So I recently heard a term that I really liked on the Chris Williamson podcast. The term he used was productivity rain dance. So I think this idea actually gets at something critical, a critical concept about work and productivity and technology. It's a concept that I think is exploring because it might explain some of both your frustrations and confusions about trying to produce stuff that matters.

So what we'll do is I'm going to load up the clip. So we'll hear Chris actually talking about it himself. Then we'll get into a little more detail what he's talking about, why I think it's a problem, and then give some advice for how to get around it. All right, so let's start with the clip.

This comes from, I think he was interviewing, Chris Williamson was interviewing Sahil Bloom, but the person talking here in the clip is Chris himself. So let's load up that audio. Look, I come from a productivity background. When I first started the show, I was chatting shit about Pomodoro timers and Notion external brains and Ebbinghaus forgetting curves and all of that.

I've been through the ringer, which is why I'm allowed to say it. And you realize after a while that it ends up being this weird superstitious rain dance you're doing, this sort of odd sort of productivity rain dance in the desperate hope that later that day you're going to get something done.

And some of that stuff does really help. And you kind of need to go through this process of, "Ah, it wasn't the 15 push-ups before I do my calls. Oh, it wasn't, it was this thing." That's the highest point of learning. All right, so that's the clip I wanted to point out there.

That studio, I've been there, Jesse. That's one of the studios Chris uses in Austin. I did a show last spring. It has the two weird fluorescent lights that look like they're coming out of you. So I love that idea, this notion of productivity rain dance. So I started looking into it.

Chris has talked about it before. So then I found this post, which I'll load up on the screen here for people who are watching instead of just listening. This post is from last summer. And he elaborates more on this idea. I'm going to read some from this post because I think it's going to help us get closer to what the key idea is here.

So in this post, Chris says, "During my interview review, I asked myself two new questions. What do I do that I think is productive but isn't? What do I do that I don't think is productive but actually is? These were surprisingly easy to work out. Sitting at my desk when I'm not working, being on calls with no actual objective, keeping Slack notifications at zero, sitting on email trying to get the unread number down, saying yes to a random dinner when someone is coming through town, organizing meetups with friends from different social groups, walking around without anything in my ears, reading, visiting new places.

After six months of reflecting on my answers, I realized I had a fundamental oversight. I hadn't been properly linking inputs to outcomes. I had basically created a productivity rain dance." All right, I think this elaboration is useful because if we think about the various examples that Chris gave here talking about these rain dances, they seem sort of different, right?

We had on one hand him talking about things like external brains or Ebenhaus forgetting curves, these sort of overly wrought productivity systems or tools you're trying to build. But here he's also talking about things like just being in his email inbox too long or like trying to get the Slacks all cleared out or just sitting at his desk like sort of acting, like playing at working because it sort of feels productive.

He is somehow unifying all of that under the same term of productivity rain dance. And the way he does it and the way I want to focus on here is by saying what unifies all those examples is focusing on input instead of output. So when you focus on input, you're focusing on activity, you're focusing on the potential for future activity.

So sitting at your desk checking email, you're doing something. The productivity, I'm putting lots of inputs into my productivity equation, that must be good. When you're doing something like building a complicated productivity system, again, this is like inputs. I'm working on my productivity function. I'm building up my opportunities for finding information or getting things done.

But what you're not looking at is the output. What am I actually producing that matters? And am I producing enough of it? So when you look around the modern office environment and see everyone frantically answering emails as they jump on and off Zoom meetings or watch the solo entrepreneur lose a morning to optimizing their chat GPT-powered personalized assistant, you're observing rain dances.

Everyone's busy, but no one is asking if all these gyrations are actually opening the clouds. So the danger with rain dances is that they are easier than the actual work of producing stuff. It's easier to sit at your desk and jump on calls or answer emails than it is to actually do something hard.

It's easier to work on your chat GPT-powered assistant. It's kind of fun. You're watching YouTube videos and scripting things together. That's easier than actually writing the thing that the assistant was going to help you with. And because it's easier, we tend to gravitate towards that. We want to spend much more time on it.

We're going to gravitate towards what's easier versus what's harder if we're not differentiating between inputs and outputs. And soon you get yourself to a point where you're super busy, but very little is actually getting done. Here's a quote from William Sim on this. "Obsessing over process while being detached from outcomes gives you all the pain of hard work with none of the actual results." So this is what a productivity rain dance is.

It's when you focus on inputs and ignore outputs. And the reason why they're dangerous is that they're more fun and satisfying or easy to get started with in the moment than the actual hard work. And so paradoxically, the stuff you produce gets worse. So what is the right response to this?

Because I think a lot of people feel the frustration of their productivity rain dances without realizing what that frustration is coming from. And there's two responses that I think are natural that I both think are flawed. We hear these often, though, in sort of online discussions of productivity culture.

The first response is to just demonize work itself. Like, I don't know, I'm doing stuff all the time. I'm feeling exhausted. Not much is being produced. Maybe the problem is like work itself is kind of meaningless. It's constructed. It's like a mirage of late-stage capitalism. It's all just hustle culture.

So you could just try to demonize work itself and lean into some sort of like quiet quitting type mentality. But this is not going to solve your problem. Just doing your work worse or just doing overall a lot less work is going to just get you into other types of problems.

Your business is going to falter. Your boss is going to move on from you. The other common response to this is to say, ah, the problem here is just thinking about productivity itself. So let's just get rid of all attempts to organize my efforts. That's the problem. But this becomes a problem, too, because if you have no attempts to organize your efforts, other people organize them for you.

Your life will just become this exhausted slurry of answering other people's requests and trying to make other people's lives easier. So we still need some organization. So what is the right response to the productivity rain dance phenomenon? I think it is, as Williamson suggests, to turn your attention from inputs to outputs.

To identify the most valuable thing you do in your job and then figure out what helps you do that better. And this should be what matters. Above all else, this should be where your focus is. Now, here's the thing. The answers to these questions, like what really matters and what is actually helping me do that thing better, you know, at a higher level of quality and output, the answers to those questions aren't necessarily simple.

But they're also different. The things you come up here are different than what productivity rain dances will produce. They're different because they're not symbolic. They're different because they're not busyness for the sake of busyness. They're instead focused on clear, measurable goals of producing more results that matter. So when you start focusing on outputs, you begin to build an approach to productivity that just works and it's not exciting and it's hard and it's simple and it's probably, you know, there's only so much YouTube content you could get out of it, but it's the stuff that actually works.

Hey, it's Cal. I wanted to interrupt briefly to say that if you're enjoying this video, then you need to check out my new book, Slow Productivity, The Lost Art of Accomplishment Without Burnout. This is like the Bible for most of the ideas we talk about here in these videos.

You can get a free excerpt at calnewport.com/slow. I know you're going to like it. Check it out. Now, let's get back to the video. So let me give you a couple of specific examples here. What are the types of less flashy, get-it-done type things that show up when you start asking what actually gets the important stuff done?

You find things like work quotas. All right, for each of these type of things I am expected to do in my job, I have a quota on how many I do at the same time. Why? Because if I have too many things going on, I get overloaded. If I get overloaded, I can't do the hard stuff well.

I want to do the hard stuff well, so I only take on one project of this at a time. I only do this many committees at a time. I only do this many pay-per-views at a time. It's not exciting. There's no AI involved. There's no cool notebook and a new pen involved, but it works.

Separating active versus waiting projects. This also works. A big concept from my book Slow Productivity. Here's the things you have to do. Just look at the first few and say I'm actively working on these and everything else is in a waiting state. And if you ask me about it, I'll say it's in a waiting state and I'll tell you as soon as it gets to active.

But as long as it's in a waiting state, I'm not doing meetings or emails or calls about it. This prevents the totality of the things that you have to do generating concurrent administrative overhead. Now suddenly you can spend more time working on the stuff that matters. Again, this is not an exciting system.

I can't build a software tool that's going to make this way simple for you. This is just a notation in your notebook. This is active. This is waiting. You can keep track of this even in your head, but it works, right? Rubber to the road works. Office hours work.

I can't be chiming in on a dozen ongoing back-and-forth ad hoc conversations throughout the day because then I can't get the important stuff done. So I have daily office hours. That's where I try to deflect more of my back-and-forth interaction. Come to my office hours. We'll talk about it.

It'll take five minutes. And the rest of my day, I don't have to be checking this inbox. Again, I don't need a special tool for office hours. It's just a declaration. It's like when Michael Scott in "The Office" just said, "I declare bankruptcy." It really is that simple, but it works.

Another type of idea that comes here, time block planning. It's not sexy. You can do it on a piece of paper. I sell a notebook for doing this. It's the oldest of technologies. You're drawing boxes on paper. But it forces you to be intentional about your time. What do I want to do today?

When am I going to do it? And if things don't fit, you have to confront that productivity dragon and say, "There's a problem here we need to fix." Again, not exciting, but it works. Perhaps the most basic thing here, the idea that comes from my sort of original book on all of this deep work, realize that deep work is different than shallow work.

And when you're working on deep work, the stuff that's cognitively demanding that matters, don't context switch during it. You have to protect that time and say, "I don't also check my email during that time. I don't also have Slack open during that time. I realize if I give this work my full attention, it'll be 2x better than if I'm sort of context switching back and forth." Like this, that's not a thing you can get a cool app for.

There is no like back-end reasoning model integration with Claude. It's just a simple mindset. But these are the type of things that work when you get away from productivity rain dances and say, "What moves the needle on the output?" Now the people who sell tools or have a lot of fancy YouTube videos about this advice online, they like productivity rain dances because they're fun.

You get to wear fancy costumes and jump around and chant things and it's interesting to watch and it's much more interesting to watch than the actual pretty boring efforts that go into trying to do things well. But for you who cares about doing things well, this is what matters.

So I wrote an essay about this recently. It's on my newsletter. If you don't subscribe to my newsletter, you can do so at calnewport.com. Here is how I ended that essay. "Productivity rain dances can be satisfying. They make you feel like you're doing your part to support a rich harvest, all while providing endless details and rituals to adjust, giving you a sense of being hard at work without requiring you to do anything actually challenging.

At the same time, however, the farmers who are most likely to succeed are those who are instead down among their crops, sweat on their brow, tilling their fields." So it's not the fun thing to do, but it's the hard things that actually produce the good work. So there we go.

Congratulations, Chris Williamson. I like that term. Productivity rain dances. We should use that more. Yeah. The thing that like I was struggling with and then when I found more of his writing on it, it made more sense to me, was unifying sort of productivity prong culture where you're building these elaborate productivity systems with busy work.

Like being in your inbox all the time. Where it's not about a fancy system, but you're just being busy. And the idea that that's all the same thing is just focusing on inputs as opposed to like what's actually being produced. Because once you focus on what's being produced, you realize both that the super fancy AI-powered productivity system isn't moving the needle, takes a lot of time, but it's not making you produce more work, really.

And also being in your inbox all the time is not making you produce more work either. Like they're both, output is not affected by them. When you focus on the output, but you know, it's not as fun, but like that's how stuff gets done. That would be a good book, I guess.

Just like do work. So for you it would be writing, right? Yeah, you got to write. Yeah. Yeah. And there's like stuff that you want to have the right tools. You know, time-blocking matters like you're doing at the right time. Maybe you have a little ritual to help switch your mindset over to it.

But you do that thinking once, it's not that exciting and then it's just doing it. Right. I think that gym, like exercising is the same way. Like you need to have tools. You need to learn like what is my workout? Like why is this going to work? Like you do need information.

You can't just like randomly go after it. But then once you have the information, it's not that interesting. It's just a matter of like going to the gym and doing the workout and tracking it and doing proper progressions. Like it's actually in the end, it's the work that matters, not the information.

So productivity rain dances, we should avoid them. We got some good questions coming up. But first, let's hear from one of our sponsors. I want to talk about the Uplift Desk. This is a topic that is near and dear to my current state. As listeners know, I'm rehabbing from an abdominal injury I had in the fall.

I ended up having to spend about two months kind of barely using my abs or my back. And now in the new year, my, you know, back is making me pay the price. Like you can't just stop using me for two months. So I'm doing all sorts of PT, all sorts of training, and one of the things you realize right away when you're recovering from something like this is that little things about your posture matter.

How you hold yourself matters. Are you slumped a little bit? Are you back more? Are you on your heels versus your midsole in terms of what muscle supporting want? It really makes a difference. And where is the place where we're sitting or our posture is probably most stably impacting us?

It's going to be when we work. This is why a tool like the Uplift Desk makes so much sense because it's at the forefront of ergonomic solutions. It helps you promote better posture and health through adjustable standing desks that are designed to help you live a healthier lifestyle. They also have all kinds of accessories to help keep you moving throughout your day.

So for example, we got the, it's like a wobble stool where you can sit on this thing, but it moves. So you could be moving and working your core and supporting yourself, but it's not going to fall over, but it moves through like a good range of motion. I also got the standing pad.

So if you're working out the Uplift Desk or it's in general like standing, so you're kind of giving your body different varieties of posture. It's more comfortable than just standing on the hard floor. So it's not just the Uplift Desk. It's the whole line of products that surround healthy posture and healthy ergonomics.

So I'm fans of what they're doing. I now appreciate posture quite a bit. So the Uplift Desk in particular this is cool. 200,000 configurations. So you can tailor your workspace perfectly to your style and needs. They also look great. Talked about this last time that the form factor is like very compact and small enough.

It's like a good-looking desk and the lifts are sort of hidden in the legs in a way that it's like not some giant contraption that you have to like turn a giant crank on. Like they look really nice as well. Anyways, care about your posture. You'll learn the hard way otherwise that you should have.

So make this year yours by going to upliftdesk.com/deep and use the code DEEP to get four free accessories, free same-day shipping, free returns, and an industry-leading 15-year warranty that covers your entire desk and an extra discount off your entire order. That's U-P-L-I-F-T-D-E-S-K.com/desk for a special offer and it's only available at that link.

So start 2025 right, stand, move, thrive with Uplift Desk. So Jesse, I talked to some of those guys over there prepping for this ad. They're really into this mission. Like they care about ergonomics. Yeah. Yeah. You cannot work at that company and be like completely slunched over in like a metal folding chair at your desk.

Like it's not going to work out. They'd run you right out of there. I also want to talk about our friends at My Body Tutor. Speaking of trying to do what your body needs, My Body Tutor, I believe is the way if you want to get fitter, it's the thing to do.

Their model, that's what I love about it, is they connect you to a coach that holds you accountable daily. So the coach helps you figure out like what are your goals? Let's get you a workout routine that works with what you have available. Let's talk about your eating and health requirements.

Like let's try to make a plan that makes sense. But you check in using their app every day with a dedicated coach. It's the accountability here that matters. The information is great. They can customize the information to like what's going on in your life. That's great. But it's the accountability of checking in every day that gets you to actually stick with the plan.

And because it's delivered online, it's going to be significantly cheaper than having, you know, a trainer in person, a nutritionist in person. So I just think it's a fantastic idea and it works really well. If you're serious about getting fit, Adam will give Deep Questions listeners. Did I just say who Adam was?

Adam Gilbert. I should mention I know the founder of My Body Tutor. I've known him forever. Great company. Adam, the founder, great guy, is giving Deep Questions listeners $50 off their first month. If you just mentioned this podcast when you join. They're also, that's the same guys who are doing the done daily we've been talking about.

So like the accountability coaching for like getting your, being productive about the work that matters. I'll throw that in as like a bonus ad. Check that out at donedaily.com. But My Body Tutor, if you want to get your body in shape, go to mybodytutor.com T-U-T-O-R. Mention Deep Questions and Adam will give you $50 off.

All right, Jesse. Let's get started with the questions. First question is from Lindsay. I currently use a Trello column for items to discuss in meetings and block time afterwards to process it. During a meeting, do you write notes of topics to remember to say or do you have a Trello card open?

All right, it's a good question. Let's just briefly define the two things that Lindsay mentioned. So one is the items to discuss Trello column. I use Trello boards to organize my obligations, to organize tasks. I have one board per role and then I have columns per each role for different types of statuses for these tasks.

And one of the key statuses is like to discuss. To discuss next time I meet this person. To discuss next time we have this standing meeting. The other thing she's talking about is the post-meeting processing block. Another thing I'm a big believer in. You need 15 minutes on your calendar.

So it's protected. You schedule a meeting, you schedule 15 minutes later than the actual meeting time. And that extra 15 minutes is for processing everything that came out of the meeting. So you can close those loops right then. The worst thing you can do is come out of a meeting with a dozen open loops that were introduced in that meeting and go right into an unrelated new meeting.

That's going to be a cognitive pile-up. That's going to be pain. All right, so to answer your question, Lindsay, if the culture of the meeting is one where people are on laptops, then I'll work right with the Trello board. As you say, it's much easier. Like, if things come up, I want to just put them directly into a card, onto the right board, into the right column.

So they're just there. I also want to edit the things that are already there. As you mentioned, if there's something to discuss in this meeting on my Trello board and we discuss it, I want to take it off right then. If there is not a culture of being on your laptop, that's when you use the post-meeting processing block to immediately take care of all of this.

So you can take notes with paper. And then when you get to that end of the meeting processing block, you go through and update your Trello. You send off the messages. You put the waiting to hear backs for those messages. But you really, one way or the other, here's your goal.

You want to come out of that meeting and post-meeting block, that sort of contiguous block of time, with nothing open in your head. No new open loop generated by that meeting that is not being processed. You want to make sure that the meeting is not just discussing with other people a given topic, but you're processing everything that was discussed to the point where you don't have to keep anything open in your head.

So however you need to do that, that really should be your goal. All right, who do we got next? Next question is from Jennifer. To what standard are various AI technologies being evaluated for current performance and future potential? How could a layman, not a computer scientist, test the veracity of the claims of an AI system or its potential?

My theory on this is that you don't need to be testing the veracity or really paying that much attention yet to claims about AI. And here's why I say this, and this is a claim that I think is more generally about technology tools. When a technology tool finds a killer app, you'll find out about it.

It will show up in your life. Everyone will be using it for a way that has obvious value. You will see an obvious value in your life. It will come into wherever you're doing your work and then that will now be a part of your life. You don't have to go out and keep up with it.

Let me give you some examples of this. Email. No one in the late 1980s was going around being like, "Look, computer networks are becoming a bigger thing. We're going to be able to do probably low-cost, low-friction digital communication using these networks." You really need to keep up with what might happen.

You really need to keep up with the advances and the technologies and the way this technology might change your work. There was no need to do that. What just happened is, once that technology hit the right elements, it just starts showing up in offices. They're like, "Oh, by the way, we're using email now because like all these other companies are." You say, "How does this work?" And they say, "You put the person's name in two and it sends them the message and it's in an inbox." Like, "I got it.

It took me nine seconds." And then you're using email. Google was the same way. You didn't have to be, most people did not have to be following search engine technology development and really trying to closely understand like what is the potential for the web for information development. Most people in their day-to-day life, there's just this point where they said, "Hey, use this tool to look things up." You're like, "Wow, that works.

I type in the thing. It figures out what I mean and it points me towards pages that have information about it. That's interesting." And you knew how to use it immediately. You didn't have to be preparing for it. I would apply that same standard to AI. I think there will be AI killer apps.

I think they are probably going to be industry specific. This has been my sort of developing claim is that different niches of the economy are going to have their own AI-enabled killer apps. When they come, they'll be inevitable. They'll be unavoidable and they'll take you nine seconds to learn.

So I would not worry about... There's a lot of hype, like attention economy hype behind AI. It's not hype in the sense that the AI is going to generate these killer apps, but the coverage is hype-y right now because they don't really have something to write about yet. This is like the so-called outcome or application gap.

The technology has defied expectations. Like it keeps everything you say it can't do. Six months later, it can do. But the actual impact in real jobs and the real economy has underperformed expectations up to now. So at this point, if you're not actually like a CIO or running a tech company where you need to keep apprised of what's happening in a techno landscape, if you weren't already doing that five years ago for other technologies, AI will let you know when you need to know about it.

It's not really your job to be a technology reporter unless it's literally your job to be a technology reporter. Tim Ferriss had Seth Godin on the other day and they were talking at the end about Claude and Perplexe and ChatGPT. Yeah. And Seth said that he used, he goes on Claude and Perplexe like an hour a day and he used it to help him fix a broken pump in his basement.

He went down, take a picture, came back up, did it, went back down. Yeah, yeah. So, you know, maybe that'll be useful. I think most people are still using Google for that. Yeah. Right? Like they're Googling it and then they're finding a YouTube video. But yeah, something like that will come around.

I mean, I think, you know, I've talked about it before. I think it's gonna be industry specific. Scott Galloway talks about it as, it's like the application layer. It's where all the interesting stuff is going to happen, not the underlying technology. I think the big companies want to have these splashy new releases of new models with like fancy names that were trained in half billion dollar data centers.

But the impact is not going to come from that. It's going to come from the apps built on what we already have. And I think a lot of the initial productivity gain, I've been on record saying this, is actually going to be not adding new capabilities to software, but giving more people the capability to use all of what software can already do.

So it's going to be a novice user of a powerful software package will be able to easily use more of the powerful functions because the AI will help them. I think that's where we're going to see the first. We're going to talk about this more in the final segment though, some more far-fetched concerns, and we'll have some fun there.

All right, what we got next? Next question is from Kelly. My company has ended all telework, so I'll be going back to the office five days a week. It used to be one day. My post-COVID arrangement did have some drawbacks. Essentially, I didn't get enough deep work time each day as sometimes I would take advantage of my flexibility and do other things.

Now that I'm going back to work more, I'll be able to work more on my phantom part-time job. How do I make the most of this opportunity? I mean probably my number one advice would be doing something like a Weekend at Bernie situation. So and this is just common sense.

You start by identifying like clothes you normally wear and getting a second set of those clothes. All right, step two, acquire a corpse. Step three, you dress the corpse in your clothes and you put the corpse at your desk and then, you know, the manager walks by and they're like, oh, there's Kelly.

Kelly's working. No big deal. You could do a little bit of Ferris Bueller days off magic, you know, where I think it was hooked up in Ferris Bueller when his mom opened the door was on a pulley. So it kind of like raised his arm and turned his body which completely fooled his mom.

Like, oh, yeah, Matthew Broderick is fine because he like oddly and spasmodically like lurched when I opened the door. So no need for me to walk in further. I'm certainly not going to talk to him even though he just lurched upright. I'm certainly not going to talk to him and the fact that there's snores clearly coming out of a tape deck that is going to be at way lower fidelity than real sound that is not where his head is.

No big deal. I'm sure he's fine. Come back and check in at the end of the day. So that's what I'm saying. Ferris Bueller plus weekend at Bernie's. I think you'll be fine. No, okay. Here's what I always say to people when it comes to Phantom part-time jobs. You got a time block, right?

So you got to know what am I doing and when am I going to do it? And it's like when are my blocks? You'd be very systematic like here's the blocks. I'm working on this other project and here's the blocks. I'm working on what's going on at work. Then you got to be on the ball in terms of keeping track of what you have to do and planning when you're going to do it.

So multi-scale planning is critical. So, you know, like what you need to work on that week. You put aside time for it that week. You see the bigger picture so you know how to get out ahead of things. So if you're in control of your time and you're in control of your tasks, you can stay on top of things which is also critical for this situation because people don't bother you as much if you're getting things done.

They don't have to stay on you. They're sort of less bothering you. So be on top of things. Be wary about your workload. It could be easy when you're back in the office to just more informally take on more things because it's social. The person's there like, "Hey, what's going on?" You're like, "Oh, I can help you with that." Keep your workload reasonable.

So it's like work you can keep up with without being overloaded. And then finally, you want to be careful about how you collaborate so that it's not, you want to avoid having too much collaboration based on ad hoc back-and-forth messaging. That really kind of locks you into a hyperactive hive mind mode.

It's hard to work on something else if you have to be keeping up with these like ongoing back-and-forth conversations. So, you know, the type of stuff I talk about in my book, A World Without Email, that becomes more important. This is the collaboration protocols I use for different type of work I do.

There's shared documents. We have a fixed schedule. There's office hours. There's, you have these different ways of things that are getting done that's very predictable so you can do your work and not have to be constantly monitoring different types of inboxes. So I think there's, all the stuff I talk about can put you so much ahead of the game with your work in the office environment that you still have more than enough time.

But also, final piece of advice, and this makes it all work, put sunglasses on the corpse. That's what they figured out on weekends at Bernie's. That was what, you know, people are like, I don't know. I think it's odd that there's a decomposing, like 230-pound man that they're dragging, but he's got sunglasses on.

So I guess he's probably just tired. Corpses with sunglasses. All right, what do we got next? - Next question's from Javier. "I'm getting my MS with the single purpose of wanting to teach at the collegiate level. I plan to teach as an adjunct instructor at first while I still work as a software engineer.

Maybe eventually this will develop in attempting to become a full-time instructor somewhere. Is there a big difference in how full-time instructors or adjunct faculty are viewed by the institution and PhD staff?" - Yes, but I think more important than whatever specific information I could give you is the bigger picture suggestion here, which is you got to go do a heap of evidence-based planning.

Right? So we talk about our two dual paradigms when we talk about how do you pursue the good life in the 21st century set up. Right? There's these two planning paradigms that orbit each other, lifestyle-centric planning and evidence-based planning. So lifestyle-centric planning is working backwards from your general vision of a life well-lived.

So you're not trying to work towards a particular goal, but work backwards from a lifestyle. And it sounds like you've done some of this lifestyle-centric planning and that you have a vision of a particular lifestyle where you being an instructor, a computer science instructor, maybe that works into this vision.

You have to pair lifestyle-centric planning with evidence-based planning. So once you have your lifestyle vision, you work backwards to figure out, given my opportunities and obstacles, how do I move closer to that lifestyle vision? This is where you need evidence. Don't use the word maybe. Don't guess, don't assume, don't write your own story.

When you're looking at a particular path to get you towards your lifestyle vision, go research that path like you had been assigned an editor from Business Week said, "Go write me a story about the reality of life as a computer science instructor." You got to go to the institution that you would like to teach at or the type of institution you'd like to teach at.

You want to take an instructor out for coffee and you say, "Tell me all about it. Like, how did you get here? Who gets hired? What are they looking for? What's the reality of your life? Like, how much work do you have to do? What's the schedule like? What's the pay scale like?" Like, get the information.

If you don't have that information, your plan for moving towards your ideal lifestyle is not a plan, it's a fairy tale. There might be some good lessons in it. The intentions might be good, but probably in real life, you know, you're not going to find a wolf in a cloak or whatever happens in the fairy tale.

You got to have evidence. So I want you, here's my instructions for you. Go talk to some real people in this position. Sandy, check your plan, get the details. Not only will this help you avoid you from traps, like, "Oh, this was a romanticized image and it's not what I thought it would be," it also can give you advantages.

Because so few people do this, like, "Let me just study this thing I want to do." In really well-known competitive fields, people do this. "I want to be an actor." There's a lot you can learn about how that happens. "I want to be a writer." A lot of people want to be writers.

There's a lot you can learn about the reality of that world. But in like so many other fields, it's much more haphazard. And if you really understand what it works, what they're looking for, the different options that are available, you're much more likely to actually succeed in setting something up that works.

So do your research, and you'll probably find a cool path. Once you have a lifestyle, you'll find a cool path there. And maybe it'll be through instructing, or maybe it'll be like part-time instructing plus a different type of like part-time software thing. I don't know. You have to actually get the details to figure this out.

But the key is don't just wander into this blindly with a general idea about what you want to be true. You want to ground this in the reality. I was just talking to someone, a friend of mine, a computer science full professor, professor. But he was thinking, he was like, "You know, it'd be fun or meaningful, I guess.

In the summer, we have a really nice community college around here, Montgomery College." And he's like, "I would love to do some, I think it might be like fulfilling to do some like CS instructing over the summer." I was like, "That probably would be pretty cool." And that's a good community college.

It is? He just built a new math and science building. I drive by it all the time. Yeah, right? I think it's a good one. But anyways, I appreciate, I admired that, right? That's a move. I mean, he's a well-known professor, but it's like a move like, you know, I want to put these skills to use not just with this one singular population.

Mm-hmm. I thought that was cool. I wonder if they have podcasting stuff over there. You know, that's what I was thinking about. When you have these like nicely well-funded community colleges, they probably have like production studios. Yeah, I'm sure. Like with nice cameras and mics and stuff. It's good for us to know about.

All right, what do we got next? We have our corner. Ooh, slow productivity corner. Once a week, we have a question related to my book, Slow Productivity, The Lost Art of Accomplishment Without Burnout, which you need to read if you haven't. But we do this segment so we can talk about ideas from the book, but more importantly, play the segment theme music.

So let's hear that song, Jesse. All right, what do we got? It's from a grad student and a mom. Same person. I'm a PhD student in biomedical sciences and a full-time mom. I need a PhD to pursue my career and lifestyle as a tenured professor at an R2 university.

My advisor wants me to do lots of experiments unrelated to my thesis during the day and read papers at night. I have mother duties at night. Is there a way to apply the slow productivity principles to my situation? How can I convince my advisor to allow for experiments only related to my thesis?

Well, let's put the advisor conversation to the side for a second, because the first thing I want to reassure you on is you can absolutely excel in a doctoral program as a mom. This happens enough that there was a it being, let's say, having your first child while you're in a doctoral program.

Just because of the roughly the age where people are when they're in this program, it's, it's, this is not an uncommon occurrence. I used to have this talk I gave. It was, I would give this talk at colleges and some of those memories are a little bit hazy for me, but I used to give talks to colleges and I, at some point, I think I was going to Duke.

Got a bunch of talks at Duke, but I gave a talk at Duke years ago and I remember I talked about this phenomenon because at the time, I guess I was a graduate student or a postdoc and there was this phenomenon that I had observed and then also my advisor had told me about.

So I think she might have been in the same situation back when she was younger, where a grad student has a baby and they become much more productive. Like, they become a much better grad student. It was like the grad student baby paradox, I think I called it. And I would ask, well, why is this, right?

Because you, you get much more constraints on your time. You know, you have a baby. Like, there's a lot of constraints on your time. And the answer was like, well, we waste a lot of time as grad students. Like, it's kind of diffuse because there's so few constraints. It takes us 90 minutes just to kind of get warmed up to start working and not until like nine o'clock at night do we really get things rolling.

And when you have these sudden, like, parenting constraints, you're like, here's the time I have to work and I can't work outside of those times. And when you're in that, you are focused and you're like, I'm getting my work done. The baby is napping, I'm on it. You know, like, my mother-in-law watches the baby for three hours in the morning.

I have to make every inch of that three hours, every minute of that three hours count. And it turns out, like, most grad students are so far from the ideal of giving things full focus that suddenly you're like a superstar. You know, your dissertation gets done, etc. So, your advisor aside, that's going to be a different problem.

You can crush being a grad student even as a mom. You just be very clear. Here's when I can work and I'm going to make the most out of when I work. It will add up. And remember, you have flexibility. This is what's nice. It's not like if one day you can do less than another, it doesn't matter.

There's not as many, you know, you have to be at this client meeting. I mean, there's a lot of flexibility there. But focused work done consistently can produce what you need in these jobs. So, for your case, maybe it's like a nine-to-five schedule. So, maybe it sounds like you have maybe some sort of childcare coverage during sort of normal business hours.

I was a nine-to-five grad student. It was very rare. But my wife worked a normal job. I got married young. My wife worked a normal job and I wanted to be home when she was home. So, I would just show up at nine. I'd come in with the morning commuters and I would leave at like 4.30.

When like most of the grad students were just arriving to start their pushes. But when I was there, I worked at like a nine-to-five job. I actually gave, I was organized and time-blocked and executed. And it was more than enough time to be a very successful grad student. My colleagues thought I was weird, but I got a lot done.

You know, because grad students, it's not actually that hard to be a grad student. Don't tell, don't tell advisors that, but it's not that hard to be a grad student. It only gets harder from there. So, you can absolutely do it. Just set your limits. This is when I can work.

And work with great focus and purpose during that time. It'll add up. When it comes to your advisor, like you got to just tell them that. I got a baby. I have a, this is when I can work. This is a substantial amount of time. This is the amount of time that like most of these other grad students are working as well.

I am very organized. I listen to Cal Newport. Like I'm going to get after it. But these are the hours I can work. I'm being a mom after these hours. And I can work on experiments throughout the day. I'm reading papers in between those experiments. I'm controlling my time and I'm doing good work.

Probably what will happen is there's different configurations that grad students have. Different relationships they have with their advisors. Some grad students have like this, much more of this employee relationship where you're sort of doing lots of stuff your advisor asks you to do. Others get more of an autonomous relationship.

That's like, okay, great. Then just, you own this series of experiments and just like do this because we need this for one of our research grants. So, you'll probably just evolve into a more autonomous relationship with your advisor. Which is what you want. It's what I did. I wrote some, a bunch of papers with my advisor.

Most of my papers were not. So, I could just kind of do the work when I want, how I wanted to do it at my own pace. I think it'll work out. Worst case scenario, you switch advisors. They're not going to kick you out. So, worst case scenario, you switch advisors to one who says that's fine.

Work on this project. Figure out how you want to do it. Show me the results. But you are fine. You are not the first grad student to have to be taking care of a baby. Many before you have found the secret to grad student life, which is under the right circumstances.

It doesn't really require a huge amount of time to do what you do well. I mean, I wrote books as a grad student. They were unrelated. Unrelated to my job as a grad student. So, it really doesn't take that much time. I mean, experiments take time. I was a mathematician.

So, you know, that was easier, but. Yeah. I had plenty. I blog, I used to blog three times a week. That was like what it was like back then. Like 2006, 2005, you had a blog. Like you blog a lot. So, I was writing blog posts all the time.

Yeah. And books. And a bunch of papers. All right. Well, we got a call. We have a call. All right. Let's hear it. Hey, Cal and Jesse. My name is Aaron. I'm a busy dad of six living overseas. I've been intrigued listening to you over the last few months.

And I particularly like what Cal has said a couple of times, that college isn't that hard when you treat it like a job. I've been enjoying applying your ideas in my own life. But I think many of them would also help my kids too. We're about to spend six months in the USA transitioning to college.

And we'll be thinking a lot during that time about college with my next two. I've, this last week, been through this "How to be a Straight A Student" and "How to Win in College." I haven't found where this particular idea of treating college like a job is fleshed out.

I want to make reading one of your books required for my three high schoolers over spring break and then discuss with them. But I'm thinking the best one is going to be "How to Win in College." But I'm wondering if there's something you might recommend more for students facing college in the modern digital environment.

Thanks. Yeah, it's a good question. I think that idea that people treating college like a job have an easy time at it. Actually came after I wrote those books. Right. But what I what I started doing, I think especially once I got to Georgetown, is I would do advising for various groups supporting non-traditional students.

Non-traditional students. So by non-traditional meaning students who weren't just sort of coming right out of high school. And I did some work with a group that was veterans. So coming back to college on the GI Bill. And another group that was working with like first-generation college students, many of whom were coming in the college, not right out of high school, who had worked or done other things.

And they used some of these, if I remember correctly, we used in some of these groups how to become a straight-A student as like an assigned text. And these students, these non-traditional students who had had real jobs, just ate it up. They're like, okay, let's get after it. What's the secret?

Do this, do that. What's the best practices? I've got kids at home, you know, I'm older, I got things to do. And they just destroyed college. They really destroyed it. And that's when I realized like, oh, maybe I shouldn't be as proud as I was about all my good grades.

Like I just happened to treat college more like a job. I wasn't brilliant. I just wasn't dumb. Not intellectual dumb, but just habits dumb. Most students, traditional college students, just kind of go to the library and put on their hooded sweatshirt and capital S study. You know, they've got their phone open and this open.

They're like, I'm going, I'm studying all late and trying to convince their parents how hard college is. The people who treat it like a job, but you know, they schedule. When does this thing do? Like when do I want to work on? I don't want to work on until two in the morning, the night before I'll start a week earlier.

When do I have time for this? They move around their schedule. They move things around. They use techniques that work. They don't waste time. Low friction studying, low friction note-taking. What's the stuff that's actually going to help me learn this? What is just nonsense structure on top of that?

So it's not actually in one of my books, but it's something I observed about how people use the books. So I'll give you a little bit of guides. There's like four books that could be relevant to students. So I'll tell you what you're going to get from each and you can decide what to do here.

How to become a high school superstar. This is a book about college admissions and high school success aimed at high school students. And it makes the argument that you can do really well in college admissions without having to be a super stressed out grind. So the whole premise is I profile a collection of what I called relaxed superstars, kids who weren't stressed by God and the good schools.

And we kind of pick apart how they do it. So that could be kind of a useful thing if you're thinking about high school and college admissions. The part one of that book has a playbook. The part one playbook which adapts a lot of my college studying and time management advice to simpler versions aimed at high school kids.

So if you have a high school kid who's, you know, has at least a year left of school, you might want to read at least part one of How to Become a High School Superstar just so they get used to a more professional structured way of doing their schoolwork.

When it comes to college, How to Win at College is like an intro to the mindset of like, "Oh, I want to be a successful college student." And it's a bunch of, it's very easy to read. It's the first book I wrote. It's 75 rules. They're each just a couple pages and just plants this idea that there's good ways and bad ways to do things.

You should experiment. You should care. Like, don't just stumble through your experience. And some of them are academic studying related rules. Some are related to how you choose courses. Some are related to things like physical fitness or keeping up with news or mental health or keeping your room clean because it's going to change the way you understand the organization of your life.

It's like a mindset book. And it's based off of, in theory, I interviewed like Rhodes scholars and Marshall scholars and Goldberg scholars and sort of based on like these ideas that like really successful students have. How to Become a Straight A Student is just straight up, here's how to take notes, study and write papers.

You got to read that at some point if you're going to college. It's how to do your academic work like a professional. That's the best-selling of those three books. I think that's quarter million, maybe 300,000 copies. Like that book is just sort of chugged along in the background. That's just straight up.

Here's how professional, here's how to study like it's your job. So at some point you would want them to read that book. The fourth book that's relevant is how to be, no, no, So Good They Can't Ignore You. This will be relevant as you go a little bit farther into your college career, but it's going to talk about how do you ultimately cultivate a career that you're passionate about.

And it breaks a bunch of myths that college students are going to hear, which is like it's all about your passion. It's all about pursuing some grand goal and it talks about the value of getting good. It's level sets your expectations for what work should feel like in your first few years out of college.

And so you will want to read that book at some point before you graduate. So I'm kind of giving you a long reading list here, but you can choose from those summaries which of those books you want to start with. All right, we got a case study here. That's where people send in stories of themselves applying the type of advice we talked about on the show to their own lives.

We can see what it looks like out in the wild. If you have a case study, you can send it to jesse@calnewport.com. Today's case study comes from Derek. Derek says you have stressed that systems and processes don't make work easy. They make it consistent because work is hard. I've always done a new partonian style system, but a year ago entered a job on a contract within my existing government organization to be a coordinator for major federal grant.

I'm responsible for receiving, processing, tracking, servicing, and reporting on four major grants. I've been doing multi-scale planning so less catches me by surprise. In my quarterly plan, I've committed to developing my obligation management practice through a big weekly capture, consistent shutdowns, and time block plans, including breaks so my brain can be on board.

On the weekend, I do a weekly plan to be on the offensive. Some of my biggest successes so far have been the one for you, one for me scheduling, and the use of task boards. I have two, one for admin and one for application processing. WorkingMemory.txt also has been a lifesaver, although I admit it's turned into a kind of diary.

I also have a centralized shared spreadsheet for applications received and have defined a protocol to track statuses and actions taken that I've shared with my team. These have been a game changer as I have over 50 applications and projects to monitor. While this has so far been a story of success, no hero's journey is complete without setbacks.

Minus correspondence, I have two email inboxes to manage and calls and voicemails to return. I think I need to autopilot these. How can I keep on top of the application project obligations while still ensuring I'm staying on top of the emails? All right, Derek, I appreciate the case study.

I think what this underscores is this idea that one of the unique attributes of office work in the digital age, right, in the modern digital environment, is the degree to which your job can become like this, where you're basically running like a complicated mini-organization where the only employee is you.

Because in this age of low-friction digital communication and information flow, it's just so easy. It's like, great, you do all of this. And because we can just give you an email address and people will just bother you and we can just say, just handle all of this. We have workloads, so quantity of workloads and the velocity at which these workloads unfold would turn the hair white of someone from like 1985.

It really can explode the complexity of work. And the only way to survive in this type of new digital environment is you have to really structure yourself. Like you have five different departments you oversee and they each have their own processes. Even though you're implementing each of these departments with your own brain, if you don't structure the information and the communication and how you go through your day, you will be swamped.

Derek would be completely swamped without these tools. And probably there's been half a dozen people who've gone through the same position who were doing a quarter of what you're doing here, Derek, and probably then still had a hard time at it. So like you have to have the structure in a way that 30 years ago, you didn't have to worry about.

So I appreciate hearing that case study. To get to your kind of question within your case study, you're worried about the volume of calls and voicemails and emails. You say, how can I autopilot this? So in other words, like have fixed times for doing this, so I'm not just in my communication boxes all day.

You have to reset these processes. You have to retrain the way the people who are communicating with you think about how they're communicating with you. You have to move them away from the paradigm of the hyperactive hive mind, which is where when they think about this email address, they think about you.

This email address is fused with you and I am talking to you and asking you something and it's rude if you don't get back to me right away. You need to move them off of that and into protocols or processes where the right information gets captured and stored and you can go through a lot of it efficiently and action can be taken and expectations are appropriate.

There's a lot of things you can do here. Again, this goes back to my book A World Without Email. There could be things like for specific types of queries people have about their applications. Don't use email. Say whatever it is. I have a shared document or a spreadsheet for collecting concerns or modifications and just go in there.

Here's like pending and here's what's done. Just go in there and you add a row when you have a question. You make sure all the information I need is in there and you can watch as things above it get handled and yours moves closer to the top of it.

You can see exactly what your status is, but it's not just an email. It's an email. Again, we conceptualize as like I just tapped you on the shoulder. Why are you ignoring me? But when it's no, I'm entering your information to a queue in this document and I see there's six things ahead of it and I'm waiting till my thing gets processed.

Our brain has a different way of categorizing that and no people won't be stressed out about it because again, I always argue what people want is they don't need responsiveness. They need clarity. You have the information. I know you have the information. I know I'm going to get an answer.

I know my status. I'm fine. I have a hundred other things to do. Don't need you to respond right away. I just need you not to forget it. You can use office hours. Hey, here are my open hours for like application questions. Just call me up. My phone is on.

It's really easy. Just, you know, you can always call me at three and like we'll get into it. I'll answer any questions you have. You can defer people, you know, when they try to email you you can defer them to that. You can have an email address. That's not your name, but it's like a project or query type that defuses the communication channel from a person and people's expectations changes when you send something to, you know, application request as opposed to like jesse@government.gov you have a different way you think about it.

Like, oh, yeah, this is going into like a system where people are going to process these requests and then you can have other sorts of protocols. Like, yeah, like this is how this works. We have the shared folder and applications. When you're ready, you put them in the shared folder.

We empty this on Wednesdays and Fridays. So, you know, if you get it in whenever you get it in by the next Wednesday or Friday, we'll empty it out. We'll send you here's the instructions. And so when people like bothering you like, hey, I want to do an application.

What's going on? You just send them back this like instruction link. It's like, yeah, here's how it works. You put this information put in this folder. We'll process it the next Wednesday and Friday. We'll send you a confirmation of that. You can then track it in the spreadsheet over here.

See, you got to retrain and redevelop your people you're communicating with and your communication protocol. So it's not just ad hoc conversation ongoing. It's critical to get away from that, especially in this type of role. IT departments learned this a long time ago with ticketing systems. You cannot just once you have any volume of dealing with people and their concerns, it cannot just be ad hoc.

It cannot just be here's my name at agency.gov. Just talk to this like you're talking to me like in the same room. It just doesn't scale. So you have my permission to try to build things that are more structured. All right, good case study. All right. So we got a tech corner coming up.

But first, let's hear from another sponsor. This show is sponsored by BetterHelp. So this is a time, a time of year, a time in history when people worry about their relationship with their own mind. Maybe you're struggling. You're feeling some anxiety. You're feeling ruminations about things. I'm worried about this or that and I can't get away from the rumination.

Maybe you're feeling like you're having a hard time just finding the normal types of joys or engagement or there's like a particular thing in your life that you can't let go of and it's really getting in the way your relationship with your mind is struggling. A lot of people are feeling this right now.

A lot of people feel this all the time. This is where therapy can help, right? If your knee was hurting, you would go see an orthopedist, right? If your brain is hurting, therapy can help you. It can really make things better. The problem, you know, when I talk to people who are thinking about therapy and they're not going to do it or they're hesitating, it's often just the vagueness and the complexity of how do I get started?

Do I just call like a therapist office like around here, especially like these days in DC, you call a therapist office like, yeah, we'll have an opening in, let me check the calendar here, 2027. Like everyone's trying to get into this office, like it's really difficult and you feel bad for even asking.

What we need is like an easier on-ramp into the world of therapy. That is where BetterHelp enters the scene. BetterHelp is fully online, making therapy affordable and convenient, serving over 5 million people worldwide. You can access a diverse network of more than 30,000 credentialed therapists with a wide range of specialties and you can easily switch therapists anytime at no cost.

So by leveraging the online delivery, it's a much more efficient system. It's, you can find a therapist that they don't have to be right local. You can switch them if it's not a good fit. So discover, you know, repair. Now is the time of the new year to repair your relationship with your own mind, with BetterHelp.

Visit betterhelp.com/deepquestions to get 10% off your first month. That's betterhelp, H-E-L-P.com/deepquestions. So I want to talk about our friends at Shopify. I know a lot of people in the same game of podcast and books and videos who sell things online. And by far the most consistent choice they use of software to do this is Shopify.

I mean, Shopify really is the standard. If you want to sell things, it is the technology that you need to be using. My friends and podcasts sell things online, but people use this in person as well. There's point-of-sale solutions. It really is the industry leader. It's sort of, in my opinion, a no-brainer.

You sell stuff, you need to use Shopify. Nobody does selling better. They are the home of the number one checkout on the planet, including ShopPay, which boosts convergence up to 50%. That means way less carts going abandoned and many more sales going forward. So if you're into growing your business, your commerce platform better be ready to sell wherever your customers are scrolling or strolling.

That's a phrase I just made up right on the fly, because I'm pretty clever that way, Jesse. On the web, in your store, in their feed, and everywhere in between. I always rhyme. I'm just, this is just off the top of my head right now. In between. I love this.

This is a good copy. Businesses to sell more sell on Shopify. And that's just absolutely true. Like you got to use Shopify if you're selling stuff. It makes it so easy. So upgrade your business and get the same checkout that everyone I know who sells things online uses and that is Shopify.

Sign up for your $1 per month trial period at shopify.com/deep. You have to type that in all lowercase. Go to shopify.com/deep to upgrade your selling today. That's shopify.com/deep. We like to do tech corners when possible in the final segment. That's where I put on my computer scientist hat and we talk about tech and the way it's affecting us in various ways more directly.

Today's tech corner. My kicking off point is I want to react to a recent Joe Rogan podcast episode where he was talking about chat GPT and he says the following sort of in passing but let's listen to it and then I'm going to react to it. GPT has survival instincts.

That makes you just go what? Is that programmed in or is that just a thing that it understands when it's looking at so large language models taking in all the information that's available on the internet. So it's like looking at patterns and the survival is a big pattern of the human experience.

Like we all want to survive. That's why death is so scary and war is so scary and disease is so scary. Other than the depressed suicidal people. But yeah, they want to be happy if they could, you know, yeah, but this this also transfers on to the things that we create and so we create them with this understanding of how we operate and it's a better version of us.

But also has those instincts of survival. The real scary thing is does it also have the instincts of success? Does it also have the instincts of acquiring resources and power because that that's where it gets real weird. Well deep. All right. So what Joe is talking about here? I think is a sentiment you hear often when people are encountering these ever more impressive language model based AI tools like chat GPT, which is like wow, these things are getting kind of scary.

What if their instincts don't align with us? So he was talking about here. Chat GPT having survival instincts. He had seen some quotes somewhere about chat GPT trying to make copies of itself, which he interpreted to be like it has a survival instinct, but it says what if it has an instinct for power?

What if it has an instinct to get rid of humans Etc. I want to react to this. Because the there's a misunderstanding plus an actual fear. So I want to talk about the misunderstanding here. So I'm going to make you feel a little better and then I'm going to talk about the real fear is going to make you feel worse.

So I'm going to temporarily make you feel better about AI and then I'm going to sort of justify some of Joe's fears and in a more technically accurate way, which might make you feel worse. All right. So can the tools like chat GPT in its current form have things like survival instincts?

The answer is no, no, they can't have instincts. They can't do things the right way to understand something like chat GPT is the following the main technical engine of a chat bot like chat GPT is a large language model. The original was GPT 3/5. We got GPT 4 and GPT-O, GPT-O mini and the GPO 2, but we have these large language models.

The right way to think about a large language model. I've been thinking about this recently. I think a good way to think about it is I don't know if you remember this Jesse, but when we were kids in the 80s and Play-Doh was like a big thing, right? You could buy the Play-Doh factory where you would feed Play-Doh into one side and you would turn a crank and it would pull it through and you could put a couple things that it would push it through.

So maybe like you went through a squisher and then it went through something that that cut things into like spaghetti strands, you know, and then the final thing might take those spaghetti strands and like you could slide it over. So it's like a star shape and then the end you would get like spaghetti strands of the original colors coming out like shape like a star.

So it's like pushing the Play-Doh through several layers of things that like modifies the Play-Doh and you could change what those things are basically at a very simple level. This is the architecture of large language models. They're a feed-forward architecture or the information which is the text. I mean it's embeddings of the text but like the text that it's reacting to gets pushed from one layer to the next to the next to the next.

There's no recurrence. There's no looping. It just goes from one layer. The Play-Doh goes through the squisher then it gets changed to spaghetti strands then it goes through the star shaper. So it gets pushed through these layers as it gets pushed through these layers more each layer does its own work.

It identifies patterns. It looks for labels. Let's I'm saying this all using non-technical terminology it labels it with stuff that the next layer can then use and look up other patterns and connect it to it, but it's moving just inexorably forward. And at the very end so I imagine like someone's turning this crank that's pulling this digital Play-Doh through this factory at the very end what comes out is like a fancy Play-Doh shape.

Now what it is with a chatbot is actually going to be a part of a word or a word or a group of words to output. I mean technically it's like a probability distribution over this but it's like a text new text output. So that's what the large language model does.

Things get pushed through the language model itself just like the Play-Doh factory doesn't change once you've trained it. Nothing about it changes. It doesn't remember last time they put red Play-Doh through here and we're going to remember that and that's going to affect what we do this time. We put green Play-Doh.

No, it's the exact same setup every time stuff goes through all this different is the type of Play-Doh you push through. Then you have a control program, right? So you have to have some sort of control program that like turns the crank, right? So it's like a with most chatbots like a relatively simplistic control program that like takes the text you typed in and puts it into the language model and the language model spits out some words and typically because it's autoregressive it'll put those words on the end of the text.

So now it's slightly longer and they'll do it again and they'll put those new words on I'll do it again. So it's kind of running the machine until it has your whole answer. If you're doing different media formats, there's more complexity. So like if you're sending a photo to chat GPT, the control program will take the photo and actually give that to a photo recognizer program that will then describe that photo in words and then it will take those words and connect them to your original prompt and then we'll run the normal chat GPT control program to like ask that question and or if you're generating a photo, that's not a language model.

That's going to be a different type of model. But this is basically how this how this works. So the the factory can produce really cool Play-Doh on the other end, but the factory itself doesn't have intentions because it's just a factory. Nothing changes. Nothing moves. It's a bunch of plates that you're pushing Play-Doh through.

If you really want something to have instincts like a survival instinct, you'll need some notion of sentience which requires a combination of four things understanding of like the world like understanding of concepts, right? You have to have some sort of actual like intellectual contemplation. You need a sense of state.

You have to keep track of like your understanding of the world that evolves as you get new information. So you need something to be reacting to you need some notion of drives. Like what is your goals, right? So there's these intentions have to come from somewhere and you need actuation the ability to actually then take actions either in the digital or physical world, you know based on the combination of all these four things.

That's how you have something like a sentience. You can think of these large language models as basically just being the understanding part. It has this understanding built in of lots of different textual things and that's how it spits out the plate on the other end can be a really cool-looking design that matches in a really interesting way the play-doh that came in the front but in order for that thing to have survival instincts, it also needs some ongoing state that's keeping track of its understanding of the world.

It needs a drive. So you have to explicitly say I want to survive again a play-doh factory. There's no thinking there you need actually some sort of program that's like is this action going to help me survive or not and then it needs the ability to take actions. That's actuation.

That would be a completely different component. So there's no way that as architected right now as sort of one of these standard chatbots can have instincts because there is no state. There is no drive. There is no actuation. It's just a feed-forward play-doh factory that spits out words and a little control program that feeds them in.

So no making the model bigger and better trained just makes the play-doh come out look fancier, but it doesn't give it any of those things. You cannot spontaneously have a feed-forward network with no changeable state and no recursion and no actuation and no drives like that by itself can't do anything.

So Joe is wrong to say. Oh, I think these things just have drives or chat GPT is going to start copying itself. The chat GPT is just a factory defined by these models are just trillions of numbers and tables that are multiplied by GPUs. But where Joe is kind of right is the hard part in this formula understanding state drives actually actuation the hard parts the understanding and these language models have a pretty good understanding of thing.

So it would not be that hard for someone to program those other three things pretty simplistically. All right, I have like a state of the world that I care about. I hard-coded in I when I'm evaluating actions, I want to do my goal is to like not be erased and then just plug it into actuation.

I wrote a program here where you know, this program knows how to do things like on the internet. It can run commands and run scripts and I you know, and network with other types of things and then I'm going to hook these four things together. So like the state will you know, gather information give it to like a run it through the chat bot factory the the language model to kind of analyze it and ask it like hey, this is this thing good or bad if my goal was survival and then you know, what would you suggest then between these actions like it could start talking so the control programs could start talking to the language models to get its understanding or ideas and then choose actions based on that and in fact.

There are systems built like this. Now, they're all right now in the category of game playing systems. But game playing AI's the advanced ones kind of work like this. Now you have like some understanding you have some planning you have a state of the world you like the state of the game or of the poker hands you have a drive like what we're trying to do is win and it talks to the the neural network the thing that's trained it has the understanding tries to figure out what to do.

Like this is how there's a the the bot that first bought the win at Texas Hold'em poker. It does something like that. It's like a combination of a trained model that just understands things plus its own simulation and its own drive and its actuation which is making bets and playing cards.

So we kind of know how to do this. Now the big companies don't care about that the money isn't not trying to build a sentient thing that's going to copy itself. The big money isn't trying to get to be the next email like we need this to be like helping people use advanced graphing features in Excel.

That's where the money is. So it's not like they're working on this. So you don't have to worry about like open AI's next release is going to be alive. No, the real worry is working with like deep seek or like a relatively small low cost model building these other pieces and letting that loose because I have this theory.

I have this theory that where you're going to get the first uncomfortable AI by uncomfortable. I mean, I kind of am worried about what it's doing or worried about turning it off. It's going to be the combination of relatively simple dynamic programs that can loop and create sort of like cybernetic control loops for doing things like actuation drives and world-state connected to an unpredictable complicated understanding of the world like in a language model or something similar.

And there's going to be this sort of runaway effect where simple control logic plus complex world understanding. Could lead to unpredictable complex seeming behavior that has real world impacts. So I don't mean to geek out too much, but like it's good news. No chat GPT doesn't have instincts. Plato factories can't there's nothing.

It just things push through it and nothing changes to ossified built in stone. But the person turning to crank. Well, if you had a few people in there who are choosing the Plato that goes in and then based on what Plato came out deciding whether not to fire a missile or not, you can start to get some really sort of unsettling type of dynamics.

So, I don't know. I've been thinking a lot about it. No one I know is building something like this, but I actually don't think unsettling AI doesn't require necessarily breakthroughs on the language model side. I think it's these other types of programs of which the language model would be one.

So something to keep an eye on the good news is by the way is in this model, the state drives and actuation is just being programmed by people. It's not emergent. It's not unpredictable. So it's you know, you know what it is you can program that what what it can do what it can't do.

You can turn it off like it's we have I call it intentional AI when I think about it more formally. We will have more control over those those hand-coated pieces. So it's not like when people have a concern as language models. We don't know they train them but the how they form their understanding is kind of foreign to us.

So we're like, I don't know how it's like making its decisions. But if you're going to build a sentient ish AI using this model, I'm talking about, you know, exactly how the state control works, how the drive control works, how the actuation works. So it's in that sense is a little bit more predictable, but I don't know.

It's an interesting world. So Joe's not he's not right. He's not wrong. That's a way of putting it. I have to listen to all that again. Yeah, for sure. But we are a technology podcast now number five temporarily the number five technology podcast in the world. So we got to do more.

We got to do more tech. Yeah, I'm going to listen out again for sure. All right. I should do some more stuff complicated world. All right. Anyways, while we are still here and AI hasn't taken over yet. I will thank you all for listening. We'll be back next week with another episode and until then as always stay deep.

Hey, if you like today's discussion about productivity rain dances, you'll also like my in-depth conversation with Oliver Berkman about productivity and productivity traps. You can find that from October and that is my in-depth episode with Oliver Berkman. Check it out. You know on on one level my my subject matter is human finitude limitation how we get things done and enjoy life in this situation of being as limited as we are.

And I guess the focus of this book is much more in my mind on action at a day-to-day level.