I'm Cal Newport, and this is Deep Questions, the show about cultivating a deep life in a distracted world. I'm here in my deep work. HQ joined as always by my producer, Jesse. Yes, we got a good show. I'm going to revisit what we talked about last week and go deeper on my advice. We got some good questions. And then because I can't help myself.
We do another AI tech corner. Sounds good. It's not always within the flow of the advice we're giving, but I've just been reading and thinking and talking to a lot of people about AI recently. And I don't have anyone to talk to about this, so I'm holding you, my audience, hostage to hear gory details. Today I'm going to talk about what does it mean when you hear the AI companies like OpenAI say, oh, our new models can reason.
Right. And they're really obfuscating what that means. Like, how do they reason? And they're always like, we're giving them more time to think. What does that mean? I'm going to break that down in the tech corner in the show. I'm going to get in the weeds of like technically what's actually happening because for some reason they don't want us to know. All right. So we got a good show. Before we get into it, it's been you brought this up here.
Back when my book Slow Productivity had come out more recently, we like to do little updates at the beginning of the shows of honors it had won or achievements it had accomplished. I haven't done one in a while, but we have one here, which I would say, and you correct me if I'm wrong, the highest honor, would you say, in your opinion, this book has yet won. I don't want to say Pulitzer caliber, but I'm not going to not say.
Pulitzer caliber because slow productivity has been selected for what is this called, Jesse? The Privy Council.
Office of Canada. The Privy Council Office of Canada has selected slow productivity. We don't have video during this section, but I can tell you that the website featuring the book... beautifully designed it looks like right out of 1997 someone was editing html code and notepad uh so it was the privy council office of canada actually i am monitoring time anyone selects the book for anything i actually got my first
royalty statement so that covering how the book is done through November and it was doing a lot better than I thought actually I knew it was doing well but I kind of just royal statement is your check well so So yes and no. So royalty statements where you just twice a year, the report of here's how many copies your book has sold in the different formats during this period. You get twice a year. Now, if the amount of money it's earned has gone past your advance amount, then.
you earn money. But if it hasn't made back your advance, you still get the royalty statement. So you can think of it as like twice a year, the year is broken into two, and twice a year you get a check-in on like, here's how your book did. The thing about it is it lags. It lags by almost a whole period. So I'm getting right now. A period's a month or? No, a half year. Oh. So I'm getting now, as we record this at the end of March, the sale numbers through November.
And like in November or like in the fall, I'll get it, the sales numbers through June. So it lags. So I'm only seeing what has happened through November. I had tuned out because I find it stressful. Should we play a theme music and just tell the audience all the numbers? We'll play the theme music, yeah. And I will honestly, yeah, play the theme music. I'll honestly tell you the numbers. Slow Productivity has sold 320 million copies. That's coming out in March. So we're hitting our goals.
We're hitting our goals. Anyway, so it's good to see. All right, enough nonsense. Let's get rolling with our deep dive. So last week we talked about new testing data that was revealing that humans may have reached their peak intelligence levels right around the time smartphones became ubiquitous. And ever since that point.
Our intelligence levels as measured by test have been going down. So my conclusion last time we talked about this is that in this new world of smartphones, we might have to start training our brain. In a way that we didn't worry about 20 or 30 years ago. And my analogy was physical fitness. That when you're in the first half of the 20th century, we didn't think a lot about physical fitness because you got a lot of exercise.
Just in your day-to-day life, whether it's on the farm or, you know, you worked in a city. You know, people walked a lot in cities, Jesse, if you think about it. People would just walk like four miles to work or whatever. But then when we got to the second half of the 20th century.
And people had more sedentary lifestyles and jobs. We had cars and offices and air conditioner, etc. We had to start exercising. We never had to exercise before. Something changed about our world. We had to start exercising. I'm arguing that we probably have to do the same thing with our brains. 30 years ago, we didn't have to worry about specifically exercising our brains. Today, we do. All right, so here is the follow-up I want to do in today's deep dive. I gave some basic advice last time.
The cognitive equivalent of telling someone, hey, make sure you do some movement every day. Go for a walk every day. But a lot of people ask me a follow-up question. What would it look like, not just a sort of... be active somewhat cognitively, but they get in really good shape. What is the brain training equivalent of those routines you see online where the people are in incredible shape?
And it's like here is how you're going to completely change your fitness, your health, and your body. What is the equivalent of that for your brain? That's what I want to get into today. I'm going to suggest some ideas for if you wanted to be the equivalent of swole.
but for your brain. I'm not going to say your IQ because this really isn't about doing well on IQ tests, but really just about the ability to hold your attention and think interesting thoughts. What would that look like? I am not recommending that everyone needs to do this. I just think it's an interesting thought experiment. What would it take to not just stave off the cognitive decline induced by smartphones, but actually become someone whose brain is a real asset? All right. So that is.
That is our goal here. Let me start just by reminding us of the advice I gave when I last talked about this because I'm now going to take that advice, that basic advice, and say this is just how you lay a base. The equivalent in cognitive fitness and physical fitness, just sort of getting your body moving and flexible and your muscles woken up so you're ready to actually do more intense training. So just as a reminder, last time we talked about this, I recommended go on walks.
And think during the walks. Don't have something in your ear. Avoid dopamine stacking. So just make a rule. I don't dopamine stack. So that means if I'm watching something, I'm not also on my phone. If I'm reading something, I'm not also checking out videos. I just do one source of stimuli at a time. Read more. It doesn't matter what it is. Find what you're excited enough about that you're more likely to stick with it, but just grapple with the written word.
And maybe seek out a hobby that requires more concentration. Like you learn to play the guitar, practicing the guitar requires concentration. So those type of activities remain good activities for laying a base of cognitive fitness. Now I have some ideas for how you can push that base into something more elite. The first idea is interval training. Now this is the basic concept that I talked about, geez, I think all the way back in deep work.
And it was something that I had actually messed around with prior to deep work when I was a graduate student, primarily writing books for students. And I was working with a lot of undergraduates to help them do better in school. Worked with a variant of interval training because a lot of these undergraduates starting around this time were starting to struggle with concentration. Sort of the distraction era was on its way in. And we would do this training routine.
where you would focus hard on something, and their case would be academic work, for a limited amount of time, and they would have a timer. But they had to focus hard for that amount of time. If their attention wandered, it was like, no, no, you've got to restart that. You need to focus for this amount of time. If you look at your...
It wouldn't have been a phone back then, but if you jump on the internet or something like that, you have to restart. And then only once they got comfortable with a given interval... I would increase it by 10 minutes. All right, now let's try getting comfortable with this new length of time. So it's interval training. You get comfortable with an interval that's a stretch, and then you stretch that interval even farther.
And I found in like a semester or two, you could get a relatively distracted undergraduate from the sort of pre-smartphone era up to like 90 minutes of concentration pretty comfortably. I can lock in for 90 minutes. So I'm going to recommend that as like the first elite training technique. The key is to have something demanding of concentration to be focusing on during your interval training sessions.
And so this could be, you know, reading a hard book or taking an online course or working on a musical instrument practice is a great one that requires intense concentration or maybe a professional problem that's just like really challenging. You need a good thing to focus on that is hard. And then you use this timer. I recommend maybe once every two weeks or so. Like if you're comfortable with something, you want to give it at least two weeks before you increase it.
And it might take more if it just takes you more time. If you find a duration like you really can't get through it very well, it might take you longer until you're comfortable. That's okay. That stretch is where your brain gets stronger. All right. The second idea. Dialectical reading. We're moving here from interval training is about just the generic ability to concentrate. Dialectical reading is now about your specific...
grappling with specific subject matters and your ability to actually work with complicated information, not just concentrate, but work with complicated information. So at the core of dialectical reading is the following. You want to avoid the engagement with content that the algorithmically driven internet promotes. So if you're looking at content on Twitter.
Or the recommendation algorithms on YouTube. It's tuned to press a button in the human psyche that's very satisfying but also sort of almost anti-intellectual. The button it wants to press is. Your team is great. The other team is terrible. And here's an example of someone from the other team being terrible, and it's a slam dunk how terrible they are. That feels great for human beings that are wired for sort of in-group, out-group thinking. It's sort of a sugar high of out-group.
isolation so you watch like a twitter thread is going to be look at these look what they're doing over it's so obviously wrong and aren't they terrible and here's an example of terrible they don't even realize how terrible it is it wants to there's this sort of righteousness of like i am just super right
Feels great. My tribe is killing it. That tribe is terrible. And you get this on all sorts of divides of, you know, it could be from sports fandom, the political divides to whatever else is going on. Okay, dialectical reading says, no, no, we're going to teach you how to actually engage with information. So what you do is you take a topic that's interesting to you, probably a topic where you feel like I have an instinct towards this is the right side. Instead of looking for outrage bait content.
That's just going to pat you on the back and be like, yeah, you're, you're, you're right. You're Martin Luther King and everyone on the other side is Bull Connors, right? Like, or it's just like an obvious setup where you're just the best and everyone else is dumb. What you do is you find a really good book. pushing one side, and you find the best possible defense of the other side of the issue, and you read both those books. That's the dialectic. So I'm reading...
two opposing but very well-constructed takes on the same issue. The collision of two smart opposing takes creates an intellectual leap forward that is 10x more sophisticated than what you get just going to your choir stalls and having the preacher preach to you. Put the specifics of the content aside, it is a fantastic... intellectual experience. I mean, this goes back to Socrates who was a big believer in this method.
of intellectual engagement. Ironically, and not to do a big aside, ironically, Plato writing in the Phaedra's dialogue in the voice of Socrates, Socrates was a character in it, had the character Socrates express. skepticism about the written word.
Like, oh, this is a problem that we're writing things down. And a lot of tech optimists point to that and say, look, we were worried about even books, which we know are great. So, of course, we worry about everything. But this is actually what Plato was capturing in the Phaedra's dialogue was not that books were bad.
But then when it comes to like developing ideas, you need this sort of dialectical interaction, right, where you're going back and forth. I'm believing this. You're believing that. Let's go back and forth and see what comes out of that collision. Well, you can do that now with books, ironically, because there are enough books. Here's a really smart person on my side. Let me read their take. Here's a really smart person on my side. Let me read their take. Smart people opposing makes you smarter.
Now, one of the fears that people have about dialectical reasoning, it's going to make you smarter. The level of sophistication with which you engage with content goes up. It's like, look, Jesse and I are here in Washington, D.C. You have a conversation with politics with just –
the average person you meet here in Washington, D.C., is going to be a completely different experience than in Kansas City. Because everyone here, there's so many experts who understand things. It's just they have way more nuance, and it's just the way they understand political information.
It's so much more complicated than the average person. This dialectical reading gives you that type of more nuanced understanding. You see the complexities and where people are coming from. It just opens up another layer of understanding of the world. People worry about it because if there's an issue they feel really strongly about, they're worried that this will somehow like trick them out of their belief. To which there's two responses.
If encountering an opposing point of view is going to change your belief on something, then maybe it's actually not that good of a belief for you to have in the first place. Maybe it's more complicated than you thought. Maybe you're wrong. But two, what tends to happen is for the things where you really are right, like your instincts are right.
Dialectical reading makes that stronger. When you've encountered the very best arguments from the other side and you still come out saying this is why I believe in this. That belief is stronger. Your ability to advocate for it is stronger. Your understanding of why your thing is right and how to advance that is stronger. So you should never fear reading different sides of things. So dialectical reading is just going to make your grasp of information of the world.
not only more sophisticated, it's going to make X or TikTok content essentially unreadable or watchable. You'd be like, oh my God, I've been watching Succession. And now I've just gone back to leave it to beaver. You're like, I just, I, this is stupid. So it really will change your brain in an interesting way. Make you less mad too. Third idea, create idea documents.
I'm a big proponent of this. Is there something you care about or you want to care about or you just find is interesting, right? So something you might want to care about is like here is like a contemporary issue going on that affects me and I really want to understand it well and have like a point of view.
And be able to support my point of view and this is a cause I care about or it's just something of intellectual interest. Like, you know what? I really want to engage more with the Socratic philosophers or the big ideas of like the American republic or whatever.
Keep a document where you take notes. Take notes on what you've learned and how you think about it and keep updating and growing this document. In the writing down of information, in the editing of information, and in the rearranging of written information, you're... cognitive scaffolding around these ideas gets more sturdy gets more sophisticated writing is thinking and so as you listen to a podcast listen to a book
Maybe do like a great courses. You're doing a great courses course audio, one lesson every day during your commute. Take notes on what you're learning. Update your sort of summary of where you stand on something. That act of writing. will get you significantly more leaps and understanding than if you just take in the information and then don't engage with it again.
To organize your thoughts on the paper is going to better organize your thoughts in your head. It's just going to make you a smarter person. It's why, by the way, like when you encounter professors, they're very articulate about the things they care about because they're always writing about it.
You know, or, you know, if you get me going on one of my hobby horse issues, like get me going on email in the workplace or something like this, right? Something I've written about extensively. I can just go, right? I can just cook.
And I can give you 15 minutes of preacher-like content with this point, that point, this point, that point. It's because I've written about this so much. It has organized and structured my thoughts. And I can access that better. So it will just make you sound smarter. If you're working with written forms of the things you care about. Next, I'm going to recommend become a connoisseur in something. So become really good, not good at an activity, but.
good at understanding quality in that activity. You can become a connoisseur of NFL football without having to become good at playing NFL football. But you can really understand, for example, the complexities of different offensive schemes or defensive schemes, like what's going on, right? You can become a connoisseur of music without having to become a musician, become a connoisseur of movies without having to actually make your own movies.
I've seen this again when you come to more elite level thinkers. When you learn how to appreciate what really good means in one field, that translates to others. Even if you're not an expert on those other fields. You have an appreciation for expertise. You have an appreciation for quality. It adds shades of nuance and subtlety into the world. It's like the number of colors available in the world gets increased. When it's not just, oh, I'm...
Music is on and I'm hearing it. But you're instead understanding like, oh, this is why this was brilliant. Like the ability to hear, see, or encounter something and say, that's brilliant. That ability to do that. just allows you to see so many more shades. The world becomes richer. It becomes technicolor. It goes from, you know, the industry standard aspect ratios to 70 millimeter projection, the cinemascope or something like that.
So it changes your world and the richness of information that seems available. So for whatever reason, that just works really well, again, if you're trying to push your brain. Final suggestion, digital diet. Start to get careful and selective about what information you consume digitally. Because again, a lot of the stuff that you're encountering is working counter to all the benefits you're trying to have here.
So maybe very wary about my Instagram scrolling habit. That might not be serving your interest in making your brain smarter. And having been more of like an elite cognitive athlete, that's kind of the equivalent of like, I really like my McDonald's milkshakes, but I'm also trying to train for the marathon. I think I have to not take the milkshakes for a while. So become really careful. You know, it's like it's books, it's TV shows.
If it's online content, maybe you're moving towards podcast, newsletters, maybe like YouTube videos that you're not surfing recommendations, but you've bookmarked the pages. Where it's someone who is more thoughtful, more expert, more interesting commentary, right? I mean there is a world of – just to name someone who's killing it right now. Think about Ezra Klein right now who's actually killing it, Jesse. He's everywhere, him and Derek Thompson with their –
With their new book, right? And he's great. There's a world of difference, right? If it's I'm going to listen to like Ezra's podcast versus I'm going to go on X. To hear people talk about the same political issues. It's just a world of difference. It's still digital content. You're going to learn a lot about important issues in politics if you care about that.
But it's thoughtful. It's thought through. It's going to be presented with nuance, for example, a completely different experience than the WWE battle that's going on if you're in something like... An X thread or like an apologist show or something like that. So digital diet matters. So start caring about that. Just like if I was giving you advice for getting like in really good shape, we were going to have to end somewhere saying you have to care about what you eat. And that's the case as well.
All right, so here it is. Let's see if I can summarize all my suggestions. Lay the base before you try to, you want to become a cognitively elite. Lay the base with stuff like walking. Dopamine stack bands, reading, and concentration requiring hobbies. Once you're sort of like used to that somewhat more heightened life of the mind, interval train your concentration. Engage in dialectical reading.
keep idea documents, become a connoisseur in at least one thing, and then put in place a digital diet. Do that for a season.
Your cognitive experience of the world is going to be different. It's going to be – you're going to seem smarter to other people. You're going to seem smarter to you. You're going to see nuances you lost before. Your stress levels will probably be lower. Your ideas, your creativity is going to be higher. I think life would just be better. Not everyone has to do – not everyone has time or –
interest in becoming like a cognitive athlete but if you do that's the type of thing i believe you need to do now this is kind of what college used to be i guess it kind of still is was like a four-year cognitive athlete training program yeah i say used to because someone sent me uh an article and maybe it's on someone's newsletter but someone one of our listeners sent me an article that was it was a professor at like a mid-tier u.s college
And he was complaining. He's like, look, I know I'm Gen X. I've been a professor for 30 years. I know when you complain about students, people like, oh, there you go again, telling kids to get off your lawn. But he's like, this is important. I need to say it. And he kind of goes through, like, here's what I've seen change in these students at the college where he teaches. And he starts by saying they're functionally illiterate, right? And he's like, they don't read.
They don't read – they don't do any of the assignments. They lie about it. I don't think they could read. I think if I gave them just like a Richard Powers book, like an award-winning book but not like Finnegan's Wake but like a book that like it's out right now or like Colson Whitehead, they couldn't read and understand it.
Like he's really worried about it. And when you get to the bottom, he's like they can't write and the punctuation and capitalization is all over the place and there's no form. And when you get to the bottom, you're like, OK, what's going on here? He's like, well, it's the phone, stupid. He's like, it's the phone. They're completely addicted to the phones. And this is like a huge problem. So I don't know if you're not at like an elite school like Georgetown where.
I don't – the kids do read and they're very smart and they're very locked in and they're like more professional than I am. Like they're great students. But if you're not at like one of these top 20 type schools, college is no longer necessarily going to be …
For sure, like, yeah, you're going to get your four years of cognitive training. That's what I think college used to be. All this reading and writing and discussion, dialectical discussion in classrooms and with other students, you come out smarter than when you started, not just because you took in literal information.
but because you trained your brain and then the knowledge economy is like, great, these people, we have, we have smarted them up and now like they can handle like the complexity of what they're going to have and like the knowledge economy. I don't know if it's necessarily the case at all schools anymore. So.
So you might have to, you know, we have to think more about training, I guess, on your own. Are you a big fan of Succession? That's a great show. Yeah. Yeah, I watch Succession. We're watching all the shows. There's like a lot of shows right now. Yeah. Right now, because White Lotus is just ending, but it's still going. Pit is great. The Pit. I've seen a couple of those. Yeah. I'm a big ER guy, so Noah Wiley. It's like we've been liking The Pit, and then there's this new show.
Oh, Severance. I realized I guess I have to watch that given what I write about. So that's going on. There's a new season of that. And then there's this new show. Seth Rogen has a show on Apple TV called The Studio.
I'll have to check that out. He's a studio. He gets put in charge of a movie studio. Looks fantastic. So there's like all sorts of shows out here right now. Did you read the article in the New Yorker a couple weeks ago about the White Lotus director? Yeah. Sean White? Not Sean. Sean something. It's White.
It is white. Yeah. Yeah. White Lotus white. Yeah. And there was also like a lot of video. He did a lot of press and there's like a lot of videos going around about his creative technique. Yeah. He got really into that. So, yes, I do watch those shows. Nice. Succession. That was a good show. Occasionally I come across people who are just very rich. It's interesting. It's like, okay, someone that's spot on. They spend a lot of time in that show.
Only like really rich people would know this. Like there's, it's this like norm core stuff. There's like a, like the nondescript baseball hat. right to like strong jeremy strong will wear it's like a be like a specific like 500 baseball hat yeah there's a lot of like it's not supposed to be flashy um but people who know know that's like a super expensive hat
And supposedly the other things – because they had consultants who were like, no, this is how you should actually act if you're really that rich. And the other thing supposedly they told them early on is don't duck when you get off and on the helicopter.
Because if you were that rich, you'd be doing so much private helicopter flying. You would know that like it's too tall. It's not going to hit your head and you wouldn't – like ducking is what you do if you've never really been on a helicopter before. Neither here nor there. All right, so we've got some questions coming up, but first, let's hear about a sponsor. I want to talk about our friends at Notion because this is a product that I am a huge fan of.
Notion combines your notes, docs, and projects into one space that's simple and beautifully designed. It allows you to easily build basically custom information software tools without having to be the type of person who knows how to build. you know, a custom software product that you're just you're setting things up very easily. This is whether you're talking about, you know, an individual setting up.
A somewhat more advanced system for keeping track of all the things you have to do. I get so many messages from readers who have taken my ideas into custom Notion systems. Or you're like a really large company that's building major information tracking systems. Notion lets you do this really well.
quickly uh i mean you you've integrated jesse you've had your share of time working with it because our advocacy had that really cool system they built out of notion yeah that made stuff a lot easier yeah and you could see all the different pieces of information that were
Applicable to the podcast. And it would give it in different views. That was the cool part. So it's like, hey, what ad reads are happening on this day? Oh, let me click on that advertiser. Show me all of the ad reads we're doing in the next six months from that advertiser.
All I remember is we were doing something before that that was much worse. And I don't remember what it was. Weren't they just like sending us spreadsheets? Yeah, like a spreadsheet as an email. I don't know. There'd be like a guy with an old-timey... Newsboy Cap would come up with a telegraph. Western Union. Western Union. Just then. You got a Udacity ad. Next episode. Stop. Make sure you mention whatever. Stop.
That's how we were doing it before, and it made it much easier. So I love Notion because I love smart information systems, and they're the best in the business. They have – here's what's cool, though. What they've done recently, which I just want to underscore quickly. is they have integrated AI.
to make your systems like immediately much smarter. The fully integrated Notion AI helps you work faster, write better, and think bigger doing tasks that would normally take you hours in just seconds. So there's a lot of ways the AI now is seamless.
Searching for certain information using AI, you can do like a really intuitive, simple search. It can go through your information and find what you're looking for. When you're actually entering information into the system, it's right there to help you write faster. It could get a first draft or brainstorm or –
Polish up notes, right? So you've – this is a big one for me. Sometimes I take notes on the computer when I'm doing an interview and it's very sloppy. If you're throwing them in a Notion system, the Notion AI can clean it up and fix the sort of obvious mistakes.
So it's cool. So they're continuing. A product I already love is using AI in a really cool way. There's a reason why Notion is used by over half of Fortune 500 companies because teams that use Notion send less email, cancel more meetings. save time searching for work, and reduce spending on tools. Notion is my type of software. So try Notion for free when you go to notion.com.
That's all lowercase letters, notion.com slash cal to try the powerful, easy-to-use Notion AI today. When you use our link, you're supporting our show. Go to notion.com slash cal. I also want to talk about a new sponsor, Kinsta, Insta with a K. Something I've spent a lot of my adult life doing is dealing with WordPress.
Every website that I ever have, going back to the original Study Hacks blog, has been loaded on some sort of WordPress host, and it really has been a major time suck. It is difficult to get good WordPress hosting. It'll be unintuitive controls. It'll be security flaws that aren't updated. And above all else, these installations tend to get really slow. This is where Kinsta enters the scene. It does WordPress hosting right.
They focus, focus, focus on speed, speed, speed. We're going to keep our installation fast so that your site actually loads, which, again, if you're doing WordPress hosting, this becomes a real issue. They're great at it. They also have enterprise-level security. You're not worried about, oh, we're out of date on the latest security fix and hackers are going to get in there and mess with our site. You can rest easy if you're using Kinsta. Fantastic dashboard.
This makes a big difference. A lot of dashboards for these WordPress hosts – and again, Jesse, you don't have to see a lot of these because you don't do as much of the web stuff. But a lot of them look like – They were basically out of the movie War Games starring Matthew Brodwick, you know, where he's like typing into the green text like goes across and the Kinsta interface is fantastic. Perhaps like.
That's my second favorite thing about them. My first is the speech. Second favorite thing, real support. 24 hours a day, seven days a week, 365 days a year. You get a real human on the line when you have an issue, right? So if you've dealt with WordPress. All of these pain points should sound familiar, and then I can point you with confidence towards the solution you should use for your WordPress hosting, which is Kinsta. I wish I had known about them. God, it's been 20 years.
I think I launched calnewport.com on an old WordPress installation in 2006. 20 years. I wish I knew about Kinsta 20 years ago. Sure, they didn't exist 20 years ago, but that's no excuse. It would have made my life a lot easier. So if you're tired of being your own hosting support team, switch to Kinsta and get your first month free. And don't worry about the move. They'll handle the migration for you. No tech expertise required.
Just visit Kinsta.com slash deepquestions to get started. That's K-I-N-S-T-A dot com slash deepquestions. All right, Jesse, let's move on to some questions. Hi, first question's from Anna. I envision a future where I live close to family, grow my own food, work from home, and have time for painting. However, job opportunities in my animation industry are dwindling, especially remote work.
Should I change careers to achieve my ideal lifestyle? It's our second animation question in a few weeks. Interesting. I wonder if that's a coincidence. I'm reading a book on another Disney book right now. And it's talking about the period where they get CalArts up and running. So this was basically like the training camp that Disney helped fund to help generate new animators. So I know more about animation than I probably should.
All right, but Anna, this is not about animation. You're asking a question about lifestyle-centric planning. So first of all, I love the approach you're taking here. This is classic LCP. You have an ideal lifestyle vision in mind, and now you're working backwards from that vision to saying how can I actually –
So when you're working backwards from a lifestyle vision, the things that matter typically is going to fall into the categories of obstacles and opportunities. What are the obstacles in the way for me getting where I am right now? To something closer to that ideal lifestyle, the opportunities are what can I leverage? What type of like unique setup? This could be like career capital or family connection or whatever. What opportunities do I have to leverage as well? So you want to.
systematically work at reducing or bypassing obstacles while taking advantage of everything that you have, all the opportunities you have as well. So that's probably the right thing to do here. The key obstacle in this analysis is you're seeing the particular types of jobs you have in the animation industry is not remote work friendly. And your lifestyle vision, when you work backwards from it, remote work really makes that work.
Because if you can be in a lower cost of living place, you could be at home and growing food and be near family and maybe not need to work as many hours because where you're living is less expensive. If instead of being near Burbank or Disney is, you're instead.
You know, where you grew up in Indiana is going to be cheaper to live. Like remote work seems to be at the core of it. So there's a key obstacle here if your current job trajectory isn't making that possible. So now you know what you're doing. Oh. I am systematically exploring and looking for insight. It might not be obvious, but I'm looking for a moment of insight about how do I reduce that obstacle.
That's probably going to come from leveraging the opportunity of the various skills you have and looking for adjacent applications. So like, sure, maybe if I want to be in like feature motion picture animation, there's like three options. They're all based in California and they're cutting back and there's not remote work options.
But maybe there's an adjacent step here. Actually, this graphic skills I have could be used maybe not in like feature film, whatever animation, but it could be used in. whatever, game development, object design, or in certain types of advertising or something. I find that adjacent field where it's very useful. And now I have taken an opportunity and got around an obstacle. So it's not easy.
But it's specific. That's what's key about LCP is that it's very specific what you're doing here. I have this obstacle. I'm enumerating my opportunities. The only... piece of like semi-advanced advice I would give is that if you're stuck, A, keep searching for ideas and inspiration. Talk to people, read things, listen to things. You're looking for that path. It could be narrow, but you're looking for the path that gets you closer.
So increasing the opportunity side is often a low-hanging fruit that can make a big difference, and that might mean learning a new skill, a relevant skill. I know how to do this. What if I learned how to do this? Someone who already knows how to do this first thing would have an easier time learning it, like looking for these adjacent skills that if you learned would open up new opportunities. So you're kind of playing the game here, but you're thinking about it the right way.
You have a lifestyle in mind. You're trying to work backwards from it. So you're doing the right thing. And don't be frustrated if it takes a while to finally find that path. Also, I'll throw in one more coda. You know what you want in your ideal lifestyle.
You can isolate some of the general properties here, like a little bit more of a slowness, a connection back to nature when you're growing your food, connection to community. Keep in mind, even while you're looking for a larger reconfiguration that gets you your full ideal lifestyle.
You could be making changes to your life as it exists now to have more of those properties. So even if you're not getting the full right off the bat, I'm on the land with the farm near my family's house in Indiana or whatever. If you're not there yet. Use this lifestyle to help isolate these things are important to me and get that into your life right away.
And maybe now you have a plot in a community garden and you've adjusted your hours. It's still the in-person job, but it's a four-day-a-week job now. You have a little bit more slowness going into it. You're more community-involved where you live, even if you can't be closer to family. Play with these things you're identifying as important and start getting them in your life now. Don't let perfect be the enemy of better.
And that's like a key thing with LCP is that it's not just until I'm living on the houseboat, I'm in trouble. It's what is it I'm identifying as important? Let's keep adding this to my life. And actually, that might open up other ideas for how to get there. So anyways. Good luck with that, but thanks for the good LCP example. All right, who do we got next? Next question is from Ruben. I'm currently learning how to code during my free time to get a job as a software developer.
I believe it would be an ideal career path towards achieving my life I want. Is this a bad idea to leave part of my career capital with my BA in psychology? I'm 24 years old. Well, I mean, the good news is you're 24 years old, so you can't really... have much career capital anyways so like you're at the beginning you have a lot less opportunity loss opportunity cost threats here so yeah this is not a bad time i'm right out of college
And I'm building out my career capital portfolio. You're starting from low stores in all of your career capital category. So I think it's completely fine. You could probably leverage the BA a little bit. Like here's something I would say. Here's an interesting piece of data that's been touted.
In the last few months, we're starting to see some of the first places in the labor statistics. So not just anecdotally, but some of the first places in the labor statistics where we think we're seeing the impact of AI. is in a class of people, technical jobs called programming, which is different than software development. So programming jobs post-ChatGPT have been suddenly declining.
Software development jobs have not at all. They're basically the same if not like up a little bit. You have to kind of go into the weeds. What's the difference? Well, programming jobs, I guess there's a very specific type of – computer programming related technical position where it's you're given very specific things to implement often with like simpler programming languages like okay go add this feature now
Can you go onto the side and change that? So it's like very – you're doing very little planning or creativity. You're not watching the lifestyle of a product. You're just sort of implementing. ChatGPT can produce that type of simpler code pretty well. And so now – The person who was tasking the programmer with that work could just sort of use AI help and might as well do it themselves.
Software dev is more complicated. You're actually – it's part of the lifestyle of the life cycle of the product you're involved in. It's creative. It's figuring out what the program needs just as much as what code it has to be in there. You might be able to leverage your BA in psychology to say, look, I – I speak normal person speak. I can think about the consumer and what they want in their product. I understand other people. I can manage other people.
There is, I think, potentially a career capital advantage to be someone who's technically savvy who also can communicate like a normal human being. And I'm sort of throwing my fellow computer scientists under the bus here, but let's just say we're not all expert communicators. Like this is not necessarily we're not necessarily it's not a bunch of like John F. Kennedy's when you're at a computer science conference. Right. It's it's we I do pretty well.
But, you know, it's a lot of leet speak, right? It's a lot of like weird sort of technical terms and jargon. So you might have an advantage with your BA in psychology. My poor wife used to come to some of these. MIT events, like holiday parties or whatever.
what's going on don't take me to these things anymore like why half the people here can't communicate the other half are very normal it's just you get a higher percentage than normal of people who are you know yeah like if you see early footage of like bill gates It's just like a lot of him like rocking and writing a simpler code in his head. Yeah. Yeah. It's not exactly – you wouldn't want a theoretical computer science student at MIT like on a –
Like a radio talk show host or something. Yeah. Like it wouldn't be a popular show. You wouldn't want to replace Mad Dog with like one of my commenters. Maybe like the performance. The Red Sox was highly illogical. His numbers against the lefties were – the performance were obviously lower. And now I'm going to recap in great detail information from the latest Star Wars or Star Trek. I don't know.
We are going to go through the 10 points that make it clear that Star Trek Deep Space Nine is superior to the next generation. I believe this is obvious. All right. I'm sorry. Actually, most of the computer science I know are cool. Programmers aren't always. So maybe you do have an advantage there. So take advantage of that. All right. Who do we got next? Next question from Bill. Do you have examples of truly fair performance evaluation systems?
How should merit be rewarded in a revenue constrained environment? I work at a national nonprofit. I think what you should do and I think the lawyers will back me up is you want to sort people into different like demographic groups and then just pay your favorite groups more. So if you like the country they're from, pay them more than no. Aren't we allowed to say that now? Aren't we all just like whatever, anything goes? I don't want my funding cut. So yeah, just no DEI.
No, there is – so it is tricky. Performance-based review. I wrote a New Yorker piece back in the pandemic from – there is a system called ROW, R-O-W-E, Results Oriented Workplace Environment. And there's a book about this that's called something like Work Doesn't Have to Suck or something like that. I mean, there's only so many business books that have that in the title, so you'll find it. It's fascinating. It is an attempt to rebuild an office.
If you apply the Roe methodology, you rebuild your office entirely around results is all that matters. And there's no expectations like, hey, when you want to work, you work. I don't care. There's no vacation. There's no hour. You don't have to be here any particular hours.
You essentially sort of like negotiate, okay, I will work on this now. Here's like the measure of success. Here's how we're going to communicate about it to make it done. And you kind of have like this portfolio of projects that you do and you're held accountable to it. It really does work pretty well when it's implemented. But the point of this article is it's very hard to implement it.
And honestly, the big issue they have – and I interviewed one of the creators of it and I read the book and talked to a CEO of a company that uses this methodology. One of the biggest issues they have is not figuring out how to measure performance. It actually gets pretty clear. Like, okay, you're doing this.
all right we know what success means here right like are you we can just look at like in a row environment i could just be like these are the three things you did last quarter does this seem valuable or not like there's no hiding the hard part actually is trying to eliminate
The last vestiges of pseudo-productivity thinking in the office, right? Pseudo-productivity being my term for my book, slow productivity for using just visible activity as a proxy for useful effort. There is a huge amount of the row. I guess you would call it like application effort, conversion efforts, is trying to get rid of pseudo-productivity speak. And they call it sludge in the book.
But it's just this idea of people are very – because of the way we built knowledge work are just very uncomfortable about like why aren't you here or I can see you. Why didn't you answer an email right away? It is a difficult mindset to get away from performing activity to generating measurable results. And it's a hard shift. They also found that they won't.
And naturally, the people who created this methodology say it all goes well. But when you talk to people who have actually implemented it, they will tell you you're going to lose a lot of people. Not everyone can do this. The CEO I interviewed for this New Yorker piece lost like 20% of his non-managerial staff and about 20% of the managerial staff. The non-managerial staff, they're like, I am actually not able to perform like this. I could be busy.
My TPS reports are handed in. I answer emails. I'm always jumping on calls or whatever. But if you just say, forget all that, just tell me what you did this last month. Some people like 20% are like, I actually am not comfortable just focusing and doing hard things. And they lost 20% of the managers because they're like, I.
Don't know how to do this if I'm not just like being able to demand quick responses and just have people jump. So it's possible, but it's hard. Read my New Yorker piece. That would have been from like 2021. And I forgot what it was called, but just search like Cal Newport. R-O-W-E, you'll find it. And in that book, Work Doesn't Have to Suck or something like that. We'll talk about that philosophy. All right, who we got next? Next question is from Ben.
In episode 345, you talked about the value of more niche online communities. What do these small alternatives look like and how do people find and engage with them if they're not hosted on these global platforms? There's a few forms, so it's more like niche. Online discussion communities, which I think, by the way, is like the right way to extract the community building value of the Internet. A few places to find these. Bulletin boards are a big one.
Oftentimes now the technology is not going to be a straight-up bulletin board technology. It might be like at talknats.com, the example I often use in my writing of a Washington Nationals-based community. What they actually do, and a lot of sports sites do this, is they'll post like a blog post and the conversation is happening in the comments of the blog post.
Straight up bulletin boards are fine. I actually look at Reddit threads as in this, even though Reddit is a large global platform, the individual Reddit threads are not algorithmically sorted and they are managed by the community themselves. Substack or other newsletter-based comments has become a big place. Like I get a substack from some writer I like.
The comments underneath the substack becomes like a community gathering place. A lot of people have Discord server set up, like you like a given show or a given type of whatever. Discord servers are big. You jump on there and you can talk and it's like a relatively smaller group of people. So all of this is out there. Podcast. So anything that's Patreon protected?
There's discussion communities around it, whether it's actually using Patreon or, you know, its own homegrown solution. So you'll see this with like, I don't know, like if you're within like Sam Harris's walled garden, there's like places you can chat with other people. So they're out there. They're out there. The key though is that it is a self-selected group of people interested in the same topic. It is a manageable number of people who are there. So you actually like know.
a lot about these people like you've interacted them with enough you recognize them and have a bit of a shared history with them even if you don't know who they are like really in the physical real world and it's not algorithmically curated it's just the content is there probably just chronological
And the standards are set by the community. That's what you're looking for, for, I think, like a richer online engagement. All right. Who else do we got? Steven asks, how can partners support each other in building a deep life? I think the key here is that the life is one that you are coming up with together. So when you have your sort of vision of an ideal lifestyle that you're sort of working towards, you need a shared vision first. Like this is where...
Our family wants to be. This is what our family life should be like in five years and in 10 years. Like if you have kids, for example, you should break this up by the sort of like pre-elementary school, elementary, elementary period.
The sort of middle school, high school period and then sort of the post high school college period. What is our family life? What do we want? Like where we live, the rhythm, what it's like. Is there someone home after school? What's our weekends like? You want to have that figured out together. Then underneath that, you can have your individual.
OK, now here's the things that are important to me, like what I want to do with my physical fitness, for example, might be different than like what my wife wants to do or my interest in my hobby interest are for sure, you know, different. So start with the shared thing. That's priority one.
We're on the same page. We're planning with that. And then you can have your own sort of buckets you're working on. And there, I think it's helpful to some degree to kind of like, hey, here's what I'm working on. What are you working on? What's important to you? What's important to me?
How can I help you with what you're doing? How can you help me with what I'm doing? And that works out as well. But make sure that there is a shared component. Like the stuff, if you're in a partnership with someone, that partnership has to extend to your vision of what the deep life is. All right. Let's see here. I'm going to suggest we're going to skip either the call or the case study just so we can roll into the tech corner with time to go today. Okay. Which one? I'm going to skip the call.
Okay. And I'm going to jump right ahead to the case study. This is where we find out that the call we skipped was, I was going to say from someone famous, but I couldn't, I don't know like what name I'm not up with like the hip references.
You know how old fashioned I am. Mr. Beast. Yeah. Well, see, my first instinct was to be like King Charles. So how to touch. And Mr. Beast called in and we missed it. It's like, I need you on my channel stat. All right. Our case study today comes from Chris. Chris says.
I was recently approached by a firm looking to fill a position that I was uniquely qualified for. It would have been a dramatic step up in pay with a profit sharing component that may have doubled or even tripled my current income in a good year.
I received an offer for the role, and part of me wanted to just go for it because the income increase was so dramatic, but I decided to put it through a more holistic, lifestyle-centric evaluation. I VBLCCP'd it. Here's what I realized after further reflecting. This role was something of a grand goal trap. I was telling myself that if I could just get to this executive level, I would have, quote, made it, end quote, and everything else would have fallen into place.
The position would have required a move to a higher cost of living area and from a fully remote role to a three days per week in office plus travel requirements that seem to grow exponentially as I really pressed the hiring team to clarify. This would have substantially reduced the number of hours I'm able to spend with my three-year-old, who I am committed to being present parent for, and would have required a big renegotiation.
of the parenting burden with my partner. It was also a high risk, high reward job, meaning what I would have been more fearful of termination if I wasn't performing with complete excellence. I also did the work to process what enough meant.
for the income requirements, my lifestyle vision, and I realized that while I do need a bit more coming in annually, I'm not that far off. Instead of taking this dramatic leap up in both income and work hours, I'm looking at other options like a more modest move to a full-time role.
That increases pay but maintains most of my current balance. That there, Jesse, is VBLCCP in action, values-based lifestyle-centric career plan. There you go. When you know what matters to you, what you want your life to look like. What's happening, like your job becomes just a tool you have among other tools to craft this life. That's a fantastic decision. I talk about that in my book, So Good They Can't Ignore You. That is one of the control traps.
That just as you get good enough to have some leverage over your life is exactly when people are going to come to you and give you shiny opportunities to take your autonomy away.
So just as you're like, I'm pretty good at this. I'm making good money and have flexibility. That's exactly when they say, hey, we want to triple your salary. But then in the quiet voice, but you're never going to see your family again. You're going to travel all the time and be stressed. But hey, you were going to triple your salary, right?
That's the second control trap in action. Lifestyle-centric career planning helps protect you against it. I think Chris is going to probably be substantially more happy on average the next 10 years.
with the decision he made than if he had just gone for the bigger job. Because why are you doing the bigger job? Like it's all about what it's serving or like the work itself. So the work itself wasn't going to be more interesting to him and the stress was higher and it was going to get in the way of multiple different goals.
other things that were important. So now he's just tweaking what he has to get a little bit more money and then they're good. So I think that's great. I just want to flag real quick how he talked about the... Processing what enough meant. We've talked about that in a prior show, but it's just worth underscoring. That's this idea of you make sure you know the numbers. Like how much money do we need?
For like this type of lifestyle we're looking for. Having the specific numbers matter as opposed to just more. If you just say I need more money, you might take the triple income job and be miserable. Where he crunched the numbers and said we're close.
We need to just fill in this. Now, let me be really, really careful in figuring out how to fill in that in a way that's not going to get rid of all these other goals. So anyway, it's fantastic. Lifestyle-centric career planning in action. All right, we got a tech corner coming up. But first... Let's hear from another sponsor. We have a new sponsor this week. This is a company I've used for. I know this company well, so I'm happy to have them as a sponsor. That is the online course company.
Udacity. Am I saying that right, Jesse? You are. Udacity. I think I should get a bell for pronouncing it. It's one of these things. I've used Udacity courses for a long time. But you don't have an occasion to say them out loud until you're actually talking. So Udacity... is the site you go to if you want to learn skills that are going to command high salaries. It's the online learning platform that has courses on all sorts of topics, but including AI, data, programming, and more.
attended or purchased multiple Udacity courses. The most recent one I was doing with my son, it was computer game programming. I did another one with my other son on like Tinkercad-based 3D design, like intro to 3D design. It's fantastic. It's a really easy interface. You can purchase the course easily. Some are free. They're video-based. Keeps track of where you are.
Knowledge is power. We talked about this in the very first segment of the show, like how do you get smarter? Online courses are a big way. I mean, you've been messing around with this. You signed up for one, right? Yeah, I signed up for a digital marketing class. Oh, excellent. Yeah. And I'm teaching one now. on Udacity about how to do a killer French accent. So it's, it's, it's been a real hit.
You have your pipe. On a related note, someone just handed me a note and I am no longer welcome in France. Oh, okay. Well, that's good to know. They just sent me a note. So I guess that got me in trouble. There's tons of options for learning. But only Udacity is consistently ranked as the top skill development platform because unlike others, it actually works.
I can attest to this with real-world projects and human experts that grade your work. You'll truly get the skills you need. When my son and I were doing the game programming, I mean, you're building the games. You're doing the work right there.
When you have a certification from Udacity, recruiters and employers take notice. So for a better job, better salary, and better skills, check out Udacity today. I saw a YouTuber, a DIY YouTuber I was watching with my kids the other day, and he said his habit.
He's a brilliant guy. He said his habit is to spend the first 30 minutes every day on an online course. He's like just making myself smarter before I move on to like the rest of my day. I love that idea. So maybe do that. Sign up for Udacity. Always have a course you're working on. First 30 minutes of your day, get smarter, get your brain working. So the tech field has always evolved and you should be too. Always evolving and you should be too. You can try Udacity risk-free for seven days.
Head to udacity.com slash deep and use the code deep to get 40% off your order. Once again, that's udacity.com backslash deep for 40% off and make sure you use my promo code deep so they know I sent you.
I also want to talk about our friends at ZocDoc. I went to the dentist this morning. All right, this true story. I went to the dentist this morning that we are recording this. How did I originally find this dentist? It was like years ago. I was... canvassing friends and someone said oh there's this new dentist in town and i i got in and i liked him and it's been great but the word got out and he's very popular he's now booking
Seven months out. So like if you want to make an appointment with them, it's like seven or eight months before there's an appointment. And so I was thinking today as they were booking my next appointment like seven or eight months out, imagine if I was looking for a new dentist now. What would I do?
Like if I ask my friends and they're like, oh, here's this one dentist and he doesn't have any openings for months. Like that is what it's like right now. Often trying to find the medical care you're working for. You get like one or two recommendations and especially in a city. When you say, I would like to actually come see a practitioner, maybe at some point during the current presidential administration, they just laugh. Like, oh, you fool. You think the dentist actually can see you?
And then there's like a lot of laughter and you hear them call someone else over and then they start laughing. Uh, and then like the doorbell rings and someone runs in and just punches you in the stomach because they're, they're making fun of you that much. This is where ZocDoc enters the scene. ZocDoc is a free app and website where you can search and compare high quality in-network doctors and click to instantly book an appointment. You can search for here's what I'm looking for and then see.
Which of these take my insurance? Which of these are actually booking appointments, right? This is how you would have found my dentist a few years ago. Oh, he's actually looking to take on new patients. Okay, great. When is their next appointment available? Let me book it right now.
Let me do some of the paperwork online using the ZocDoc interface. Actually, my dentist uses ZocDoc, so it's interesting. It makes it so much easier and so much less stressful and honestly humiliating to try to find because no one's laughing in your face for... daring to ask if they have actual openings. So this is the way that you find medical care. Appointments made through ZocTalk also happen fast, typically within 24 to 72 hours of booking. You can even often score same-day appointments.
So stop putting off those doctor's appointments and go to ZocDoc.com slash deep to find and instantly book a top-rated doctor today. That's Z-O-C-D-O-C dot com slash deep. ZocDoc.com slash deep. All right, Jesse, let's do our final segment. All right, because I can't help myself, we're a tech corner segment. We're going back to AI. There's so much to keep up with with AI right now, and I want you to be up to speed.
All right, here's what I've been rabbit-holing on recently. This idea that these new AI large language models like DeepSeq and the new O family of models from OpenAI can quote-unquote reason. You give them like a math problem or something and they can sort of work through the steps to get to an answer. They're doing much better on like math tests or crossword puzzle tests or something like this, right? Like they are reasoning. I'm really interested in this.
All right, how do these things reason? Well, let's ask the companies. They're giving incredibly vague language, and a lot of journalists are just repeating this language. So like OpenAI, when they talk about their new models, they say the new models, they...
They slow down to think more, and that's how they're reasoning. And this is the way it's being reported in mainstream journalism. We're talking about – there was just an article out in The Times a couple days ago about how do these new models reason. And it's just saying like, yes, the new models, what they do is they slow down. They slow down and they think more. And so they can produce more reasoned answers. What does that mean? Right?
Jesse, when you think about a model slowing down and thinking, what does that mean? It feels like it's an actual person that's like, what, taking a deep breath and I'm going to take my time? What's slowing down, right? So I call the source and... talked about it. Like, Hey, walk me through this tech, point me to the right papers. Here's what's as best we can tell is actually going on. Okay. Uh,
There was this idea that actually precedes ChatGPT, right? This is an idea from the Academic Community Studying Language Models 2022. You get this paper that says if you want to increase the reasoning performance of language models. you should use something called chain of thought prompting, C-O-T prompting. And what they were showing in this paper is in the question you're asking the language model,
You basically tell it, I want you to show your work. Don't just give me the answer. Like, show me your steps. This is what I'm looking for is walk me. I'll give you an example of it. See how I'm solving this problem here, step by step. Give me an answer to this new problem with similar style of step-by-step explanation. And what happened is this would give you better answers.
Because language models, of course, they're trained to expand text in predictable ways. They're trained on taking real text, taking words or sequence of words out of them and having the language models try to fill those in. And the closer they get to real words there, like the more they get – rewarded in their training. So like what was happening is you would give a language model like a logic problem.
Jesse has this many apples, and then someone comes along and takes this many apples, and then the person who's taller gives you back this many apples. How many apples does Jesse have? And what the language models would sometimes do is give an answer that sounds like what an answer to that problem would look like.
Like, yeah, I'm saying Jesse has some number of apples. Like that's a – for a language model's perspective, that looks like a reasonable response to this prompt. It's talking about Jesse. It's talking about apples. There's a number. But that number could be arbitrary.
Chain of thought prompting says, no, no, you have to really ask the language model. Walk me step by step through your reasoning for how you get to the answer. And I'll give you an example of me doing that in my prompt. Now you do something similar. When it walks through its reasoning in the answer, it's more likely to get the correct answer because as it fills out its reasoning, it's actually accessing relevant circuits.
that do different parts of the reasoning. And it's actually more likely, you're more likely to activate like the relevant circuits that can do the different calculations and put them together in the right way. And you're more likely to get the right answer. That's chain of thought prompting.
Okay, so the issue is now we get chatbots and they're getting popular. We really can't rely on prompt engineering as something that like the average user is going to do, right? This is like the big push with consumer-facing. large language model tools is like we can't expect the users to write these incredibly complicated prompts to get their results. Researchers were doing this, but we can't expect the users to do this.
So what seems to have happened is they said, can we basically use reinforcement learning that kind of force one of these models to always act as if it's being prompted to show its work, right? So this is the key, and I'll just be incredibly brief on this, but like the key that got like the original chat GPT so much notice is...
They took the first really large, large language model, which was GPT-3, which could do amazing stuff, but you really had to write incredibly careful prompts to get it. And the big innovation, second innovation OpenAI had, the first being just make the model large.
was this thing called, we think of it as like a reinforcement learning-based fine-tuning, where they took this model that had all this knowledge in it, and if asked correctly, could produce all sorts of stuff, and then they started running it with a bunch of prompts.
And every time it gave a good type of answer, they would zap it with a happy zap. And it was sort of like, ooh, whatever weights we're using here, let's make those stronger. And whenever it gave an answer that they didn't like, they would give it an unhappy bad person zap. And it would be like, ooh, let's kind of...
reduce some of these weights that gave us this answer. And so for like chat GPT, they were saying, we're going to run, we train GPT-3 and we're going to run a bunch of prompts. And when you answer this, like an agent, you're chatting with a human.
We're going to give you a happy zap. And then so it got good at answering questions without you having to write really careful prompts. I think what is happening, according to my source, is that they're just doing this now for chain of thought reasoning. So they take one of these models.
They train it up, the old-fashioned, you know, with all the text on the internet and just like any of these other models. And then as they start running through answers to it, giving it questions, when it shows more of its work, they give it a happy zap.
And when it just gives an answer, they give it a negative zap. So this reinforcement learning fine-tunes this big model like, okay, okay, okay. I get it. You want answers where I explain my work. You could do this for anything, right? This is how they – stop certain types of content from coming out of these language models it's completely trained and then they they give it some questions that might elicit
responses that are like violent or something they don't like and every time it gives a bad answer they zap it and then like over time it's like i'm not going to give those type of answers so this is what i believe the reasoning models have done is they take a standard big train model
And then they do a reinforcement fine-tuning step where they're saying, show your work, show your work, show your work. Give us big, long answers. And when we know this since 2022, pre-ChatGPT, that when the chatbots, when the language models rather have to explain their work. they're more likely to give accurate answers. All right. Now, part of what they're doing here in addition to that is these answers are so long, they're now hiding a bunch of it.
So like what's really happening is the chatbot is giving this like incredibly long explanation for its answer, which helps the accuracy of it. But the explanation is so long that if you're a user who asked a question, you're like, oh my God, I don't want to see all of this. So they're hiding a lot of that. And we know this because if you actually look at the user agreements, what's happening now if you use these reasoning bots is they say, don't just count the tokens that you see.
There's something called hidden tokens that it generated part of its answer that we're hiding from you and the user because you don't need to see all that. But we have to charge you for it. Like, I mean, it produced all of this. So you can kind of quantify that, oh, they are.
generating these really long answers and they're hiding a lot of it so that like what you see is just some of the steps and then you get the final answer so that's what reasoning is it's not thinking slower the reason why it takes longer is that it's just generating a much longer answer. That's what the slower is. You see this when you use ChatGPT. It takes a lot. Each token they produce...
ChatGT still produces token by token. DeepSeek gains efficiency by doing a few words at a time. But you notice how the words come in one by one. It's because for every one of these things they're generating...
You have 500 billion parameter transformer-based model pushing through to generate just that word. So it takes longer to generate these reasoning model answers because what it's really doing is writing like... pages of explanation all the way up until its final answer it just takes a long time to do all that they don't show you all that work but that's what the slowness is is it's just giving really long answers so that's what reasoning is
It is different than when I talked around The New Yorker last year about planning and how language models can't plan. And they said ultimately what language models will need to be flexible planners is planning engines outside of the language model. And you need some sort of coordinator that is going to talk to a language model and talk to a planning engine. The canonical example is like playing chess against an AI bot.
Like ultimately you can't just have a giant neural network that can be hard coded to, to look ahead. a hundred thousand possible moves because that's just too big of a neural network to hard code you you need some sort of like systematic simulator to check those and then maybe like the neural network to look at its quality etc so i still think there's only so much this helps answers It's not the really sophisticated how from 2001 style reasoning is going to require non...
language model components that can actually recurse and do loops and have state and actually explore things more systematically. But that is what I think is going on with reasoning. It's the same language models based on the same type of original training that we've seen recently. 500 billion parameter models plus reinforcement learning that gets it to show its work. And when language models show their work for certain types of problems.
they're more likely to be accurate. So that is what is going on. I can't find this explanation in almost any common press article about this, but I think that is what is really happening. All right, so I can't help myself, Jesse, but there we go. Are you going to write an article about it? I mean...
It's not really like a New Yorker article. I don't know. This is more of like a tech explanation article. You did a chat GPT article, didn't you? I've done some, yeah. It's kind of boring. Yeah. So I'm making you guys listen to this instead. I think it's interesting. I'm just kind of frustrated that none of the people who write the Tech Explainer articles, all the big names, they're all just falling back on this.
It's just taking time to think more. And that's how the reasoning works. It's like, well, let's like actually understand, like, is there something, is there a new architecture? Is there a new technology? Like what's happening? Yeah. I had to read academic papers to try to just get that explanation, but I think that's what's going on.
All right. Well, speaking of slowing down, I think that's all the time we have for today. But we'll be back next week with another episode. And until then, as always, stay deep. Hi, it's Cal here. One more thing before you go. If you like the Deep Questions podcast, you will love my email newsletter, which you can sign up for at calnewport.com. Each week I send out a...
new essay about the theory or practice of living deeply. I've been writing this newsletter since 2007 and over 70,000 subscribers get it sent to their inboxes each week. So if you are serious about resisting the forces of distraction and shallowness that afflict our world, you got to sign up for my newsletter at calnewport.com and get some deep wisdom delivered to your inbox each week.