Ep. 359: Should We Fear Cognitive Debt?

I'm Kyle Newport, and this is Deep Questions, the show about cultivating a deep life in a distracted world. So I'm here in my Deep Work HQ, joined as always by my producer, Jesse. Jesse, I have a question for you. Have you at any point ever used AI to help you while you were writing something? No, a lot of people have been asking you this, though. Have been asking me or been asking you? You. Yeah. Through email. This, I think, is the point. Most people right now.

know about tools like ChatGPT, but are not actively using them for things like to help their writing. But more and more people are. It's this sort of growing movement.

So this is what we're going to get into into our deep dive today. There's a blockbuster new article that came out of the MIT Media Lab that did some really interesting experiments to figure out what happens when you're writing with AI. What is the impact on deep work? So that's coming up. I can have a special surprise guest join me to help me dissect that article and understand this relationship between deep work and AI. So that should be fun, Jesse. Then we've got questions.

including a really good case study where you're going to get a sort of bullet point description of the systems for time management and attention management that this particular individual do. And I think it's a really smart system. for those looking for the deep life so that you'll like it. And then our final segment, I'm going to do a twist on my normal what I'm reading segment. I'm adding a twist to it, what I'm not reading. This is just an excuse, Jesse, for me to rant about something that...

has been bothering me recently. I will preview more importantly, we have a little bit more Cal Network art in that final segment. This time, a completely undoctored photo. So we should look forward to that. If you stick around for anything, it is to see the next beautiful piece of Cal Network art. So we've got a lot to do. So I think we should just get rolling.

and start with our deep dive. So last month, Ezra Klein went on the How I Write podcast. The host, David Perel, asked Ezra about using AI for his writing, and Ezra's answer... generated some controversy. It began when Ezra said the following, and I quote, I think it is very dangerous to use chat GPT in any serious way for writing.

Ezra then goes on to give some reasons for this claim. One of the reasons is that AI can help you rewrite or polish or check what you've written, but it can't tell you if the ideas itself are good or not. As Ezra elaborates... You have to be attuned to that voice in you. It's like, not right, not right, not right, not right. You're not trying to bypass that or get around it or get to where it's soft. You're trying to get to the point where you're like, ah, got it right.

and that's usually intellectual labor. Ezra later adds, ChatGPT can't identify fundamentally wrong ideas. When later asked specifically about AI and journalism, Ezra says, I'm not completely against anything, and I have not, and not for lack of trying, and I think not for lack of being informed or interested in the issue, found a way that I consistently use AI in my work. I'll sometimes use it right now as a replacement for Google searches.

And it's valuable for that. A lot of people appreciated Ezra's points, but some people, including several I know and who wrote into the show. thought that his stance was nostalgic and out of touch, like sticking with a typewriter in the age of word processors. A few cranky people might do it, but it's basically just a worse process that makes you slower at your craft.

The future of writing, they're convinced, must include a symbiotic relationship with AI. So who is right on this point? Ezra or AI's defenders? Well, this question has been on my mind.

And that's when I saw that my good friend and longtime friend of the show, the author Brad Stolberg, sent me an Instagram post, which he had recently published, about a new paper that came out of MIT that took a closer look at the impact of using AI. in writing this post hit a nerve online and it caught my attention to this paper i would argue points us towards a stronger broader argument about ai writing in our culture that shows ezra is perhaps on to something anyway here's the good news

In a coincidence of timing, Brad is here in D.C. visiting his family. So I asked him if he would swing by the HQ to talk with me about this paper and get into it on this topic for our deep dive today. And fortunately, he agreed. So we are joined today by Brad. Brad, thanks for coming. Hey, it's great to be here, Cal. So tell me about this paper that you wrote about. What is it? Who wrote it? What's the big points?

The paper is called Your Brain on Chat GPT, Accumulation of Cognitive Debt When Using an AI Assistant for an Essay Writing Task. The paper came out of MIT. The highest level summary is for four months, a team of researchers from the MIT Media Lab followed 54 study participants who were using ChatGPT. to heavily assist them in writing essays. And the top line finding, and I'm going to quote the researchers directly.

is that LLM users consistently underperformed at neural, linguistic, and behavioral levels. So those are the three things. Those are the three things that they evaluated. On the neural level... EEG brain studies found that there was a 47% reduction in brain activity with the heavy chat GPT users. So their brains were using 47%.

less neural connections as they were quote-unquote writing. I guess in this case, they were getting a lot of help from the model. Behaviorally, researchers found that 83% of the heavy model users couldn't quote, anything from what they'd just written. This compares to about 10% from the group that used no technology and very similarly about 13% from the group that used a simple Google search to aid them in their writing.

And then on the linguistic level, they had neutral arbiters give subjective, qualitative analyses of the writing in words that came up repeatedly to describe the AI writing. was, and I'm quoting, soulless, empty, lacking individuality, typical. Now the last thing that I'll say is perhaps the biggest finding from this study, and it's right there in the title, this term cognitive debt.

And what the researchers found is that, yes, using ChatGPT to write for you is absolutely more efficient than writing yourself. You don't have any cognitive strain. It's not hard. You just put the prompt in, you massage it a little bit, and it's done. the time required and they finished earlier of course they finished earlier because they're just prompting the model so yes you have this short-term uh benefit if you will of being much more efficient but

What you pay in the long term to get that benefit seems like tenfold the cost in decreased cognitive fitness and cognitive health, and that's the term cognitive debt. They essentially argue that when we use AI to write for us, we... pick up a fair amount of cognitive debt. We are mortgaging our future cognitive fitness, really our ability to think for ourselves for this sort of short-term efficiency.

I mean this is interesting. It reminds me of the article I told you about that I wrote last year for The New Yorker. And it was about students using AI for writing. And I looked virtually over the shoulder of a graduate student writing a paper. And that's what surprised me, and this was one of the points of that article, is it wasn't about speed for the student. It wasn't about producing text they otherwise couldn't. Like, oh, this is somehow going to produce writing I'm not ready to produce.

It mainly seemed to be about strain reduction. That's what really seemed to be going on is that writing is a pain, quite literally, like cognitively. It doesn't feel good to write. And they would go back and forth with the what about this? What about that? And it gave them all these moments of release from strain or sometimes it would take them a long time to get a sentence by just asking the model again and again. But it was better than.

from a strain perspective, the de novo construction, right? It's, it's easier to ask questions of a model than to produce stuff from scratch with the brain. But okay, but this is interesting. This cognitive debt idea, it's arguing. There is this huge benefit that comes from writing, and yet we often think about productivity enhancing tools and making things easier. How do you understand? What is the dividing line between activity where you want it to be hard versus?

The hardness is a bug, not a feature. I think that's at the core of this argument. I think you're spot on. The metaphor that I used in the little mini essay that I wrote for Instagram.

is this notion of fitness so if you think about physical fitness there's a whole manner of reasons that one would train their body but let's just focus on physical health we know that lifting weights running pushing yourself physically is good for your health. You live longer, it is associated with reduced risk of certain diseases, both physical and cognitive, and so on. Now, if you went to the gym,

with a forklift or some other apparatus to lift the weight for you, you could spend the same hour in the gym, but it would be much more comfortable. You wouldn't be straining. And that gym session would be a lot easier. but you wouldn't accrue those health benefits. And I think essentially what's happening when we over-rely on a tool like ChatGPT to write for us is we're sacrificing our cognitive fitness. We are replacing the strain of thinking.

of writing with a forklift. In this term, we're replacing it with a forklift, and in this case, it's the model that does the work for us. And I think the result is a decline in cognitive health and cognitive fitness. And that's precisely what this paper showed. Another point I'll make real quick, Cal, before turning it back to you, is the researchers also found that when they took chat GPT away.

from the group that had been using it for four months, they wrote worse than people who had never used ChatGPT before. This is scary. Exactly. That's interesting. So you're losing familiarity with the mental processes.

Yeah, required to produce thought. I mean, something that comes to mind here is when I wrote that article for The New Yorker, I went deep on what happens in your brain when you write. And it is really complicated because writing is unnatural. We're not evolved to write. Linguistic communication is a cultural invention, not an evolutionary invention. So in order to write, we have to hijack a lot of different parts of our brain and get them to work together.

There's like visual cortexes involved. There's spatial reasoning involved. I didn't learn this until I read this, but there's a spatial reasoning component, the same part of your brain you use to keep a map in your head. Like, how do I get from here back to the watering hole?

That gets used to spatially position the outline of your arguments in your heads. You're trying to understand where you are in an abstract argument. The visual cortex brings up images. The linguistic centers bring up like words. The auditory centers like convert those words. the sounds, you can think about how to put them into words. I think I used the term as like a symphony being conducted in your brain. This is probably all very beneficial. We know this in the context of reading.

So there's research on what reading does for us, and writing is just reading in reverse. You're producing the words instead of reading it. But we know reading, the research from Marianne Wolfe, among others, argues... We develop these deep processes, deep reading processes, which creates stronger connections between all these brain areas. Once they've been formed.

then you can access them more easily. Now you can access and string together different parts of your brain and it allows for more advanced reasoning, more sophisticated reasoning, more empathy, et cetera. So I guess this is what they're seeing probably when EEGs.

activity goes down using chat gpt is you're probably taking whole parts of your brain offline that you otherwise would be using during the writing process and so those connections aren't getting strengthened so like the equivalent of the muscle getting stronger is probably literally neurons

wiring together because they were firing together. And so if we think about it like cognitive weightlifting, yeah, I think you're absolutely right. Like those muscles aren't getting worked. But what is the – that's what I was trying to figure out. What's the context where we want to get stronger?

Because we do use forklifts to follow the physical analogy. Like if I'm in a warehouse, a forklift is great because we used to have to use the longshoreman. would have to pick things up and hurt their backs or use pulleys and forklifts are better. We can move things onto it. But in the gym, I want to do the work. So how do we think about this? Like, how do you think about it in intellectual work?

I think it's a trade-off between what capacities do we want to maintain and strengthen versus what capacities are we okay letting go.

By no means am I personally an anti-AI purist. I think there are so many good use cases. And I do think it's true that whenever there's a new technology, there tends to be this kind of hysteria or moral panic. It happened with the fire. It happened with the wheel. The Industrial Revolution was so beneficial for so many different reasons. However, there's tons of public health data that shows that the invention of the vacuum cleaner, for example.

helped people to clean a lot faster. And as a result, you are a lot more sedentary because you're not having to crouch down and stand back up and cover your whole house on your hands and knees to clean. Is it good that we have vacuum cleaners? Yes. But if we don't replace that...

then you have these negative health consequences. You have obesity if you're just sedentary all the time. And I think that when it comes to something like writing, which really is maybe the best proxy that we have for thinking. If you want to think well, then it helps to write well. I don't think that we should be so flippant with outsourcing our ability to think. Our ability to think is a central feature of our humanity. In many ways, it separates us from other species.

we can very quickly become like Pavlov's dog putting something into the AI. It responds. We either like it or we don't. We give it a thumbs up or a thumbs down. We ask it the next question, and then we're sacrificing so much of our humanity. Now, if AI was going to be able to, let's say, run a gazillion different permutations to somehow get to the bottom of cancer, well, then the cost-benefit is pretty clear. Let it do that. Let it use all of its computing power to figure that out.

But in the case of writing, we're essentially letting AI think for us. And I just don't think that that's a smart thing to outsource. I like this idea, by the way, of a moral panic around the invention of fire. I like to think of like a Stone Age John Hyde.

When these kids are looking at this fire, they're not out there learning how to mammoth hunt. And do we really want kids who can't mammoth hunt? Come on, fire. Fire is a problem. So then what if we say – I'll throw out a proposal. You tell me what you think. School. Don't use LLMs. It defeats the whole purpose. School is the closest thing we have to an intellectual gym. So why would we put a fort cliff or echo skeleton in the gym? That's.

The whole point of writing in school, it's not getting us to something else that we're trying to get to and making that more efficient would be good. The writing itself would be the case. I'm going to say school. Don't use LLMs to write at all. I would say maybe if you're a professional writer too. This seems to be Ezra's argument. It's like I do this for a living. If you listen to this, I'm trying to be really, really good at this.

parts of my thinking my thinking is what i'm paid for like that's the whole ball game this doesn't make sense to me like why would i want to outsource any all i do is write these scripts for my episodes that's my whole job

To make that a little bit faster, like what am I trying to do here? This is not – it doesn't get me to something else faster. So, okay, do you agree with those or am I missing something? School, professional writing, you shouldn't be using LLMs in the actual – like crafting a prose process i don't know am i am i getting this right i'm sort of doing this on the fly here i'm doing it on the fly with you i do think you're getting it right i think that you you would add maybe composing music

or if you're going to be writing emails or notes to family members that feel important, essentially, if you step back and you look at this maybe more philosophically or spiritually,

If there are activities that require strain in the short term, but in the long term, you derive satisfaction from, then those are activities worth protecting. thought experiment that makes this so simple because it's an extreme is imagine that you could press one button and perhaps we'll get here in a few years, who knows? And you could say, I want a Grammy award-winning folk song.

And you press that button and the song comes out and you spend maybe 30 minutes going back and forth making edits. And then you win the Grammy. Would you be as satisfied with that Grammy than if you worked on that album for two years? in the lion's share, if not all of it came out of your brain? I don't think so. I think that you'd be much more satisfying if you did the work. Now, there could be a world where a part of doing that work is using this as a tool.

But that's very different than relying on it or having it replace the actual act. So I think it's twofold. I think it's exactly what you said, which is you don't want to use it in areas that are core. to how you live your life, how you make your living. If you get paid to think or you're a knowledge worker, you want to be really careful about protecting your ability to think. And I think we just need to be careful of this.

you know, this is a tale as old as time, this tradeoff between short-term convenience and long-term satisfaction. And I personally am hesitant to outsource things that are hard in the moment, but give me satisfaction, make me content, make me satiated.

allow me to feel like I'm falling asleep after a good day's work because I put in the effort. I don't want to click two buttons and have that done for me. I have a term I just invented. You're so good at naming things. All right, what do you think about this? It's an adjustment of existing term, type two thinking.

Ooh. Right? So type two fun is this well-known term for activities that aren't fun in the moment while you're doing them, like hard mountaineering or something like this or a really hard workout. But it's very satisfying and fun when you're done. You're glad that you've done them. We should have the same thing for thinking. Type two thinking is difficult in the moment. The blank page is difficult in the moment. But having composed something that you're proud of is like deeply satisfying.

And long term, you know, it's going to make you better and smarter. And so you think about type two thinking like type two fun, like, yeah, the hardness is the feature here. Maybe that'll catch on. I don't know. I love that. And I also think that there's there's a chance that in the future we'll have. essentially cognitive gyms where like you set aside time and you say during this hour and a half, I'm not going to use the model at all.

I'm going to walk on a treadmill instead of drive the car. I do have one question for you, and this is going to be perhaps some pushback from your listeners. I know that people who read my work are probably thinking this. I think the pushback would be this is a tool.

And it's all about how you use it. And what if you get the type too hard by getting better at using the tool? So why would you spend two hours writing if you could sit there and prompt the tool back and forth? And you're still going to have that cognitive strain. It's still going to be really effortful. And it's going to be, as you said in the opening, a symbiotic relationship where the skill that you're developing or the competency that you're developing.

is prompting and using the tool i mean i think it's a much more probably the responsibility is a much more narrow cognitive activity right like there is There's some sophistication. You're like, look, I'm in the gym and I have the animatronic that's lifting the weights for me. There's a lot of – I have to control it and make a lot of decisions and move this lever and that lever. It's not easy, right, to get it. It's a different type of hard, but it's not.

The type of hard that you're hoping to get out of the gym, which was getting stronger. So prompting. carefully to try to work back and forth to get out stuff you can use is probably a much more narrower band of your mind you're using it's a very sort of like tool functional type we're already good at that by the way like we're

We are very good at like using these type of consumer-facing tools, which are made to be really easy. That's not necessarily something we have to work really hard at, whereas the connections we get in writing.

Those are hard won. They make you smarter. They make you more empathetic. You're able to better understand the world and manipulate ideas, which I think is more of a superpower. I mean I don't – I'll ask you where you draw – you would draw your line. I think I'm probably similar to Ezra here.

I think these tools are fine as a better Google search. We're kind of already used to doing Google searches. Sometimes it's like a Google search that will do some summaries for you. I have to be a little bit careful about it because it hallucinates a lot, but that can be useful. Sometimes if I'm writing an article on deadline, I'll do some copy editing with it. Not that I can't copy edit myself, but that – if I'm in a hurry, I'll sometimes be like, hey, I'll let –

LLM help me copy edit because if this is cleaner grammatically, it's just going to make the next editing pass faster. Like I wouldn't necessarily bother with a book because we have time. But if it's on a tight deadline for an article, I might do that. I'm wary about deep research unless it's really, hey, here's some quantitative data. I kind of know where it is. Will you O2 or something put this into?

spreadsheet or into a format like actually do data manipulation in a way that makes it easier to work with like that makes sense to me I'm a little bit more wary of like hey go summarize everything on this topic because I just don't think it does a good job. And also that's part of the work of journalism is knowing like your sources and following those trails. I think it doesn't do a great job. I don't know. That's where I am. But maybe we're being arbitrary here.

But I draw the line where it's like the actual construction of the ideas or prose, like I'm filling the blank page. When you have your sources, I think I agree with Ezra there, you grapple. You make all these connections. You write. It goes slow. 20 minutes in, it starts to pick up speed. That's where a lot of interesting stuff happens. But your cognitive – let me go back to your cognitive gem idea. In some sense, that's what we want to avoid.

Like what we want to avoid is getting in a world where – I mean it's exactly what happened, as you said, in physical health. We became much more sedentary and then had to invent activities just to try to get some exercise back. We could do this intellectually.

We kind of outsource all of our thinking. Our brains are much worse. And now there's like a small number of us that have to like go to cognitive gyms to like try to keep the brain up. Or we could just not outsource the intellectual activities because having my student paper written faster.

It doesn't make my, it doesn't make open up new opportunities or like open up new economic opportunities or growth. It's just sort of convenience in the moment. I want to, it would be nice to avoid the cognitive gyms. And so thinking about type two thinking, maybe that's the way to do it. Unsurprisingly, I'm going to completely agree with you here. We can continue to build this metaphor out in a way that I have found really helpful.

in relation to health and fitness. So we talked about the production of writing or thinking, and we made the parallel to lifting weights. And sometimes you do want to outsource lifting the heavy thing at the warehouse. But if the goal is to maintain strength, in this case, if the goal is to maintain the ability to think for ourselves, we certainly don't want to outsource that. I would say at all, but we don't want to become reliant on it.

The metaphor extends to what we consume. So at the same time, we underwent the Industrial Revolution and we became more sedentary. There was also this massive change in food technology and the creation of ultra processed or highly processed foods. And highly processed foods are easy. They're convenient. They taste great. They have a kind of.

addictive quality about them. The more that you eat them, the harder it is to enjoy more nourishing foods. And what happens when you eat too much ultra processed foods. is you have all sorts of metabolic issues and dysfunction, and you tend to struggle with overweight and obesity. And we've got a built environment around us where not only are we sedentary, but there's ultra-processed foods everywhere.

And we spent a lot of time talking about the production of writing, but I also think there's something to say for the consumption. And if you are consuming a bunch of short form AI generated content on TikTok, on Instagram, on Twitter. Wherever it is, the same type of content that these researchers coined as soulless, empty, lacking individuality slop.

Well, then not only are you not exerting energy to build cognitive fitness, but you're also consuming all of this highly ultra processed information that has a very synthetic feel to it. And I think. That is my biggest fear is that we enter a mental built environment that is similar to the physical built environment where we're not exerting ourselves and we're consuming ultra process content. I mean, this comes up in the reading research that when you're.

reading like on a screen, lightweight material that sort of just immediately is interesting, but it's not challenging. it doesn't develop what they call deep reading processes. So there actually is a distinction. It's not all reading is reading. It doesn't matter whether it is on your phone or if it's in a classic book.

The harder the thing you're reading, the more your brain has to work. And the real intellectual advantages, the reading research says, and I think Proust and the Squid is a great book on this if you want one, says it's really when you're challenged. You're grappling with the ideas or the complicated.

character tree or complicated structure to what's going on, that's where the brain gets stronger. So, yeah, I agree with this a lot. I mean, my prescription, like if I was going to be follow this analogy through to. It would be you need to read hard things, you need to write, and you need to consume the sort of ultra-processed content sparingly. You have some things you pursue. Go on the smart end of it.

have like Brad's Instagram feed, which is all textual by the way that he writes. It's like the thing we're talking about today. Like, sure. You have some of those. Or if there's like a particular – I always talk about like Instagram is good if you need motivation for a particular pursuit you do. Like some people are like, look, I'm a triathlete. It helps for me to see videos from other triathletes. It gives me like –

Hell yeah, that's right. I need to keep doing this. This is your digital minimalism to a T, right? It's set aside specific time, call the people that you follow, use it for a particular purpose. Don't just go in there and go with the default flow because the platform is going to own you. You're not going to own it.

So one of the things I discovered when I wrote something about this in the new book I'm working on, and I was looking at the history of how we came to understand physical obesity as an issue, and it's interesting. It took a long time. And so I'm wondering what – I don't think we have our arms fully around what the equivalent cognitive obesity is really going to be. So we don't always see the problem when it's developing.

Now we see the problem in the physical world. Oh, okay, obesity, we can measure. We have a very specific BMI type of way of talking about it. We know exactly what the chronic health conditions it really makes worse, and that's understood now. It wasn't understood in 1982. When the graph started, the hockey stick up. The calories in went up. Sedentariness went up as well. We're in like those early stages of cognitive obesity, and I don't think we know exactly what it's going to look like.

But I don't know, I'll tell you, man, when I'm on like an airplane and I see the 21-year-old in the seat across and it's the TikTok is scrolling. And you look, I don't know if you've seen this, you're not a TikTok guy, but three seconds swipe. One second swipe, like six seconds swipe. Like, what is this doing to the brain? I don't know. I have a funny TikTok story as an aside. It's actually not funny. It's terrifying.

But maybe two years ago, I was told that TikTok is going to be the next thing for authors like us because there's this book talk. Yeah. And I got a TikTok account and I was on TikTok for a day. And the only way that I can describe that day is it felt like if I had eaten a hundred bags of Skittles and then at the end of eating a hundred bags of Skittles, my whole body just felt like synthetic yuck.

And I remember telling the marketing person that told me I should be on TikTok that I just can't do it. I'm happy to forfeit whatever fame or book publicity I'm going to get. This is not good for my brain. And it's precisely for. for that reason. I think another piece of prescriptive advice that I would have is it's just the importance of being aware of the trade-offs when you're producing or consuming this highly processed information.

and then having some rules and constraints, as you mentioned, Cal, about how you use it. The metaphor, or to continue the analogy here, it's very hard to just merely exist in the modern food environment and be healthy. A huge part of the obesity problem, in my opinion, is a lack of education on a carbohydrate, a protein, a fat, added sugars. How many calories do you actually need?

People that are able to maintain decent health in a terrible food environment, nobody just goes with the flow. Because if you just went with the flow, you would eat fast food because it's easy and it's on every corner and it tastes great. You have to have some level of knowledge.

And then with that knowledge, you have to have some sort of constraints and some sort of rules. People can take this into overdrive with cockamamie diets and all kinds of restriction, but you have to at least think and be intentional and deliberate about what you eat.

when you eat it, and why you're eating it. And I think at the very least, we should approach our information environment similarly because there's going to be fast food restaurants popping up on every corner in the world of thinking. You're saying people who are healthy. Often there's – it's not like it's hardship, but there's hard rules they follow and there's points of friction all the time.

Exactly. I'm not going to have this. I'm at this. I'm not at the poker night. I'm not having the whatever. Like it doesn't dominate their life, but it's a constant feature that like there's. I'm aware of what I'm doing. So if you don't have something like that in your digital life, you have to be worried that how do I know I'm not going to end up with cognitively obese? If it's like, hey, just go with the flow. Like this is interesting. Why not use that? This seems to be making things easier.

Why not use that as well? Like if you can't point towards here is where I draw the line, here is where people look at me weird, then that's probably... That's probably a problem. It's like I always tell my wife when we talk about financial budgeting. I say if you haven't said no to spending money, if we haven't said no recently, we don't have a budget.

Right, because a budget doesn't exist unless it's actually inducing constraints that change behavior. Like we would have normally spent money on this, but we didn't. If that's not happening, you can say you have a budget, but you don't really have a budget. So yeah, maybe there's a cognitive analogy there. All right, final thing.

How do we think about writing where we don't think – it's writing we do where AI could, I guess, help. I'm thinking like emails, right, like business logistical writing, where we don't think about that writing as being – key to cognitive development, right? It's not like writing a paper in school or an article as a professional writer. This is something I've been grappling with a few thoughts on AI in professional writing.

where it is the writing is fully functional. It's just, look, I need to coordinate this team to get a meeting on the books. I need to get a summary of these notes to this person so that they can like integrate them. into the slide deck or something like that. That's the other major use case of AI and writing right now. I think I have some thoughts. Have you thought about this at all? About where we don't think the writing is cognitively important.

Should we just be having AI write those emails? How do you think about that? You are more the expert on this than I am. So I'm going to defer to you. But you asked me the question. The only thing that I've thought about. is I do think what will happen is we'll have a world where my AI is talking to your AI. And the AIs are just going back and forth about when to schedule the meeting or any other number of administrative overhead tasks.

So one question that I have is if you're going to have AI write it for you, then do you really need to be writing it in the first place? Because you kind of have to assume if you're going to have AI write it for you. then perhaps the recipient is going to have their AI read it or at the very least have their AI respond. And now you get robots talking to robots. Does there even need to be a human in the process at all? Yeah, I mean, I think I agree with that.

An overwrought AI email. Yeah, what are we doing here? Linguistic communication in the workplace is a... crude solution to a more complicated problem there's usually a coordination problem that needs to be solved this information needs to get to this person a decision needs to be made people need the right information to make that decision

And we happen to have this tool. We have ways of like typing text back to each other that like we'll use this tool as like a. crude proxy for reaching this decision but you're right it seems like if there's a more direct way to do it that's probably what we should do if i'm having ai write like a really complicated email to try to figure out what information you need for a meeting man wouldn't it be easier if my ai could just

go to your AI and get the information that it needs, or they could go back and forth directly. I also think when it comes to things like email communication, there is an art to it. And it's not about making your brain smarter, but I think it's about... I don't know, efficiency at a broader level. Like what is it I write in my book, A World Without Email, I get into this. I mean it's a lot about like what's the end goal here? What is this communication serving? Oh, we want to get this.

We want to get a meeting on the books. And it's actually non-trivial to sit down and be like, well, what's the best way to do this? And let me capture that in this message in a way that kind of gets us there. as effectively and efficiently as possible. And there's a real art to that. And all that art goes out the window when it's AI back and forth, right? The other way I think people are using AI for professional writing is they just don't trust their writing ability.

And that goes back to like probably because you're not writing. So there's a sense of I don't really trust myself to write this in a way that the person on the other end. is going to see as professional or something like that, right? In which case that probably goes back to it. That being said, I think there's a lot of places in professional communication where it just straight up saves time and it's probably fine. Like, oh man, I got to go through these notes.

and summarize the five points from the meeting and this is just drudgery hey i'll do that for you just just like it's just like i'll eat a cliff bar or on the way home from my son's baseball game if it's eight o'clock and we haven't eaten We'll stop at McDonald's or Burger King. Now, if I built a diet on McDonald's or Burger King, I would be very unhealthy. I wouldn't feel good. I'd lose physical health and physical fitness.

But if I'm intentional and deliberate and there are specific times when it just makes sense, I think that's fine. I think the analogy holds. You look at nutrition purists who are like, I shall never do this. And they have all these rigid rules. Those rigid rules end up consuming them.

So I'll come back to that and say by no means in the little piece that I wrote am I saying that we should shun AI. I just think that we should be really deliberate and intentional about it because we weren't with smartphones and the adoption of social media.

And it's becoming pretty clear that there are a lot of detrimental negative impacts. Most people are probably listening to this podcast on a phone, which is also proof that there are so many great things about this modern technology. But we went into it. Just kind of arms open, let's see what happens. And I think with AI and AI writing and AI consumption or consumption of AI written material.

What I'm arguing for, I think what you're arguing for, is let's learn from our mistakes with the adoption of social media and smartphones, and let's be really deliberate and intentional about how we use this technology instead of just going wherever the current takes us.

All right, so it sounds like we're kind of on Ezra's page here. Then I'll summarize then the advice that seems like we're saying, or at least I'm saying. If you're a student, don't use AI to help your writing. The whole point there is to make you smarter, make you a better thinker.

Writing is an incredible tool. It's the pull-up of the cognitive world. Don't use AI. If you're a professional writer, that's your whole job is to have a nuanced, subtle grasp of this. It's like an athlete needs to actually do a lot of training.

Don't use AI to do your writing. If you're in a professional circumstance and there is a sort of tedious – work to be done of i have information in one form i need to move it into another or i have a long cc email chain where people are talking about their availabilities and i want to suggest three times that works for everyone fine that's maybe a good use of ai because that drudgery is not helping you

It's annoying that we have to do this via email in the future. We wish like my AI could just schedule with your AI, but in the short term, fine, use it if it's going to automate that. The final thing I would add, if you're in a professional context and...

You're nervous about your writing quality. Maybe you're sending out an important email. You're like, look, AI can help me. You can check the tone. Grammarly does this well. It'll actually say like, here's places you could make this sound more professional.

I would say in the short term, if you're worried, hey, that could be helpful. Long term, that should also be a signal. I should actually deliberately work on my writing in lower stakes situation till I get to the point where I'm not so nervous about it. I think it's sort of like.

You're you go to the national park and you're like, I don't know that I can actually make these trails. I'm going to like drive a scooter around. The scooter will get you to the end of that trail or whatever in that context. But you should come away from that saying, like, maybe I should get in better shape. So like next time I go to a park.

I don't have to take the scooter around. So all right, those are my four pieces of advice. So Ezra, I think we're on your page here. This is not using a typewriter in the age of word processors. This is someone who uses their brain for a living. worried about their brain, wanting their brain to be as strong as possible. So I would say not nostalgic, not moral panic, but actually pretty reasonable. Am I hitting it? What am I missing?

I think you're nailing it. The only thing that I would add is in that fourth example, if you're someone who's worried that your writing could be better and you're using the tool to help. I wouldn't do it on autopilot. I would pay attention to the edits that it gives you. If you think it sounds better, I would stop and think, strain a little bit. Why does this sound better? How did it change it in a way that is better?

And now next time, let me try to get closer to that on my own. Maybe even say, tell me if this is professional or not or like rank it on, you know, but don't give me a better version. Let me now try a different version and see if I can get that. that ranking up i mean do that with caution because look it's not an actual person and it is very confident in what it tells you it doesn't mean that it's right but yeah you could even use it as a training mechanism that's fine but

Going back and forth about your sentences. Yeah, you got to be worried about it. All right. Brad, thanks for coming in. I think we cracked this problem. It's interesting. Thanks for turning my attention to this paper. This is obviously a much bigger issue that we both write about in our essays and books, and I'm working on this in my new book, and I'm sure we'll touch on it more in the future.

But cognitive fitness, cognitive debt, cognitive obesity, AI, all of these issues are wrapped up together. We'll have a lot more to say about in the future, but at least this gives us an interesting view at what's going on right now. So thank you, Brad. Thanks, Cal. This was a pleasure.

Let's take a quick break to hear from another one of our sponsors. Notion AI is an all-in-one AI powered by your work in all in one place. It automatically captures meeting notes, instantly finds the exact content you need, drafts detailed docs for you, and lets you chat with the best AI models. Notion AI just became...

Twice as powerful for teams, making it the best AI tool for work. A little background here. Jesse and I have long been fans of Notions. We have used various custom tools built on Notion over the years. Clearly, our listeners love it too. The caller we just heard from mentioned, his custom Notion task system, right? So this is something that we've known for a while. But they've added these new features recently.

that I think are really cool. Jesse and I were experimenting with one of these new features, which really caught our attention. They call it enterprise search. But here's the idea. You can use keywords or an open-ended question. For a single powerful search experience across all of your connected tools and information, you're unifying now your scattered knowledge right within your workplace and can get quick AI summary of those results.

You can use something called AI connectors to have the search be able to connect into other work tools you are already using. For example, Jesse and I use Google Workspace for running the podcast here. So with Notion Enterprise Search, we can do a search that might combine information we have stored in Notion with like the episode scripts that we have stored in Google Drive. And they can find, hey, look for this example.

from some episode and summarize what we said, and it can find it and summarize it and connect it to other information. It really does bring the power of AI to your information. and not just some sort of like general chatbot like you might be used to using. Notion uses the top AI models built in, and you can choose which ones you want. GPT-4-1, Cloud 3.7 Sonnet.

Chat with it directly in Notion. No separate subscription needed. You don't have to go over to a separate tab and type into a separate window. Look, we're not the first to discover Notion. It's used by over half of Fortune 500 companies and teams that use Notion. send less email, cancel more meetings, and save time searching for their work. That's all music to my ears, of course. Check out Notion, the best AI tool for work, right now at notion.com slash cal. Now, you need to type that in.

all in lowercase letters, notion.com slash cal to try the powerful all-in-one Notion AI today. When you use our link, you will be supporting our show. So go to notion.com slash.

Cal, I also want to talk about our friends at ShipStation. If you ask Jesse where all of these packages he orders from e-commerce stores come from, he will say, and this is absolutely true, Jesse, you can back me up on this, a stork delivers them. He told me this just earlier today. He thinks that's how they come. But for those of you who actually run an e-commerce store, you know packages don't arrive by magic for a lot of companies. It is a huge source of labor and stress.

This is where ShipStation can make your life so much easier. Last year alone, over 700 million orders were fulfilled with ShipStation, half of which were... Creatine shipments going to Jesse. With ShipStation, you can sync orders from everywhere you sell into one dashboard. This is key. You have one dashboard. And you can replace manual tasks with custom automations to reduce shipping errors. And all of this is at the fraction of the cost you probably were already spending for fulfillment.

Here's a couple things about ShipStation that caught my attention. It's the fastest, most affordable way to ship products to your customers because it has discounts up to 88% off UPS. DHL Express and USPS rates and up to 90% off FedEx rates. It also seamlessly integrates with the services and selling channels you are already using. So you can manage all of your orders over multiple different channels.

and systems in one easy-to-see dashboard. You want to hear something cool? During the time I've been reading this spot, 1,400 packages were shipped with the help of ShipStation, and none of them, Jesse, were delivered by a store. Come on. Upgrade to a smoother shipping experience. Go to ShipStation.com slash deep to sign up for your free trial.

No credit card or contract is required, and you can cancel any time. That's ShipStation.com slash deep. All right, let's get back to our show. All right, first question is from Antonio.

Can AI models be creative? My naive take on this is that AI models can explore very well the space that is well described by the training data, but they are not very well suited to explore beyond the training data. This capability of going beyond the training data is what I would call creativity. Well, Antonio, this is really one of the key questions or key goals in AI research.

Some of the terminology used for this is called generalization beyond the distribution. So it's sort of a holy grail of sort for artificially intelligent systems. The idea would be that you... train a system what they would hope to happen is you would train a system on some sort of data and in doing so they would learn generalizations about how the various things that data work interact with each other

And then using those generalizations, you can produce new understanding and new ideas and new output that... doesn't directly reflect the type of things you saw in the training data. So humans do this. We're pretty good at generalization. So maybe we learn looking at a particular example how gravity works.

And then we can understand that same rules of gravity are going to apply with these things we've never seen before in a new context, right? So we can generalize and apply that generalization. This is a big debate right now. How much generalization is happening in AI models based off of large language models? The sense I get from the research community seems to be not as much as researchers would hope.

Some of this comes from just AI researchers I've talked to for recent articles. They would like there to be more generalization, but it takes a huge amount of data and they don't know how much this is happening. There's an important new paper that came out a couple of weeks ago. Apple put this out.

where it was looking at the reasoning capabilities of cutting edge large language models. And they were basically arguing the reasoning, quote unquote, really doesn't generalize. It can't go beyond applying directly.

rules that were inferred during training so it seems like the way that language models are working right now for the most part is if you give it a bunch of examples of stuff during training it can build pattern recognition into a transformer architecture for sort of understanding this is one of these examples.

And they can infer some sort of general rules about how – and I'm using rules here not in the traditional AI sense but in my own informal sense. They can sort of embed in their actual neural network architecture some sort of rules about how these types of examples go forward. So if you give them a novel version of that example, it will recognize it as that and it can sort of apply the rule to get some sort of answer. So for example.

If it's seen a lot of simple arithmetic problems, the pattern recognition piece of the neural networks can recognize, hey, there's an arithmetic addition problem in this prompt I'm seeing. And there could be some then rules embedded informally into the architecture that captures like different numbers being added together. Here's what the result would be. And it can then see your novel problem. And even if it hasn't seen that exact.

math problem it sort of knows um how to do that math like that seems to be what it's doing but you're not going to get like a sort of a generalization about the properties of numbers beyond it so No, language models, at least as we're seeing it, are not scaling into the generalization we would hope. We're kind of running out of data on which to scale them in a massive sense. So to get that type of creativity you're talking about, Antonio, is probably going to require other types.

of systems. Maybe they have language models plus symbolic reasoning and other types of more explicit type of thinking and ontological organization in there. You're not going to get real big generalization from even the very best models, though some people disagree with me on that. But it is possible with AI, just not the systems we have right now. All right, what do we got next, Jesse? Next up is Michelle.

I feel like I'm incredibly disorganized in my work life. What's the smallest possible first step I can take to generate real improvement? All right. That's a good question. A lot of people ask about this. What is step one if I want to get some sort of... minimum effective dose, get some sort of sense of organization out of it. I thought about this a little bit. Here's what I would suggest.

You're completely disorganized at work and at home and you constantly feel like you're behind things and maybe you're in your email inbox a lot or on your text threads like reacting to crises. I think you need some sort of master list. You need some place where you write down everything. that is just in your landscape of things you might have some responsibility for or need to deal with. This is probably step number one. And you have to tend to it, right? So at the end of every day, just dump.

What came up today that I'm just keeping track of in my head? What text message did I get that I'm kind of forgetting about? What is coming through on some emails that now I realize there's a bigger project I need to do? Just get it all written down somewhere. Do it on like a legal pad so you can like carry it with you or maybe like a Google Doc or something like that. That is probably like the minimum thing you can do. But what it does give you is permission.

To not have your brain be constantly trying to refresh and remember everything that could cause a problem. Your brain does not like stress. It does not like that negative subjective experience of. oh man, I forgot this. And now they're calling and saying, where are you?

We don't like that. Your brains don't like that. And if it thinks that this could happen unless it remembers something you're forgetting, it will put in a lot of cycles to try to prevent that negative experience from happening. So if you can reassure your brain. You don't have to keep track of everything anymore. We're not forgetting something. It's all on this list, and I just have it, and I throw stuff on it, and I look at it a few times a day.

And I don't know how I make sense of it, but at least I see like, oh, God, I forgot about that or I should do this or I'm never going to do this. But at least I wrote it down and I'm reading every day. That's going to make a huge psychological difference. And that's the foundation on which real time management can then be built. The hard part is.

How do you figure out what to do and when you're going to do it? How do you organize these into bigger projects and make sure that progress is made? But that's more complicated. The master list is the time management equivalent of saying just get up and make sure you walk 5,000 steps a day when it comes to trying to get yourself back into shape. Just get used to movement from sedentariness.

is a big first step in the physical world. So full capture to use the David Allen term versus haphazardly keeping track of things spread between emails, chats, text, and your brain is like a really big step.

And you're going to feel an urge after a while like, well, I have this all down here. I look at it and I keep it up to date. And it's reasonable. It's not much friction. Like it's not some big lift that I'm going to believe this is impossible. I'm going to give up on. Eventually, your mind is going to start saying, OK, well, maybe we could.

have a couple of rules for how we make sense of this, or maybe we're going to make a list for the day, or maybe we're going to look at our calendar and sort of organically you get this bottom up time management that'll emerge from it. So I think that's the minimum effective dose. It's not going to give you a fully.

It's not going to give you like a fully organized life, but it is going to save you from the worst of the stresses that can be caused by disorganization. I'm thinking about, Jesse, a podcast topic. We might actually do this next week. I'm not quite sure. But I'm thinking about like what is the minimum time management organization system that can kind of keep your head above water for a while?

Because I'm interested in this question of like in the summer, for example, if you're a very organized person, it takes a lot of cognitive load and effort to maintain a intricate time management system.

So what is like the minimum effective dose system that you can downgrade to? Something that if you tried to do it for five years, like you're going to start missing stuff and it's not sustainable, but you could do a couple months in the summer and not come out of the summer feeling like you're really behind and stressed out.

So I'm thinking about that now. I have some ideas for it. So stay tuned. I might do that for next week's podcast episode. Is this something you're going to try or? Maybe. Yeah. Yeah. Yeah, maybe. Maybe that's what I'll do. Maybe I'll try it. And by try it, I mean I'll make you try it. And I'll warn you, it does involve semi-regular electric shocks. But this way, like, you don't forget. It's all Pavlovian, guys. If you get shocked...

Every time you forget an appointment, you're not going to forget those appointments anymore. All right, who do we got next? Next up is Duncan. I have lots of projects I would love to tackle in my life outside of work, but struggle to ever find enough time. What can I do to fix this?

Okay. I'll tell you, Duncan, I'm going to connect this to an unexpected source maybe. But it makes sense for me because I just got back from Disney. And I've read a lot of things about Disney.

But because I was at Disneyland, I'm rereading this book I first read five years ago about the history of Disneyland itself, okay? Interesting analogy from that book or interesting case study I guess, right? It's the 1940s.

Disney is like a big concern at this point. They've had a bunch of hit movies. They have been – were really involved in the war effort. It's a big, biggish company. Walt Disney is pretty famous. Also like a little bit – stressed out and dealing with some depression, he begins to put a lot of time into the development of a scale size train for himself.

For himself. So he buys a new house that has five acres of land for this train idea. Then he has a custom-built locomotive, actual steam locomotive. but it's scaled down. I don't know the scale specific numbers, but it was just big enough that you could sit on it and like kind of have your feet up front. So think about like a...

Hot wheels – not hot wheels, power wheels. What's the – the plastic cars kids can drive? Like that's sort of like that size, right? And he had this thing built and he built rolling stock and then they built this track through the property. that he bought. I think this was in like Beverly Hills or something like that. And it was elaborate, right? They had a train bridge on it with trussles.

trestles that was so big it was big enough that it fell under like the jurisdiction of the county for you actually have to have it inspected like any other bridge you would have they he built a tunnel that did an s turn So that when you go into the tunnel, so it's a big enough tunnel that like people, grownups riding on a train can go.

When you go into the tunnel, you can't see the light on the other side because it goes through a turn to get to it so that you have the effect of like I don't know how long this tunnel is going to go. So they're like buried over, completely landscaped this whole thing. I forgot the exact total length of the track, but maybe it was something like a quarter of eight, half a mile or something like that. Anyways, here's what hit me about this. That's an incredible amount of time.

This is a hobby, so to speak. And it was like an incredible amount of time and treasure to build this thing. And he is a very busy person. He's the CEO of this like large movie company that was doing lots of different projects and there's lots of money things moving back and forth. So I was wondering, like, how did Disney have time?

To build like such a big project. Now, part of the answer is, well, if you own the company, you know, you have more flexibility. That's true. Part of the answer is also in a pre-digital world. You had a lot more flexibility. You didn't have to be on demand in the same way because you couldn't be on demand in the same way. There was no email, right? There were no Slack messages.

People were used to more autonomy just because of actual obstacles to access. Every time Disney needed to go talk to financiers in New York, that's three days on a train. And then three days back from the train, when he goes on vacation to Europe to rest, that's going to be six weeks.

It just takes that long to get across the country, to get on a boat, to take the boat across. So we're more used to it. It wasn't as fast-paced as things were back then. But still, he was a busy guy, and he found time to do this massive project over about a two-year period. So part of what's going on here, I think, is... The mindset. So mindset is, hey, this thing is really important to me. You find time.

You find time. There's a lot of gives in the schedules. You come home early. You're working on this at night. You kind of build a lunch break around doing it, right? When something becomes important to you, if you could find something that sparks you in the right way, you'll be surprised by the time you find. Now, we've seen this effect a few weeks ago. We did a podcast on data out about the four-day workweek.

People had that counterintuitive, unexpected response where they just as in these experiments just said this day of the week is gone, like no big prep work for it. And it was fine. It turned out there was all this give in the schedule. All along, their schedule could have been four days instead of five. They had plenty of time, but we sort of fill in time. We make things take longer. We add more meanings to it. So that would argue the Disney argument here would be.

Find an important enough project that you actually really want to make progress on it. You'll be surprised by how much progress you can make. Here's my second thing I want to say about this. See how much time you get back when you don't use screens.

It's another thing that is using an inordinate amount of people's time and they don't realize it is this sort of default to like, let me put on a streaming service and or look at my phone.

And you don't realize how many hours of your time outside of work are getting eaten up by that. So a good experiment is do something like the digital declutter I talk about in my book, Digital Minimalism, and say for a month I'm not going to use these tools at home. I'm not going to look at my phone for social media.

I'm only allowed to watch a streaming thing if it's with other people. So like family movie night, put those type of rules in place. You're going to be kind of bored at first. Like, wow, I have a lot of time. I woke up. I don't have to leave for another 45 minutes. What am I going to do for those 45 minutes? Right. I'm home. It's five.

I'm not really going to start dinner for another hour. Like, what am I going to do? I'm bored. And you let that boredom drive you towards projects like, oh, I make I get make progress on things. This is, for example, how Walter Isaacson. wrote many of his initial historical bestsellers, his Einstein biography, his Franklin biography. How did he write these? I've heard an interviewer ask when he was also CEO of Time Warner. This is a busy, busy job like Disney.

And his answer was, I don't watch TV at night. He would come home. Instead of watching TV, he's like, I'm going to write. And then you sort of got used to that and you would ride into the night a little bit and you sort of got used to it and just like, hey, stuff gets done. You'd be surprised by how much can be produced. So that's what I'd put together. Make your project exciting enough that you want to fight to get it done.

And then take screens out of your life experimentally. Put those two things together. And I think you might be surprised by how much you get done. What happened, Jesse, with the train to Lily Bell was named with the locomotive. What happened with Disney is the train.

They had a crash and it scalded like a little girl, like not in a bad way, but like there was an injury. And he was like, put it away. We're done. I'm done with the train. They just put it away. And he went on because that was that gave him the inspiration he needed for.

uh, Disneyland. And he was like, I learned something from this. And then he turned his attention to creating Disneyland. So it was interesting. How depressed was he? You know, it's hard to tell because they didn't have, they didn't talk a lot about it back then. So you really had the sort of.

pick this out from the way that people who worked with him talked about like what was going on with him there's a lot of like uh what do you think i mean i think he was depressed because well hey we didn't have mental health back then but it was the war So the war – you have this company. World War II, right? World War II. You have this company. It's growing. You have Snow White, huge hit, right? Then now you have this – you're a movie studio.

Things are going really well. The war comes and now your European market's gone. And that's a big part of the revenue. And so now suddenly they have to cut back. They have to fire a lot of people and change – put most of their efforts towards government contracts. So like you're having the biggest movie in the world. Government charge.

contracts that make stuff for the war it was like animations for the like the war effort like uh like propaganda propaganda stuff yeah like it would be like i've seen a bunch of it it would be like oh it's you know animations of like

The fighter, the planes flying around the globe and yeah, like propaganda videos, like U.S. propaganda videos, which is not what he wanted to be working on. And so I think he just kind of got depressed. He got paid for that though, right? They got paid for it, but it was just to try to keep the lights on because this is not the same as having like a hit movie.

So it's like they had to fire a lot of people. Then they had a lot of labor issues, right? The animators had gone on strike. This might have been pre-war. That really disillusioned him. Like he had all this like disillusionment happening for years. And so I think he just, when you're really driven like that and you suddenly are taken out of your life, the things you were working on and succeeding on, I think he just got.

depressed disneyland took him out of it he's like we're gonna build this park and everyone thought he was crazy he's like we're gonna make it happen um And it was. It was very successful. And that was actually sort of saved Disney with these parks eventually. They generate a lot of money. I don't know if you know this. When you go to Disneyland, they have abundant opportunities for you to spend money.

I don't know if you know about it. It's almost as if they're trying to make money off of their guests. I always say, what are the two most powerful ways to make money off of people? A, convincing them that they might make a lot of money, so that's casinos, or B, their kids. I want my kids to have the right experience, and so you.

write these checks i had an aha moment and not what else would you buy though other than a ticket and like food and stuff so souvenirs so this was an aha moment i had uh Not me, but actually the guy who does my hair had to figure this out, right? So we're in Star Wars world or whatever they call it at Disneyland. I don't know what it's called. They spent a billion dollars on just like this part of the park, and it's all like Star Wars-y.

And we come across this thing where there's like a guy out there, they're promoting something, and we're like, what's going on here? And they're like, oh, it's a make-your-own-lightsaber experience, right?

And like, oh, that's, that sounds interesting. And he's like, yeah. And you know what? We have some spots. It's reservation. We have some spots like, oh yeah, this is interesting. Like, this could be fun. You have spots right now. Great. He's like, great. So how much is it? He's like, it's $237 a person. Weird number. Yeah.

So we're like, well, we're not doing that. Right. Uh, you had five people. Yeah. And we're not going to spend that money. My wife's like, we're not spending a thousand dollars on lightsabers. Yeah. He's like, he's like, I'll tell you. He's like, yeah, I get it. He's like, you know, at this gift shop though, you can build plastic ones and it's like $30 a lightsaber. Like, oh yeah, that's much better. Okay, good. Here's the insight my hair guy had.

There was nothing behind that wall. There was no $237 build a lightsaber thing. The entire point of that is to get you to buy the $30 lightsabers, right? Because that'd be brilliant. They're like, hey, your kids want to build lightsabers? Like, yeah. Like, it's going to be $250. Like, boo. And then they're like, well, I guess you could do this $30 experience. Well, some people must pay the $250. That's probably what's really happening. But I like this idea that it was like just for.

This is a very effective way to get you to spend $30 on a lightsaber. No, it seemed like it was... But some people wouldn't shake a stick. Some people were. I'm trying to say it nicely. I think... There were plenty of people in the Star Wars portion of the park who seemed like they would spend the $237. Yeah. And I would say they were – they're in their 20s.

probably live with their parents and don't spend a lot of time in the gym. That's like, that's probably how I described it. There was a certain demographic. See, I think about it totally differently. I think I could see like people who are broad, like flying private over there and be like, all right, no problem.

Like whatever. Yeah. But are they that interested in a lightsaber? Probably. If they fly there from like Japan or something. Yeah. Okay. So you're like, this is nothing. We spent so much money. Yeah. Yeah. I get that. We spent so much money to come here. Like, why not just add that? Yeah. Yeah. Yeah. Because you can do like private tours there. It's $800 a day or something like that. I mean, think about if you fly private there. I mean, that's 50 K right there. Yeah.

And then you do the private tour where you spend like a thousand bucks for your party and you get to go to the front of every line. So it's like a private tour. But all it is is they don't want to just say if you're rich, you don't have to wait any lines. So it's like, oh, we have a private tour and our tour guides are able to bypass the lines. But all they do is just.

Let you get on the rides right away. So yeah, you're spending a thousand bucks a day on that anyways. Yeah, you're right. The 237 for the lightsaber. Whatever. Point is, who was the original question from? Duncan. Duncan. Point is, Duncan, beware of lightsaber scams. I think we handled that one perfectly, Jesse. All right, who do we got next? Next up is Fred.

I have an internship in mainframe development with a real and valuable skill. Very few people have the skill set within my organization and many of them are nearing retirement. However, it's long hours and often overlooked by management. This fall, I have an opportunity to transition into a data analyst role for a third internship. However, there are drawbacks as well. How should I manage my internships? You know, I don't care as much about internships, right? It's not your job.

So the default there would be, yeah, do the data analyst one. You're just gathering data, right? See what that's like. Maybe there's something about that you like. Like you're gathering information with these internships. I'm not going to overthink that. When it comes to actually choosing what you want to do for a job after the fact, especially if these internships open up, each thing you did an internship in opens up a job opportunity.

The question is always twofold when it comes to these career capital moves, right? The building up rare and valuable skills and using them to construct your career. Is there an opportunity in this job to build career capital? So are there rare and valuable skills that I could develop?

For the mainframe job that your mainframe internship could lead to, the answer seems to be clearly yes. You describe it as a rare and valuable skill. Most people don't know how to still work with these mainframes, but you need to upkeep them. So, yeah. Yes. The second question you have to ask is, will there be an opportunity for me to cash in career capital if acquired to gain some autonomy or control over how my career unfolds? That's often the sticking point.

And that's what you would really have to assess here. So if for whatever reason this company is very rigid about people in their mainframe development group, no, this is just what that job is and the salaries are here.

And that's it. And we don't really care. We feel like you're expendable. Then the answer to that second question is no. And you'd be wary about it. If on the other hand, you're like, look, yeah, they don't realize the value of this. But if I got really good at that and was like, hey.

I'm keeping up 10 of these systems and now I want to change so that I'm doing this remotely or I'm here once a week or whatever you want to do. And they're like, oh, OK, yeah, we don't want to lose you. If you think there would be a chance to apply your capital to get leveraged, then I think it could be good.

whether or not they're recognizing the value right now. The classic place where the second question trips up people is law partners. As an example from the book, So Good They Can't Ignore You. In law... especially like big law. So like working at the big law firms in the big cities, you for sure are building up a rare and valuable skill, right? Because you were mastering a specific part of the legal code that is literally very valuable to clients and not a lot of people can do it.

You're using your law degree in your brain and your ability to like really work hard to learn it. You build up a huge amount of career capital in law, but at big law firms. They give you a very limited number of options for investing that career capital. Like really the only option they give you is investing it into having more salary. If you get good at this.

You can become a partner and then you can become a managing partner and what you're going to get in exchange for that is more money. And they are very rigid about that is it. If you're going to be at this firm, here is the path and here are the expectations. And the only thing you can open up by getting better is moving to the next step. And so it's a classic example of a place where, yeah, you can build up a lot of career capital.

But they make it very hard for you to have flexibility in how you cash in that career capital to take control of your career, and that's a problem. And it's why a lot of lawyers end up with good bank accounts but really unhappy because they got really good, but they have no choice with what to do with that except for just to make more money.

And for a lot of people, other parts of their ideal lifestyle are then being trampled on because they don't have control over it. So you always have to get both of those questions. So that is how you should evaluate the career opportunities that come out of these internships. Can I build a rare and valuable skill? And will I have options if I do? And that's what you're looking for is whatever has the strongest affirmative answer to both of those.

When it comes to your internships, yeah, do the other internship, right? Like might as well open up more opportunities, learn more things. That information is not going to hurt you. What's going to matter is the choice you make for what job you go after. And then once you do, once you have that job. So that would be.

That would be my advice. I mean there are some options for cashing out your law career capital, but they're hard. You have to leave the big firms. You have to try to do your own thing. You have to renegotiate your situation.

I know someone, for example, who successfully renegotiated. I'm not going to leave the partner track. We can keep my salary where it is. And I'm going to do this many hours. It's billable. So I'm going to do this many hours. So actually, maybe the salary is going to be lower.

than i would be getting if i was doing 80 hours a week at billables or whatever but i'm going to live remotely i'm going to work remotely and live somewhere else and where i'm living what you're paying me for this like actual 40 hours of work i'm doing each week

is way better than what most people are being paid here for working 40 hours a week. Sure, it's not as prestigious and it's not on the track and I'm not going to make a million five, but it's a good salary for a reasonable amount of work. And now I can live in this other part of the country. So I've seen that before.

You've also seen people like try to put out their own shingles and then you can have some control. But law like famously is – I think it's a classic example is you build a lot of capital, but you only have one thing you're allowed to invest in. It's like the company store back in the old –

days of mining companies like they would pay you a good wage to be a miner but the only thing you could do with it is spend it at the company store you had to live in the company town you had to live in the company housing and you had to buy from the company store and they kind of just got all that money back that's how i think about some of those uh

partnership track, sort of elite jobs sometimes. All right, who else do we got? Next up is Joel. I've noticed that much of Cal's goal-setting philosophy aligns with research on goal hierarchies, top level.

subordinate goals down to specific goals for the current day has cal ever seen this research or explored similar evidence-based frameworks and shaping his approach to goal setting or behavior change i mean i don't know i find goal research to be so boring There is this research that often comes out of business schools and there's often long acronyms.

And like here is the whatever 17 letter long word type of goal setting paradigm. And it's really not that interesting to me. I don't find the research interesting. Like you give someone a framework. You're giving them structure to their thinking and planning that they didn't have before and like it does better. Like, look, this is a good framework. So, no, I kind of find that research boring. But I would say this. The type of things you're talking about, like multi-scale planning, et cetera.

I don't see that as goal setting. I just see it as time management. Like what is time management in the end other than making intentional decisions about what should I do next? Like that's time management. And there's a particular segment of people. Sort of this modern, like educated knowledge worker, spend a lot of time on a computer screen type people. Where you have this tension between.

The bigger picture things you want to work on that are going to be long-term important for you and maybe for your professional prospects and the dizzying whirlwind of digital distractions that just sort of like makes up your day-to-day. And so you have to be very intentional.

Otherwise, you're just in the whirlwind and nothing gets done and you get stuck and you get stagnant and you're basically trying to prove your worth through your pseudo productivity, which is a young man's game and something that's not going to lead to like a sustainable life. And so you need some way of. Making a smarter decision in the moment about what to do next beyond just like who wants my attention right now. And so having multiple scales of thinking about what's important.

Helps you trickle big picture ideas down to small picture decisions about what to come next. To me, this is just common sense and it works well in practice. But I don't really see it as goal setting so much as I see it as making smart time management decisions in a current digital work.

where it is very hard on the fly without structure to make smart decisions about your time. Other people don't have this issue, right? I mean, if you're in a situation where you're not in a whirlwind of digital distraction workplace, it might be much easier. If you're working on one big project.

you could run like Oliver Berkman's schedule, which is basically deep work for three hours on the thing you're doing that's important to you. And then just like do your best with the rest of the day, see what you're in the mood for, keep up with stuff that's urgent and just try not to work too much.

Like that's fine if like maybe you're a professional writer, for example, or an independent thinker or something like that. But if you are trying to build career capital and find meaning and move the needle in a sort of busy knowledge work job, I just – I think …

Time management requires some care about what you want to work on next. It requires some care in terms of how do you make the decision about what's the right usage of my time and multiple scale seems to make sense. So I don't know if there's a lot of research on it, but. There is a lot of good experience of real people who have tried this. All right. Do we have a call this week? We do. All right. Excellent. Let's hear it.

Hello, Cal. My name is Jamie Chalmers, a long-time listener, first-time caller. I want to thank you for your books. They've been tremendously helpful in my creative career. I've got several different projects that run concurrently. I was listening to your recent episode about Inbox Zero, which I thought was fascinating. In particular, the method that you use using Trello and your working memory text file to capture information out of email and plug it into your card system.

I've listened to that episode and you talking about that about half a dozen times throughout my head around it. And while I've had some success in using the working memory file to get emails out of my inbox, as it were, I've not had a lot of success in translating those into my Notion task system. And so I kind of wondered whether you could just dig back into that a little bit more, explain some of the mechanics, perhaps the head spaces that you get into when you're focusing on.

translating that information into action. That would be super helpful because I can see that it would be a really useful way to go about things. And, yeah, I'd love to find out more. So thank you once again for everything that you do. Massively important in my life. I appreciate you and look forward to hearing from you. Thank you. All right, Jimmy, it's a good question.

And I think part of like the hang up here might be that I go from inbox to working memory dot TXT. And then from there, there's a lot of directions where that information can go. Only one of which is a task system. So like, let's do this again real briefly. So the basic idea here is how do you clear an inbox, right? I don't recommend... Going message by message necessarily and trying to dispatch completely is each message before moving on to the next.

The context switching there becomes a real cognitive load. It becomes very exhausting. You know the strain. Everyone knows it. Why can't I just keep going through these messages? Why am I so resistant to it? It's because your mind can't keep switching. from one topic to another so quickly. It's mentally really dragging on it, right? And so what I do is I go through my inbox. I don't handle the message. I just...

put a summary of like what that message demands of me into a text file. Really just as fast as I can type, right? It doesn't have to be, you know, clean or interesting or something like that. And then I have all of those in my text file. I can organize them. In fact, I was thinking what I'm going to do, Jesse, is I'm going to load up an inbox right now while we're talking. So I can just give an example of how I'm going to read real.

messages from my inbox right now and talk about like what would be my summary in my my working memory.txt file okay so like i'm loading up i have so many inboxes this is crazy this is my personal It's a personal inbox, an inbox that we do kind of behind the scenes stuff for the podcast and for like my books or whatever. All right. So I'm looking through this now. First email is a newsletter. So it doesn't matter.

Next one is a meeting scheduling for an appointment with a trainer. So what would I put there? I would just type down schedule training. All right. Next one, ironically, is from a PT I used to work with checking in. So I'd be like, get back to their name. There's something from you, Jesse. Put that in ignore. That's immediately where that goes. Button I press there. I got a scheduling email about a club I help run at my kid's school. So I just put, again, get back to.

blah, blah, blah about whatever. I have an invoice, pay, pay invoice. Do you delete these emails after you make the note? Yes. So then I would be delete, make the note, archive. Okay. Yeah. You are archive. Right. And so there's enough information. So when I say like, get back to blank, I put their name. So now I know I can just type that name to Gmail and that, that email will come back. Invoice to pay, pay invoice.

okay uh respond to it's a note from actually a well-known podcaster who i sent a note to and he got back to me and um rogan it's not rogan yeah he doesn't like you doesn't like me He's like, I'm convinced he doesn't like you. Like, why don't you tweet me your move, buddy? Tweet me. Um, mailing list, mailing list. Someone sending me a galley of something. Uh,

South Korea publicity thing put on calendar, right? And something to schedule. It's a note to myself. Put such and such a schedule for two o'clock on Thursday. So that type, that's like what I'm writing. Oh man, you emailed yourself? Because I was somewhere else. Okay. And then you're going to note that that's actually kind of funny. Yeah. Because I was, I was out and about, like if I'm getting my haircut and we're like, oh, let's schedule the next haircut while I'm here.

You had a lot going on at this hair appointment. This is a Disneyland emailing yourself key in today's episode. You would like this guy actually real. Anyways. I'll email myself on the fly because I know when I next clean my inbox that I'll move to workingmemory.txt and then I'll take care of it. All right. So those are the type of things I'd write. Then you sort them by type. Right. So like a bunch of these were schedule. I'm just going to like.

Copy and paste. When I say sort, I mean, I'm just copying and pasting text within a text file, plain text file. I'm going to put like the scheduling things all in a row. So I just have a bunch of like scheduling things like in a row.

There's like getting back to people things. I can be like, all right, getting back to people. And I'll put those like next to each other or whatever. So I'm kind of like grouping these things by type, by type of message. Or if there's a bunch of things on the same subject matter, like if I had multiple.

I did a lot of foreign press this week for some reason. So like I would have like multiple things about foreign press. I'd put all those things together. I'm just kind of like sorting things. Then I will go through and tackle these things by group.

And it's just easier when the groups are all the same type of thinking. So now I have like in that list there, four or five scheduling things. I'm going to open up my calendar. I'm going to open up my inbox. I'm like, all right, now we're doing scheduling. And doing this all at once has like a lot of advantages, right?

Most of those, first of all, are not going to end up on my task list. I'm just going to go through one by one and do the scheduling. And some of these, there's a link to a calendar. I'll do it that way. And other ones I email back, like here's the time I want to suggest. I'm just handling it right there.

So I'll go back and find those again in my inbox and like respond to them as needed. I'm handling it right there. But because I'm doing all this scheduling together, it allows me to have some extra efficiencies as well because I might be like, man.

This is going to pepper my, I don't want to pepper my week with these things. All right. You know what I'm going to do? I'm going to make like Friday afternoon, my like appointment call time next week. And I'm just going to offer that time to everybody. And that way the rest of this week will stay clear, right? So seeing them all together, or I might say...

I'm going to take these two things. I'm going to punt them and be like, you know what? Let's get at this later in the summer. I was like, this is feeling like there's too many things. When I see how many things I'm scheduling, I'm like, this is too much. I'm going to punt two of these things that aren't that important. When I see it all together, I can make that decision more clearly.

Right. And then I'm like, let me get back to people. And now I'm like in that mindset. Again, those are things where I'm just responding. And some of these things will require actual putting a task. Right. Like the South Korean publicity thing. Oh, this is like a non-trivial amount of work I need to do here.

I'm not going to do it right now. So maybe I'm going to add this to my Trello as a task card. Maybe I need to actually connect this to a deadline. So I'm going to put the deadline on the calendar. Like this is like. The drop dead deadline for getting these like answers back to this question. And I'll paste the questions into a Trello card and move on from there. So it's moving things from the inbox into.

hastily type lines in a working memory.txt, sorting those things manually into like groups and then dealing with those groups where I'm either ignoring them or dealing with them right away. scheduling something on my calendar or creating a task where the information I'll point to the information. And I still do it the old fashioned way where I just copy the subject line and put it in the Trello card. So I know what to search for when I get to that task and I'll be able to find it right away.

I know people have been telling me there's ways to link directly to the message in Gmail. People keep emailing me this and I keep ignoring it. But yeah, there are more advanced ways of doing it. So hopefully that's helpful. Does that make sense, Jesse, like what I'm talking about here? It feels like an extra step, but I'm telling you, like this is much easier mentally than if I just went from each email to each email.

And tried to answer it until I was done with it. Like I'm scheduling this appointment. Now I'm responding to someone about this. Now I'm trying to get back to this person about this. Now I'm scheduling an appointment. It just becomes. overwhelming and it's surprising that it does, but underneath the covers is just a context switching. All right, we got a case study here. It comes from TAF. All right, TAF says...

I was listening to you talk about the Thoreau schedule today, and you essentially described the working schedule I've managed to carve out for myself over recent years. It resulted due to a variety of factors, including COVID and having a young family. I have a pretty mid-range marketing job for a CPG brand, managing a small team. And my current role of...

Current role the career capital I've developed enables me to take ownership of my schedule because the management team knows that I will deliver strong work regardless of where and when I work. I'll stop right there briefly to point out this is a common... application and career capital. If you have very valuable skills and you deliver, people trust you. You do the things you say you're going to say when you say you're going to do them.

So you are now a super rare commodity and there will be a lot of accommodations because, oh my God, here's someone who actually does what they say they're going to do when they're going to do it. Like we do not want to lose this person. You have a lot of control there.

Here now is the schedule that TAF created. So that being said, my schedule borrows from many concepts you've discussed. All right, so here's the schedule. I wake at 6 a.m. and read, but not on my phone. I go to my office at 7 a.m. and start my day.

clearing emails, and checking in on core project timelines to make sure there's nothing urgent on the cards. I find that starting at 7 a.m. means I'm working before most of my colleagues are logged on, which means there are very few distractions or meetings booked during this time, and I can just focus on what I need to get done.

Then I have time blocked my calendar from 7.30 a.m. to 10.30 a.m. for my highlight. This is a term that comes from the time dorks, where I focus on the one major priority deep work project that is most critical to my day. Then at 12 p.m., having completed five hours of work, I stop and work out. I'm outside and that allows me to think. And then once I come home, I cover off the last two to three hours of the day with any shallow work admin meetings and sign off at 4 p.m. to be with my family.

In the evening after my daughter has gone to bed, my wife and I read for about one to two hours before bed. I always have a number of books on the go, a novel, something nonfiction, something theological, and a book of poetry so that I have plenty of choice for whatever mood I'm in.

I basically read until I fall asleep, and I like the idea of reading at the very start and end of each day. Some additional habits that help. I have no social media apps on my phone. I use Trello to organize my work and personal life. I have one central board with my priorities for today, which contains anything critical. The rest fall into priority two or backlog. Anything that sits in backlog for a long time eventually gets archived if nobody raises it as important.

I treat my Outlook as if it's an actual mailbox, so I close it once I'm finished with it and then go back to check it at specific intervals in the day. I've found this reduces the impulse of reading emails as they arrive and constantly being distracted. I also have a two-folder system for my email. My inbox has incoming mail or emails I need to address, and then filed has everything else. My reason is that once I've read an email, it can be filed, and if I need again, I can search for it.

This has reduced all email admin and a necessary folder assignment or rules. Finally, I time block my work days with recurring meetings so that I'm automatically blocked off every day for my morning highlight, my lunchtime.

10K run, and my time with my family at 4 p.m. This means every day has the same structure and my calendar availability is prescribed for anyone booking meetings. I've never had any issues or pushback from colleagues with this approach and people rarely book meetings across my time blocks. So there we go. Someone that it's not a sexy story of I became.

like the only nuclear physicist in the world that could handle this, and therefore I have a job where I work one day a week and surf all day. It's not a sexy story like that. It's just someone who does their job very well. They deliver. They're reliable. They do high quality.

They combine this with lifestyle-centric planning or lifestyle engineering or lifestyle architecting, whatever we want to call it, to work backwards from their ideal lifestyle. And it's not something that would catch your attention if you hear it described, but it is a really good lifestyle. Reading in the morning, done at work at four, doing a long run and workout in the middle of the day. Get stuff done, deep work every day.

Meetings are constrained. Email checks are not all the time. People are okay with it because they deliver, and the life is really good. So I think that is a really good example. I appreciate you sending in that case study. All right, so we have a good final segment coming up here. We've got some Cal Network artwork and a What I'm Not reading segment to get into as well. But first, let's take a quick break to hear from some sponsors. Jesse, I have a question for you.

How many people do you think stopped me when I was at Disneyland recently to compliment how good my shave looked? Three. It was 1700 people. All right. Now, whether or not that's actually true, I will say I have been enjoying shaving more recently because I have been using Harry's. Harry sends the best quality razors right to your door for a fraction of the price of the big brands.

We're talking about really good German-engineered blades, these nice, comfortable sort of rubber-coated handles. I really like them. But also shaving products. You get, like, excellent shaving cream delivered right with your razors. One I like is that you can also get a richly lathering skin softening body wash in scents like Redwood, Wildlands, and Stone. Interestingly, Jesse, richly lathering skin softening is how a lot of people describe.

this podcast. So there you go. They send this all to your door. It's automatic. You set it up so you don't have to remember to go to the store. You don't have to add another task to your to-do list. I love that about them. Before I started using Harry's, I was basically in the Stone Age. When it came to shaving, I might as well have been scraping my face with a piece of sharpened flint. Take your own grooming out of the Stone Age. Get Harry's. Normally their trial set is $10.

But right now, you can get it for just $6 if you go to harrys.com slash deep. That's our exclusive link, harrys.com slash deep, to get a $6 trial set. I also want to talk about a new sponsor, our new friends at Small's Cat Food. I grew up with cats. I don't know if you know this, Jesse. I grew up with two Siamese cats, Singha and Nitnoy.

Also, little known fact, a full grown African lion. So that went well. The lion ate and killed some people. But the cats, the Siamese cats, I love and I miss having cats. So I'm still kind of tuned in. to the world of cats and cat food, which is why I have been noticing more and more people who do have cats to this day mentioning Small's cat food. If you have...

digestive issues with your cat, maybe they're throwing up their food or you just don't like what you're feeding them and you're like, I think we could do better, you should check out Small's. Small's cat food is protein-packed.

It uses protein-packed recipes made with preservative-free ingredients that you'd find in your fridge, and they deliver it right to your door. I love the delivery stuff because I don't want extra tasks on my task list. That's why Cats.com names Smalls their best overall cat food. To get 60% off your first order plus free shipping, head to smalls.com and you can use the promo code DEEP for a limited time today. Look, Smalls was started in 2017.

A couple of guys who are home cooking cat food in small batches. It's grown from there. They work with the humane world for animals. They've donated over a million dollars worth of food to help cats. Like they know cats are important and what cats need. Not everyone is like this. Jesse, I don't mean to throw you under the bus.

But you tell me all the time, hey, when it comes to cats, I feed them mainly unmixed cement powder and garbage. That's what you told me. And there's people like you that don't understand. The value of actually good cat food, and that is what you're going to get from Smalls. After switching to Smalls, 88% of cat owners reported overall health improvement. That's a big deal. The team at Smalls is so confident.

Your cat will love their product that you can try at risk-free. That means they will refund you if your cat won't eat their food. So what are you waiting for? Get your cat the food they deserve for a limited time only because you are a deep questions listener. You can get...

60% off your first Smalls order plus free shipping, but you got to use my code DEEP. That's 60% off when you head to Smalls.com and use promo code DEEP. Again, that's promo code DEEP for 60% off your first order plus free shipping.

at smalls.com. All right, let's get back to the show. All right, so in our final segment, I want to do a twist on my what I'm reading segment that's called what I'm not reading. But first... Jesse, I want to quickly cover a new piece of Cal Network artwork that came in. I do have to sort of ding this person a little bit because a lot of people have been making sort of like custom artwork.

In this case, they just took an existing photo of me and added to the existing photo a copy of a fake Cal Network book. But I still think it is interesting. So here we go. We'll put it on the screen here for people who are watching instead of just listening. And what we've got here is a – this is my author photo actually. I don't know if you've seen it. This is what's on my book jacket. Cal Network with his Georgetown tank top. That's what I wear to teach. Veining out as I constantly do.

with Austin Aviator Sunglasses, holding Cal Network's hit bestselling book, I'm the Boss Now, which is his career advice book. So there we go. It's my author photo with... a mocked up book. So I always appreciate some good Cal network artwork. All right. So to our final segment, I have been doing recently what I'm reading where I talk about like interesting articles and books.

that either cover interesting ideas you might like, or I think are just good books or articles for those looking for some more depth. I want to rant briefly here today. I'm going to call this segment what I'm not reading. It's a particular type of article that's been common recently, and I want to give you permission to ignore every single one of them because they're nonsense and are going to cause unnecessary stress.

I have loaded here a sample of one of these type of articles that I'm going to suggest you don't read. I'll put it on the screen here for those who are watching instead of just listening. The article is from Fortune, and here's the headline. OpenAI CEO Sam Altman says AI can rival someone with a PhD just weeks after saying it's ready for entry-level jobs. So what's left for grads? This came out on June 20th.

This is an article. I saw this because like Google notification just showed it to me on my phone. There are a lot of articles like this where there is some sort of. Over the top alarm is headline, which makes anyone who's just loosely following AI be like, oh my God, we're so screwed. But I want to look a little bit deeper at this article to try to explain why you can ignore this type of thing.

All right, so if we read this article, I have a couple quotes here. Here's the first paragraph. Earlier this month, OpenAI CEO Sam Altman revealed that the technology can already perform the task equal to that of an entry-level employee. Now on a podcast posted just last week, the chat GPD mastermind went even further saying AI can even perform tasks typically expected of the smartest grads with a doctorate.

Soon after, it says, as companies like Amazon have admitted, they will soon cut their corporate ranks thanks to AI and Anthropic CEO Dario Amadei, warning that the technology could wipe out half of all entry-level white-collar jobs. It begs the question. What jobs will be left for those tossing their graduation caps into the air in the coming years? There are a lot of articles like this where tech CEOs are saying pearl-clutching, groin-tightening, scary type of things like this.

This is all nonsense. And let me try to explain to you why. Okay, first of all, let's start with the claim here that, and I'm going to read this, AI can perform tasks typically expected to the smartest grads with a doctorate. What are they actually talking about here? What they're actually talking about here is the fact that OpenAI carefully tuned one of their foundational models to work on a specific type of math competition question. So they hired...

We got some correction on this from a listener. They used a data set from a company that hired math PhDs and paid them $100 an hour to write. math problems of this type with step-by-step solutions and then they could use reinforcement learning techniques to try to tune one of these foundation models to do well on this very specific type of math problem.

A professor involved in this project said, oh, these are hard math problems, not the type you would assign an undergrad, but the type you might assign a graduate student. From there we get... AI can now do PhD-level workers' jobs. That is a nonsense leap from a very specific type of math competition problem. That is like of the type that might get assigned in a graduate student problem set to AI is doing graduate level jobs. And not only is it.

a huge exaggeration. But as we went over in the what I'm reading segment a few weeks ago, even those results are highly contested because open AI was like, look, these are hard problems and we can do them. Independent research firms said, great, let's take these models and we'll put them on other similar math competitions that happened recently. And they did terribly.

leading to the idea that they had been very, very fit to this very specific type of problem that happened to be really well suited for the reinforcement learning techniques that we know how to use now on AI. So that is a massive exaggeration. Another massive exaggeration that's happening with... a lot of this reporting is conflating of the post-pandemic tech downturn, which is leading to lots of layoffs. It's getting a little bit better now, but it's leading to lots of layoffs because...

There was a huge boom in the tech industry during the pandemic and post-pandemic. We're cyclical. It's a hard job market there. Companies are cutting back. We put a lot of money into a lot of these areas. We need to cut back because we need our profit. Our profit ratio is to be higher for the stock market. And just what happens? Companies are cyclical. So there's a cutback.

And there's a lot of this sort of disingenuous conflating by these reporters that make it seem like without maybe necessarily 100% claiming it that these job losses are because AI. is automating the jobs this is nonsense it is not what is happening yes amazon uh is cutting meta cut a bunch of people microsoft is cutting a bunch of people why

Because they're bloated. They cut those revenue numbers. They cut those expense numbers go down. Their profitability goes up. They got too big. They are not replacing. Hundreds of thousands of people or tens of thousands of people with AI that's automated them. AI cannot do that yet. So I think that is very disingenuous. I saw an article the other day that said, look.

CS majors are cratering. They're going down and it's because AI is taking all the jobs. It's like, hold on a second. CS majors are down because the tech industry is down. The same thing happens every time. The tech industry goes down. When I was a computer science undergraduate, we had the same issue. Post .com bust in early 2001, the tech industry contracted.

majors went down because there was less jobs. And then majors went back up again when there was the Web 2 boom, brought a lot more investment and there's more jobs again. There's a contraction post-pandemic. jobs that made jobs for except for you know specialized ai jobs like if you're a machine learning person are contracting some so majors go down and yet people will conflate it and say oh it's because ai is taking the jobs ai is not

replacing software developers. Most of the cuts of these companies are not software developer jobs. So I think it's really disingenuous reporting, but there's a lot of this going on. Later in the article, we see some of this evidence. So here they say, for example, they do note, well, okay.

However, I'm reading from the article now later in the article. However, in the tech industry in particular, volatility in the jobs market is nothing new, said Art Zale, CEO of the tech career platform Dice. After all. Nearly 600,000 tech employees lost their jobs between 2022 and 2024, according to layoffs at FYI. Yes, the tech industry has really post-pandemic contracted. So yes, there's a lot of job cuts. That's not because...

In 2022, before ChatGPT was out, AI was taking those jobs. It contracted, and then it will grow again once more investment capital comes back in. Look at like Dario Amadei saying, technology could wipe out half of all entry-level white-collar jobs within like two years, whatever he said. All right, here's the reality. It's not going to do it, and he's just blowing hot air. And Sam Altman knows that AI is not taking... PhD level jobs because it's not really taking any jobs right now.

They're blowing hot air. Why are they blowing hot air? Because it requires a massive amount of investment capital, ongoing investment capital to keep these places afloat. OpenAI is desperately in need of this. massive loan from SoftBank, this huge mini, mini billion dollar loan, which is happening in tranches to like go through and they need excitement around their tools. And the more you feel like these tools are.

going to in the future be immensely disruptive the more you overlook right now that they're losing three to four billion dollars a year so it is it is in the interest of the ceos of tech companies to push any possible narrative of very large world-shaking disruption because you will overlook any issues with their business right now.

If you think where they're going is going to be the biggest disruption that we've had sort of like in the history of humanity. So they are now just spewing anything they can get attention for because they know they will get articles like this. a a model fine-tuned to do well on math problems but not really because when other people tested it didn't and say

Well, now even PhD-level jobs are gone. That is a nonsense leap, just like it was a nonsense leap when Sam Altman talked about how AI was going to help us build Dyson spheres around the sun, the power, the growth of the universe. Just like it was a massive leap when Scott Alexander very confidently says in Project 2027 of like, well, yeah, I mean, pretty soon the AI is not only going to be able to program its own super intelligences, it's going to build self-replicating robots.

the data centers and this will all happen in the next three years that's not going to happen so anyways i wanted to do a little bit of a rant and say

You can feel confident right now when you see like these like super alarmist headlines that don't reflect anything you've seen in your own life or your own industry, your own career. All these jobs are gone and it's quoting some sort of tech CEO. You can essentially ignore those for now. Follow the more like tech reporting, tech journalism, but also just wait to actually see major changes in your own world. I think the hype coming out of these CEOs is becoming almost like parody.

This has now entered my list of what I'm not reading is alarmist articles about whatever the latest chaos, crazy, nonsensical hot air thing that Sam Altman has said and will be widely repeated by outlets looking to get people to click. All right, there we go. That's my rant, Jesse. Wait, so do you have notifications on your phone? I don't know what... Okay, let me rant about this. So you have to have this Google app on your phone.

They force you to have it for, I had to get it because I couldn't consistently log into Gmail. You have to have this Google app. You have an Android. No, it's iPhone. Oh, you have an iPhone. Yeah. So something about like the two-factor authentication, you have to have the Google app, the use like Google things. And so that app just like shows me, I don't know.

I guess like it just will pop up like these articles will pop up and like little bars on my iPhone. But I can't get rid of the Google app because then I can't use my email. Can you turn off the notifications? I don't know. Is that something I can do? I don't – It cracks me up. I have no idea. Let's see here. I don't know. Probably. But anyways, because I read a lot of AI content and my job is like a tech – when I do tech journalism, it's always showing me these nonsense articles.

It's always like Dario Amadei says, you have 17 minutes left to live before AI-driven robots harvest your brains to lubricate their self-replicating robot machines. Just like these type of headlines again and again and again. So I see a lot of this nonsense and it's what I am not reading because, again, every time you look at it, it's like some super exaggerated claim like here or they just make it up. Amadeus is just like.

Half the jobs will be gone next year. The reporter's like, oh, that sounds bad. I'm going to get a headline about that. That would be really bad, right? It's great. It's crazy. And you're like, well, how is that going to happen? I don't know. Key unlocks. It's going to happen. I don't know.

I could write these articles. I should just do reports like that's the deep media LLC head and be like, we're going to be able to fly robotic pterodactyls next month. All right. Why not? Like, that would be a big deal. Bees with lasers are going to start murdering our dogs by 2028. Sure, right? I mean, I don't know. It's possible. Lasers exist. Bees exist. There's the RoboBee project that happened at Harvard 20 years ago. It's possible.

I sometimes feel like this is what – there's so much interesting stuff happening with AI and so much particular impacts and good stuff and bad stuff. There's so much serious reporting happening that I hate that all this junk is getting out there. I mean the robot bee thing is true though, so be careful about that.

All right, that's all the time we have for today. We'll be back next week with another episode of the podcast. 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.