, put up another comment here, put on the screen. This one's from Leonard Eusebi. He says, hi, Cal. I just recently discovered the show and I'm enjoying it a lot. I think you are misinterpreting what Weinstein said about doing experiments. The way I hear it, he's referring to the way the RL post-training works. Let the LLM generate a bunch of different outputs and reinforce the ones that work. This is similar to doing experiments and learning from them, but it only applies during training. Well, it's a good but nuanced comment, Leonard, so I'm glad you brought it up. I wouldn't bet a lot of money based off of the discussion that Weinstein had with Rogan that he's really deep in the weeds on how sort of post-training, post-pre-training RL-based tuning actually functions and the different ways it runs.

1:14:31I don't know that he was thinking that closely, but it is a good point that you're... So let me respond to you about this because I think it's worth talking about. Let me tell you the way I think about reinforcement learning. That's what he means by RL. Let me tell you the way I think about reinforcement-based tuning. And it's a bit of a simplification, but to me it kind of gets at the core of what's going on here. So during pre-training, which is that really long period of time where you use real text and you knock out the words and the model tries to guess the word and then you do gradient descent to adjust its weights that make its output be a little bit closer.

1:15:02So this is where you keep doing this until the average loss on the token guessing game testing set is practical. This is where you run this for days and days in these giant data centers with 50 ,000 or 100 ,000 GPUs. That's where you're optimizing it to produce guess tokens. If I give you real text with a word knocked out, you do a pretty good job of guessing what that word is, right? That's this massive pre-training. This was, by the way, the thing that was making these, this is my New Yorker piece from August. The thing that was making language models more capable over the last few years until more recently was they made them larger and they pre-trained them longer.

1:15:43And so for a while, we had this sort of scaling effect that as you made language models bigger and you trained them longer and on more data, they seem more generally capable on all sorts of different things. And what that article was about in August is that those scaling benefits actually really started to die off about a year ago. And now just purely making it bigger and pre-training longer is not making the models better. There's actually like a big figure from OpenAI's past, Ilya S., because I don't know how to pronounce it. I don't have his last name handy. He actually came out recently, Jesse, and said, yeah, the age of scaling is over.

1:16:12He had this kind of nice quote where he's like, we had this long period of research and then a five-year period, 2020, 2025, which is about scaling. Now we're back to a period of research because scaling is not giving us any more rewards. So what did they do once just making the networks bigger and training them longer didn't make them more capable? They turned to tuning, which happened after all the pre-training is over. Without getting too much into details because there's different ways to do this, the one way to think about the more common types of tuning is that you have prompts and correct answers to that prompt.

1:16:42Here's a question and here's a correct answer. That's very different than how pre-training works. It's unsupervised. You're just guessing a missing word and you know what that word is. So I guess that word is the correct answer. But in tuning, you might have a math problem and you might have like a detailed proof as like the correct answer. Or you might have a question about building a bomb and the correct answer you paired with that is I'm not allowed to talk about that. And what you do, very roughly speaking, is it's like you load that prompt into the language model and it's activating all of the relevant sort of circuits and attention and pattern recognizers that are sort of parsing and figuring out what is this question and what are the types of things we could say next.

1:17:23and then you used a correct answer at that point in a reinforcement learning-based algorithms to sort of zap those final logics and please don't yell at me ai people this is just simplifying but to kind of zap those final logics that help you select among the right words to come next once you've identified like what type of answer that should be given here to uh really so let's reinforce the logics that are closer to this particular right answer now if you do that uh It's like, so you're not teaching it how to understand. You're not teaching it in tuning. You're not making it smarter in some sense, right?

1:17:57You're not, it's not in this phase. It's not the type of training where it's like learning whole new concepts or it's learning to recognize things I couldn't recognize before. What you're doing in tuning is saying like, okay, you have all this pre-training that was very expensive to recognize and understand. You know what you're doing. You know this is like a joke and that you're at the punchline and that the punchline has to refer back to this thing that was a cat. Like all the pre-training has all these circuits like lined up. And then if you come back in and you're reinforcement learning with like, I don't do cat joke answers, you're then going to have like these final wires at the end of all these layers.

1:18:28When all of these other things are activated, really push it towards whatever token is next and saying like, I don't do cat jokes. And if you do that reinforcement learning on enough examples, you're kind of like brute force training it. Hey, whenever you find yourself with a cat joke again, just say, I don't find, I don't like cat jokes. Right. So it's, it's like a final layer of tuning. so you're being more precise about how it should use the smarts that it learned in pre-training. That's roughly how I think about the different forms of reinforcement learning and the impacts they have. No, so this is not, I really don't think about this here as a language model experimenting to figure out more about the world.

1:19:07I think about this more like a trainer coming in, like in a dog training example, and shocking the dog to prevent them from doing certain things. Or like you better always go and chase after the lure when it's thrown because when you don't, like I hit you with a training crop. It's like more of that type of thing. So it's not a language model like it's out there like learning and trying to understand how different things do. It is having its behavior curtailed, pushed down particular avenues and shutting off other avenues. We don't want you going this way. Go this type of way. This is what like a lot of the reasoning reinforcement learning was about.

1:19:42it was give answers that show have lots of steps in it as opposed to just giving an answer so kind of shuts down the circuits that one of its opportunities and all the different ways that might answer a question there's ways where it might answer briefly with like yeah the answer is seven like don't let's turn down the weights on those types of circuits and turn up the weights on the circuits you already had that like you're going to show your work and so it's all about aiming the way these things behave towards the end and that's that's really simplification but to me that again i don't see this as um running experiments to learn more about the world the learning really happens in pre-training the reinforcement learning is more about now constraining and and focusing what you want to do with those smarts it already has but in all cases here's the key thing yes but that's when humans are building this thing once deployed it's static it's a lawnmower i don't you know i don't care how many like super um advanced laser studies of grass you did that you could build in these sensors to the lawnmower to change the different modes it goes in like the lawnmower is always just going to like statically do these things every time you run it so to me that's the key the key thing is weinstein is clearly thinking about this as an active agent that's learning as it learns more who knows what it's going to do and it's static once it's trained but if we even look at how it's trained and we're careful about that and this is semantic but i don't see that as a machine running experiments to learn more and There are machines that do that, but language models don't.

1:21:07That's not what reinforcement learning is. All right. I don't know. There's my technical answer. All right. Let's see what we got here next. Gubbiji Punk'd 7028 says, Your definition of consciousness appears strictly functional, whereas Brett seems to be discussing qualia, the subjective experience of what it is like to exist. The philosophical consensus regarding whether synthetic structures can possess qualia is currently agnostic. We simply do not know. asserting that AI cannot be conscious based solely on the difference between biological and artificial information integration is philosophically premature.

1:21:38Well, Gubbie G. Punk T728, that's not what I'm arguing. I am also agnostic on the idea about whether different non-neuronal substances can have the emergent phenomenon of awareness that we, you know, in cognitive science and brain circles, they call qualia. I'm agnostic on that as well. I don't think it's impossible for, for example, a computing machine, an artificially intelligent machine to maybe have something like emergent consciousness. That's not my argument. It's not about the substance. And it's not about the underlying operations that happen in the substance. My argument about consciousness, and I think almost any consciousness researcher except for – there are philosophical theories of consciousness.

1:22:22It's basically like everything has some amount of consciousness and it's all on the spectrum. It's a little bit more out there. But the brain researcher side of consciousness thinkers, they would all agree. Like, look, we don't know exactly all the things you need to get recognizable consciousness, but we know you're going to need some sort of dynamic collection of ongoing dynamic interconnected processes. And we don't know exactly what they are, but some sort of world model or updatable state and drives and memory. And you need some sort of sensing and action and be able to react to action and update information.

1:22:57and all of this has to be pretty deeply interconnected in the right ways. And if you have the right processes and the right dynamism, then if you get somewhere, you're going to get some sort of emergent consciousness. My argument about language models is it doesn't have nearly enough of those things. I mean, nothing in the language model is dynamic. It is static. It is statically deployed. It's not even deployed on a single machine. How can you talk about interconnection? They're too big to fit in the memory model of a particular GPU. So now you have precast weights for different layers spread out over all sorts of different GPUs, which are just running through multiplications for multiple different queries on those layers for multiple different queries at the same time, and then eventually passing those numbers back around.

1:23:34They're integrated different. There is no even like hole. You can't even point to like this is where the computation is happening for this word that was just generated. But more importantly, it's static. There's no state that upstates. There's no world model. There's no drives. There's no actuation or reaction. There's no online learning. So my argument about LLMs not being conscious is a purely architectural argument that's engaging specifically with the architecture of LLMs, which are static feed-forward unchanging. You've got to have a lot more things that are dynamic and deeply interconnected before you have a chance of the subjective experience of consciousness arising.

1:24:10We don't know where that happens, but almost no one would say a static table of numbers that are divided among a couple hundred GPUs are dealing with a lot of things at the same time, that this somehow has any sort of notion of emergent consciousness. If that does, like everything does, every data center does. Google does. Every distributed system does. My phone does, right? Like that's too loose of a definition. So I have a very particular functionalist, architecture-based argument for consciousness here. I'm not falling into reductionist traps. All right, this is fun. What we got next? Mabel 4367.

1:24:44The biggest risk of AI today is people and organizations not understanding the technology and applying it to problems they can't handle in a satisfactory way. Like profiling people to determine things like credit ratings or profitability of committing a crime. It's kind of interesting. So these examples, they might hit like the average listener who's just been watching or listening to a lot of LLM content online, super intelligence, all jobs will be automated. Like what are you talking about? Credit ratings or profitability of committing a crime or profiling, et cetera. Where this comes from, and it's just an interesting story to know, is sort of pre-LLMs, There was a robust academic AI safety community that was looking at AI.

1:25:30Now we're talking, this is much broader of all these different types of AI systems that have been around for a long time, decision systems, recognizers, neural network-based recognizers, machine learning-based tools. A lot of these systems existed. They weren't particularly sexy, but a lot of them existed. This is a field in which we've been building lots of tools. And traditional AI safety was like we should be nervous about tools that make decisions using like machine learning or AI because they're often obfuscated decisions. We don't know how they're making the decision. So how do we know that they aren't encoding things that we would never want like a human to do?

1:26:05Like how do we know they're not encoding some sort of something that we would think if a human was doing is like clearly discriminatory? We don't know if we're just – it's dangerous just to suck up a bunch of data, build a model. That data may be biased, and then it builds up a model that's discriminatory, and now the AI is spitting out loan decisions, and you've just encoded discrimination into like AI. It's a real issue. It's also like a perfect topic if you're a humanist, critical theorist, or someone who cares about these issues, right? You're like, here's technology going to – you're being too fast, and my expertise can help make sure we don't create trouble.

1:26:39And it was like a really good issue, very niche, very academic, whole conferences on it, but not something that had like a lot of discussion in like the broader population. And the LLMs came along and it kind of sucked for the AI safety people because now the discussions were like, is this an alien mind that's going to liquefy our children to power its server farm batteries? Like it became this like outlandish, are all jobs going to disappear within a year? and now like the safety researchers are like we've been very carefully like building these frameworks about you know how to evaluate data from like a comptian perspective and now you have you know people going on the dorkish podcast and it's i think if we use like wooden tip spears we can get past the terminator robots human sensors when the wars begin when the data center wars begin next year it's just like this crazy outlandish you know whatever and they kind of got left behind a little bit.

1:27:30But those issues are still out there. That community is still out there. That's what this commenter is speaking about. And yeah, those are real issues. Again, most AI is not language models. It's not stuff that we think is going to automate our jobs or become alive, but it's integrated into lots of things. And so those are still interesting questions, but it's a hard time, you know, it's a hard time to be talking about these pre-LLM systems. All right, let's see. Let's do a couple others here. um felix md5cy says if llms could be compared to the speech centers of the brain is it possible to design complementary mathematical systems that model other brain regions and integrate them into a unified whole great follow-up question what i was saying before my architectural argument against llm consciousness says like we know or at least we suspect that consciousness artificial or natural requires like a sufficiently uh sufficiently diverse collection of dynamic processes that are deeply interconnected so felix md5cy is basically asking So what if we just build, you know, LLMs are one component of our artificial brain and we build these other types of processes digitally and we connect them enough, can we get consciousness?

1:28:31The answer is maybe yes, but we don't really know how to do this for so many of those components. It's like not something that we're close to. Part of the problem here is LLMs, which are focused on language understanding production, we have a lot of data to train them. We can use the whole corpus of human written language. this whole massive corpus of humans recording like thoughts and ideas we can train them on that and that allowed us to get like something that's really good at that purpose the vision is similar we have a lot of good data sets where it's basically pictures and descriptions of what that thing is they get these off the web because there's often tags that explain what an image is or they look at text near an image and so you can train models to like recognize images what's in an image or if you use these sort of the diffusion models you can reverse it and have it learn how to go the other way and produce an image from text which is like how these image producing ais work or how soar works they the way they train those is um there's a lot of different ways they do it but like they'll also they'll they'll um one way is you get a picture and you take a bunch of pixels out and they try to fill in the missing pixels and then another way is you have them like go from text to like an image you compare it to what like i have a real image of that thing there's all sorts of different things you can do um so we have data for vision we have data for language but like for all these other things like that's we can't just do unsupervised learning with data that's the problem how do we build a world model that like generalizes how like the world functions like the roboticists are working on this the lm people like oh it's easy we'll do it from text nah it's not working we train these models with 100 000 gpus and all the texts ever written by humans and they have no real consistent model of the world the roboticists are trying to figure out how to do this from vision data and experience data this is really really hard and i you know i've talked to a bunch of roboticists about it it's hard and they're not close to having this answer um how do we model things like short-term motivation or drives we're not sure you know i don't know uh so so yes i think it is theoretically possible in fact like maybe one day we'll do it i bet we will but it's hard most of the other systems maybe are much harder than language models to build so it's probably better to think of language models is a great breakthrough on the thing they do really well.

1:30:43It's like we built an artificial language processing center of our brain. That could be really useful. But that language processing center from our brain, if we destroy the rest of our brain, he ain't conscious. Things aren't going to go well for you. You can destroy the language part and probably still have enough things left for an Oliver Sacks case study. I have a hard time remembering things I can't read, but I can still have some subjective experience. But if you destroy the whole brain but the language processing part, I don't care how developed that thing is, he ain't going to be conscious.

1:31:10All right, move quicker here. User Gizmo says, Cal, aren't you stating a version of the Chinese room argument? The counter to that is something like, if you look at individual neurons, it's just a cell passing electrical signals, so it can't be conscious, but of course it is at the macro level. But I do agree with you nonetheless, there's some key things missing in LLMs like continuous learning. I like that you agree. The first part, no, this is not like John Searle's Chinese room because I'm not making a reductionist argument about the actual underlying substrate as being relevant to consciousness.

1:31:37Again, I have an architectural argument. enough sufficiently dynamic systems doing the right dynamic things sufficiently interconnected llms are static and only to really do one thing aren't going to satisfy that chinese room argument is interesting it has a lot to say it's very relevant in the in the age of llms uh it's it's a product experiment jesse where you're basically translating i guess from like english to chinese and there's a person inside this room and he has all these like manuals he can look through to give him steps and rules that he follows. And the other end, he has like a well-written, the proper Chinese version of what came in.

1:32:12But you wouldn't say that anything in that room understands Chinese. That was John Searle's argument about like, what does it mean to know something? It's a good argument, but it's not what I'm drawing upon here. All right. Gragas 1337. I feel like these comments are getting longer as we move forward. Eventually, there's going to be like four or five columns and just like continuously scrolling. People who think about AI are verbose. My God. All right. Gragas1337 says, Cal, you're making the classic reductive argument of LLMs. It's just a bunch of numbers that go through matrix operations, et cetera.

1:32:41All right, I guess I'll stop there because at least that particular question, I'm not making the reductive argument. I don't care about the substrate on which LLMs are built. And it has to do with like specifically the process of the implement and whether they're sufficiently diverse, dynamic, and connected. And they're not. I don't care that they're numbers or ions. All right. Spider-Man in London. him he says uh language models don't run experiments but he's quoting me language models don't run experiments and then he says but chat gpt pro literally performs multiple parallel operations then compares them to pick the most right answer that's what distinguishes pro from the standard model all right let me just stop right there what he's talking about there is what's known as inference time compute which is a again one of these like host of features along with a lot of reinforcement learning-based tuning that was introduced after scaling failed in the last year to try to find some way to get some sort of measurable improvement on something out of these models so they could keep trying to keep excitement and investment going.

1:33:38Inference time computing, so this particular type of inference time computing says the control program that calls the LLM, calls the LLM multiple times, gets multiple answers, and then decide which one is best. This is like wildly impractical from like a financial perspective. Like people don't just feature is not widely used because it's it super uh increases the cost of inference since you're doing the inference like four or five times on certain benchmarks that open ai cared about hey if you have an answer four times and for example one of the ways they did this was like ask it five times and whatever answer it gets the majority return that hey these things hallucinate 20 of the time you get a better answer that way because it's unlikely to hallucinate in the same way uh three times in a row or something like that right and so you can on certain benchmarks The performance was better if you ask the same – the control program asks the same question three or four times and then returns the most common answer or sends all of those answers to another LLM and says which one of these looks best.

1:34:36But it's a marginal improvement and it's expensive, so it's not widely used. But no, that is not the same as the LLM runs experiments to better understand the world because they're static. They're static. Once they're optimized and tuned, they cannot change. They cannot learn. They cannot, the LLM does not learn. It doesn't know that you asked it to run four times. It's just a table of numbers. You get a bunch of answers and another program chooses one of those. That's not learning. There's nothing that's stored. There's no model of the world that's updated. So, all right, let's roll on here. Now we get a couple, I think we're ending with a couple more interesting or funny ones.

1:35:12No more arguments about reductive models of consciousness. Jules 5347 says, this all reminds me of the Kostrup quote. these people are simulating a kidney and expecting that with sufficiently complex model p is going to come out of the cpu a funny quote uh we got here mark dev one brother if you continue down this path you'll have to mention on a pod that you would never kill yourself took me a second to understand that what he's saying is the computers and or open ai are going to come after me to kill me so i better make it really clear if they try to make it look like a suicide look to sam altman if something suspicious happens to me.

1:35:51If I die of severe electrical burns coming out of my computer, go see where Sam Altman was up to, all right? Or maybe some of these other commenters. Baptized Imagination says, anyone who has to use these things at their job knows they aren't conscious in any way. That's a good way to probably end this discussion. It is true. Weinstein's fears decide. You actually use these, you realize there's something weirder going on here than just conscious beings. So there we go. I don't know. I think these are really smart questions. jesse i think like the comments are really smart we're getting the deep things i don't know if i'm right about all these but it's a cool back and forth um and it belies the idea that internet comments i guess are always you know just you're a poop clown and then it's a gif of taylor swift you know it's smarter than that so you know take it a grain of salt some of my stuff might be right so it might not be i love the comments i love getting into it and i want to emphasize the thing I said in that podcast episode too is I wasn't picking on Brett Weinstein.

1:36:49He actually was saying things that a lot of people are saying. He just talks really clearly. It's like this is a great, and he was on a big platform. So this is like a great way to get into the summarize a common thought about fears about AI. So it's not like Weinstein has some particularly pathologically bad thinking. He just stated really clearly something I think a huge percentage of AI users, if you push them like yeah probably something like that's true so there we go alright that's all the episode we have time we have for today I need to write my next episode which is going to be three hours on the architecture based model of substrate agnostic consciousness I just feel like there's now going to be a demand for that maybe I'll get my friend granny to come write it with me alright that's enough of this nonsense see you next week I guess it's coming out after Thanksgiving so I hope you had a good Thanksgiving we'll see you two weeks after Thanksgiving Living Deep and Thriller December.

1:37:41Until then, as always, stay deep.

1:37:48Hi, it's Cal here. One more thing before you go. If you like the Deep Questions podcast, you will love my email newsletter, which you can sign up for at calnewport.com. Each week I send out a new essay about the theory or practice of living deeply. I've been writing this newsletter since 2007 and over 70 ,000 subscribers get it sent to their inboxes each week. So if you are serious about resisting the forces of distraction and shallowness that afflict our world, you got to sign up for my newsletter at CalNewport.com and get some deep wisdom delivered to your inbox each week.

1:38:40Thank you.