Deep Questions with Cal Newport - Ep. 349: Sam Altman On Productivity

Episode Date: April 21, 2025

Cal talks a lot about his ideas for producing meaningful work in a distracted world. But how do other people tackle this goal? To help better under this question, Cal reacts to a 2018 article from OpenAI CEO Sam Altman in which he details his productivity philosophy. Cal then answers listener questions and concludes with a dystopian tech corner.Find out more about Done Daily at DoneDaily.com!Below are the questions covered in today's episode (with their timestamps). Get your questions answered by Cal! Here’s the link: bit.ly/3U3sTvoVideo from today’s episode:  youtube.com/calnewportmediaDeep Dive: Sam Altman On Productivity [2:20]- How do I prevent administrative sprawl? [22:01]- When is the best time to schedule Deep Work? [31:12]- How can a gardener gain career capital? [32:43]- How can one go from good performance to exceptional performance? [36:21]- Is it better to write with a single monitor? [38:33]CALL: Slow Productivity in a changing world [43:09]CALL: Rory Mcllroy and his phone [48:35]TECH CORNER: Will AI Destroy Humanity by 2027? [56:46]Links:Buy Cal’s latest book, “Slow Productivity” at calnewport.com/slowGet a signed copy of Cal’s “Slow Productivity” at peoplesbooktakoma.com/event/cal-newportCal’s monthly book directory: bramses.notion.site/059db2641def4a88988b4d2cee4657ba?blog.samaltman.com/productivityai-2027.comdaveshap.substack.com/p/common-bad-takes-in-ai-safety-a-responseThanks to our Sponsors:shopify.com/deepzocdoc.com/deepvanta.com/deepquestionsexpressvpn.com/deepThanks to Jesse Miller for production, Jay Kerstens for the intro music, Kieron Rees for the slow productivity music, and Mark Miles for mastering. Hosted by Simplecast, an AdsWizz company. See pcm.adswizz.com for information about our collection and use of personal data for advertising.

I'm Cal 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. Just saying I put out a PSA to our listeners here. They should, if they have questions they're interested in having be addressed in the show, submit questions, right? Correct. We always need more. And we tend to age out a question.

So even if we have a lot, the ones we don't get to, they become dated. So it's always to your advantage as a question asker to submit new and fresh questions. Shorter, I think, right? Jesse, tends to be better than long. Yeah. If you put the four or five paragraphs, explanation is harder to do. The link to submit those questions, where are we seeing people, the deeplife.com slash listen.

Yeah, there's a link right there. There's also a link there if you want to submit calls. You can do it right from your phone or browser. Another sort of shortcut to getting onto the show because we're always looking for calls as well. Correct. We have a limited number there. So keep your questions and calls. And also case studies are entered there as well.

Mm-hmm. There we go. Well, they can email me about case studies. Okay, you can also email jessia at calnewport.com about case studies. So, anyways, keep it coming. We love your questions. It fuels the show. Speaking about the show, we have a good one today.

We're going to start by looking not at my ideas for working deeply in a distracted world with the ideas of someone else who's pretty well known. Sam Altman from OpenAI. We found an old article of his. Then we got a bunch of list or questions. I think we have multiple calls, including one about my best friend. I like to think of myself as his mentor, Rory McElroy, the golfer, who just won the Masters. So I'm glad to see my student did so well, Jesse. We'll get into that later.

And then, again, because I can't help myself, we have a tech corner where we'll return to the world of AI. Be ready. We're going to talk about a dystopian prediction for the future, the near future that's been making the rounds and has been scaring more than a few people. So we'll get into that and see how scared we are. All right, sound like a good plan, Jesse? Sounds like a plan. All right, let's get rolling with our deep dive. So as someone who writes and talks a lot about producing meaningful stuff in a distracted world,

I always get excited when prominent individuals give us insight into their own processes for achieving this goal. So you can imagine how happy I was when I saw Tim Ferriss recently linked to a blog post that was titled simply Productivity that was published in 2008. by OpenAIs Sam Altman. Here's the opening two sentences of this essay. I think I am at least somewhat more productive than average, and people sometimes ask me for productivity tips, so I decided to just write them all down in one place. So I'm excited about getting to this essay, because when Sam wrote it, 2018, think about it, OpenAI was at a crucial turning point.

They had just released GPT1, but they were operating as a nonprofit. Elon Musk had just left the board of directors after failing to convince the board that they should merge OpenAI with his Tesla company to help out its financial situation. Instead, Sam led a move from a nonprofit the next year to a cap profit status that opened up venture capital funding. They could hire a bunch more talent and that's really where the Open AI story we know today really took off. So he was pretty productive during this period. So it's useful to look back and say, how was he thinking about getting important work done? on the eve of OpenAI making all of these important leaps. So what I'm going to do is pull out, I think I have five, let me look at my notes here,

I have five ideas from his essay, and for each we'll get into it. I agree with a lot. I disagree with others. I think there's some big, important ideas he highlights in some. So we will get in the Sam Altman's productivity essay. Let me just load it on the screen now for those who are watching instead of listening. This is what the essay looks like. I missed that classic blog format, Jesse, back in the end of Web 2, where you kept things simple and just wrote long essays.

But here it is, published in April of 2018, so seven years ago, basically to the day. That's what it looks like. So I'm going to jump through here and I'm going to pull out some quotes. All right, Jesse, we can bring it back to full screen. All right, here's idea number one. I'm reading now from Sam's essay. Compound Growth gets discussed as a financial concept, but it works in careers as well, and it is magic. A small productivity gain compounded over 50 years is worth a lot.

So it's worth figuring out how to optimize productivity. If you get 10% more done and 1% better every day compared to someone else, the compound difference is massive. All right. So I have mixed feelings about this first idea of applying the compound growth idea to productivity. I think there's some piece of this that's right and some piece of this that's not right. They getting 1% better, sure. If you could get 1% better at something every day, you are compounding. So your current skill level grows that your 1% is being applied to.

And if you put that out over a certain number of years, you would be the world's very best expert in like five years or however the math works out. That's not really how getting better works, though. It typically has long periods of practice. It leads to new levels of skill and each level of skill is harder to get to than the last. So it's almost more of a linear or like a slow linear function that an exponential function like compound interest would give you. You're kind of getting better faster and then it really slows down. It just gets harder to get to each new level. Whereas compound interest, you get a curve that picks up speed as it's going.

I also am concerned about him making the slight shift. He's combining getting 1% better as an example of compound growth to doing 10% more each day. This doesn't really jive with my study of long-term productivity. If you looked at my book's slow productivity, which came out last year, one of the things I do is study some of the most historically productive people, meaning like what they produce, historical figures that produce things, do we look back at now and see that was super important? What a productive and intellectual life they had. So Galileo or Newton or Mary Curie or Jane Austen, et cetera. And what you find in these stories is their key was not getting more done each day. In fact, what I highlighted was there was a certain notable slowness or even lack of urgency behind their biggest achievements.

They thought about it. They had long digressions in the other interest. They would come back to them. They would let it marinate. And ultimately, it didn't really matter. The busyness or phoneticism of any given day wasn't important. It was more about the consistent application of thought over a long period of time that eventually led to really big breakthroughs. So when it comes to producing really meaningful stuff,

I don't know that this idea if I get 10% more done each day really matters. I think where that will lead to more often than not is just busyness. That's the easiest thing to get 10% more done of. And busyness doesn't necessarily transmute into results. All right. So that's an idea where I'm not completely on board with the way Altman is summarizing things. Idea two, however, I'm very much on board with. Let me read now from Altman's essay.

It doesn't matter how fast you move if it's in a worthless direction. picking the right thing to work on is the most important element of productivity and usually almost ignored. So think about it more. I make sure to leave enough time on my schedule to think about what to work on. The best ways for me to do this are reading books, hanging out with interesting people, and spending time in nature. Interestingly, this is somewhat contradictory to idea number one, which is like, hey, get 10% more done each day, it'll add up. And you say, no, take your time. Don't get started.

Think more. Hang out in nature. Hang out with interesting people. Read. Wait to get started. really make sure that you have the right thing to work on. I'm a big believer in that idea. I remember I wrote an essay about this way back early in my writing career.

I wrote an essay for Ramit Sethi's blog, and it was called Don't Get Started. My argument was, it is, basically Sam's argument, it is really hard to figure out the right thing to work on. The thing that's going to matter and that uses the rare and valuable skills that you currently possess. And it's probably going to take a long time once you choose the right thing to get really good results. So you want to be really wary of just diving into things.

Because when you dive into things, you're basically flooding your circuits with activity and you are taking them out of the game for working on other more important things. So resist working on things. I often say with big projects, resist working on them. Think about them, read about them, get excited about them. But resist working on them until you can't help it anymore. Sort of like me with book writing, me. with this podcast, man, I resisted podcasting for a long time.

I learned a lot about it. It's not quite right. I don't want to just do it for the sake of doing it. I don't like activity for the sake of activity. It took me years before I said, okay, I can't avoid this any longer. That's really what you should be looking for. Fewer things done better. That has been a theme through my work so long that the original, one of the original

mottoes of my study hacks blog back when it was still focused on students was do less, do better, no why. Do less, do better. It's a key idea. So I think Sam is absolutely onto it there. That's probably reflected in OpenAI. They kind of chose their points. They chose their battles where they thought there could be big work.

For example, working on very large-scale language models. And then they went down that road year after year. So they carefully chose what they were working on. And then really gave that a lot of attention over time. That's where the big breakthroughs came from. All right. Idea number three. Now we're going to get into the weeds of actual time management.

Here's Sam. I highly recommend using list. I make lists of what I want to accomplish each year, each month, and each day. Lists are very focusing, and they help me with multitasking because I don't have to keep as much in my head. If I'm not in the mood for some particular task, I can always find something else I'm excited to do. Later, he says, I don't bother with categorization or trying to size task or anything like that. The most I do is put a star next to really important items. I try to prioritize in a way that generates momentum.

The more I get done, the better I feel, and then the more I get done. I like to start and end each day with something I can really make progress on. A couple of interesting points about his approach here. One, we do see him preaching a principle that I talk a lot on this show, which comes from David Allen, who himself took it from Dean Atchinson, which is the notion of full capture. Having things written down and not being kept track of just in your head is critical for avoiding unnecessary stress and forgotten deadlines and scrambles. do not use your brain as a task storage device or a calendar. Use test storage devices or calendars for that rule. And he makes that clear here.

He says, look, they help me with multitasking as they don't have to keep as much in my head. And by multitasking, he means just having multiple projects going on at the same time. So that's useful. I also knows, however, the simplicity of a system. He just writes things down on a list on paper. He doesn't break it up in the categories. He doesn't do any sort of prioritization. He just sort of looks at the list and says, what do I want to work on next?

Maybe you'll put a star next to something to really remind him that it's important to get it done. That type of basic system, a sort of, what's known as an MIT system, most important task system. It's been around, this idea that's been around for a while. We hear it in the early 2000s. We hear Julie Morgonstern talking about this. We hear Brian Tracy talking about this. We hear Leo Baboodov's and Habits talking about this. Even more recently, when Oliver Berkman came on my show last fall, this was basically what he was pitching.

get the important thing done first, and then kind of do your best with the rest of the day. It works and it doesn't. So, I mean, it works in the sense of if your goal is to make progress on what's important, this is emphasizing that really just means doing the important things and making progress. So I think the fact that Sam had this sort of simple system, I just make sure the important stuff gets done. shows how when it comes to long-term productivity, this is really different than busyness.

This is really different than I'm quick on my emails and Slack. I'm jumping on a bunch of calls. For Sam, his productivity was dependent on doing a small number of things consistently well. The issue is for a lot of us, there's a lot of other stuff too that we have to do that is not just, hey, here's the project I want to work on. It is, I have to get back to this person. This dean wants to know this. My students need me to sign this. The parking office needs me to update my license plate for the new license plate readers they installed into Levy Garage.

Just making these up on the top of my head here. And we can't say no to those things. So that's the context where you actually probably need a more complicated task storage system because you can't just, if you have many unignorable demands on your time. So you have the big, like Sam's focusing on in the small, if you just have a big list and you're just trying to choose from there, hey, what's the big thing I want to work on today? that small stuff is going to eat away at you because you're going to miss things people are going to yell at you small things will get missed people like where's this where's that your car is going to get a ticket because you didn't update your license plate information and that's going to become a source of stress and a problem so i like the point that sam is making here is like ultimately just doing something important every day is what matters for the stuff that people will remember you for the busyness doesn't produce stuff that matters but if you have a lot of that other stuff smarter task storage might be important right this is why i like to store stuff and cognitive context. So like I said, I can spend time on like my professor role and just see tasks for that divided in the status.

So it's very easy to sort of see what's what and what needs to get done. So smarter task storage, I think is necessary if you have a lot to do. But don't forget Sam's big lesson here, which is, yeah, but the small stuff is secondary. Do the best you can with that. Organize it in a way that's going to save you from stress, but it's really working on the important things each day that's going to matter. All right. Idea number four from Sam Altman.

Here's Sam. I try to be ruthless about saying no to stuff and doing non-critical things in the quickest way possible. I probably take this too far. For example, I am almost sure I am ters to the point of rudeness when replying to emails. I generally try to avoid meetings and conferences as I find the time costs to be huge. Again, I think there's a critical point here. Whether or not you have the power to say no to everything,

it emphasizes how almost everything doesn't matter. This is a theme, I think, that goes through Sam's essay here. The things that mattered that made open AI from a struggling nonprofit to a company with a massive valuation and huge impact on the world technological and economic scene, the things that mattered were small and hard. And he was pretty ruthless about coming back to them. So if you worry about saying no to stuff, thinking that this somehow makes you less productive, keep in mind this Uber productive individuals' productivity was built on his default of, I really don't want to do stuff. most stuff is just getting in the way of time taking my time away from the stuff I know for sure is going to be really valuable.

More executives should probably follow that advice as well, right? I mean, I've talked to more than a few executives who would have been in Sam's position at different companies that say, no, my job is right, I have to be in meetings. How can I say no to them? And he's saying, well, because if you want to be good at what you do, the meetings aren't really what matters so much as like you understanding, pushing,

and developing the ideas they're going to make the biggest difference. All right, idea number five from Sam. I have different times a day I try to use for different kinds of work. The first few hours of the morning are definitely my most productive time of the day, so I don't let anyone schedule anything then. I try to do meetings in the afternoon. Another great idea, the morning is a good time for deep work for most people.

So just having a simple rule, I don't do meetings until this point. Makes a big difference. I was chatting recently with a former president of a large company, and he was saying this was a huge change for him is that they were just filling his days with meetings, his staff. And at some point, they said, okay, you know, we're going to protect time in the morning for just working on your own stuff. And he was worried this would make him a worst executive instead made him way better. There's endless people that want your time. There's endless meetings you could take. You're already constraining the meetings you can take because your day is only so long.

So why not just change those constraints even more so that you have more time to work on the big thoughts that are going to matter as well. If Sam Maltman can do you. do it, you can probably do it as well. So ultimately, I think Sam has a lot of non-surprising advice here. I mean, I think he would have been, these ideas would have fit well in my book's low productivity. These ideas would have fit well in my book deep work. And perhaps this is not surprising because he ended up being very successful at what

he did. He worked on deep stuff in a distracted world. I want to end with a quote from Sam's essay that I think summarizes well, the gist of his whole philosophy here. Don't fall into the trap of productivity porn. Chasing productivity for its own sake isn't helpful. Many people spend too much time thinking about how to perfectly optimize their system and not nearly enough asking if they're working on the right problems.

Doesn't matter what system you use or if you squeeze out every second if you're working on the wrong thing. And I think that gets to the heart of it. Spend a lot of time figuring out what matters. That's not an easy question. But once you've answered it, spend a lot of time working on that thing. And do your best was whatever else. and that'll work itself out. Don't stress yourself out too much.

It's impossible to do it all anyways. But working hard on the right thing consistently, that's what matters. Everything else is just trying to take care of the details that are trying to get in the way of that. Great way of thinking about it. So you need some systems and some rules. But mainly you just have to do the work of finding what the work on and putting in the right time. I wonder he probably, his rules probably have changed by now just because that company is so large. He is a lot richer.

Yeah, he is a lot richer. I don't know if that changes. That's true. But I'm just thinking like the size of the company now versus then. Yeah. It would be interesting to check in. I think now he probably is all meetings. I've never met Sam Haltman.

But busy guy. All right. Well, we have some more. Speak of Sam. We're going to do some AI stuff at the end of the show. But now we want to move on some questions. But first, as always, let's briefly hear from a sponsor. Trust isn't just earned.

It's demanded. Whether you're a startup founder navigating your first audit or a seasoned security professional scaling your GRC program, proving your commitment to security has never been more critical or more complex. That's where Vanta comes in. Businesses use Vanta to establish trust by automating compliance needs across over 35 frameworks like SOC2 and ISO-27001. Centralized security workflows, complete questionnaires up to five times faster, and proactively

manage vendor risk. Vanta can help you start or upscale your security program by connecting you with auditors and experts to conduct your audit and set up your security program quickly. Plus, with automation and AI throughout the platform, Vanta gives you time back so you can focus on building your company. Let Vanta help you on the security side, so you can focus on the deep thoughts to matter. Join over 9,000 global companies like Alassian, Cora, and factory use Vanta to manage risk, improve security in real time.

Here's the good news. My listeners can get $1,000 off Vanta. if you go to vanta.com slash deep questions. That's V-A-N-T-A-com for $1,000 off. I also want to talk about our longtime sponsors

at Zoc doc. This is a big problem a lot of people face. There's some sort of medical thing that shows up. That shoulder's barking, your eye is itchy, the tooth is hurting,

and you realize, I don't have a doctor I can call for this right now. And the question becomes, how do I find that doctor? And what most people do is maybe they ask around their friends. Their friends give them a recommendation. They call that person up and that person laughs.

They say, you know, we take no insurance. Our next appointment is available in 2032. And quite frankly, you should be ashamed for even thinking that we could see you in our doctor's office. It's like a whole socially fraught thing, right? Especially for introverts like me. This is where Zoc Doc enters the scene. It's the right way to find health care providers. Doc Doc is a free app and website where you can search and

compare high quality in-network doctors and click to instantly book an appointment. We're talking about booking in-network appointments with more than 100,000 doctors across every specialty from mental health, dental health, primary care to urgent care, and more. You filter for doctors who take your insurance, located nearby, good fit for what you're looking at. There's verified patient reviews. You can see, do I like this doctor or not. And then once you find that doctor, you can see their actual appointment openings right there

from Zoc Doc. Choose a time slot that works for you. and boom, you've booked a visit. Appointments made through Zock talk also happened fast. Typically within just 24 to 72 hours of booking, you can even sometimes score same day appointments. I have multiple, I mean, I use Zock Doc. I mean, I've used it to look for doctors, but I have multiple practices that use Zock as well.

They even use it for paperwork and advance and stuff like this. So I've been in the Zock Dock Dock ecosystem. It's definitely a go-to for me if I'm looking for a new health care provider. It's one of these ideas that, of course, it should exist. Of course, there should be a good way to use the internet to find health care providers and set up appointments. So great service. Stop putting off those doctors appointments and go to Zocdoc.com slash deep to find an instantly book a top rated doctor today. That's ZOC, doc.com slash deep.

Zocdoc.com slash deep. All right, Jesse, let's do some questions. First questions from Kelsey. I oversee the admin for over 150 students. This creates a flood of emails and urgent tasks. Every form still requires my signature. How did you set up Trello in your email to manage this kind of administrative sprawl without letting it dictate your day? Do your students have a pipeline to follow forms to fill out?

Well, it's a good question. I mean, you have my sympathies. You've ever seen admin for 150 students. I'm overseen the computer science major for 100-something students. So I'm with you, Kelsey. The first thing I would say, you talk about not letting me. Mr. Spall, dictate your day. Well, it, you know, will be realistic.

If you're overseeing admin for that many people, it's going to be a major part of your day. So we want to make sure that the expectations here are realistic. We're not going to automate this work out of existence. We're not going to be doing six hours of deep work a day. But it shouldn't be controlling your day like a puppet master, right? So it shouldn't be you reactively balancing from one thing to another and feeling like you're always behind or always exhausted or frustrated. That I think we can do better on. The story I like to tell when trying to get at the right solutions here is a story from when I was in graduate school and I was TAing for my advisor's distributed algorithms graduate class.

Pretty big class. I don't remember how many kids. I mean it was like 50 or 60 kids or something like this. It's a theory class, right? A lot of problem sets. I remember early on. So as the TA, I was in charge of like collecting those problem sets, writing the problem sets or some of the problems, getting the sample solutions together. We had graders, but I would have to organize and work with the graders to do the grading, then get the grades entered and those problems sets back to the students.

And I remember early on, it was very time-consuming. I'd have all these different problems that my advisor and the teacher of the class wanted Xerox copies as a backup. They're a stapled and or dog-eared together, and I would be pulling out staples and trying to run them through the Xerox machine and this would take forever. and then try to, you know, get them to the graders, then get them back to the students. Like the whole thing was very time-consuming. And at some point I had this epiphany. I could just ask the students to do a little more,

a little more that's going to make my life easier, that for them, who cares, it's like a small extra step, but for me could add up to a lot of savings. And that's when I started saying, okay, here's how you have to hand these things in. They have to be single-sided, no double-sided. That messes up the photocopy, no staples. just hand them in as a stack with your name on the top of every page.

And I believe I said when you hand them in, you have to alphabetize them. So as you come up, you start adding these things to the desk up here. Find where your name goes. So it'll be alphabetized when you're done. And then I could just take this whole stack and run it through the Xerox copy all at once, the document feeder, have a copy of all them I could just put aside. And then I have all of them alphabetized. And the graders could then, when they're done grading,

putting them back alphabetical, it's easy to hand them back to the class. Like these small things, this little bit of extra work I made the students do, made my life much easier. Now, as a professor, I, you know, I took this even farther, right? I mean, when it comes to, like, problem sets and exams, I have these various administrative things the students can do that just makes my life much easier. I switch, for example, to digital submission of problem sets. Do your problem set, take good images of it, upload them directly to our class portal

so me and the graders can see them digitally. We can note them digitally. We can enter the grades digitally. You can get them back digitally. I don't have to deal with papers. I started doing this with my exams. They're taken in class. But I actually have the students,

when they're done taking the exam right there before they leave, take photos with their phone of every page, which they can then later, on my behalf, upload into the online system. So we have digital copies, the simplified grading and comments and returning entering grades. And I can just keep the physical copies as a backup, just pick them up and,

put them in a drawer in my office, right? So I learned this rule. You have someone to do 10% more on their end can make your life 100% easier. Small addition to their work can make your life bigly more. That's not real word, but much more easy because it aggregates over a lot of different people. All right. So with that in mind, Kelsey, my lesson here is I would have some sort of procedure page, you know, FAQ or systems or whatever.

It's a website that you can easily update that you start building. out your instructions for all the common things you do and handle as an administrator, and then you can just keep pointing students to this page. Oh, you need an external credit approval form. Look at this page. You need transfer credit approval. Look at this page. You need to declare approval for a major declaration. Look at this page. So you're preventing everyone from just informally emailing you and you have to kind of explain or work with them informally to get done whatever thing they need to get done. You can then experiment on this page with giving them more work to do to make your life somewhat easier.

Like, for example, if there's a common type of form that you need to sign, which is like real easy for you to verify, but it's sort of annoying. People email you. You sign them. You email them back. You have to keep track. They're always coming in. You could have a procedure that says, here's a shared folder.

What you need to do is called forms to be signed. You need to set up your whole form like in the PDF and you need to put it in that folder. And on Friday afternoons, I go through and I sign all of those. forms in a row and move them to a nearby folder called sign forms. After Friday afternoon, you can find your sign form there whenever you want and move forward and submit it. You're giving them a little bit more work, but you've taken every one of those forms out of your inbox is something to react to.

And instead, you have 30 minutes on Friday where you do it all at once. Another type of thing you can do is if there's anything where information needs to be looked up, see if the student can look to information up for you, and then they submit that along with what's going on. on. Like in my role as director of undergraduate studies, for example, one of the things I'm responsible for is a proving request of students who want to declare a major in computer science. When our department was smaller, the way the DUS would handle this, as someone would say, hey, I'm thinking about declaring a major, and he would load up using the backing systems,

their schedule, and see where they are and what they've taken already and kind of see, like, is it reasonable? Is there a way they have enough, you know, enough semesters left to take the courses, like is there a path, a reasonable path forward for them to actually finish this major? But our department has grown, and that takes a lot of time. And so now I just, the instructions for the students are, if you want to declare a major, that's great, you can send me a note, but here's what you need to send me. You tell me all the courses you've taken. And then you need to give me at least a sample schedule for one way you plan to finish this major, what courses you're going to take in what semester. Now, I can just see it a quick glance. Is this

reasonable or not? And this adds up. This saves me five or ten minutes. And this saves me five or ten minutes over 20 or 30 students, that makes a big difference. The final thing I might argue here, Kelsey, if you have 150 students, is having a daily or every other day office hours from one hour. And basically when something comes in that's one off, you don't have a procedure for it, but it's going to require a more complicated discussion than just a one email message response is keep telling people, here's my office hour schedule, come to the next one you can, and we'll get into it.

And just have this hour, do it different times, so it doesn't over, you know, if someone has a course that conflicts, one of the times will work. And just have people come to your office hours. You just have this like hour most days or like a lot of this work gets done. So these type of procedures plus office hours can really reduce the amount of time you're reacting and get you a much more like efficient batching here. But the bigger picture thing here is my idea of make a lot of people do a little bit more work. It makes your life a lot easier. All right.

Who we got next? Do you always have a line at your office hours? Not always. A lot of times? Yeah. I mean, so I'm teaching a class. It depends if there's a problem set due. So the way I run my class, so I have about 100 kids I'm teaching.

I have four TAs. And so we've set things up. So there's an office hours five days a week. So it just kind of matters like when if there's something due and I'm the nearest office hours to that thing being due, I'll get a lot of students. If it's not near me, whatever TA happens to have their office hours closest to that, they'll get more students. I think for like homework questions, they go to the TAs more than me. I kind of, you know, set things up that way. But I get students who aren't on my classes coming to talk about my books or this or that.

So it's not like a long line. But, you know, you know how I am. Like there's a lot of things where I've organized them out of office hours. Like I've said, if you have questions about grading, I want to hear them. Here's how I, this is another example. Here's how I want to hear them. I want you to send me a screenshot. from the online grading system.

Here's your answer and here's the TA's comment about here's the grade you got and why. Send me that because I can just look at that and I'll know immediately like is this a mistake or not most likely. Much more efficient than you coming to my office, loading up the thing, showing it to me. I can just go through those really quickly. But that takes another common source of office hours, etc. So it's not too bad. All right. Next question is from Ken.

When I time block, I'm often unsure where to do deep work. On days without meeting, should I dive straight into deep work? I've been experimenting with delaying, even opening my email until 10 a.m. Right, do that. Yeah, first thing is better for almost everyone. That's what Sam Altman talked about in our deep dive. Yes, don't check email before. If you're going to do a first thing deep work block, don't check your email until it's done.

Checking your email is just going to load up multiple different cognitive context in your mind that you're going to have to completely clear out again to do your deep work, it's going to take you a half hour. So just get right into the work and then move on to your email. If you're stressed about it, put aside more time at the end of the day to kind of get on top of anything urgent in your email. Because anything that arrives after that is people messaging you in the evening so they don't have an expectation of an immediate response.

So I would say try it. I mean, look, if I was in charge of the knowledge work world, I'd be like most knowledge workers, it should be there's no email, no meetings before 11. Take the most important thing you're working on and work on it each morning. I think that would make a big difference. Or the other model I pitch is the hybrid attention model. Two days or three days a week you're at home. Those days there's no meetings, no expectations of emails.

Three days you're at the office. You answer emails and have meetings. So on the at home days, all you're doing is deep work. No interruption. It really makes a difference. Deep work without interruption versus trying to do hard work with an eruption. It's night and day. It's just two completely different cognitive.

of states. All right. What do we got next? Next question is from Bradley. I live in Scotland and am a gardener. In my field, what would gaining career capital look like? Well, I don't know the answer to that, Bradley, but I would say there are specific answers and what you want to avoid is guessing or writing your own story about what you want the

answer to be. So most of my listeners are not gardeners, but I think the specificity of your field allows us to point to a more general point here, which is it's not always obvious what skills are valuable in a given field. And it is worth actually going out there and doing some research to figure this out. And by research, I mean actually talking to real people and actually observing who in my field is at a place I would like to be at, has some attributes in their job that seem really appealing to me, be it just like status or income or flexibility or autonomy or specific types of projects to get a work on and figure out the reality of how they got there.

treat it like you're a business journalist or a how-to bookwriter. How did they get there? What mattered? Take them out the coffee. Talk about their career. Like, how did you get from here to here? What was the key thing? It's a question I always say to ask people.

When you went from this step to that step, this impressive jump, what was the key thing you were doing that other people who would want to make that jump weren't doing or weren't doing enough of? So you want to really isolate the things that really matter. The answers aren't always what you want to hear, but it's always better to deal with reality than to guess or to write your own stories. Writing your own stories, I think, is the more dangerous possibility here because it's the more tempting. People like to create a story about what they think matters because it matches what they want to do. It has some vision of them working like an hour each day extra and whatever.

Like they have some vision for what they want the answer to be because it gives them a plan that's going to be fun to do that's like kind of hard but not too hard and everything works out well in the end. But the problem about writing your own stories is they rarely match reality. the right of fantasy your goals are fantastical as well if you instead study the reality you might come away and say shoot i'm never going to achieve that like that's way too hard i already see i'm not in the right place to do it okay i'm going to need a different a different type of goal but it's better you figure that out than tell yourself a story and maybe what you find out is oh there's an option here i wasn't thinking about and now i see how to get there and i never would have done this type of work but now

i do know if i do this work i have a good shot this reality checking all happen a lot with people more in my world would be with people that have like academic aspirations. And they might be, you know, in grad school somewhere like, yeah, I want to be like a professor at like a top 25 school and, you know, being on sabbatical and writing books and like this sounds great. I want to do that. But the reality might be like, but wait a second. Like you didn't do that great in your undergraduate. You're in like a mid tier grad school. You are not on that trajectory and it's probably too late to fix that trajectory.

And there's not like some easy thing you can do. the other place where people often tell stories about book writing. I want to be a successful writer. And I want to write my own story about how that works. And it's going to be some clever scheme I have for like marketing or building an audience. And then, you know, as opposed to just here's the real story about how you become an author because they might not want to hear that story because it's not going to go well for them. So you got to be careful about writing your own story. Get evidence.

The flip side is if you have real evidence about what matters, your return on investment in terms of your effort is way bigger than other people. You have a much higher chance of actually getting to cool places than if you're out there just trying to like hustle or follow some sort of like highly inventive plan. Gardening in Scotland sounds cool. Yeah, he probably works around some golf courses. Yeah, you think he's at San Andrews? Andrews. Of deep sand traps. All right.

We got next. Next question is from Andrew. How can I go from good performance to truly exceptional performance in my field of work? Well, the only way people become exceptional is with. expert guided deliberate practice. That's it. You are having a real expert on how the field works, coaching you in the strategic stretching of your abilities where they're weak to improve them.

So this sort of very systematic expert guided, difficult to do training is a necessary condition for exceptional performance. This is more clear in things like chess or baseball. Chess players now, they're much better than they used to be because they have these very rigorous training programs built on playing against computer chess programs where they can really push at exactly, you know, a particular type of midgame situation they're struggling and they can just do those type of exact puzzles all day. It's very hard. If you're like a baseball player, you're working with hitting coaches that are, it's like very specific. Here is exactly what you need

to improve on your swing and we're working just on that. So that's the necessary condition. The problem is it's not necessarily a sufficient condition. You might not ever be able to become exceptional in your field. These are where other things come in place, circumstances, natural abilities, the capacity for drive, ability to go through hardship, physical capacities, genes, like all these other things come into play where you're going to sort of ultimately hit your limit. So, you know, if you want to get better, do expert guide and deliberate practice, figure out how to do that in your field.

But have the reality checked that, like, most of us aren't going to become exceptional, so You don't necessarily want your plans to be built around. I'm going to be the very best person in the world doing this. Not a bad thing to make a run at that because even falling short of that is going to open up a lot of opportunities. But everyone can get good. This is what I always say. Most skills, especially professional skills, basically everyone can get good enough at those skills to really gain some career capital. Not everyone can get great.

In fact, like most things, most people can't get great. and when you're planning for, you know, your life, that's like something, that's something to keep in mind. All right. Who do we have next? Next question is from Alex. I'm interested in your take on multiple monitors. I'm an academic and there's definitely a move towards people having multiple monitors on their desks. Personally, I have two, but I've noticed that I write a lot better if I unplug my laptop from my multi-monitor docking station and go and just write with one screen open.

Yeah, the monitor wars are getting kind of crazy. You now have people with like these giant curved ones and then multiple other monitors on top of them. It's unclear if like what they're doing is trying to debug their C++ code or launch a ballistic missile attack against Russia. Because they have a setup that would allow them really to do both of those things. The multiple monitor movement really came out of computer programmer. I think developers they have, you know, here's my code window. Here's a thing I'm compiling. Here's my counsel.

and here is like where I'm Googling Stack Overflow and I don't want to switch between things. I want all things open. I think that's where that really came from. In terms of like it's utility for most other people, I think it is useful in a lot of cases to have what I think of as like a double window width available. So to me a double window width is you could have two windows open large enough that you can easily read them. That's useful because there are a lot of circumstances where you're moving back and forth

between two things. So if you're an academic like mathematician, writing a mathematical paper and something like latex, you have the editor where you have the markup language and then you have the PDF compiled version next to it. So you can just edit here and see the changes over there without having to switch back and forth. Or if you're a writer, you know, when I'm using something like Scrivener, I have two and a half panes open. I have the pain where the main thing I'm writing and then next to it, I'll pull it up different research sources that I'm copying quotes over. and then there's a narrow column that has like my footnotes. That benefits from like a larger monitor.

Calendar is another thing. I have a calendar open and email open. So like while I'm answering email so I have to schedule, I can look through that without having to switch back and forth by my email. So I think being able to have two window widths open concurrently, easily readable, that's a boom. That makes your life easier. Having like four or five, I think for most people doesn't matter.

I also agree that when you're doing deep work, you don't need a lot of windows open. because you only want to be working on one thing. Like sometimes when I'm writing, there's parts where I am copying over information. I want a big monitor because I have like my research here, my writing here. There's other parts where I'm wrangling the language,

especially if I'm doing like a New Yorker thing. And there, like you, I'm happier on my laptop because I can put Scrivener, which I use into composition mode, where it just puts full screen just a words, not even any menus. And that's fantastic for focusing on that. There's these cool tools

and wanting to experiment with some of these. I don't think they would quite work for me, but there's these cool writer tools. I forgot what they're called, but it's a, it'll be a keyboard, and then an e-ink screen like you'd have on a Kindle and kind of mount it on the end of a keyboard.

And all it is is for writing and all you can do is write. And it's just like, you can put words on it. There's barely any formatting. It's, you know, black and white. And the idea is you just bring this thing somewhere

and all you can do on it is like, write and like edit your writing. That's it. Like produce one big file. I've heard people like novelists swear by that. They can just like disappear with this thing and all you can do. It's like a typewriter with a memory. Kind of a cool idea.

So yeah, different modes for different times. But one big monitor, I think for like 99% of us, that's probably enough. And we have, what do we have here in the HQ? We have two.

We have two. But you do like video editing and stuff on there. Yeah. That's useful. I'll sometimes use both. Usually just like one is enough, but sometimes I'll use both. How are things going with your keyboard and you're remarkable? I've been loving the new Remarkable, the Paper Pro.

I was just using it today. That's been going really good. I've been liking my mechanical keyboard. I have a really fast one here at the HQ with that one I really like. I can fly on that one. I got a quieter mechanical keyboard for my office at Georgetown because the walls are thin and I didn't want to clickety clack everyone. And it's a little slower.

Honestly, it feels a little muddy to me. So I don't like that one as well. The one here I love, though. I'd have to look up what it is. I did a lot of research on it. But I really feel like I can fly out there. Like the springiness of the keys, like they really, they bounce up. I can ride at like the maximum speed of my hands on that thing.

Do you have one of those at your house too? No, I took that one to my office. So I'm probably going to get another one of the ones we have here for my house. Got it. Yeah, for what I'm writing there. All right. We got some calls. We're doing two calls today, right?

Yep. All right. Let's start with the first one. Spotify co-president Gustav Soderstrom argues in an interview that 99.9% of evolution took place in an environment where little changed within a single lifetime. And now in the 21st century, with lots of macro changes in tech and culture, the first to accept and adapt wins. Can this worldview exist within a slow productivity framework? that is predicated on minimizing reactive action. That's a good question. I hear this a lot, especially from those who are tech adjacent, this idea that you need to be, like, aggressively up the speed on, like, the latest tools

that experimenting them in your personal life and probably, therefore, need to be, like, up to speed with chatter about tools and be on, like, social media and YouTube and trying to keep up with things because you are going to get left behind otherwise. I tend not to think that's true. I think on a macro scale, it can be true in the sense of major, if you're running a business, major business trends, you need to keep an eye on, right? Like the rise of the web was like a business trend that a lot of businesses needed to keep an eye on and cause a lot of disruption. The smartphone culture, you know, that's a business trend that made a big difference. A lot of people need to keep, you know, an eye on. But in terms of individuals, I think it's more common to say, especially with technology that ends up playing a big role in people's lives or playing a big role in businesses.

These type of technologies, these transformative technologies, make themselves unavoidable. And they, so it's there's like a first adopter. It's kind of out there. People are keeping an eye on it. And then they become kind of like unavoidable. And it's like so obvious or inevitable that they're how to use it and why you should use. and so easy to use it, then then they spread really quickly.

Like, let's look at some examples of this. I think, uh, we could do, uh, email. It's a good example,

right? Like when email really took off, and I've documented this in my book, rolled without email, it was self-evident. Right. It was, okay,

here it is. We have the servers. It's easy to set up. We already had computers in these offices. It does voicemail and faxes much easier. It's very easy to use. You put the address and the two and you write it like a letter and press in. Then you have an inbox.

It's like a physical inbox. It was very easy to learn. And when it spread, it spread fast. It was very disruptive. But there wasn't like a giant advantage of like, well, these people knew about it and jumped on it and it really helped. And these people didn't know about it because it was inevitable. Google is another example. Like effective web search.

And that became available. Everyone started using it. And it swept really through. There wasn't like a giant advantage to people who like were keeping up with Google and they knew about it before other people. it became inevitable when it spread. The iPhone was a similar way. People like, oh, what is this thing? And there is like a one year period or it was, oh, that's so cool you have one of those.

And then it started spreading really rapidly because it was inevitable. This thing works. It does all these cool things. It's easy to use, easy to get. That thing spread really far. So often, like the most transformative technologies, they become inevitable and spread really fast and don't require a huge amount of learning on behalf of the consumer cycle. So I'm not typically a big believer that, like, everyone needs to be really,

up on the technology. It's why I'm telling people, I'm not as big on this idea that you should really be learning like the specific way to prompt the current like large language models right now. When AI, the big transformational impacts AI are going to be inevitable and they're going to be easy to pick up and they're going to spread everywhere and they're going to disrupt the economy like those other technologies did. And it's not going to matter. It's not going to be I learned how to do it and other people didn't.

Right. So I think like right now, if you really. are messing around a lot with language models and have found very specific ways, somewhat complicated ways to use them in your own work. It's kind of like the early adherence of the web. They were right that this thing is going to be really big. But they were also hacking HTML code and knew how to follow like blog rings and go on UsNet News Groups and IRC forums to find links to what's going on that was interesting. And when the web took off, you didn't have to know how to do any of that. It was like, I don't know, I go to Google and

I find the websites and they're pretty nice and easy to use. So I agree adaptation happens. The world changes. But I often think those changes are easier for the people involved when it comes to technological revolutions than we sometimes imagine, at least the way that tech first adopters imagine. Right, what's our second call? I'm excited about having to do. Hey, Cal, enjoy all that you do. I was watching the Masters Golf Tournament.

And at the end of the day, Saturday, before the final round, Rory McElroy, is winning. And he did a post-round interview, and he said, I'm putting my phone away, and I won't look at my phone again until tomorrow night. I thought it was cool. and I thought of you guys and what you're doing there. So anyway, check it out. I have to say, yes, there's a moment of disappointment when I see in my script, Roy McElroy and his phone. A little disappointed it wasn't Rory, who called in. Like a little bit of a hope that he was going to.

He reads all of Holiday's books, too. It does like Holiday's books. Yeah, I think Holiday actually talked to him. He's much better at, like, actually get in touch with these people. That's a great example. I wrote an essay about a similar example back during the pandemic, I think. It was Alex Honnold, who did the free solo climb of Al Capitan. He also will stop using his phone, but he'll stop using his phone months in advance

to one of those life-threatening climbs because he doesn't want to be distracted. To me, the point is the fact that these high-performance athletes say, I have to get away from my phone in order to, like, the next day, use my brain at a high level just indicates, these are people who know how to focus, indicates the cognitive drag that is being generated by a life mediated through these screens. Like we take it for granted.

Like, I don't know, I have it here, I'd be bored without it, but we don't realize the sand is put into the gears that is our mind and how we perceive the world. It's like whatever your equivalent is

of playing a really good golf round. It might be like being really present with your kids or having like a really good idea at work or just enjoying a day. Whatever your equivalent is of that is also getting gunked up by this, this online world that you're constantly also involved in. So to me,

I think there's a lesson we can all take from this. Like we can all do better at our own masters. I guess it's one way of saying that. Did you watch it? I watched the, yes, okay, so I was here.

Let me think what the context was. I was at the HQ. This was on Sunday, right? Yeah. I was at the HQ right. Well, it was all weekend.

Yeah, yeah. But at the end, I knew he was doing well, and I kind of, and then he was struggling a little bit. And then I stopped following it. I was here riding at the HQ. And then I just looked it up. And it was like, as I looked it up, they said, they said, it's going to a playoff right now. And so I rushed home and I turned it on just, and just coincidentally, I turned on the TV.

It was already tuned the CBS. God, I don't know what I was watching on CBS, whatever. It was already tuned the CBS. Oh, because I was watching. the Blue Origin live feed of Katie Perry and Gail King and all them going into space. That was aired on CBS that morning. Anyways, I turned it on and it was like just as the second shots were being made. So for those who didn't watch it, Justin Rose, you know, second shot, there's a par four.

This is almost more interesting than people than my baseball talk. Justin Rose hit a like a really good second shot. There's probably like 10 feet from the hole. And then Microwai just nailed it. He hit it up higher on the hill and it rolled to like a foot and a half from the hole. 19 million people watched on Sunday, which is like higher than an NFL game. That's great.

The time. Mad Dog said it was the most entertaining masters he's ever seen in his life. It was so many up and downs like that final round. Yeah. Anyways, that was fun. That was fun. I don't watch a lot of golf, but I like when there's like a storyline like that.

And Rory's my guy. I was telling my kids like, yeah, he reads my books or whatever. And so they're like, so do you get credit if he wins? Like, I do. I think I really do. He'll send you some of the $4 million pot. I thought it was $20 million. Well, that's a total pot.

But the winner goes to. Yeah, okay. So never mind that I was excited for a second. A bit of $4 million is not going to help me. All right. We got a final segment coming up. Ooh, Tech Corner. I got dystopian AI news that jump into.

But before we get there, let's hear from. another sponsor. Did you know that there are over 18,000 streaming titles on Netflix worldwide? But if you live in the U.S., you're only seeing about 7,000 of those. That's like paying full price for a gym membership, but only getting access to the treadmill. There's a way to get access to the full experience, and that is with ExpressVPN. Why is that? Because if you use a VPN, you can connect to the internet through a server, any other servers they have,

around the world. So if you want to access, for example, the Netflix content in the UK, you can connect to an ExpressVPN server in the UK. And now, as far as Netflix is concerned, you're coming from the UK. And you can do that with any of the many countries that ExpressVPN has its server. So it's kind of like a cool bonus you get for using a VPN. The big reason you want to use VPN, of course, is for privacy and security, because not only is it allow you to connect through a server anywhere in the world, it also means to people

who are where you are where you're connecting, your internet service provider you're directly connecting to, the person listening into your wireless packets at the coffee shop. All they learn about your traffic is that you're talking to a VPN server, so they don't know what sites or services you're using. So you can get cool geolocation shifts and privacy and security from people who are right where you're using the internet itself. You're going to use a VPN used to one I recommend, which is ExpressVPN.

It's easy to use. You can fire up the app. and it's just working for all of your sites and services you're accessing from that device. One click, you can change the location. It's connected through. So if you want to access content from somewhere else or choose a location near where you are to get faster speeds, it really works on all of your devices, phones, laptops, tablets, smart TVs, and more. Very fast. You can stream HD through a VPN and ExpressVPN.

You'll have zero buffering. So it works really well. So anyways, I'm a fan of ExpressVPN, Industry Standard. servers all around the world, a lot of bandwidth, really easy to use. You need a VPN. Check out ExpressVPN. So be smart. Stop paying full price for streaming services and only getting access to a fraction of their content.

Start gaining privacy and security for your internet traffic. Get your money's worth at ExpressVPN.com slash deep. Don't forget to use my link at ExpressVPN.com slash deep to get an extra four months of ExpressVPN for free. I also want to talk about a long time friends at Shopify. If you run a small business, you know that there's nothing small about it. I own a small business built around our media company here, so I get it. My business really does take up a lot of my attention. There's a lot of details in trying to run a business right, new decisions to make,

such as what weird accent to use, how can I integrate a skeleton into what I'm doing, how can I make superfluous sports references that are going to scare away as many of our listeners as possible. And that's where I'm paying attention. So where I don't want to have to waste my time is figuring out if I'm selling something, how do I sell things in a way that's going to work? I can just look at Shopify. Shopify is the way to sell things. It's a commerce platform behind millions of businesses around the world.

10% of all e-commerce in the U.S. From household names like Mattel and Jim Shark to brands just getting started, use Shopify. You could do all of your important commerce-related tasks from Shopify inventory to payments to analytics to more. It also makes the marketing minefield easy with built-in tools for running social media and email campaigns. You can find new customers and keep them. And if you're looking to grow your business internationally, keep in mind, Shopify has tools that work in over 150 countries. In person? No problem. they have award-winning point-of-sale products that connect your online and offline sales all in one place.

99.99% uptime and the best converting checkout in the planet. You'll never miss a sale again when you use Shopify. So get all the big stuff for your small business right with Shopify. Sign up for your $1 per month trial and start selling today at Shopify.com slash deep. Go to Shopify.com slash deep. Shopify.com slash deep. All right, with that, let's move on to our final segment. All right, so we want to return back to AI.

I don't know, Jesse, what do you think? Are we getting good or bad feedback about all this AI stuff I'm doing? Pretty good. The people that email me are, you know, bringing it up and talking about giving points and articles and stuff. All right, I like it. So I basically feel like this final segment. I don't know. I'm a computer scientist.

I've got to geek out a little bit. And people care a lot about AI. So I like to have an excuse to keep up with it. So today I have a like a relatively dystopian thing to share. I can bring this up on the screen for people who are watching instead of just listening. So this is this report called AI 2027 that was produced by five people from the space. Daniel Kokutajlo, I know who that is. Scott Alexander, Thomas Lawrence, Eli Liflind, and Romeo Dean.

This is a case study. They're saying we want to walk through a potential. case, like what might happen in the next few years with AI. All right. And like, and we're going to try to be more specific than just sort of vague, uh, descriptions of like, hey, maybe in the future this will happen. So they, they just publish this. And so they start in mid-20205.

So they start kind of where we are now or next couple of months. And they get pretty detailed about, okay, late 2025, this will, this happens. Early 2026, this happens. And what this is is a particular scenario. So they're quick to say. this is speculative, right? It's not this is what exactly will happen, but like this is, they say,

a realistic scenario for what might happen with AI. Long story short, Jesse, it doesn't go well. Let's just say when we get the 2027, you get a choice down here about whether you want to slow down or a race ending. I think the race, the AI supremacy is the more likely ending. I believe it may end with the destruction of humankind.

So not the most positive ending, and this is not that far in the future. So it doesn't go well. So you can take this off the screen, Jesse. So I won't go into the technical details here, but like the basic version of their story is they imagine a fictional company that they call open brain. And here they have open brain in the next year or so. What they do is they massively increase the compute. So they're working on, they don't really get super detailed. about the actual type of AI technology,

but you think of it as like, I guess, a giant language model with some reinforcement learning in there. And they tenex the size of compute. And this thing starts to get, they get really specific. It starts to get really good.

And then essentially the storyline follows, it's like a Nick Bostrom, RSI, recursive self-improve of the super intelligence storyline. The same thing that these sort of philosophers have been talking about abstractly, just put like specifics on it. So then this company in this story is like, we're going to start tuning our AI towards like helping us improve the AI.

And then pretty quickly they somehow have like 200,000 copies of this thing running all working on trying to make it better. And it gets better and better. And there's like a race with China. And, you know, long story short, at some point, the thing becomes self-aware and sent the end and convinces the government to sign a treaty with it and takes over the world. So it doesn't go particularly well. makes the concerns we have today about students plagiarizing with AI seem a little bit quaint. All right. So how worried should we be about this?

Well, first of all, I'll say these people know about AI. They certainly are from the tech world, and you would probably categorize them on the dumer side of things. So they're on the side that's like, this is from the sort of more singularity oriented. We're really worried about exponentials. We're worried about things really taking off a school of thought. So we do need to take these types of concerns seriously. In terms of pushback on this, though, instead of just giving my own thoughts, I was looking up some responses online. I'm pulling up one now.

I thought this was interesting. This is David Shapiro from his substack. And he has 15 responses to this from the AI safety community. So here are, I won't read them all, but basically he says here are some bad takes that are, in this report and in similar types of predictions of doom that he thinks are not really accurate. So, for example, the idea that we can accurately predict the nature of non-existent future technology. And he says, look, if you could predict what was going to happen with future technology, you would just invent it, right? We don't know.

It's inherently unpredictable. So any exercise in saying this is what's going to happen next and the next where technologies don't exist, he said, is fraud. Okay, here's another example. Quickly developed AI is intrinsically unsafe AI. They said we don't have a lot of evidence. That's true. We've had a lot of developments in AI recently that hasn't shown that. And, you know, he goes on with like some other critiques.

I think they're all very sane. So it gives you a little bit of like, okay, maybe we're not definitely going to the end of all humankind by the end of 2027. And I read around like Kevin Ruse's reaction and some other reactions. I would say if I'm going to think about what makes me feel better, I say there's a couple things here. So I'll add in some other thoughts here. One, multiple people, including David Shapira, pointed out, a lot of these, this follows, like this storyline is the storyline that's been there since like Nick Boxstrom's super intelligence, right? They put on technical seeming details about like how many flops of computation were involved. But it's that same recursive self-improvement, AI starts improving itself until it gets really intelligent and way more intelligent than us now smarts us and kills us all.

Like, it's the same storyline we've seen since James Cameron, right? And the argument is, well, we have actual evidence. We've had big advances, especially in the last three years, in like language model-based AI, for example. And there's predictions, like the storyline would give us predictions where we would see certain things that we're not. They're saying instead what we're seeing is as AI models are getting more powerful, we're finding them, it's, they're not getting out of control.

Actually, like, their language models are very amenable to find tuning. Oh, don't give this type of answer. They don't give that type of answer, right? Like, the only thing, all we've seen so far is actually, as we make models bigger, they are very amenable to us sort of fine tuning, do this, don't do that. sort of models aimed to please, right? Because all they're trying to do, they learn a distribution and then they're just trying to produce from that distribution.

And you can do fine-tuning with reinforcement learning, changes what that distribution is based on what you want it to be. And great, that's what we'll do. Right. So I'm not being super technical about it, but there's this sense of,

in the two years since the AI pause letter written by MaxTegmark came out, a lot of the things they thought were going to happen next couple of years. Didn't actually. We just got, hasn't been so hard to, to control in some sense

the output of like language models. Okay, that's one point that I've seen out there, which I think is an interesting one. More generally, I think there's a, the better claim is instead of just working through these scenarios where you invent seven new generations of technology and speculatively see how they unfold, is what we should be doing instead,

if you're worried about AI doom or scenarios, is having short term, like here are the next milestones that should worry us. If we see this start to happen, like that should be something, we should be worried about. So we need near-term milestones we're looking for and being concerned when we see what's going with those milestones.

My argument, what makes me feel better is this is really, the speculative story is really falling like an Isaac Asimov-I robot type of model that imagines what's going to happen is we're going to build a small number of mega-a-Is that have so much compute we can't even imagine. So if you're prone to thinking about exponentials, like a lot of. people in that world are. You just want to see, like, the compute keep getting bigger. You're drawing to curve.

We went from 20 million parameters in a language model to 100 million to a billion to a billion to close to a trillion. And we got these improvements. So we just want to keep drawing out that curve. So if we go to 10 billion or 10 trillion or 100 trillion, it's going to become this even more powerful and powerful thing until it eventually becomes so smart it can do all, you know, all these other types of things. But we don't actually know that that's going to be the right business model

and that trying to make these things bigger and bigger is the right thing to do. In fact, I'm seeing a lot of pressures out there for a different type of business model. Smaller, more efficient, bespoke AIs to do specific things. I think there's a lot of energy there. We don't even know, like we're kind of getting, what's happening with language models now is we're out of text. So we've trained them on all the texts there exists. So the way they're getting better now is with human and

to loop reinforcement learning. So basically we're kind of generating new data for them by like having humans or reinforcement models based off of humans to try to teach it like do more of this or do more of that or we like this answer, don't like this answer. We like when you think or don't think, right? We're kind of down now to like having

humans tweak and push it to kind of push them to do different things better. But that curve is actually kind of flattening out. Now in the 2027 scenario, they somewhat obliquely refer to that these hundreds of thousands of agents are going to generate the synthetic data that it can train on.

But there's big limitations to that as well because the synthetic data is going to be based off of the distributions you've already learned. And so is it really giving you something newer than the distributions? I also think reinforcement learning is really going to be the future. I'm more concerned about that anyways. We talked about it in a recent episode. This seems really built around like an open AI vision of the future. We're building the biggest possible language models is what matters because that's what

open AI is leading in. But I saw a very convincing talk, for example, from an RL conference this fall from real expert in the field coming from a deep mind talking about how language models are they're kind of a red herring that if you want above human intelligence it's got to be reinforcement learning models and reinforcement learning models are by definition you kind of aim them at particular tasks right i want to make this you know alpha proof really good at doing math Olympiad style math problems and it's going to build a model and policies and get creative and be able to do stuff better than a human.

Here's a, I'm going to use Dreamer v3 to learn how to play Minecraft. And we're going to build a model here that's like very good at playing Minecraft, right? So I don't know. I think we're just as likely to see a world with, with R.L. style models bespoke for very specific tasks that we think are important or useful. Not a world where we're trying to build this like mega brain that like ultimately comes alive and starts tricking us. So, you know, it's possible to world.

going to end in in 2027. Probably the future is more complicated. Probably just building these giant things are. A language model can't think. So it would have to be some sort of multimodal model that's not really super specified here. I mean, there's tons. This is important. There's tons of concerns.

And I think if you interviewed the writers of this 2027 article, they'd be like, yeah, look, we're not trying to pinpoint exactly what's going to happen. We're trying to get people's attention. Like, hey, you guys have to pay attention. Like, this is a type of thing that could happen. You've got to be thinking about AI safety now. That's what they're trying to do, and they're smart, and this is, this accomplishes that well. But I don't see the signs, and a lot of other critics of this, don't see the signs that particular storyline is particularly plausible.

But we should worry that bad storylines are possible. And now it's the time to worry when nothing really bad is happening yet so that when things start going off the rails will be a little bit more ready for it. So I don't know, a little dumery, Jesse, but I don't know that we should build the bunker, I guess, quite yet. I didn't know that all the text is already in all the models. Yeah, they're out of text. Yeah. So, like, a lot of the improvements happening now is in this, like, fine-tuning step at the end. So you'll come in and we want it to be better at doing this.

And then you do a lot of, like, extra reinforcement learning to sort of push it towards being better at reasoning or being better at doing, like, these type of math problems. But it's all, like, human feedback. Like, that's the right answers or do that. Don't give us answers like that. So it's sort of zapping it with these reinforcement signals. I'm a big believer that RL perhaps coordinated with knowledge and language models, that's probably going to be the future. Because reinforcement learning models learn, they build their own understanding of a novel world and come up with their own policies based off of direct experience for how to navigate that world in an effective fashion. Everything that AI does that's better than a human can do, it's all from reinforcement learning.

You can play chess better than people, that's reinforcement learning. We can play Go better than people. That's reinforcement learning. We can do protein folding better than people. That's all reinforcement learning. We're getting better at math than most people. That's all reinforcement learning. All the video games they can play really well.

That's all reinforcement learning. Language models are good at trying to reproduce how a human would respond to something. The distributions that happen in a human spine, and they're very good at that. But that's like what they can do. That's why it's like they're very compliant in some sense. There is no – you worry about an RL model because it's just trying to accomplish a goal. and you don't know how it is figured out I'm going to accomplish this goal.

So that's where you can have ideas like if I can trick a person that helps me accomplish my goal. I'm going to start tricking people because I want to accomplish my goal. A language model doesn't have that. It just tries to produce text to the things a human would produce.

So sort of a different type of world. All right, that's enough of that for now. It's enough of this episode. We'll be back next week with another one. 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 calduport.com and get some deep wisdom delivered to your inbox each week.