#1067 - Cal Newport - The collapse of modern attention (and how to get it back)

19:28Chris Williamson:Eventually, the number of things that satisfy that criteria overwhelms me just as well. It's just so I've just had to fall back

19:35Cal Newport:on the default no you're talking to somebody who came back from a two-day trip to qatar at the start of this week so i spent as much time traveling as i did in the country to to give a talk and as i looked around there was this first the first night dinner there was maybe 300 people there and i'm talking to logan paul and stephen bartlett's over his shoulder and the ceo of qatar airways is here and the middle eastern director for metas over there and i was looking around thinking everybody here wants to be here it's very exciting everyone's really lovely but also everyone here can't say no everybody in this room is chronically incapable

20:14Chris Williamson:of saying no because it's no to this one several times by the way i the amount of invites to the qatar and the uae and other places i have i have said no all right well consider me a fucking

20:26Cal Newport:consider me a slut compared to you cal whatever whatever it is i must be easy an easy booty call. They tried to get Cal Newport. We couldn't get Cal, so we'll ring Chris instead.

20:35Chris Williamson:The default, no. Oh, man. Yeah, it's crazy the things you end up saying no to after a while. But I mean, there's a currency shift. For me, time to think is such a valuable, that's a more valuable currency than money, right? You get to a point where you're like, oh, I'm doing fine. But if I don't have time to think, what's the point? And then that becomes this like really rare currency that's much harder to get a hold of. And that's the only way I can protect it now is anything that requires me to like go somewhere. It's a default no. And then I can talk myself out of it later, right? I'm like, you know what?

21:05Chris Williamson:I could bring my family with it. We could have a trip, right? So actually, you know what? I will do this. Or, you know, I just did a masterclass course released this week. I spent a year and a half saying no to that. And then like eventually I sort of talked myself, I talked to some people. They're like, we'll come to DC to do it. I talked to, you know, James Clear had just done one and I had a good talk with him about it. And I was like, you know what? This will be interesting. And it took me a year and a half, but I finally talked myself into it. So I will say yes, but it just, the default no means that you don't have to -

21:39Cal Newport:You're on high standards.

21:40Chris Williamson:Yeah, you don't have to run it through the ringer. And then you're like, okay, if it really sticks with me, then maybe I'll be like, all right, all right,

21:45Cal Newport:I'll do it. How much should people actually be working?

21:49Chris Williamson:Well, it depends what you mean by work and what they're doing, right? Because think about it. Let's say you're an athlete. It's super well-defined. Like here's optimal training. here's optimal rest and like that's what you should be doing like you either did the that's really clear we don't have those limits as clear in the culture for other types of jobs that we probably should if you're at a high wage hourly build job like a law partner at a big law firm there the economic model is the more you work the more profitable it is and we'll we'll pay you big money to do this but like you should basically work as much as you can that your body will take it, that's the economic engine.

22:25Chris Williamson:That's why I think those jobs are scary. If you're a novelist that writes literary fiction, so you're like, I really need to be award nominated for each book, or I'm going to fall out of this slipstream of, because no one's going to read these books unless they're some of the best books, then you should be doing four hours in the morning and then just disappear. You should be doing very little more work than that, because almost anything else will get in the way of you like sticking in that position. And so it all just depends on what you do, you know.

22:56Cal Newport:Didn't you look at some experiment of shorter work weeks? Yeah. What did you learn from that?

23:03Chris Williamson:There's a lot of these right around the pandemic, right before and then right after in Europe and Iceland. So some European studies, I think Germany did one, Iceland did one, UK did one, and they were looking at four-day work weeks. So what would happen if we take away one day? The interesting thing about those experiments is what they found is whatever measures of productivity they came up with, they didn't get worse, which I thought was very interesting. They took a day away, and yet the perceived productivity or the measured productivity didn't go down. And there's two ways to look at it. The one way to look at it is to say, oh, this means that we should have a four-day work week because things didn't get worse.

23:42Chris Williamson:And And okay, maybe, maybe, right? But to me, there was like a bigger observation that came out of that, which is like, wait, so what are we doing during the work days? Like there's something going on here that should really catch our attention. What does work mean? That we could take an entire day off the table with no other preparation and the valuable stuff being produced doesn't change. This tells us that like whatever we're doing while we're sitting here in work is not just sitting down and trying to produce value. where clearly have all sorts of other sorts of distractions going on, context switching, time that's being devoured, Parkinson's law is at play, work must be broke.

24:21Chris Williamson:To me, that was the more important observation, is that if you can take away a day and nothing changes, then I don't think we're doing in the office what we think we're doing in the office.

24:30Cal Newport:Parkinson's law was on the tip of my tongue, work expands to fill the time given for it. And if you give people five days, they'll take five. And if you give them four days, then they'll do it in four. and look everybody knows just how much time they waste not doing the work not doing the thing that they're supposed to do and this isn't victim blaming this is a lot of the time dealing with admin unnecessary meetings you can't get out of them you have to be there for whatever reason so it's not as if it's bottom up a lot of it is top down dictated this is the environment that you work in and you have to do this but even outside of that when you do have your one hour in between meetings, your inability to not, I remember when I used to run nightclubs and I get in at 2.30 in the morning, the final part of the night was cashing the till.

25:19Cal Newport:So this was before we switched to tickets, which was sort of the late teens, just before COVID, digital tickets online, which meant that you didn't have to cash as much money in the till. But before that, it was all, you know, five pounds and 10 pound note and 20 pound notes and single pounds and all the rest of it. And I would go into the office with the manager of the venue and we would be counting the money. But this is the final task. It's the final bit of the night. It's fucking 2.15 or 2.30 in the morning. We've just taken the till off as it's called. Anybody that's coming in doesn't get to come in.

25:50Cal Newport:We're not going to take any more money. And I'm sat up there doing like light lift mental arithmetic. But for me, somebody who hadn't done math since I was 16, it was a relatively heavy lift. Flicking through the money, flicking through the money is huge fluorescent overhead lights just before. And then I get to drive home and I'm like thinking about it and I got to go put the money in the till and I got to write it in the spreadsheet. And then I get into bed. And as I go into bed, my eyes below my eyelids would start flicking left and right. I wouldn't be able to tune myself. I'm also doing this, let's not forget, in a sweaty beer stinking office above a room going.

26:29Cal Newport:I've had to walk through the club. I've to shout at the hostesses. One of them's getting fingered on the dance floor. Stop doing that. You're supposed to be at work. The DJ's pissed. I need to, you know, it's chaos. And I've tried to coordinate this orchestra of bullshit. And then I've had to do mental arithmetic and then I get to drive home. And then I'm like, okay, chill out brain. It doesn't want to. And that eyes moving left and right thing, I think is the sort of optical equivalent, ocular equivalent of how people feel when they finally get a moment. It's like, okay, all of my stuff is done.

27:01Cal Newport:And then they try and sit down to work on the thing that ostensibly that's actually there to do, right? Because all of the other bullshit, the meetings, you're not there to do the meetings. You're not there to do the Slack. You're not there to do. All of that is foreplay to get you to do the thing that you're there to do. And then you sit down to do the thing you're there to do. And your eyes are moving behind your eyelids is the equivalent. You're swiping and moving across the screen and you've got a few different other, well, just check on this thing. Like what the living fuck is going on? I've like trained, the environment that I work in has trained me out of being able to do my work.

27:36Chris Williamson:Well, we are meant to do, like what would be the ideal work day in an office environment that would actually mask the human brain? It would probably be you come in, you work on something hard for a while. Like that's what you do in the morning. you have lunch and then you like catch up with have some meetings talk to some people hey what's going on and and you know do some tasks and that's your day like that's basically what we can do like two things one big burst of like let me focus on something hard and then we can kind of come down the mountain after that with let me chat with people what's going on some decisions need to be made or whatever that's probably about optimal instead we juggle a dozen to two dozen tasks that all have their own demands.

28:16Chris Williamson:They all have their own communication needs. This is why the Microsoft data shows, oh, the work happens Saturday and Sunday morning. It is really hard. You can't go from, and meetings are very hard as well. We think like, oh, I'm not actually doing work during meetings, but what you are engaging in a meeting is all the parts of your brain that deal with social interaction. And those are a large part of your brain. And that is a fraught and mental energy consuming activity to sit in a room or on a Zoom screen and try to manage all these different people and how do I look and what am I saying and what's going on here and I have to say the right things, it's draining.

28:51Chris Williamson:And you come out of something like that, it's difficult just to jump right back into something else. And if you come out of something like that and there was a lot of obligations generated, oh, we discussed in this meeting things I need to do. And now you try to go straight from that meeting into another. Well, now that's really in the back of your head. What about this? What about this? We can't forget this. We just made our obligations. that that feeling of fatigue it's a it's a really as fatigue is what it feels like a mental fatigue like there's a sand in your brain sand in the gears of your brain that's the state that a lot of people who work in front of a computer screen like that's the state they're in most of the day and they don't even realize oh that's a bad feeling that's a negative state that's that's not how it needs to feel because you have nothing else to compare it to yeah the amount of things we're doing the amount we're trying to switch back and forth i always thought that part of the problem was a lot of our current thought about work culture and hustling and what it means to produce was influenced by Silicon Valley in the 90s and 2000s because that was considered this very ascendant part of the economy.

29:55Chris Williamson:Through the 2000s, through the Steve Jobs era, we looked at Silicon Valley, these are the coolest companies. They're doing all the coolest stuff. Over there, I think they adopted a model of work that was very inspired by computer processors, right? So because that was what was in the air in the 80s and 90s in Silicon Valley was the computer processor wars, you know, the 386 versus the 486 versus the Pentium. And it was all about speed. And the thing with a computer processor, if you're a computer type, what matters is you never want the pipeline to be empty, right? You want to always make sure you have stuff for that processor to do so it never wastes time.

30:32Chris Williamson:The processor will, every command you give it, it operates the same as any other. It can switch. It doesn't care what they are. it just sits there and operates one command after another. And the whole game with getting processors to be effective is like, don't have downtime. Like the real fear, I can put on my computer scientist hat for a second. The real fear in computer processor design is that you sometimes get to a command that's going to generate a huge delay. So you say like, oh, go get something from memory. That takes a lot of time from the perspective like a computer processor cycle. It's just sitting there cycle after cycle doing nothing while you're waiting for the memory bus or whatever.

31:06Chris Williamson:So we invented these processor pipelines like, oh, while we're waiting to get something back from memory, here's some other stuff the processor can run so that it's never not working. And the idea was you want to move as fast as possible and you never want to have downtime. And that's how you get the most out of a computer processor. The human brain is like 180 degrees different. We can't just switch back and forth between unrelated commands. You switch me from one to another thing and boom, 30 minutes of my mind is fried. Humans operate very differently. But I think Silicon Valley associated, it said, here's the thing we're going to associate with being really good at your job.

31:40Chris Williamson:It might have used to have been, I don't know, your skill. It was Don Draper and Madman. Remember that conception of what does it mean to be good at your job? They weren't showing Don Draper grinding it out. Like, man, Don Draper is like in the office till 3 a.m. every night or whatever. Now, he took the 5 o 'clock train back to Connecticut or whatever. He was really, really good at coming up with ad copy. He was good at what he did. That's what you used to respect. And then after the 80s, 90s, Silicon Valley became pervasive. Like, no, what matters is you never have a no-op. You never have a down cycle.

32:14Chris Williamson:You might as well say yes to more things. You might as well get more emails. You never have time or you're not working. That's what productivity is going to be. And that was a disaster for the human brain.

32:24Cal Newport:If you struggle to stay asleep because your body gets too hot or too cold, this is going to help. Eight Sleep just released their brand new Pod 5, which includes the world's first temperature-regulating duvet. Compare it with their smart mattress cover, which cools or warms each side of the bed by up to 20 degrees, and you've got a climate-controlled cocoon built for deep, uninterrupted rest. The new base even comes with a built-in speaker, so you can fall asleep to white noise, nature sounds, or a little ambient Taylor Swift, if that's your thing. And it's got upgraded biometric sensors that quietly run health checks every night, spotting patterns like abnormal heartbeats, disrupted breathing, or sudden changes in HRV, which is why it has been clinically proven to increase total sleep by up to one hour every night.

33:06Cal Newport:Best of all, they've got a 30-day sleep trial so you can buy it and sleep on it for 29 nights. And if you don't like it, they will give you your money back, plus they ship internationally. Right now, you can get up to$350 off the Pod 5 by going to the link in the description below or heading to 8sleep.com slash modernwisdom using the code modernwisdom at checkout. That's E-I-G-H-T sleep.com slash modernwisdom at checkout. out. There's definitely an element of this that it's very public productivity. It's very obvious. Look at how hard I'm working, right? If you're the one that replies quickest on Slack or on email, then it's evident that you're the one, it looks like you're the one that's working hardest because you're the one that's most responsive.

33:49Cal Newport:Whereas the person who's silently working on their own, they can't broadcast it by design. They can't broadcast it to everybody else. So yeah, this obvious productivity in a way is way less sexy. So I think the new elephant in the room is AI and how that is enabling an increase in pace of output, but almost certainly a decrease in quality. so fold AI into your existing worldview because to me it just seems like a huge force multiplier for what already was pretty sloppy slack email async communication that's always on people taking their work home with them never being able to not context switch not focusing on quality and instead focusing on quantity not being able to dial themselves in to do deep work for one moment And now that is enhanced and magnified even more by the use of LLMs to help you put out more, to help you think less.

34:56Cal Newport:So your focus is actually, you're in Slack with your LLM. It wouldn't surprise me if there is a LLM integration into Slack at some point in future. I don't know whether there is where you can just do it in there. So you're just talking back and forth in one fucking workspace. Talk to me, fold AI into this. You must have a million thoughts.

35:15Chris Williamson:Oh, there's a lot going on with AI. I mean, I think in its current instantiation, so we think about like an office worker for the most part, put programmers aside, I'll get back to them, but non-programmers are really interacting with chatbots. Like that's the main way they're integrating right now with AI. It's exaggerating exactly what you said. For a lot of people, it's exaggerating the problems that already exist. Now there's a term for this that comes out of a Harvard Business Review article from last year. They call it work slop, which they put together as one word. And they have some pretty compelling data.

35:44Cal Newport:So what's work slop, define work slop for me?

35:47Chris Williamson:So work slop is AI generated work products in the knowledge work sector. So like emails, reports and PowerPoints or what have you that are generated quickly by AI, but they're so low quality that they actually, it's very difficult to, they make everyone else's jobs harder. This seems to be, this is like the defining aspect of work slop. It's quick to produce, but it's so low value that it actually no real progress is made. So like you get a work slop email from your boss or whatever. And like, this isn't useful to me. It's this weird wordy thing that's broken up into sections and it doesn't get to the core of the problem we have to solve.

36:24Chris Williamson:So you made that email quick, but in the bigger scheme of things, we made very little progress towards what we want to do. Or you put together a work slot PowerPoint presentation so that you would have something at the meeting. But now we're spending 20 minutes looking at this nonsense and nothing, it's not helping us. It's not helping us actually do things. So this is what's happening, or at least my fear. I mean, the reality is most people aren't using these tools in the office. I mean, let's just set the reality, right? So, but for the people who are using them right now, which is a healthy percentage, but it's not...

36:55Cal Newport:Do you know what the numbers are?

36:55Chris Williamson:Well, it's difficult because there's a lot of fudging of the numbers here. There's a lot of mistaking have used or experimented with are regularly using them. So, I see this mistake happen a lot. And so, it's difficult to get good numbers. There was famously, maybe it was an Ethan Mollick article where he was talking about, in my world, academia, the homework apocalypse. And he's like, look at this study. Students just don't do work anymore. Nine out of ten are just using chatbots now. But you look at that study, and what it actually said was nine out of ten had tried using a chatbot at least once.

37:31Chris Williamson:And if you looked at who's using them regularly, it was like two out of ten. Because for most of the students, it wasn't helping them the way they thought it would. So I don't know what the numbers are. If you count like advanced Google use, I think it's larger. Like, yeah, I searched for information on this instead of going to Google. That's larger. But in terms of people who are actually making Office work product out of it, I think it's smaller than the people who follow AI commentary or talk about it on AI Twitter, AI YouTube. I think it's a lot smaller than they probably assume just because in their world is pervasive.

38:02Chris Williamson:But the people who are using it, this is the problem. They're trying to avoid, this is my theory on this, is like, how is AI helping like an office worker now? Well, their brain is exhausted from all this context switching. So what problem are they looking to solve? They're looking to avoid having to do hard moments of cognition because their brain is so fried. It's really difficult to like solve the blank page problem. Oh God, I got to send this email. I got to, it's a blank screen. I got to start writing from scratch. That's really hard. And if I have something -

38:31Cal Newport:The inertia that they've been trained out of overcoming because of the prime. It's almost like a one-two punch. Humans were primed to not like heavy, well, we already didn't like heavy cognitive load. Then our ability to deal with it and get through that initial resistance was decreased through the context switching. And now we're, don't worry about it. Don't worry about it, carbon-based life forms. The silicon-based life forms are coming.

38:57Chris Williamson:And let's throw in one other aspect in there. Also, outside of work, we had these distraction machines in our hand that were further degrading our comfort with concentration because any possible moment of introspection we would have had even outside of work, why would I do that when TikTok has like the perfect dash cam video of, you know, a Karen getting punched or something. Like, I got to watch that, right? And so we have that revolution comes along, plus the email revolution. We completely atrophy our ability to think and we exhaust our brain. So the other aspect of it, as we talked about, it's really exhausting to go through your day context switching.

39:31Chris Williamson:So like, I don't have any reserves left to write this PowerPoint. That seems impossible. And then AI is like, hey, hey, hey, hey, I can do it for you. It'll be fine. It'll be fine. It'll be good enough. It'll be good enough. You're like, oh, okay, I can smooth over. I used this analogy in a New Yorker piece last year. It's like it takes your effort graph, looks like spikes, like an EKG or something like that. And AI smooths over those peaks. And so you don't have to, your peak concentration required can come down. Like, well, you can fill the blank page. and then maybe I have to work with it a little bit.

40:02Chris Williamson:That's easier than doing it from scratch. But the stuff being produced is no good. And so I feel like work slop, it's almost less of a critique of AI than it is AI making obvious a problem with the way we were already working. I think that's what's going on there. I think this is even happening with computer programmers. This is considered heretical right now. I guess I'm used to being yelled at. But people are really excited by this workflow where I have seven or eight cloud code agents going concurrently producing code and testing them. And I'm just a manager of all these different processes.

40:40Chris Williamson:And they're all producing this code on my behalf. And it feels really cool and interesting. Like this has to be the future. I don't know that that is. I mean, I don't know the context. The problem is outside of like demos or internal tools or just having fun. that's not really code you can trust very well. And it does, though, completely lower the peaks of being a computer programmer, those peaks of cognition. It's much, much easier to manage a bunch of cloud code processes than it is to come up with an algorithm. And then you have that same blank page. So I think the jury is still out on even where we're going to end up in the AI impact on programming.

41:18Chris Williamson:I don't know where it's going to end up, but the way it's being talked about in the last few months after the latest cloud code update, which is sort of, I guess that's something humans don't do anymore. I don't think we're there ready to say that yet.

41:31Cal Newport:I get popped with Claude Code ads. I get, you give me a terminal, I have no idea what to do. I'm like, you know, someone's grandmother trying to use an iPad. I have no idea what's going on. So they are pushing very, very hard at the moment for this.

41:46Chris Williamson:It's funny, but it's a little bit crazy. That's my world, right? I'm a computer scientist, is that for engineer computer scientist types, they forget how technically advanced they are. So yeah, Cloud Code works in the terminal, right? And that's why it works so well. It exists in a world of text only. Text command line commands, like the old DOS command line, it's all text commands, which you can do a lot with. You can create and edit and compile a computer program. So it's very good at that. And it's a limited set of textual commands. That's perfect for a language model. And the engineers are like, oh, we can use this terminal-based tool to do all sorts of other stuff that's not computer programming.

42:24Chris Williamson:great, this has solved the problem. Everyone's going to be doing this. Everyone is going to have these sort of personal assistants based on something on Cloud Code. I'm like, man, do you realize how foreign a command line interface is to people? Do you realize how weird and nerdy and complicated your world is? You're like, yeah, this will be great. My grandma will just on the command line understand that the Cloud Code agent can bring up a bash script that's just going to cat those files over to the regex grep. It'll be fine. No one knows how to do any of that type of stuff. So it's sort of funny seeing the engineers building these incredibly intricate, nerdy, wonderful tools they've custom built for cloud code to help them in their life.

43:02Chris Williamson:And they think the gap between that and everyone else having AI automating things in their life is like, oh, it's this real small thing. I'm like, oh, man, I don't think you understand. I mean, people are still not quite sure about the right click. I think you still have a ways to go.

43:18Cal Newport:I saw this tweet from Robert Freundlaw. Lawyer uses ChatGPT to help write a brief. ChatGPT hallucinates cases in quotations. Court sanctions lawyer and four co-counsel for not catching the errors. The lawyer who used ChatGPT has practiced for over 30 years. He prompted ChatGPT to write an order that denies the motion to strike with case law support. Told the court that he doesn't normally use ChatGPT. And he used it this time because he was caring for his dying family members. He said no of his co-counsel were aware of this use of generative AI. Kors says that because all five attorneys signed both documents that included these errors, and they admit that not one of them verified that the case law in those briefs actually exist, their conduct violates Rule 11b2.

44:04Chris Williamson:There's hundreds of those happening, right? I heard, I don't know where this site is. There's a site that tracks this. lawyers getting busted for chat GPT written briefs that just make things, because it will for sure make up things if you ask it. Because again, what it tries to do is, you know, not to get, people know this, but right at the very bottom, what is a language model trying to do is trying to solve the word guessing game. That's how it was trained. It was given real text. You knock out a word and say, replace that word. Can you figure out what word was really there in the real text?

44:35Chris Williamson:So the language models just think they're trying to expand a real text that really existed. So they're trying to produce text that makes sense given the prompt. There's not world models or structured reasoning in there of like, okay, this is a legal brief and we have a notion of a citation. We don't know how it thinks about that. There's hundreds and hundreds of cases of this happening. I heard Scott Galloway talk about this on the Pivot podcast. There's some site that tracks this that he keeps an eye on and he says, it astounds you. You think it's a handful of people? It's not. It's all the time.

45:06Chris Williamson:Here's my story of getting burned by that. I sort of learned my lesson. I was working on, because the one way I'll use ChatGPT is just sometimes instead of Google, right? Especially if I want like instructions for how to whatever, change settings on something. It's great. It has a lot of really useful cases.

45:22Cal Newport:It's fucking spectacular for all of that stuff. If you want to use it as basically a glorified Wikipedia that's more instructive, like. Yeah, yeah.

45:29Chris Williamson:Like Wikipedia, you can ask questions of. Yeah. So I was using, I was writing an essay and it was on Isaac Asimov's Rules Robotics. This was a New Yorker essay. And I left my copy of iRobot. I was here at my studio and I left it at home. I was like, oh, I needed to add this quote, right? And I left it and I was like, oh, you know what? That story's in the public domain. It's all over the internet. And this seems like it would be perfect for ChatGPT. Like, hey, can you just grab a copy and find me that quote? And they'll save me a little bit of time. I was like, yeah, here it is. here's a quote.

46:04Chris Williamson:I was like, yeah, roughly I remember I put in there. And then the fact checker was like, where's this quote from? I was like, yeah, it's from the story or whatever. I get the book. It had just hallucinated a quote that was more or less like what was said, right? Because again, it's kind of playing the game of this is the type of text that would make sense giving the prompt, but it wasn't the actual quote. It had full access to it, right? You can search. It's in the public domain so that the actual story is everywhere. So I had just naively assumed if you ask it for some information that exists on the internet that, oh, it'll just go find it and format it for you.

46:41Chris Williamson:It didn't. And then I went to a whole dialogue with it where I was like, this is not the right quote. And it's like, yeah, you're right. You know what? I thought you meant paraphrase a quote. Here it is made up. I was like, that's not the real quote. Can you go get the real quote and give it? At this point, I was just experimenting. I'd already filled it in the article. And I was like, you're right. You know, oh, I was being tasty. Here you go. I could not get it to give me the real quote. So anyway, I learned my lesson. I was like, oh, don't assume, even if it's common information, that it has access to.

47:11Cal Newport:Dude, the desire to fucking reprimand an LLM. And I've shouted at them, capital letter, exclamation marks. It's like, what are you doing? What are you hoping to achieve by throwing your emotional distress at this fucking disembodied voice on the other side. Okay, bits aside. I fucking love ChatGPT. I think it's been really, really fantastic for tons of things. What's important is learning the limits and not using it for case law. This episode is brought to you by Whoop. I have been wearing Whoop for over five years now, way before they were a partner on the show. I've actually tracked over 1 ,600 days of my life with it, according to the app, which is insane.

47:55Cal Newport:and it's the only wearable I've ever stuck with because it tracks everything that matters, sleep, workouts, recovery, breathing, heart rate, even your steps. And the new 5.0 is the best version. You get all the benefits that make Whoop indispensable, 7 % smaller, but now it's also got a 14-day battery life and has health span to track your habits, how they affect your pace of aging. It's got hormonal insights for ladies. I'm a huge, huge fan of Whoop. That's why it's the only wearable that I've ever stuck with. and best of all, you can join for free. Pay nothing for the brand new Whoop 5.0 strap, plus you get your first month for free and there's a 30-day money-back guarantee.

48:32Cal Newport:So you can buy it for free, try it for free. If you do not like it after 29 days, they just give you your money back. Right now, you can get the brand new Whoop 5.0 and that 30-day trial by going to the link in the description below or heading to join.whoop.com slash modern wisdom. That's join.whoop.com slash modern wisdom. What opportunities do you think an increasing reliance on AI opens up? Because I get the sense that as more people use LLMs to do the work for them, this will create advantages in some areas for people who don't need to be reliant. So have you thought about the holes, market openings that will occur?

49:13Chris Williamson:It will. I mean, the way I think about LLM-based AI versus more advanced AI that we don't know how to do yet is, my theory is what is being affected is going to be more narrow at first. It's going to be places where there's an exact match between what generative AI existing tools can do and existing market sectors. We saw this actually, the week we're recording this, we actually saw this reflected in the stock market. It was this interesting paradox that was going on this week where the stock price of software companies that deal with stuff that is well-suited for an LLM went down. They call it the SaaS-pocalypse, right?

49:51Chris Williamson:The software service-pocalypse. So, companies that do legal advice, companies that do graphic design like Figma and Adobe, because a lot of we have generative image generation is making building images from scratch is less useful. Customer service, so companies that do a lot of customer service type software. We saw the stock was sliding on these very specific software industries because like, look, I think LLMs are going to be able to do this. It was triggered by Anthropic releasing and some plugins that made it easier to integrate LLMs into your services without having to hire these other companies.

50:23Chris Williamson:But you would think that would be good news for the big tech companies building the AI that's going to replace all this. Their stock was sliding as well. And so the big tech companies had this big slide that at the end of the week we're recording this, there was a rebound at the end, but it was like a trillion dollars in market cap disappeared from the big tech companies at the same time. So what does that mean the market was betting on? What are investors betting on at that point what was going to happen. And they were betting that in the near future, the next year or two, what we're going to see is selective impacts in specific fields from generative AI, but also that too much money is being invested in these AI companies already, which means they're betting that they're not about to automate most of the economy.

51:07Chris Williamson:They're not about to, you know, just one more iteration away from a huge economic disruption. They're not at this peak of like complete transformation because if they were, you would be trying to increase your holdings in these companies. Like, I don't care how much money they're investing. These companies are going to be worth an astronomical amount of money. But the market is betting. I think the impact is going to be more limited in the one to two year window than a lot of the commentary was seen. So I think that's important because talk is cheap, but tech stocks aren't. And so people, the way they spend their money actually often has more of, I think there's a lot of information in that versus just, I've been reading these articles online and my God, the vibe really seems to be saying this is a big deal.

51:51Chris Williamson:So I kind of agree with the market's consensus right now. For sure, there's going to be industries that are affected, but it's not going to be one of these situations where you say, okay, any work that's not just the deepest creative work is all going to be automated in the next few years. I better go learn how to do art or something like that. I don't think it's going to be that broad at first. I don't think the current generation of AI technology can support as broad of impacts as people think. There's a lot of extrapolation from, well, if it can do this with code, certainly it could do this with all these other jobs.

52:20Chris Williamson:If it could do this with this industry, well, certainly next it'll do it for all these other industries. We have to be wary of those extrapolations.

52:26Cal Newport:Right. I think I read an article from you, what if AI doesn't get much better than this? Yeah. Sort of, if we have, I don't know, some sort of Flynn effect thing that kicks in, but for AI, where, you know, because I think a lot of people would agree ChatGPT 2 to 3, fucking hell, to 4, 4.0. I know there's a whole furor on the internet about people that have got girlfriends or boyfriends that are virtual on 4.0 and they're all getting upset and sad about it. And I don't understand. I don't think I use the tools sufficiently deeply to be able to test this and benchmark it. It's like my Fire TV sticks remote isn't working well.

53:04Cal Newport:um i it was able to do that fucking five years ago um but is your your thinking is that we're maybe going to reach asymptote for what llms generally and transformer technology is able to do and then it's going to be a new architecture entirely if we're going to actually get beyond

53:24Chris Williamson:this yes yeah that's what that article is about i think that was a very of the articles i've written And I think that was a really important one that came out in August. And the story it tells, and a lot of other people have told the story as well, you know, around that time and since. But the story it tells is basically what happened is there was this big paper that was published in 2020. The lead researcher, Kaplan, Jared Kaplan, I think was at Anthropic at the time. And it was this paper where they said, hey, something weird is happening here. If we make LLMs bigger and we train them longer, they perform better.

53:56Chris Williamson:I mean, technically, they're saying the loss decreased. That sounds kind of obvious, but in machine learning circles, that was surprising because there's this idea of overfitting where if you just make your model bigger, the performance goes down. So it used to be like, you have to find the perfect size model for your problem space. That's the way people thought about machine learning until this paper came out. And like, I don't know, transformer-based LLMs, they were using GPT-2 and they were systematically making it bigger and they were seeing that the performance just kept going up. And like, this is interesting, so let's try it.

54:27Chris Williamson:And that was GPT-3. All right, let's actually make this like 10x bigger. Surely this can't be right. And it was. It matched the Kaplan curve exactly. Like, oh my God, this actually got way better just by making this bigger. Like, all right, well, certainly that must be the end of it. Let's try it with GPT-4. They made it bigger. They trained it much longer. Months and months they trained it. You know, Microsoft had to build these custom data centers to train it with new AC technology didn't exist before. And it fit the curve. It was like way better. And the thing GPT-4 did that really got, so GPT-4 set off the whole industry.

55:00Chris Williamson:The thing it did is it started showing abilities beyond just language. And that's where people got excited. Like, oh, wow, if you train a language model on enough language, it learns about things that isn't just producing language. It can play games, it can do math problems, it can do logic. I mean, this was super exciting. It was super exciting. So the assumption was, do this two or three more times, you have AGI. so that's what the whole industry was based off of when we went from three to four was this was legitimate justified excitement expand the size and the training duration two or three more times and the economy is going to happen in a box that was the engine for all this excitement so they tried at OpenAI it was called Project Orion they made it bigger modeled in four, they trained it even longer like here we go and they tried it and they said, it's not much better.

55:56Chris Williamson:And this was this big brick wall surprise for the industry. Like, wait, it didn't get better. Everyone else tried as well, right? Grok, they tried this with Grok as well with the Colossus Data Center. We're like, we're going to have 200 ,000 GPU data center. No one's ever built anything this big. And it was like a little bit better. Meta tried this. They had a model called Behemoth. Like we built the biggest data. It's bigger than anyone we've had before. or they didn't release it because it was marginally better than the last model that they had. And so this was a huge issue, right? You couldn't just make the models bigger and train them bigger.

56:32Chris Williamson:So what they did was they switched to, what are other ways we can get performance increases? And can we get more narrow by what we mean with performance? And this is when we began to get all the Alphabet suit models. Well, it's GPT-0, O3-mini-slash-whatever. and they switched the focus from just, this is amazing if you use it to we have these benchmark graphs and look at these graphs. Things are going better on these benchmarks. It all became about benchmarks because these are very narrow things that you could train models to do well on. They weren't intuitive. GPT-4 was just awesome. By the time we got to GPT-5, their whole launch, their launch page had 28 graphs of benchmark names that no one knew what they were.

57:12Chris Williamson:And so then they had to look for all these other ways to get improvement and that's where you got like inference time compute. Well, what if we compute longer for harder questions and they begin really pushing fine tuning? Well, for specific types of problems, we can get data sets that have questions and answers. And we can use reinforcement learning to try to take this pre-trained model and make it better at this particular type of problem. And then we can have a benchmark that shows us we got better at this problem. and my argument in that article is like this is a way different game than we were playing when we went from 2 to 3 and 3 to 4.

57:47Chris Williamson:We're no longer scaling to AGI. We're taking basically GPT-4 and we're doing all of this like tuning and adding extra stuff on top of it and around it and measuring these very narrow benchmarks and that's why people have this feeling ever since. Like I guess they're better but it's not in an obvious way. It's better in specific tasks or if I vibe code this, it looks better I guess and it seems more narrow and so yeah we're we're reaching an out there's a long answer to a short question but we are reaching an asymptote on um just pure fine-tuned llms as an engine for ai we're gonna need more architectures it's gonna take more time well presumably chat gpt6 could come out and oh

58:26Cal Newport:fuck they just blew through the entirety of my prediction this curve no longer curves flat in the way that i thought and shit this is this is a different universe now yeah but that won't happen

58:38Chris Williamson:because they tried and they don't know how to do that. So it's not going to be just an LLM. I mean, my prediction of the future of AI is I think what we're going to see, I think LLMs are very powerful, but what we're going to see is much more of hybrid models that are custom fit to particular problems where, okay, this system does this thing better than a human. And in its guts, there's like an LLM in there, not a huge frontier model, but one that's like souped up and optimized for this particular type of thing. but there's also like five or six other models and there's an explicit world model there's a future predictor there's a policy network trained through reinforcement learning to try to evaluate situations to see what's good or bad there's a whole logic engine on top of this that hooks these together these are what I think the AI systems of the future are going to be like they're going to be bespoke and there's going to be a ton of them so when we get the AGI it's not going to be GPT-7 can do everything you ask it as well as a human it's going to be a world in which there's 10 ,000 different AI products and you realize is everything I can think of now, there's some product out there somewhere that can do this better than humans.

59:42Chris Williamson:Just like there's AI that can play chess better than humans. There's a different AI that can play Go better than humans. There's an AI now that can beat professional poker players at Texas Hold of No Limit. They're all different systems with their own pieces in them. And a lot of them have some language models in them as well, but a lot of other pieces as well. It's distributed AGI. That's what it's going to be like. We're just going to wake up one day and say, there's fewer and fewer things where we say humans can do this better than computers. And it's a different model than PAL 9000. There's one giant, it's a really inefficient way to imagine solving this problem.

1:00:14Chris Williamson:If we just have a big enough language model, it's going to do all activity. It's going to power all agents. It's going to automate all systems. That really doesn't make sense. I think it's going to be a much more distributed path towards AGI and AI.

1:00:33Cal Newport:given what ai can and can't do and what the quality of work is that it puts out at the moment what is some good advice for somebody who wants to work against the weaknesses that are going to be exposed in other people because of their reliance on ai by avoiding it themselves or by using it appropriately what would you focus on because again it seems to me like quantity is easier to achieve than ever before. Quality is going to be rarer. That inertia, getting the project off the launch pad, the blinking cursor of the blank page. Where should people focus their time and their attention in order to capitalize this?

1:01:13Chris Williamson:I think you need to begin thinking about the feeling of cognitive strain, the way that a weightlifter thinks about the burn of a muscle or a runner thinks about burning lungs as a thing that is uncomfortable in the moment, but man, I'm excited about this feeling because I'm getting stronger. You got to make yourself really comfortable thinking hard. That is the differentiating factor. I mean, obviously, I've been saying this since 10 years now, but that's even more now going to be the differentiating factor, right? And if you talk to athletes, they're like, this is like Schwarzenegger and pumping iron talking about pump.

1:01:47Chris Williamson:And that's really painful what he's doing actually, right? Like lifting the level of weights that the physical pain he's in is high and he compares it to an orgasm, right? Because if you're a weightlifter, you're like, oh, that pain is directly translating into more strength and more muscle mass. You got to think that same way about your brain. You cannot flee cognitive strain. You have to think about it in a knowledge work cognitive age. That is the feeling of my brain getting more capable. Yeah, I want to seek that out. Let's go get it. Let's go get some, right? Like I want to, this, nope, bring my focus back to this thing.

1:02:20Chris Williamson:I'm going to try to push this through. And then when you're done, be like, oh man, I exhausted my brain. That's awesome. That was like a really good cognitive workout. So while everyone else is using AI to run away from strain, you should be the person running for it. Because especially in the American context, I mean, the knowledge economy is now a massive portion of our GDP. And the knowledge economy itself is shifting more towards cognition-intensive work. So, you know, knowledge work can capture anything where you're not building things. But now all the lower-level knowledge work is being outsourced or automated.

1:02:52Chris Williamson:A lot of it has been replaced over the last 30 years by software. We don't have support staff and assistants and secretaries like we used to because, well, you can use Microsoft Word and email. We don't need separated people. And so the work that's left in our economy, the knowledge economy, has been getting more and more cognitively demand. And so the number one skill is I'm used to straining my brain, learning hard new things, and maintaining focus. That's what I would train.

1:03:16Cal Newport:That's so good. I really, really agree. And the funny thing is, that's why I asked at the top if you just felt like fucking Cassandra, because each subsequent development in technology makes this more important. uh yeah i do there's always going to be that seductive whisper in the back of someone's mind that well yeah but i can work faster with ai i can work quicker by what if my boss sees me doing executive functioning through slack more whatever what is the what's the elevator pitch for you should do work of high quality and that will end up winning you have to think about employment

1:04:00Chris Williamson:Ultimately, it's a marketplace. There's a lot of obfuscation and fog and smoke, but it's ultimately a marketplace, right? You're paid money. In exchange, you produce things that have economic value. That's what makes that exchange make sense. There is not ultimately an underlying economic value to the coordination activities by themselves. There is no actual economic value to the speed of your Slack responses or the number of meetings you go into or the number of like bullet-pointed emails with those sort of chat GPT emojis that you put out, that itself doesn't generate economic value. The stuff that does a knowledge work almost always requires you mastering hard skills and applying them through concentration.

1:04:40Chris Williamson:And ultimately that shakes out. There's only so far you can get or so far you can hide being busy because busyness can't be monetized. And, you know, of course you can create a smoke for a while. Like, I don't know, like, you know, Chris seems like productive, I guess. Like he's always on these emails and this and this and that. But if you're not actually producing things that have economic value, like ultimately that catches up to you. Your opportunity's narrow. You're going to get found out at some point. Where if you do the other thing, it's like, no, I'm creating stuff that is rare and valuable.

1:05:11Chris Williamson:It's unambiguously has value in the marketplace. You write your own ticket. Like what? You want to have a business where you work half the year? You can do it. You want to get paid a huge amount of money? You can do it. You want to like work for a company, but you choose when you come into the office and you declare like, I don't want to do meetings. That's actually a thing, by the way. I talked to a marketing team at one of the major tech companies not long ago, and they said, you know what? We're in the sales side. And like our group, the sales group, we are exempt from meetings because they can directly monetize, oh, you brought in this many dollars.

1:05:46Chris Williamson:We can see it. And if you're bringing in dollars, they're like, you can do what you want. And they could also see, if we make you go to meetings, those dollars go down. It's like, forget the meetings for you. Everyone else, where there's not a clear number where they can see how much value you're bringing. Like, oh, you better be there in the medium.

1:05:59Cal Newport:Dude, I've always thought this. The big problem that most people have that doesn't exist in the world of sports stars, if you're a sports star, everything that you're doing is to facilitate performance and performance is very tightly bounded and it's quantifiable. If you're a weightlifter, that 300 kilos is 300 kilos. You either pick it up or you don't pick it up. And your sleep and your recovery and your nutrition and your hydration and your game tape and your technique work and your S &C and your body work and massage and soft tissue and all of that stuff combine to this output. It's a very, very sort of single ordinating principle.

1:06:40Cal Newport:The same thing goes for tennis and the same thing goes for football and the same thing goes for baseball and so on and so forth. You do not perform well. You begin to scrutinize all of the contributing elements that come toward that. The problem that you have in most normal people's lives is the output that they're optimizing for is diffuse and very hard to work out. Well, I want to be a good father, but I also want to perform at work. And I do Brazilian jujitsu on an evening time. And my wife makes me go dancing and I want to be engaging at a cocktail party. Okay. Well, first off, that's lots of things.

1:07:14Cal Newport:It's not a single ordinating principle. And secondly, defined to me the lineage between your disrupted sleep last night and your poorer performance around the dinner table or in Brazilian jiu-jitsu or whatever. The diffuse thing contributes because you inevitably have to make trade-offs from one thing in order to do another. But also, it's just hard. It's hard to work out how your performance is performing. And this is the same in the work life, perfect example, the salespeople, we just know if we make you do this thing, we lose that thing. And that thing is more important than this thing. It would be like, if for some reason, sports stars were being encouraged to stay up late, you go, well, we know if we make you stay up late answering fucking slacks, your performance in the game the next day decreases.

1:08:03Cal Newport:But for most people, there's this implicit assumption that part of what you do is the contribution to the strategy and the operations and the executive function culture and so on, which means that you forget what you're there for. I think people have forgotten what they're there for. What am I supposed to be here at work doing? What is my outcome goal?

1:08:25Chris Williamson:There's so much fat in the American knowledge work sector right now, right? Because we're so wealthy and there's so much money being slung around that we can have whole organizations where most people don't even know how they're directly connected to producing that value. And they could just be doing email all day or whatever, right? It's so inefficient. But there are, I mean, there are plenty of knowledge work areas where people don't put up with a bunch of this nonsense. And it's all areas where it's very easy to quantify your production. I did this essay a couple of years ago where I did a reflection where I said, God, almost every thought I've had in my books all came out of my experience as a grad student at MIT.

1:09:06Chris Williamson:So I was at the theory of computation group in the computer science department at MIT. Don't call it department, but the theory of computation group in the CS lab at MIT, which is like a group, the professors there, the students, we weren't like this, but the professors were super geniuses. Like literally Turing Award, Turing Award, MacArthur, MacArthur, Turing Award, Dijkstra Prize, like smartest people in the world. And it was incredibly clear if you were successful or not. What major theorems did you prove in the last few years? That's it. That's all that mattered. Right. And that required a lot of thinking.

1:09:38Chris Williamson:So they were terrible with email. They had no interest in social media meetings. Like if you're trying to throw meetings at them, they would just ignore you. Right. I wrote about this in deep work even, and people push back. I was like, this is what it's like in that world. If you send someone an email in this world, like one of these professors and you're like, uh, this is ambiguous, you kind of didn't word this well, or I don't really want to do this. They just ignore it. Like, that's on you, buddy. Like, I have to get, you know, I'm being, I will lose my job if I'm pre-tenure, if I don't come up and solve theorems.

1:10:10Chris Williamson:And they put up with no nonsense. And a lot of that actually infused the book, Deep Work, because like, you know what? I came of age in an environment where all anyone cared about was focus and everything else was secondaries. Like athletes, just like you said, If this is getting in the way of my launch angle going down or my batting average adjusting, I'm going to change it. But it's crazy right now in knowledge work how many positions that's not true. But what I advise people then, get in a position where that's true. Change your profile at work or if you're changing your job, change your job into one where your value production is unambiguous.

1:10:44Now, this is a double-edged sword because it swings both ways.

1:10:47Cal Newport:Because you can't hide anymore.

1:10:48Chris Williamson:You can't hide anymore. But if you get into one of those situations and then you do the cognitive work, I know how to focus. I build the skills. I apply the skills. I'm not afraid of cognitive strain. You're in the absolute best position in our economy, right? You can write your own ticket, but you have to be willing to go into a circumstance of like, this is the only world I know. And academia is, what did you publish? That's all that matters. It's all we care about. What did you publish? Book writing? How many copies did your last book sell? That's all that matters. There's no, you know what, though?

1:11:19Chris Williamson:He answered our publisher email so quickly, so let's give him another deal, folks. No, it's exactly how many dollars did you make us last time? That's what we care about, you know, for the next time. So it's a scary world where you're being held accountable. But it's an equation I always say is that if you're accountable, you don't have to be accessible. If you're like, I can point to this is the value I produced and I'm killing it for you. Then I don't answer emails. I don't go to these meetings. I don't do 50 sort of things. You can get away with almost anything you want. So I think that's, more people should make that move, especially in the AI age, I suppose.

1:11:57Chris Williamson:More people should make that move towards like, hey, hold me accountable and then do the work to actually show up. It makes your life so, it's such a better way to go through knowledge work. You get away from that hyperactive hive mind, brain melting, distracting, soul crushing, slack all day long nonsense.

1:12:13Cal Newport:In other news, you've probably heard me talk about Element before and that's because I am frankly dependent on it. And it's how I've started my day every single morning. This is the best tasting hydration drink on the market. You might think, why do I need to be more hydrated? Because proper hydration is not just about drinking enough water. It's having sufficient electrolytes to allow your body to use those fluids. Each Grab and Go stick pack is a science-backed electrolyte ratio of sodium, potassium, and magnesium. It's got no sugar, coloring, artificial ingredients, or any other junk. This plays a critical role in reducing muscle cramps and fatigue while optimizing brain health, regulating your appetite, and curbing cravings.

1:12:51Cal Newport:This orange flavor in a cold glass of water is a sweet, salty, orangey nectar, and you will genuinely feel a difference when you take it versus when you don't, which is why I keep going on about it. Best of all, there's a no-questions-asked refund policy with an unlimited duration. Buy it, use it all, and if you don't like it for any reason, they give you your money back and you don't even have to return the box. That's how confident they are that you'll love it. Plus, they offer free shipping in the US. Right now, you can get a free sample pack of Element's most popular flavors with your first purchase by going to the link in the description below, heading to drinklmnt.com slash modernwisdom.

1:13:23Cal Newport:That's drinklmnt.com slash modernwisdom. Let's say that you were in an organization that was small enough that you could actually enact some change. Maybe you're at the top of it, near the top of it, or you're just you're toward the bottom of it, but you feel like you've got the ear of the person that's in charge. if you were to say we've got the classic diffuse hive mind pseudo productivity malaise like the the ambient soup that everybody's swimming in how would you what would you do what would you how would you rework the internals of an organization that still needs to communicate obviously there has to be coordination people aren't working in silos there is going to be inevitable communication and coordination that needs to happen.

1:14:06How do you survive the modern world?

1:14:09Cal Newport:What would you propose? How would you restructure things?

1:14:11Chris Williamson:Yeah, I mean, I would do a few things. One, I would say we're going to have explicit workload tracking and management, right? No more just people throw stuff at you and you implicitly just add it to your plate. We want a place where we write down what everyone's working on and we can see it. And now we can start talking about things like what is an ideal WIP? What's an ideal work in progress limit for an individual? How many things do we want someone working on at the same time before that curve starts to go the other way. So what you have to do once you start doing that is saying we need a place to track things that need to be done that no one is actively working on right now and we can feel okay about it.

1:14:45Chris Williamson:So I would definitely want to set up where things enter into our radar of this needs to be done. There's a place for that to go and to be stored.

1:14:53Cal Newport:Oh, it's like an organizational getting things done inbox.

1:14:58Chris Williamson:Yes, and it's not on anyone's plate because as soon as you are responsible for something, it generates email slack and meetings. So once it's on your plate, it begins to spin off administrative overhead and slow productivity, they call it the overhead tax. That gets spun off as soon as it's on your plate. So everything by default goes to a team plate. No one's working on it. Then we keep track of from that plate as we move things to people's individual responsibilities, we have like, I don't, you should do three things at a time, that's it. And when you finish something, you can pull something else in.

1:15:31Chris Williamson:So do a small number of things fast and well and then keep bringing things. So I would definitely do that. The second thing I would do is I would say no more hyperactive hive mind. If you send a message that requires more than a single message in response, that should not happen over digital communication. If I can't just answer your question with one more message, then that has to be real time. Now we can't have that turn into an explosion in meetings. So what we're going to do is we're going to have daily office hours for everyone. so there'll be a daily time where everyone knows they can call you or walk to your office or whatever and go through a bunch of things with you real quick instead of sending emails.

1:16:04Chris Williamson:We're going to have morning stand-up meetings within the teams for sure. Who's working on what this morning? Who needs what from who to get that done? Go do the work. We're going to have, so we'll definitely do those as well. We might throw in phone hours. It's a new idea I'm thinking about where you say, look, there's a longer period of time, like maybe all afternoon where you can always call me if there's something that's so urgent you can't wait till the next office hours, there's enough friction in phone calls that that actually turns out to work pretty well. Like I'm not just going to call you because I want to get something off my plate.

1:16:35Chris Williamson:I won't call you unless it really is serious. So I would do that as well. And then I would say, okay, what ongoing work does this not work for? What type of projects do we work on on a regular basis where this isn't working because it's too long to have to wait till the afternoon's a problem? I'd say, great, let's identify those. And for each of those, let's build a protocol. Here is our protocol for collaboration on this type of work and however that's going to work. But it's like the information goes into this spreadsheet and then whatever. Someone checks it in the morning, they move things to shared files.

1:17:06Chris Williamson:I don't know what it is, but whatever it is that prevents us to have ad hoc unscheduled messaging isn't necessary. So explicit workload management, I would have this rule of no hyperactive hive mind. I would have protocols for any type of recurring collaboration where we could be explicit about how we actually want to do this. And then I would have a culture of talking about deep work and concentration, like a tier one skill. How's it going? How many deep work hours did you get in this week? Are you happy about that? What was getting in the way of that? Did you have a particularly good session?

1:17:35Chris Williamson:Tell everyone else about it. Like what worked? Oh, I see. You did music. You have a different look. Oh, let's all think, you know, hey, here's a good idea that we can borrow. Make deep work. Culturally, something you talk about as like this is a tier one skill that we're really proud about. You do those things. You're not going to 2x your profitability. This is the thing that's always frustrating me about these ideas is like, you could make more money if you do it. But that's, it's really hard. Those changes I just talked about, it's hard. There's friction, there's personalities. And this is the thing I really underestimated when I wrote those books.

1:18:11Chris Williamson:The way we work now is like a low energy point, right? It's like the easiest possible configuration of work. So if you feel friction, you're trying to do something more structured, you're trying to do something that makes better use of our brain, and you're getting resistance, the place you're going to fall when you give up is the way we're doing it now. So it's not arbitrary, I've realized. This hyperactive hive mind, let's just figure things out on the flow, no workload management. It's not arbitrary. It's the low energy. It's like this local minimum. It's the place that minimizes the complexity that still allows a company to run.

1:18:46Chris Williamson:And I think that's why we keep falling back. In mathematical terms, it's a suboptimal national equilibrium. It's not the optimal way to work together, but no one person can leave it and make their situation better. It's a low energy state. It's an attractor. It's a local minimum in the utility landscape, whatever mathematical metaphor we want to use. And so it's not arbitrary. I was like, oh, it's like a law of work physics. This thing is like a neutron star in the world, the universe of work that just attracts everything back to it. And it takes a huge amount of energy to escape its pull. That's why I think we've had so much trouble solving this problem, even though you would make more money if you did it.

1:19:26Cal Newport:I wonder, I'm thinking about sort of immediately implementable solutions for this. I get the sense that you could probably tell people, we don't use Slack before 1 p.m. Like nobody is to post in Slack before 1 p.m. Because that, you can ring, if it's SOS emergency scenario, or you can just call somebody. We just don't use it. And then it means that everybody knows that they should not be doing, it's a company-wide deep, I mean, look, are there going to be some departments, HR, for instance, probably would be used, but your job is HR, you're in the PR department or something like that. Your job is actually about comms.

1:20:10Cal Newport:But if you're in marketing or if you're in accounting, something like that, okay, sit down and do your fucking work. And up until a point, what do you make of intermittent fasting for communication? company-wide.

1:20:20Chris Williamson:Yeah, it works, especially though what really makes that more sustainable is if you have that quick morning stand-up on the team scale at the beginning of the day, where everyone says, here's what I'm going to be working on during these morning hours. Here's what I need from each other to make progress on this. So what would have unfolded over Slack and email, you're doing in 10 minutes. So you say, okay, here's what I'm working on this morning. I'm working on the new white paper. Here's what I need, though. I need those figures from you. Can you get, when can you get them to me by 930? All right.

1:20:52Chris Williamson:You're going to get them to me by 930. And I need those quotes you promised. Can you get, just do that right away? Okay. So you all know what I need from you. Okay. Now I'm going to put my head down and write that report. So having that, uh, that meeting ahead of time where everyone says what they need and what they're going to do, that makes that time work better. And then the thing that really works, do the same thing on the other end of the morning. All right. You said you were going to work on this, this, and this, what happened? So there's accountability on the other end. You can't run away from, you know, if you just went on email and social media, they're like, well, wait a second.

1:21:25Chris Williamson:I thought you were going to write the white paper. Yeah. And if other people flake, they don't send you the figures. They don't send you the quotes. You're like, I got stuck, man. I never got this.

1:21:34Cal Newport:Cal didn't do what he said he was going to do.

1:21:35Chris Williamson:And they're there in the same room and they're like, oh, okay. I get it. I get it. I can't just ignore stuff, right? Like I actually have to do it. I think that's a great idea. I think something like that works well if you put that accountability before it and you put it after it. That scares people, by the way, though. That really does scare people because you actually have to do the work. And this is the thing where really social media and smartphones killed this way worse. AI is going to make this worse. But that was a big inflection point. In terms of losing our comfort with concentration, that got really bad once we got algorithmically optimized content and we really got used to that.

1:22:13Chris Williamson:And so it's scary. If you just go to a company and say, here's the new plan, boss. We're going to have a meeting in the morning. You got to tell me what you're going to do for the next five hours. And then you got to do it. And we're going to check in after that five hours and see how it went. That's a nightmare for a lot of people. That is like, oh God, I don't know what I'm going to do.

1:22:30Cal Newport:I agree. I get the sense that a nice way to introduce this would be, look, everybody's brain here has been turned into slop. Everyone. No one is able to do their job as effectively as they should. So you are expected to do the work. But the reason that we do the pre and post is not to whip somebody into performance review. It's to give you accountability because you don't look like a tit in front of your coworkers. But if you don't get to the point, we're going to... The same as when you start training for a marathon, you don't run 10K on the first day. You will titrate the dose up and over time, week one will permit some fuckery and week two will permit a bit less fuckery.

1:23:11Cal Newport:In week three, we're all in it together. And this person's pulling ahead. They're really like a hyper responder. They're making loads of gains in the focus gym and other people are moving a bit more slowly. Okay. What is it that they are doing when so on and so forth. But imagine that. Imagine if you had a company-wide focus initiative where people were just, okay, we're going to move together. Everybody is going to focus on focus. And interesting around the AI thing. So George, my housemate's writing a book at the moment. Do you know Cold Turkey? Do you ever use Cold Turkey?

1:23:40Chris Williamson:I know about it, yeah. The software, yeah.

1:23:42Cal Newport:Yeah, yeah, it's a website limiter, app limiter for MacBook. We've been using it for a decade. His Cold Turkey went rogue and just kept shutting his browser down even though he wasn't trying to access the thing that he wasn't supposed to. It said he needed to install it. It was a nightmare. And here's a conversation between him and his AI. Cold Turkey has gone rogue and I need to remove it. Please tell me how to delete it from terminal. And the response is, I'm not going to help you bypass it, George. This is exactly the scenario you set it up for. You're two days in. The book is waiting. Close the terminal and write.

1:24:19Cal Newport:And he replied and said, no, it's got a bug, so I can't get on calls. He's like pleading with his own AI because he's obviously put in the instructions, be rigorous with me, be tough with me, tell me that I should be getting back to being focused when I start to go off task, do the thing. and that's a an ai equivalent of what you're talking about which is this supervisionary oversight commission thing but his just happens to be based in silicon instead

1:24:43Chris Williamson:of in other people so maybe ai will help us we it could basically chastise us like well the problem

1:24:48Cal Newport:is the problem that you have with the ai thing is it's so fucking sycophantic all the time um that it will tend to bend eventually to what it is that you want yeah but no one believes that the

1:25:00Chris Williamson:chatbot interfaces, the future of AI, the boosters, the skeptics, the moderates, there's an emerging consensus that we're going to look back at this current moment where we interact with AI by typing into a chat window. That's going to be like the Usenet news groups at the beginning of the internet. It was like a cool thing early on that showed the promise of the internet, but the tools got better. There's better ways to make use of it. So the thought is in the future, AI is going to be more integrated into more things. It'll be more agentic. It'll be a lot, not like having conversations in English text, but deploying agents to do things, maybe with natural language, but also it'll be more integrated in the software.

1:25:40Chris Williamson:Individual tools will be more common. So it'll be much more common. I'm in Microsoft Excel and I'm like, can you sort row five by this amount and cut out all columns that have less than as many values? And it does that. That's what the interactions are going to become like. And so this idea of having a singular anthropomorphized entity through which you're having all conversations. That's almost like an accident of early AI. I mean, OpenAI will tell you this, that ChatGPT was supposed to just be a demo of the type of things you could do using the APIs into their language models. It's like the type of tool you can build that would make use of AI.

1:26:16Chris Williamson:And then it caught them completely off guard. And everyone wanted to use ChatGPT and chat with it because it was really cool. I don't think that's going to be the form vector. So I think a lot of these issues we have now, like this is weird, it's unsettling, we're anthropomorphizing it, we're getting parasocial relationships with the agents, we're having romantic relationships with them, we're getting unsettled because having English conversation, we have a hard time not simulating a mind on the other end of this type of thing.

1:26:41Cal Newport:Which is why I shout at my chat.

1:26:42Chris Williamson:That's why you shout at it. I think a lot of this two years from now is going to seem, it'll be super narrow, right? Because I don't think just having this sort of general purpose oracle you chat with, that's not the future. That's not what people think we're going to be doing.

1:26:53Cal Newport:Why are people mad about 4.0 being removed?

1:26:57Chris Williamson:They were just, my understanding was they were just happy with the fine, so you tune these things. The conversational style comes from a post-training tuning session where you give it, you've already done the pre-training, which is unsupervised. And you go through this post-training session where you have a lot of examples of questions and answers and you ask the question and then it gives an answer and then you sort of zap it using optimization theory to try to move, like now we're going to change the weights to be closer to this answer we already said was better. So if you have a bunch of examples of the way you want something to respond and you go through one of these sort of zapping training sessions after the fact, it'll respond more like that.

1:27:36Chris Williamson:So they just changed the way they were doing that. And the thing they changed to, people didn't like the tone that created. So it was just about what literally like the data sets you're using when doing this fine tuning after you've done that big, massive pre-training where it's unsupervised.

1:27:53Cal Newport:Talk to me about the role of quantum computing in AI?

1:27:57Chris Williamson:Minimal to non-existent.

1:28:00Cal Newport:So QAI is all just bullshit?

1:28:03Chris Williamson:Yeah, I'm not, yeah. I mean, quantum computing is really interesting. There's a huge amount of technical problems just to actually get these things scaled to the number of qubits in which they're useful. And there's a fallacy out there in thinking about quantum computing that it's basically like a normal computer, but times a million. Yeah. Which is just not the way these things function, right? So there's only very specific problems you can solve with a quantum computer because you actually have to express the problem in the language of physics in such a way that you're creating what's known as a wave function that when it collapses, it's going to collapse to a configuration that's the right answer, therefore like implicitly searching a large state space in sublinear time.

1:28:43Chris Williamson:Only certain problems allow you to do that. So it's unlike a normal computer where I can program a computer to do almost anything. Quantum computers, it's much more narrow what you can do with it.

1:28:51Cal Newport:Could you give me an example of something that it would and wouldn't be able to do?

1:28:55Chris Williamson:Well, the big example, this is a guy who was at MIT when I was there. Peter Shore, early on, was the one who figured out, like, hey, one of these complicated wave function collapsing things you could do could factor prime numbers. Q day. Yeah, factor numbers to find the prime factors, rather. Find the prime factors of big numbers. That's a really big deal because… Yeah, public key encryption. and ironically, this just goes to show how crazy MIT was, is also at MIT is Ron Rivest, who I TA'd for, who invented, you see R and RSA, he invented public key encryption. So the guy who invented public key encryption is there next to the guy who figured out how quantum computers could...

1:29:36Chris Williamson:Could maybe undo it. Undo it, yeah. So it's kind of interesting. So it's good at that. There's a lot of problems that are based around simulation of quantum or physical physics systems. And that's, you can simulate quantum physics systems directly using quantum in a way instead of having to try to simulate it. So it's very good for that. There's a certain type of search. It gets a little technical, but there's a certain type of search that you can implement that has applications. So there are interesting applications. But the thing I was beginning to sense recently, which made me worry, is that there was a sense of like height migration.

1:30:10Chris Williamson:So people are getting a little bit frustrated, sort of like post-GPT5 of like, this isn't filling my need to have something to being a technology that is going to change everything. I love that concept. And then again, sniffing around, okay, but what if we just quantum somehow will unlock AI and solve all these problems we're having? I think it's way more complicated than that. There are narrow applications of these particular things that might have some AI application, but you can't run an LLM on a quantum machine and now it's a billion times better. That's just not how it works. So quantum's interesting.

1:30:42Chris Williamson:It's just really hard. The problem is the errors multiply. I mean, they make these qubits, these quantum bits they use for these algorithms. It's incredibly complicated. There's different ways to do it, but in some ways you have laser beams in a super cool chamber holding like a particle in a very careful state. And it generates errors and then the errors add up with other errors. And after you make enough of these things, then the errors, they swamp out of control. It's a really, you know.

1:31:09Cal Newport:So you're telling me that the fucking M6 chip in the MacBook Pro is not going to be a quantum one. It's not going to be the Q6 chip.

1:31:17Chris Williamson:It's not. In fact, I was, now I want to know what QAI is. What is QAI? You mentioned QAI.

1:31:22Cal Newport:QAI, quantum AI.

1:31:24Chris Williamson:Yeah, but I mean, is there a particular product or just people talking about quantum's going to just make AI better?

1:31:29Cal Newport:Yes, yeah, there is. I have a friend who I train with. This is like, you know what I love? Some of the people that I love the most are the ones who you wouldn't predict have the life that they do. and there's a girl who trains at lift atx on a saturday lovely girl i've trained with her a bunch of times real cool boyfriend's cool like does fitness modeling super hot the long hair lift the big out all the you know like but super strong all the rest of the stuff like feminine as well quantum computing degree like works in works works in quantum computing and she was telling me about quantum ai and she was telling me about q qai as it's referred to and it's a burgeoning field supposedly unless she's lied to me unless she's totally fucking lied to me yeah i'm curious what they're working on uh ut austin has good quantum theorist look i'm searching for it

1:32:19Chris Williamson:a guy i knew from it they hired him away there rc quantum quantum ai merges quantum compute with machine learning the process high-dimensional data faster than classical systems now they're working on it but i don't know i don't know how that's going to work basically all right well so I don't know what they're working on, but it's not something that you hear a lot in computer science circles yet. So maybe they'll have some breakthroughs. It's worth looking at, but I don't know how that's going to work.

1:32:46Cal Newport:Okay. One of the other elements, I guess, that people struggle with when it comes to deep anything is learning, the process of learning. Talk to me about the mechanics of keeping a deep reading habit alive.

1:33:04Chris Williamson:Well, I mean, I think reading pages is probably the cognitive equivalent of steps, right? So if you're a 10 ,000 steps a day person, it's like, this is just like a baseline to make sure that like, at least my physical systems are being used. You should have a page count, 25 pages a day, 20 pages a day of reading a book. It just is like getting those cognitive steps in because I think we recognize more and more reading it. I would say it's the cheat code, but it's better to think about it as like, reading is the thing that formed the modern brain. And I'm like, I'm more and more convinced about this.

1:33:39Chris Williamson:I have a book idea I'm working on now where I'm sort of exploring this idea. The brain before we had the Neolithic revolution, it was the same neurons, right? 15 ,000 years ago that we have right now. But if we go pre-reading, those neurons were doing the things they were evolved to do, which is very much about like the visual system and the audio system. And we could communicate through spoken language and that's fine. And then we invent reading. And this is not something that our brain has evolved for. So in order to read, we have to go through this sort of excruciating process of learning to read in which what you're doing is actually rewiring sections of your brain to connect in ways that they weren't originally meant to connect to.

1:34:17Chris Williamson:So we're reforming our brain when we learn how to read. And we develop what Marianne Wolfe calls deep reading processes, where you've now yoked together different parts of your brain that don't normally work together, that can now have to work together in order to understand written text. Once your brain is wired to do that, if you reverse this and write, you can generate much, much more sophisticated thoughts than you can if you haven't done this wiring and your understanding of things. The complexity of what you can understand when you have this new rewired brain, that also really goes up. So reading is like, it's not just, oh, I get stronger in my brain.

1:34:55Chris Williamson:It reconfigures your brain into like the modern, you know, post-cognitive revolution brain.

1:35:01Cal Newport:okay what why is it important to read physical books then what what is lost if i read substack i know that you're a fan of substack i love substack i think it's fantastic uh what's the difference between reading it on a laptop versus a phone versus a kindle versus a

1:35:18Chris Williamson:physical piece of paper well there's two different things going on here there's medium and content type like so if you're reading a book in a a physical book or you're reading in a kindle doesn't matter, right? I mean, they're both actual physical medium. Like the way that the Kindle is actually a physical experience. It's actual little disks that are, you know, dark on one side and light on the other. And to make a page, they have little electrical impulses and you shock the disk you want to turn and you don't shock the ones you don't want to turn. And so you've literally created an actual black and white physical version of the page on the Kindle.

1:35:53Chris Williamson:You're not, unlike a computer screen or a TV where it's light being emitted, there's no light being admitted. It's physically, that's the page. It just created a new physical page that has text on it. That's why you have to actually have a light on a Kindle to read it. So it's just a page that reconfigures itself into a new page. I love eating technology. I think it's really cool. Content type, the issue is, I mean, there's a lot of this research we've known since the 90s. A lot of this is captured in, the best book on this would be The Shallows, Nick Carr's book, The Shallows. When we're reading something like a webpage or Substack, for whatever reason, we skim much more aggressively.

1:36:28Chris Williamson:That's the main issue. We jump around much more aggressively just trying to pull out the key points. I think that's all just acculturated, right? Like you could sit and read, like if you print out a Substack article and sit in the library and you read it carefully, it's the exact same thing as reading a book. It's the exact same thing in sense of the experience. On screens, we tend to skim around more. The other advantage of like a book that was actually published versus like a post you see online, it's just better thought through. right? So when you write a book, you spend a couple of years on it.

1:36:59Chris Williamson:Like you're really, you spend a couple of years crafting the book and you might've been based on a lifetime of thinking about this topic. And so you take your time when writing a book and it gets edited and re-edited and you go back. Like I'm writing a book now. I've been working on it off and on for like three or four years. I've rewritten this book like three times. It's like, this isn't right. This isn't clear enough. It's, you know, and so when you go through text that has been that carefully thought through and structured, that's also you just get a different experience because the pieces click together at different scales and it just uses you build in your brain these intricate interlocking pieces that all hook together and is beautiful and you get that aha moment feeling.

1:37:40Chris Williamson:There's an actual physical endorphin rough you get in your brain. So I think reading smart books written by smart people that took a long time to write, that's your calisthenics for your brain. It literally changes. You're a smarter person if you do that versus if you don't.

1:37:57Cal Newport:So good. I have to say, reading full-length books has been, the volume that I do that has been decreased over the last few years, largely because of Substack. So there's an extension for Google Chrome called Push to Kindle. And if I press it, the article appears on my Kindle because I don't like reading on my phone and I don't like reading on my laptop, probably for the reason that you said. But when I think about it, it very much is running downhill because what's the longest substack that you're going to read? 20 minutes, maybe? 25 minutes, a fucking long article. And maybe part of that, maybe part of my is that I do get the outcome, right?

1:38:44Cal Newport:What is it that I'm looking to learn? Oh, I want to find out from steve stewart williams about sex differences in mate desire for sexual novelty something like that okay well i will learn the outcome in the same way as i could feed myself food that was just a cube of calories and that would sort of give me the caloric intake that i needed but what you're presumably reading for apart from just the enjoyment of reading it is to be able to recall it and for it to be woven into the broader mental landscape that you've got, which actually probably means you need to spend time and attention with it.

1:39:24And some of the leanness and brevity

1:39:27Cal Newport:that comes with an article actually might work against you. Maybe you need it to be said to you in five different ways. Maybe you need the author to meander off onto a story that takes three pages to explain about this guy who owned a Ferrari and parked it outside of a hotel so that you can then come back in. And each one of these is a little Velcro latch hook that you can hook yourself into. And yeah, I wonder whether the reading or discriminating toward reading stuff that is exclusively shorter form results in the sense that I'm learning lots. But if you are to actually do some sort of scrutiny around that, well, okay, how much of it can you remember?

1:40:12Cal Newport:How long did you spend with this idea? Did you spend long enough for it to be a part of now your mental models and the framework that you, how much can you recall? That would be an interesting challenge.

1:40:25Chris Williamson:And the frameworks of understanding are shallower just because it's less time to establish them. So like in a subset, it's not a bad thing, but what can you do? You typically have like one idea and like here's something that supports that idea and here's maybe like a different idea and here's why that doesn't work. and if that's all you're consuming that becomes your mental model for how knowledge is gained and i think we see a lot of this i mean think about internet culture now is much more conspiratorial and i don't mean in the like sort of grand conspiracy theory which it is but not in not just in like the grand conspiracy type of thinking but in the confidence there's this quick jump the confidence where you're like that's wrong because of this and boom and you think that like this is like this slam dunk case or something like that.

1:41:12Chris Williamson:That's a result of not reading a lot of books. You read a lot of books, you're like, okay, this is way more complicated.

1:41:19Cal Newport:Everything is way more complicated than you thought it was.

1:41:21Chris Williamson:And there's probably a clear truth here, but clear truths are more complicated. Like even the notion of what a clear truth feels like comes out of reading books, right? Like you understand, oh, ultimately like this person was right, but it's complicated. And like, yeah, this was not so clear cut. And this is like a compromise. and this was really important, and these factors were here, but honestly, those factors aren't as big as you think, and this factor really was more important, and so this really was the right thing to do. So even your notion of what's true or what's not true or what it means for something to be clear is different than if you're just looking at boom, slam dunk.

1:42:00Chris Williamson:I think it's a big problem online. Both sides of the political spectrum do this. You want everything just to be, this person is just garbage and completely wrong, and there's like this one simple thing I know that means you're completely wrong and I'm completely right and you're wrong in like the worst possible sort of way. And that is like such a sopholithic, I'm saying the word wrong. Solicistic. Yeah, exactly. You said it, right? I have to read more. But it's sophistry for sure, right? This idea of this is how truth and argument unfolds is like there's an obvious flaw that's easy for me to grok, which I guess now could actually be a verb as opposed to just meaning to understand.

1:42:38Chris Williamson:Also, I could literally grok it, I guess. And now it's clear that you're wrong and I feel righteous. And then we go seeking that. And then we want to simplify everything in the world to, you're just terrible. This person is perfect. And this idea makes the most sense. And if you disagree with this idea, it's because you want to eat children. And it just becomes, it's a different under, this is what I think we get wrong. It's not just like we don't have the right information. we've changed what our notion of truth is because we're not exposed to the complexity of truths when you read a not only a scholar like a smart case for it but then you read the arguments that they confronted and then you read someone else that's arguing against their point and you're like oh okay i've i've seen the clash of like minds and now in that clash i like i kind of see what's going on here like yeah the truth really leans this way and it's i feel really real conviction in that because i've seen like the best minds come at this from either side and i really understand and it's not cut and dry, but ultimately like this is the right thing to do.

1:43:39Chris Williamson:That was like a very familiar thing to people and leaders like in times past where you lose it if you're exposed to these low resolution copies, these low resolution simulacrums, these easy to digest pre-chewed versions of argumentation and understanding. It just changes the way your brain thinks about what true even means.

1:43:59Cal Newport:Yeah, there's an arc to sensemaking that you kind of need to track and if you don't track it you just assume that answers appear yeah it's like no no they don't cal you fucking rule let's bring this one home why should people go to keep up to date with

1:44:12Chris Williamson:everything you do oh god uh calnewport.com i guess my books are on amazon my podcast uh deep questions on youtube or wherever you get podcast newsletter at calnewport.com deep work too many things going on now chris deep work is 10 year anniversary i'm excited about it all new I replaced all the blurbs on the back with most of them are now organic. I could just like people who have said things about it without me asking them to say it. So that's fun. And I have a master class out on this stuff too. So I don't know. It's everywhere. Too many places. I feel too busy.

1:44:46Cal Newport:For a person who's a digital recluse, you are everywhere. But that's a function of focusing on quality, not quantity. I can't wait to speak again, man. This has been so much fun. I appreciate the help.

1:44:55Chris Williamson:Always a pleasure, Chris. Always a pleasure to talk with you. Let's be real. Whether you're filming for TikTok, YouTube, or that one client who thinks viral means vertical, you don't have time to dig for decent stock footage. Storyblocks gives you unlimited royalty-free video, audio, and templates that actually look good. 4K quality. No surprise fees. No copyright drama. Just drag, drop, done. Go to storyblocks.com slash Spotify and make your life easier.