24: Linus Lee - Engineering for Aliveness

[00:00] Welcome to Dialectic, episode 24, with Linus Lee. [00:04] Linus works in the world of software. [00:06] Building, researching, engineering, designing, and exploring how technology and software can amplify us rather than diminish us. He hopes to create what he calls instruments for super agency, leaning into the notion that technology at its best should make everyone more human and more capable. These days, he's focused on AI at Thrive Capital, a VC firm where he builds internal tools, researches, and advises. [00:36] DataWorks and Replit across engineering, research, and AI. And he's also prolific in his own time with over 100 personal software side projects and extensive writing, much of which is incredible. One of the most foundational things he explores in his work is how language and knowledge can be codified, explored, and expanded by way of software. We talk about all of this and more, including one of my favorite framings he has, where he distinguishes between what he calls [01:06] others help us more deeply understand and move toward mastery. Running through the conversation, as you might guess, is the state of LLMs and how they affect software. And we also talk extensively about his work at Thrive and how he thinks about bringing an engineering mindset there. [01:20] I hope you notice it throughout, but we also end the conversation specifically talking about how humanity and technology don't need to be at odds, and in fact, how technology can help us dream and help us wonder.

[01:31] I hope you enjoy the conversation and are as inspired as I was by Linus. If you enjoy this episode and you enjoy dialectic, [01:39] Please share it with a friend. Every little bit counts, and it means the world to me. [01:44] With that, here's Linus. Linus Lee. [01:47] Good to see you. [01:48] Thank you for having me. I'm excited. I like to interview multifaceted people, and I also like to interview writers, but I sometimes joke like the easiest or the ideal interview, in some sense not really ideal, is somebody who has like eight really good essays, and you have a lot more than eight, and you also have so many projects and so many things. So we're going to cover some things today, but not everything. [02:10] Yeah, I'm excited. I probably do have seven or eight really good ones, but... [02:15] In the process, I've also created a lot of exhaust that is maybe... [02:19] Not perfect, but it's still fun. It's part of the process. [02:23] I want to start with. [02:25] Actually, a quote from you, something you wrote, I think in just in one of the tidbits in your stream, so not even a tweet or a post. And you were talking about conferences. [02:34] You say, when I write a talk, I almost always just want you to walk away thinking about the technology you create as an instrument for advancing your values and a lens through which to view the world with those values. [02:47] I think, as I mentioned, you are wildly prolific, you're polymathic. [02:52] creatively, technically, professionally. [02:54] But I do think it's clear that values are underpinning almost everything you do. [03:00] And one of those values that I know you care a lot about is agency. Yeah. Yeah.

[03:04] Agency is... [03:06] a popular topic, especially on Twitter these days or recently. But I think the way you approach it is sort of this meta view that technology – [03:16] can be [03:17] power consolidating or widely empowering. You have a frame in particular that I really like. And so my question is, how does technology extend or extinguish agency? [03:29] Yeah, absolutely. [03:31] I think there are two big ideas that I want to, and I think this will be an overarching theme that comes up over and over in our conversation, but there are [03:38] There's maybe a broader idea and then a narrower personal idea. [03:42] The broader idea is that [03:45] technology is, if you're building technology that is such an amplifier of technology, [03:50] your personal work and impact in the world. [03:53] that [03:53] I think it's... [03:55] Good to be thoughtful about... [03:57] the impact that the technology is going to have in the world, not just in terms of [04:01] the economic impact or anything concrete like that, but more [04:06] Anything that you put out into the world [04:09] is going to push the world in some direction. [04:11] And you should make sure that first, you're intentional about that direction, that it's going to push the world in. And second, that you're... [04:19] thoughtful about [04:20] building technology from the [04:24] knowing what direction you want to push the world in rather than sort of [04:26] you know, deriving your ideology as a function of what you end up sort of stumbling on into building. And so regardless of what my personal values are, I think my push to anyone building technology is to be thoughtful about.

[04:38] the fact that [04:39] any direction isn't just forward, it's also in a specific opinionated direction, pushing the world. [04:44] With that said, I think [04:45] For me personally... [04:47] agency... [04:49] as a concept in the sort of recent past, at least for people who are very online, I think means something very specific around [04:56] You know, things like, quote, you can just do things and so on and so forth. I take maybe a more kind of platonic and more... [05:03] slightly higher level view, which is... [05:06] If you have a particular point of view, having more agency means you can more freely go to the things that are and push the world in a way that's that's a reflection of of you and what you want to see. And technology is sort of definitionally like an agency amplifier. [05:22] But... [05:23] I think by default, [05:25] Technology tends to amplify people's agencies in society. [05:29] again by default in a sort of biased way. So, [05:32] One way to view technology is that technology is a way to... [05:36] turn capital into impact. Capital can be human capital, it can be money, it can be other resources, but technology tends to be, you know, [05:43] The shape of any technology is like, I have something that is a resource or money or people or team, and I can turn that more efficiently into something that I want, a result in the world. And so because of that, I think technology by default has a tendency to... [05:58] give more impact or concentrate more impact on, [06:01] towards the people who already have a lot of resources. But I think there's another way to [06:06] But we can't stop building technology.

[06:08] because they don't like it, right? Because technology is also the reason for so many good things in the world and so much progress in the world. And so I think there's a way to build technology in a more opinionated way, where you layer on your own opinion about what technology should look like and how it should be usable, right? [06:20] that maybe counterbalances that default posture, instead tries to... [06:25] be designed in a way and distributed in a way that helps individual people sort of all equally [06:31] kind of egalitarian way. [06:33] distribute their impact in the world rather than just sort of following the default gradient of of. [06:38] Technology at large just, I think, has a tendency to concentrate power. Yeah, there's a piece of that, too, which is equal is a challenging word, but at the very least, the accessibility of it is widely distributed, which I think is really important. Or that it be widely distributed feels like a really important part. [06:55] Yeah, I think... [06:57] there's a lot of discussion, especially regarding AI, of... [07:01] kind of [07:02] in some equitable way distributing the impact, but I think it's also really important that [07:07] the [07:07] the capability for technology to amplify people's will and people's values in the world also be, uh, [07:13] To the extent that anything can be equitably distributed. [07:17] You have one frame on this. [07:20] that maybe it's just a double click or it's the same idea. The phrase instruments of super agency is, [07:25] which I think neatly kind of sits next to the super intelligence idea. Yeah. We talk a lot about super intelligence. What do you mean by that? [07:33] I really like phrases that [07:35] have a lot of loaded connotation in them. And so this is one of those that are like really high density.

[07:41] And there's like instruments and then there's super agency. Super agency may be a little bit more straightforward. [07:47] Super intelligence, as traditionally defined, [07:50] or used is [07:52] You know, it's like awards are all kind of weird. I think superintelligence is also weird because it implies that there is some sort of like... [07:58] finite level of like quote unquote normal intelligence. And then there's like super intelligence, which is the level of intelligence that exceeds it. [08:05] I think that's also kind of a problematic point of view. [08:08] But, you know, to the extent that you subscribe to that way of looking at intelligence, I think Super Agency you can look at in a similar way where... [08:14] there's maybe some normal amount of agency [08:16] Bye. [08:18] an ordinary human has in the world by way of them personally interacting with other people or organizations or with the physical world. [08:25] And superagency would be... [08:27] the concept of giving that person [08:31] wider impact, having that person be able to push the world more in whatever direction that they desire, [08:38] with less effort, much farther. And then instruments, I just generally love as a word, like tools and instruments. Instruments I particularly love because it, in my mind, conjures this image of something that... [08:49] is quite intricate and has a lot of depth and requires... [08:53] some [08:54] practice, substantial practice to gain mastery. And so, [08:58] Instead of this thing where you like, [09:00] write three lines of code or you spend five minutes downloading this thing and then now you've mastered this thing, an instrument like a violin takes, you can take a lifetime to master, it can take centuries to master. [09:10] and

[09:12] that [09:13] There's almost a cost. [09:14] There's a cost. [09:16] Uh, cost maybe. There's also another way to describe it could be like, you have to grow into it. And you can maybe get some benefit by spending... [09:24] three months on it or six months on it, but there is so much... [09:28] deepening layers, so many deepening layers of benefit, the more time you put into it. Like you can't, [09:34] finish violin in the same way that you can like finish a video game like there's there's always more to learn yeah and i like that as [09:42] I think really great tools, at least of a certain kind, which we can also talk about, but I think really great. [09:47] engaging tools have that trait of like, there's always more to get out of the tool and more depth that you can... [09:53] find. I like that a lot. [09:54] You were leading me here. [09:56] one of the things you write extensively about, [09:59] in the context of tools and broadly interfaces and technology is this idea of instrumental versus engaged interfaces. [10:06] One of the first things... [10:07] I remember talking to you about when we first met. [10:10] was this metaphor you have of maps versus GPS navigation. Those two things, I think it can be subtle, [10:17] They have a similar goal, but they can do it in totally different ways, fundamentally different approaches. There's a lot to go into on this, but I think at a super high level, it would be helpful to have you explain instrumental versus engaged. And then maybe specifically talk about why some amount of friction can actually be good for enabling agency in a tool. Yeah, definitely. We've talked a lot about. [10:39] so far at a conceptual level, and this is going to thankfully take me down to a more personal level.

[10:44] When I first started... [10:46] really thinking a lot about tools. The way that I got into it was that I was a huge productivity nerd, probably a lot of [10:52] A lot of you listening. [10:53] And I would try every new to-do list tool, every new note-taking tool, you know. And there was a time in the tech industry when, like, note-taking tools are very hot and it was, like, the sexy thing to work on. [11:03] And so I got into thinking about tools that way and amplifying agency that way. [11:07] And. [11:09] I think a lot of people who love thinking about this kind of stuff [11:12] have some intuitive bias for... [11:15] Like... [11:16] some intuitive bias that like working for something and needing to work for something to get something in return is like inherently a good thing. Like they want to see more detail. They want to be power users like power. There's a maybe I'm grossly oversimplifying, but. [11:32] a lot of people, including myself, that really like thinking about these tools, [11:36] want to be power users and think that [11:38] The world would be better if there were more power users of all these tools. [11:42] And I spent a few years kind of building tools with that assumption, right? [11:47] And I think one of the ways in which I've grown as a person who thinks a lot about building tools is that, [11:54] Like, actually, Ivan at Notion at one point in like a company, All Hands, said this thing, which is like Notion's all about. [12:00] democratizing the ability to build tools for yourself, which is like, oh, that's like such a me idea. Like, I love that. But then he also said this thing that was like, [12:08] Normal people don't wake up in the morning wanting to build software. They just want... [12:13] to do have fun. They want to like help somebody. They want to solve some problem in their life. And building software is this like incidental thing that they sometimes have to do to like make that a little bit easier.

[12:22] And I think one of the ways of ground concretely is [12:25] For most people... [12:27] include it like this is partly a function of people partly a function of what they do like even for me there are certain things in my life where like i really don't care how it's done i don't care until like [12:36] like interface deeply with the texture of the task or whatever people say. It's like, I just want something delivered to my home. I want to like end up somewhere where I'm going to get an advice chair and [12:44] I just want the result. [12:46] And so... [12:47] The taxonomy that I make when I'm thinking about [12:50] building tools these days is [12:53] Sometimes for certain people, for certain goals, [12:57] The way that you want to get there is just by describing what you want and then getting the result. [13:02] And doing that as cheaply and as quickly and as effortlessly and predictably and reliably as possible. [13:08] And that is like an instrumental... [13:11] point of view of tools. The tools are there to take some specification of goal and deliver the result as quickly and cheaply as possible and reliably as possible. [13:20] But then obviously there is this other kind of tool whose job is to like [13:24] get you as deeply engaged with whatever you're doing [13:28] And like, [13:29] Musical instruments are a great example of this. I think maps are a great example of this, which we'll talk about in a moment. I think, like... [13:36] Lots of... [13:37] Crafty tools like IDEs are another great example of this. Their job is to put you face-to-face with... [13:44] all of the requisite complexity of whatever you're dealing with. Yeah. And, and like, [13:49] The complexity is actually good in that case, because whatever you're doing requires you to contend with that complexity. Like if you were...

[13:55] trying to perform a sonata, it would suck if you, like, had to just press a button and then, like, listen to whatever was generated. Like, you actually want to perform to your fullest, like, with as much nuance and detail as you can put into this thing. And it's great to have an instrument that lets you express that. [14:11] And very concretely, if you're buying a more expensive piano, [14:15] One of the reasons that better pianos can be better is that they let you express the [14:20] more deeply, but then you also have to think more about what you're doing. Yeah, more mastery required. More mastery required. [14:26] It's not just like a Guitar Hero thing where you do the thing and there's like a ceiling to like how good you can be at it. [14:31] Maps and GPS, I think, [14:33] or the most kind of canonical example of this that I use where – [14:37] Sometimes GPS is totally the right thing. Like if you want to just like get in a car and end up somewhere, you're just like in a rush. [14:43] GPS, great. Self-driving car, even better. [14:46] Sometimes... [14:48] I went to Catalina Island recently. It was like a very spur-of-the-moment trip, and I went there, and there was this, like... [14:56] tour guide station thing and I went in and I bought a physical map. [15:00] Uh, [15:01] And I had a phone, but I bought a physical map because it had some hiking trails. And I used that map to run around. And I felt like it gave me... [15:08] such a [15:09] It was, like, fun to... [15:11] Maybe this is a tried example, but it was fun to learn about the physical space around me by actually working with this physical object. And I wouldn't have – the point of me being at the island was roam around and learn about what was there and not to get somewhere. And so in those cases, when you're composing music, when you're like –

[15:28] Sometimes in your writing programs, which maybe we should talk about, but there are lots of areas and lots of different domains where it's good for the tool to force you to contend with the complexity. [15:37] And in those cases, I call those engaging things. [15:40] interfaces or engaging tools and [15:43] It's not that... [15:45] Like, a common fallacy that I... [15:48] Sometimes I've fallen into the past is to say that [15:50] Like either that or [15:52] Some people want instrumental tools and some people want engagement. [15:55] that there are these power users versus some people are lazy and some people are. Yeah. Like that's, I've fallen into that fallacy at times of all, I've also fallen into the, another kind of fallacy at times, which is to say like certain tasks require mastery and certain tasks don't, which I think is also not true. I think it's really a function of all of these things. Like, [16:12] For some people, at some moments, you just want the result. [16:15] And it may be for the same people in other moments and other tasks, they actually do want the mastery and the complexity. And it's just very station-raditional dependent. And so even if you're building, like, [16:25] a calendar tool, productivity tool, an IDE, what have you. [16:28] Depending on who you're selling to and why that person is trying to use that tool, sometimes the right thing to do is to make it as cheap and easy to get the results as possible, and other times... [16:37] the [16:38] revealing the full complexity. [16:40] of the medium is the right thing. [16:42] Thank you. [16:43] One of the things that sort of feels like we're moving towards, and I actually talked about this with Jeffrey Litt on the podcast as well a bit, [16:50] This is a quote from you. "The ideal instrumental interface for any task or problem is a magic button that can one, read the user's mind perfectly to understand the desired task and to perform it instantly and completely to desired specifications.

[17:02] You might call that an agent, which is obviously very top of mind with regard to LMs. [17:08] We'll talk more about the specifics later. [17:11] from a super broad standpoint, [17:13] it seems that we're sort of on... [17:15] a trajectory towards all utility needs being instrumental. [17:20] Over time. [17:21] And I'm curious for you to either... [17:23] Challenge that. [17:25] Or even just like think like, do you do it seems that at the very least the trajectory of the slope is going that way. So to go back to your the simplest example, if I'm just trying to get somewhere, I don't need a physical map. I probably don't even need to turn my turn now. Now, I just want Waymo. Yeah. And if I'm on a vacation and I'm trying to like hang out, have fun, leisure hour, whatever it might be. [17:46] I'll take your complexity. [17:49] Maybe that's an oversimplification, but I'm curious what you think. [17:53] I agree with your... [17:55] intuition that [17:57] this kind of thing that people are building that [17:59] I guess we've just had us call agents. [18:01] is... [18:02] trending in the direction of like the perfect instrumental interface that I talk about. [18:07] I could imagine some [18:09] future indefinite point where this thing can read your mind and perfectly execute whatever you want. Yeah, before you even realize you thought it. Yeah. [18:18] The thing that I'm concerned about, [18:20] is that... [18:23] the kind of things that engineers and people who build things build is partly a function of what they think should exist. But there are also lots of other factors that influence what people build, like... [18:32] other things that influence what people build include like, is it cool to build this type of thing? And like, is it easy and all these things. And the thing that I'm concerned about is that,

[18:40] because agents [18:42] feel very... [18:44] sexy right now and it feels novel and it is in some ways I think [18:50] the easy way out to build a new type of thing that gives you new power. Yeah. [18:55] that because of these reasons... [18:57] problems that [18:59] maybe better solved... [19:01] by forcing the user to contend with the complexity of something are going to instead be solved [19:07] by these instrumental tools that take away agency of the user. [19:10] One example of this might be like... [19:14] Actually, software is a good example where there are a lot of [19:17] agentic coding tools around. [19:19] And... [19:20] I think there are ways to intentionally use them that are really powerful. On my team at Thrive right now, our designer has been... [19:27] using a lot of these coding agent tools. And it's been really fantastic just seeing... [19:33] the amount. [19:34] Not just that we can build more, but then we could build better looking, better feeling, better working things. [19:40] So there are great ways to use it. [19:42] But... [19:43] Because Asians are... [19:45] the sexy slash maybe the, the like, [19:48] creatively easier thing to build right now, I think, [19:51] maybe too many people are using agents too much of the time and building software systems that they don't fully understand or they're not... [19:58] they're not [19:59] pushed to fully understand. [20:01] And I think that'll probably come back to bite those people later. [20:05] We'll talk more about some of the merits of the instrumental side. [20:09] later, but I want to talk about

[20:11] the engaged part and specifically what you call a technology representations. Yes. [20:18] you have a definition of engaged interfaces. [20:21] or at least a description of them. [20:23] oriented around two ideas, seeing and expressing. [20:26] You say a good, engaged interface lets us do two things. One, lets us see information clearly from the right perspectives. And two, express our intent as naturally and precisely as we desire to see and express what creative and exploratory tools are all about. Obviously, alluding to the earlier part, creative and exploratory tools being different than purely... [20:45] need-based ones. [20:48] Obviously maps are a great example of this. [20:50] You also say a representation must abstract. [20:54] And so what you're pointing to here, I suppose, is in my question, I guess, is, [20:59] Why can... [21:02] Highly accurate faithfulness to reality be a bug and conversely lossiness be a feature put another way. [21:10] Does complexity always reduce agency? [21:13] Oh. [21:15] Maybe those are two separate questions. I think those are two separate questions. I think there's actually three different... [21:20] things here, which I will take a moment to write down because they're all good. [21:25] Okay, so I'm going to go back to the first one, which is this idea that good engaged interfaces have two jobs. It's to help you see what's happening. [21:32] And [21:33] help you express your intent as fluidly as possible. [21:37] This is one of my favorite sentences of all time that I've come up with because I remember I was...

[21:43] I had like a bit of a mini existential crisis where I think I was in like Berlin at the time. [21:47] And I had just come back from like a conference where I had like talked about tool building. [21:52] And somehow I got it stuck into this like conceptual rut of being like, [21:56] okay, I work on interfaces. Like I talk about interfaces all the time, but like, what is a good interface? Like, how do I know, how would you grade how good an interface is? [22:03] And I like started melting at this like realization that I talk about interfaces and how to make them good and actually don't know what good interfaces are or what makes them good. So I was like, oh, my God, I need to like. [22:15] I need to figure out what my definition of good is. [22:18] And what I came up with... [22:20] was at the time I didn't have this like instrumental and engaged vocabulary, but [22:24] in current terms, [22:26] what I ended up coming up with then was like, see and explore. [22:30] good technologies, good interfaces, [22:32] let you very clearly see what's happening. They don't obscure things that don't need to be obscured. [22:36] And then they let you take action on it, which is the mechanism by which your agency impacts the world. [22:41] And you can sort of express your intent. [22:44] towards some arbitrary level of precision that is needed for you to have the desired, like, [22:48] level of agency. [22:50] And so I really love that phrasing. [22:52] An inherent part of that is like, [22:55] Obviously, you can't see everything. If you're looking at a map, if you're looking at a diagram of anatomy, if you're looking at a chart, if you're looking at even like a code base through an IDE. [23:04] a design tool [23:05] You can't see every little detail. [23:07] And in that way, everything is kind of a map. [23:09] And so critically, I would add, like the map not being the territory is –

[23:13] such a feat. You have this amazing excerpt from something where you're talking about like the Empire built an Empire sized map in the Empire. Yeah, it's like perfect one to one. And it's so useless. I believe that is a Borges short story. Where? [23:29] it [23:29] Yeah, many of you may have read it before, but it's about this kind of [23:33] group of scholars that decide to build a map and they desire to build more and more accurate versions of maps so that the map has to get larger and larger and larger until the map is exactly the size of the territory and it covers every inch of the territory and then the map is totally useless. It's like an evapid exercise in like, like scientific accuracy, which brings us to abstraction. [23:51] I think abstraction... [23:54] is necessary for [23:57] Agency. [23:59] It feels kind of intuitive to me why, but if I had to articulate why, it would be that, like... [24:05] I mean, put it in a pithy way, like if you had a map that was the size of the United States, you would not be able to carry that map or look at that map to go somewhere. [24:12] it would just be like unwieldy. Maybe more concretely it means that [24:15] If there's too much detail and you can't [24:18] The point of an abstraction is to give you a model that you can fit in your head and work with. Yeah, it's almost compression. Yeah. Like if I want to understand the transformer or if I want to understand some part of the economy or a company, the point of having a model or an abstraction is that I can just fit that in my head and then sort of like do things with it to figure out. [24:34] how to understand it or what action I need to take to push that [24:37] that reality in some direction that I want. And if there was infinite levels of detail... [24:41] The point of having a model would be lost because then I could just like mess around the things in the real world. But it would take forever to understand every load of detail.

[24:50] And this is true of all of scientific models, and I think also abstraction... [24:54] generally and also in the software sense. And so I think good abstractions [24:58] necessarily lose some detail. [25:00] And this is... [25:02] There's some room for opinion here, right? And so, like, if you... [25:05] If you go to Google Maps or Apple Maps, there was a bunch of different options of which abstractions do you want to use [25:11] your map with. You can do the terrain, you can do the transit map, you can do the road map. [25:17] And these are different models of the real world. None of them reflect reality fully, but they're useful for different types of things. Yes. You're also not likely turning them all on at once. Correct. Yeah, that would also be a very hard map to use. Yeah. [25:30] One of the things I think this leads into is... [25:33] this juxtaposition between [25:37] constraints and abstraction and [25:41] more complexity and the way that that balancing between those two things can give or take away agency. You've said we need diverse and accessible representations. [25:50] But then you've also talked about [25:52] Actually, introducing complexity is almost being a way to take the user seriously. Or you think about the musical instrument, the more complexity, the more you can do. I love that. Take the user seriously. Yeah. Yeah. [26:01] forcing the user to contend with complexity, in your words. [26:04] You also said, though, on the note of constraint, reducing the number of choices the user has and contextualizing the input UI to shape the behavior. I think that's me paraphrasing you, but talking about interfaces. And then as maybe a last example, [26:19] an excerpt from you where you're talking about video games that I think captures this tension. You say,

[26:24] A video game, for example, can sometimes be better by being more realistic and easier to learn. But this isn't always true. [26:31] Sometimes the fun of a game comes from the challenge of learning its mechanics or strange surrealist laws of physics in the game world. A digital illustration tool is usually better off giving users more precise controls. But there are creative tools that lead artists to discover surprising results by adding uncertainty or elements of surprise. And so a needlessly challenging question, maybe, but. [26:54] Can you square the circle in this? How do you think about when we actually want to add complexity and friction versus abstract and make things simpler for someone? [27:04] Yeah. [27:05] This is a hard question. I think [27:09] First, I would go back to the top of how we opened and say... [27:13] Some part of this is about... [27:15] as a creator of a tool, let's say you're [27:17] building a drawing tool. [27:19] That's all for making images. [27:22] As. [27:23] The designer of that tool... [27:25] You may... [27:26] have some aesthetic that you want to proliferate into the world that says, [27:31] You could say, I think the world would be more interesting if... [27:35] we pushed artists to create with imprecise tools or take into account more serendipity. Like this is a this is an opinion you could express through the vessel of your the tool that you're making. Yeah, it's almost a prompt. [27:44] You think about a writing prompt or any kind of prompt. It's actually like, and it can be really effective for helping somebody get creatively started. Yes, exactly. So this is a reflection of...

[27:56] your values or your aesthetic. [27:57] End. [27:58] Again, going back to the top, sometimes you make things, I think sort of always, ideally, you make things as a reflection of your aesthetic or your values. And other times, you may want to just give people as much power as possible. And then the precision and the reflection of reality may be useful. [28:12] So that's one way of looking at it. [28:16] Another way of looking at it... [28:18] Maybe. [28:19] that sometimes you're building instrumental tools and you want to just [28:24] help the user elicit what they want out of themselves as easy as possible. [28:28] For different tasks also, there are maybe different [28:31] correct levels of abstraction. And so... [28:34] if you are... [28:36] trying to research a company for a high school research report. [28:40] You need a different level of abstraction than if you're trying to do like a leverage buyout or something. [28:45] And actually, there's there's [28:47] Even on this side, on the more engaged, more detailed, complex side, there is still room, I think, for... [28:54] tool builders to express their preferences or values, I think. [28:59] One interesting avenue of this is... [29:02] There are a lot of tools... [29:04] There's a handful of really interesting tools out there that try to help researchers understand excellent literature about a topic, like medicine and machine learning. [29:12] And [29:14] Some of them, they stretch a gamut. [29:17] Some are... [29:19] more sort of [29:20] power usery and some are more tell me the question and I'll deliver the result. And I think depending on

[29:26] And all of those are whether intentionally or not. [29:29] expression of some value that the creator of the tool has about what the right level [29:34] of detail is for the reader or the user of this tool to have to contend with. [29:40] A tool may give you the answer or it may say, here are, you know, 20 papers that agree with you and 10 papers that don't. And like now deal with this. Yes. [29:48] depending on who you are, you may want to like [29:51] force the researcher to deal with this conflict in the literature. And so [29:56] The fact that the spectrum exists, I think, gives people that build [29:59] Tools. [30:00] room to like express what they think is the correct level for that particular task for that particular person. Totally. It also sort of seems almost that the instrumental engaged idea is a gradient in and of itself. [30:10] And that maybe as tools become more adaptable over time, like a good video game. There's a notion I really love about games from this guy, CT to win, where he talks about like video games designers sculpting agency. [30:21] If a game's too easy or too hard, it's almost adapting with you, which I think is important. Yeah. Yeah. [30:27] There's a related... [30:29] kind of round that I've... [30:31] given to some of my friends recently, which is... There's this, like... [30:35] And with any given level of technology, there is this frontier of like, [30:39] If you want to build something that is easy to use, you need to sacrifice some level of complexity... [30:44] And then if you want to make something really, really, [30:46] Accurate and detailed, you may need to sacrifice some like learnability or usability. [30:51] So there's this front here. [30:53] But this frontier... [30:54] changes over time. And this frontier itself is a thing that you can move. So I'll give you a

[31:00] I was talking about, again, coding AI with a friend recently, and... [31:05] I was... [31:06] So one kind of advancement that you can make is to build systems that are better at using current technology to build software. And so I went on this... [31:13] thought experiment where I said, let's imagine that we're in an alternate timeline. [31:18] where [31:20] We have built... [31:22] super intelligent, super capable coding systems, code generation systems, but the best programming language technology that was available was C. [31:31] And so you have this, like, super coding AI. You can tell it, please build Google Chrome. And it'll, like, write a perfect C program in one shot that is Google Chrome. [31:41] In this world, [31:42] would you still want to invent Python? Would Python still be useful? And for some people, no. Like, if you're just, like... [31:49] a normal person going around your day, like not building software, then like it does not matter. [31:53] Bye. [31:54] If you're a person whose job is to think about what software to build and how to build robust, resilient software systems in the world, then actually Python is a huge advancement. [32:02] because... [32:03] at the same level of complexity of software, it lets you build it much more easily. And conversely, at the same level of ease of use of software, it lets you handle much more complexity. And so it's... [32:15] I think, a notational innovation or maybe an innovation in how we represent and model software. It's an innovation and abstraction. [32:23] that pushes this frontier trade-off between [32:28] between kind of simplicity and complexity forward.

[32:31] So that. [32:32] you can, I mean, in a way, it goes back to like the seeing and exploring thing, right? Like in [32:37] the system is doing and how it's composed and you can express your intent [32:40] much more fluidly without having to worry about things like pointers. I love that example. It's, [32:45] a little zoomed out, but it makes me just like, [32:48] it feels not that far from the notion of like, if, if we could communicate telepathically, uh, [32:53] would we have needed to invent language? [32:56] And it's kind of silly that the first example that comes to mind is there's a sci-fi trilogy called The Three-Body Problem. And it's a slight spoiler, but like the aliens... [33:05] we find out like can't really lie because they communicate more or less like they have no difference between thinking and talking. Yeah. And so it's really interesting to think about like. [33:14] There's just so many ways you can back into the interesting ways that we shape our tools and our tools shape us. [33:20] Yeah, and language and abstraction or core tools. [33:23] The next thing I want to talk about is... [33:26] Maybe I'll start with a few quotes from you about [33:29] Thinking Tools. [33:31] You say, despite its ubiquity, the most interesting and important part of creative knowledge work [33:36] the understanding, coming up with ideas, and exploring options part, [33:40] still mostly takes place in our minds. [33:43] With paper and screens serving as scratch pads and memory more than true thinking aids, there are very few direct manipulation interfaces to ideas and thoughts themselves, except in specific areas. [33:54] constrained domains like programming, finance, and statistics, where mathematical statements can be neatly reified into UI elements. And you go on to say, in the best thinking tools today, we still can't play with thoughts, only words.

[34:06] While building tools to solve hard problems for humans, we should strive to also improve people's depth of engagement, going back to earlier conversation, with those complex problems and their solutions as a way to preserve human agency when working with increasingly capable aids for our work. Otherwise, we risk losing touch. [34:22] touch with, and therefore understanding over critical decisions. Mm-hmm. [34:27] Tools for thought is a can of worms as well. Talk about agency. [34:33] But... [34:34] Nonetheless, it's an area you've spent a lot of time, and it's, I think, an area that a bunch of [34:39] curious people are very interested in. And I think it ties really nicely back to our conversation about representations. [34:46] Can you talk about why... [34:48] tools for [34:50] or rather representations for thinking, maybe it's just notation, per the thing we were just talking about. Why is that such a compelling dream? [35:01] One thing I'm really... [35:02] Personally... [35:03] Afraid of? [35:05] As we... [35:08] Build more technology is... [35:11] I'm really afraid that... [35:14] As we build stuff that absolves us of our need to really understand what's going on, that we are going to be pushed to not care about understanding what's going on. [35:24] again, this comes back to this like instrumental versus engaged thing, right? But like, [35:29] I think for all critical... [35:32] systems in the world [35:33] like things related to money, things related to health, things related to how we govern ourselves,

[35:38] We should have like really smart people who are who really deeply understand and try to advance the frontier of our understanding of how these systems work. [35:46] And I think a core to my belief is [35:48] maybe just based on intuition, is that [35:51] agency or ability to understand and influence what's going on is really important. [35:55] And that... [35:57] To have that level of agency, we need to also have like a full and detailed understanding of how these systems work. Like if you want to... [36:02] Right. Economic policy. You need to have really deep understanding of the economy as a machine. If you want to like write software, you better have a really deep understanding of how computers work and mechanical level. [36:12] And so I don't want to be [36:14] intermediate or automated out of understanding. [36:17] Maybe more concretely, this like... [36:20] kind of [36:20] dream of like a UI where... [36:24] I'll describe it as [36:25] visually what I imagine when I talk about this like [36:28] I think I have a blog post that's called like runtime for structured thought or something. [36:31] This dream came out of... [36:34] At some point I was building a note-taking app because it's like what I do. [36:39] I had this idea at some point that I could not let go of, which was like, okay, I dumped all these notes into my notes app. And then, but then like, [36:46] in the process of actually writing these thoughts down, [36:49] all the thinking is still just happening in my head. Like, [36:53] If you're doing long division, [36:55] There is some part of the long division thinking process that is taking place on the piece of paper. Yeah. Like there is a mechanical thing that's happening with your paper and pen that is like, [37:07] Not a thing that's happening in your head. I don't know if that makes sense. It's almost between your head and the paper. Yeah, it's between your head and the paper. We've invented a way to externalize some kind of thinking into this way of writing things down. I think programming can sometimes feel this way at times. You've found a way to mechanically externalize something that used to be a part of just stuff that happened in your head onto paper.

[37:26] And I [37:28] Really like the feeling of that. I think that's so cool that you can do that at a slightly lower, slower latency or it's like on a slight delay. [37:35] Like it captures a little bit of that, but it's more like, oh, looking what you've already written. Like... [37:39] Writing is different than thinking, but there's like a lag. It's more in that direction. What I really wanted... So the cool thing about Long Division... [37:48] is that it's like, it's really obvious. It's more obvious than without it. When you've done something wrong, or there's certain kinds of like invalid things. - Yeah, yeah, yeah. - That you cannot. - Yeah, yeah, you can't like do if you're like writing down long division. [38:00] or if you're like, you know, doing math on paper. Same with certain kinds of algebra, but this is really restricted to math. [38:06] And I wanted to. [38:08] some way to like think out loud on paper, where if I wrote something that was obviously logically incorrect, [38:14] I physically would not be able to write it down. That would be so cool to have. And so, like, if I... Little devil or angel on your shoulder. Yeah, except the way I wanted this happen is not that I would write, like, two sentences down and then... [38:28] some chatbot on the computer will say, "Hey, I think these are incompatible." Clippy is like, "No thanks." Yeah. Instead one, just like, [38:36] For me to write these things down and then for me to like some I don't know how this would happen, but it would be so cool if like as I write my second sentence down, I like run out of room or like it's like geometrically not possible for me to continue writing the sentence that is incorrect because it's just not compatible. [38:49] And I want this kind of [38:50] fitting the puzzle pieces together [38:52] feeling for thinking out loud on paper, or the way that you write things down is like mechanically helping you think better.

[39:00] And that's like an aesthetic thing that I want to proliferate in the world. [39:04] And so that-- [39:06] That's kind of ultimately what I'm saying. And the reason that I got into... [39:11] At some point, we might talk about embeddings. The reason I got into researching embeddings is because it felt like it was the closest thing that we have so far to embeddings. [39:21] the stream of... [39:22] turning... [39:24] Conceptual compatibility and incompatibility and coherence into like geometric things that can exist in the real world. Mm hmm. [39:31] when I was in the... [39:33] depths of [39:35] latent space [39:36] stuff in 2022 and [39:38] I would tell everyone I could about this idea of like, [39:42] Imagine that in the future, the way that you read a book is not by opening a bunch of pieces of [39:48] the trees, but instead you walk into a room [39:51] And and [39:53] against all the walls, it's a huge room against the walls, are these like sculptures. And each sculpture is like a big idea. And different ideas [39:59] have different shapes that you can tell from a distance. And as you get closer, you see more and more the detail of the shapes. And that detail corresponds to the details in the claims. And maybe you in your head have your own sculpture that corresponds to kind of your belief in the world. And it's like very obvious visually where they're incompatible because they look like things that wouldn't fit together if you like try to match them together as puzzle pieces or something. This like reification of... [40:23] not just like the words, because the words are just like, they're like what we sound like when they come out of our mouths. But like something structural about thoughts.

[40:30] How can you express it visually? And it felt like, [40:33] Embeddings. [40:34] could capture a lot of that where similar ideas are physically closer together and maybe you could turn them into shapes in some way. And I, [40:41] There's some research that I want to continue to do over time in that front of like turning the meaning of ideas into shapes. [40:48] and thereby like [40:50] Yeah, like thinking should be like putting puzzle pieces together. It's pointing at a notion that I think... [40:56] I believe, which is that [40:58] We are fairly underexplored when it comes to spatial interfaces. [41:03] And I think there might be a lot of heavy lifting that, [41:06] broadly AI might be able to do to help us better investigate them. And there's other things that are going to have to happen. And maybe you need to have a VR experience or whatever, but, [41:14] But it almost feels like those two categories of technology are things that might feed off of each other over time. [41:19] We're like, [41:20] there might not have been that much work to be, I don't know, your example makes me even think of something that, [41:26] didn't require computing technology for people to sort of come up with like a memory palace. But it sort of felt like we stalled out on like really pushing the boundaries of spatial interfaces for a little while. And I wonder if that will be something that we return to in the near to medium term. If you can just like generate a world really quickly based on. [41:43] the embedding of a book. [41:45] Absolutely. I think [41:48] What happened, I think... [41:51] the way that [41:52] humans interact with [41:54] information of all kinds. [41:56] has been... [41:58] taken over by... [42:00] writing.

[42:04] I'm not very deep in the pure math world, but an interesting kind of [42:08] I don't know, discussion tidbit that I heard about was like, there are certain kinds of proofs [42:13] that you can express as a diagram. [42:15] Like if you want to prove that [42:17] When you have like [42:18] I'm about to lie. I'm literally about to describe something that can't be put into words, but in words in a podcast. But if you have like a triangle and like a line that's like a tangent, I don't even know if I can do it. But basically, there are there are. [42:29] certain kinds of statements about geometry of triangles and angles, where if you draw a diagram, anyone can look at it and be like, oh, that's so obvious that this is the case. But to prove it in writing requires a lot of mechanics. [42:44] There's an interesting discussion. Different representations. Different representations, and some are more efficient at certain kinds of claims. But like the culture of... [42:51] A lot of the culture of pure mathematics, as I understand it, is predicated on [42:54] written proofs and proofs and prose. And it's a particular kind of written tradition. [42:59] that's biased against [43:02] this other way of arriving at conclusions. [43:05] And I think in general... [43:07] we are so used to [43:10] the, like... [43:11] conflation of [43:13] rigor of thought and... [43:16] writing lines of text on paper [43:18] But that doesn't have to be what thinking feels like and what communication feels like. [43:23] And I think actually that the, [43:25] The way in which this rubs me wrong of like, oh, we've like turned everything into reading and writing text. [43:31] is the same thing that I, it's like the same kind of

[43:35] frustration that I feel about everything turning into chat. [43:38] Some things are fine as chat. Some things are the best expressed by writing things down. But [43:43] in like writing and in books and in the way that like, [43:48] Thinking has... [43:49] come to look like so much of it is just reading a bunch of text, loading all this complicated state into your brain, and then like, [43:55] doing a bunch of abstract, amorphous stuff, and then writing more ideas out. [43:59] That. [44:00] If you have to turn that into the shape of a software interface, that is what chatbots are. [44:04] And instead, I want to work with things like [44:07] charts and tables and plots and... [44:10] diagrams. [44:12] that are just totally other still very rigorous ways of expressing ideas and working with them [44:16] Or maybe even these sculptures and things. But that are... [44:20] more direct and feel more like they exist in the... [44:22] universe of like things that the human body is good at working with. Yeah. There's a lot there. I mean, it's, it's, I couldn't help but think while you're talking like, [44:31] It does feel that we're nearing the end of at least the written tradition being the sort of dominant... [44:36] way of media being consumed and created. [44:39] and we're moving to something that's actually more oral, like audio and video. But it's experienced through, on one hand, that's experienced through a two-dimensional screen, so there's maybe less of this spatial interaction. And then simultaneously, as you mentioned, [44:51] We have. [44:52] large language models showing up at maybe the tail end of the text itself. [44:56] World, society, whatever – [44:58] grounding us back in text, at least in some part of it. But yeah, it feels that all of that is at least very flat. [45:04] which is interesting.

[45:05] Bye. [45:06] Yeah, there's a lot there, as you said. I'll add maybe one more bit onto that, which is... [45:11] Again, this, like... [45:12] Coming back to... [45:15] Building technology that's a reflection of how you want the world to look like. [45:18] Even in the history of... [45:21] the written tradition. [45:23] Technology has changed [45:25] further and further constrained what writing looks like. [45:29] Obviously, in the beginning, we had... [45:31] Like... [45:32] handwriting for a long time we had handwriting [45:35] And then we had sort of like... [45:37] what I'll call like more structured handwriting, which is like, okay, you're like writing, you know, like manually copying down the Bible or something. But like you have to let, you know, roughly like you have to put things in lines and there are these like guardrails around the paper and all this stuff. [45:51] But you're still like handwriting. If you make a mistake, you just write over it. And then we have typewriters, and we have typesetting, and now we have digital documents. [45:59] And like, [46:00] If you're writing an HTML document, [46:02] it's really hard to make stuff that's just like free form, where if you look back at [46:08] really old books. I went to [46:10] The London National Library, Public Library, British National Library. It's one of those things. They have an exhibit of old books. I totally butchered that, but I still love the library. [46:21] Where the really cool thing about these books is that they were so... [46:25] expensive to make that there were like [46:27] heirlooms and like yeah it's almost like a painting yeah exactly and [46:30] But these are also fully handwritten. And so... [46:34] there was no such thing as like margins and like lines and line height and spacing and all this stuff. You just like wrote stuff where there was space.

[46:41] If you look back at books where this was like the way of producing books and like there was no preconception of. [46:47] any of this typographic stuff, you just kind of wrote things where there's space. And over time, we've [46:52] And also, if you wanted to draw a diagram in the middle of your writing, you just like draw a diagram. If you want to like if you want to emphasize a word, just like write it bigger, you know, and then it's gotten more and more difficult. [47:01] to express to like take advantage of these other richer axes of expressing yourself. And so instead you have to like. [47:08] you know, italicized in one specific way or bold in one specific way. Yes. And we have kind of like the richness of writing has calcified into the set of markup that we have today. [47:18] Again, like doesn't have to be that way. [47:20] And it would be cool to... [47:22] Like someone should write a... [47:24] Actually, this is even continuing today. [47:27] Markdown is so ascendant. [47:30] I remember when I was in high school, [47:33] Markdown was not so everywhere. [47:35] and [47:36] like most people are on Google Docs and Word, and Google Docs and Word, it's so easy to make any text, any color, any shape, any font you want, and you can put it anywhere. [47:43] Markbond, you can't call your text. And so people... [47:47] I think there's a real cultural shift, again, where the aesthetics of this really messy Word doc or Google Doc is kind of getting sucked out. [47:55] In some settings, that's really nice, but in general, I think that's kind of a loss. Yeah, I didn't realize that was the way you were going to go. I'm very pro-markdown, but that's a great – and there are a lot of benefits, but – [48:04] Yeah, it's really interesting. [48:06] We're circling representations. A few more... [48:09] kind of ideas on representations at a super high level and maybe leads to where we're going to go later in the conversation. Um,

[48:15] Hanging over so much of how we think about stuff today is LLMs. You maybe got at this a little bit with the embeddings note. You say language models decouple the way information is stored from the way information must be presented and consumed by humans. [48:30] What does that imply about how we interact with information? [48:34] I'll turn the question on you. What does that imply? [48:39] I think a lot of ink has been spilled around the fact that [48:43] Language models are good at translating one form of information to another [48:47] I think that's correct. I think that's also maybe not the most exciting way that I look at LMs. It is a thing that LMs are great at. Yeah. [48:54] I think many of the people listening will know that like, [48:57] the transformer architecture was originally invented for [49:00] translation and that's the form of data transformation. [49:05] at least in the, like... [49:07] Language modeling realm. [49:09] So this data transformation way of looking at models is fine. The more exciting thing to me is like, [49:16] again, going back to this embedding stuff, the more exciting thing to me is language models in the process of being [49:21] trained. [49:22] they have to somehow... [49:25] internally derive some way of like [49:28] doing a thing that looks like thinking in some mechanical way. [49:33] Like. [49:34] I always love physical metaphors and [49:37] If you imagine a language model [49:40] The most visceral metaphor that I have for, like, why a language model, my understanding language models are cool. [49:44] is that imagine you have like, I love that, if I was like infinitely wealthy, and I just had a bunch of side projects I could spend money on, one of them that I would do is to build a physical version of GPT-2.

[49:54] where imagine you have like [49:56] a building that is maybe like [49:58] I don't know how many layers distributed to like, [50:00] 12 layers. They imagine a 12-story building [50:02] And it's like, you know, one of those like marble [50:05] rolling down the tracks things like imagine it's just like one huge rube goldmerg machine and you put like a token you put a token at the top that's like a ball with like the token number like 32 or something and it's like rolls down and somehow there are all these like [50:19] rails carved into the building such that maybe it takes like three days for the ball to roll down but by the time the ball rolls down it like rolls down very neatly into like the next token like [50:28] There is a, the reason I think language models are cool is because there is a mechanical process that they, like, [50:35] Or... [50:36] it's like fundamentally geometry happening inside the model that somehow embodies the [50:40] All of this stuff. [50:42] And obviously it would be really hard to turn it into a physical building, but the intuition of like, okay, there's something like physical and geometric feeling that is much more physical and geometric than like what writing feels like that is happening. Oh, this is so cool. Yeah. There is. [50:56] But one kind of data transformation that language models do is like inputs to outputs, which is fine. [51:00] And that's sort of like [51:02] maybe less surprising because that's what the models are trying to do. But a more surprising kind of data transformation they do is they – [51:08] like, [51:09] Language models. [51:11] Use the first few layers to [51:12] translate [51:13] Again, kind of simplified, but they use the first few layers to translate the infotokens into some representation space that... [51:21] is more geometric that the model can contend with, and then spends most of its compute working with ideas in this domain is like shapes.

[51:29] And only the last few layers are... [51:32] used to then translate that, its own kind of internal thought, back into the human writing. Yeah, yeah, yeah. And so this idea that there's this other way of thinking about stuff that is [51:41] way more mechanical than writing. [51:43] that exists. [51:45] that we can kind of like do science on and figure out is so cool. [51:48] Yeah, they're sort of like distilling it into essence... [51:52] space and doing a bunch of work with it in essence space and then like giving us some, there's a, you've talked about so many different, you've built and written about and talked about a lot of different examples of ways we might create new representations and new tools from the prison project to, um, [52:10] broadly, like what would a synthesizer for thought be? I think a video you shared with me early on was this great [52:16] talk on liquid art. All capturing a lot of these ideas. There's a tool called Loom that you refer to. Loom's brilliance is that it lets the creation happen in a different perspective. Painting in time first, then filling in the details in space. My phrase for this perspective shift is a perspective transformation. [52:35] because it reminds me of coordinate transformations. Synthesizers, you say, because synthesizers are electronic. Unlike traditional instruments, we can attach arbitrary human interfaces to it. This dramatically expands the design space of how humans can interact with music. All circling this idea. There's one final one you're talking about. Your friend who is working on Flora. [52:55] The founder, they told me about a mission statement I always found really inspiring. He wants to allow artists to speak beauty into existence. I love this phrase because all of the focus here is about the precision of the words and the ease with which the words can conjure ideas into being.

[53:12] There has got to be a way to get there by finding better ways to speak of beauty rather than by mechanizing the means of beauty production. Oh, I love that. I know I said it, but it's like and like, man, I mean, we could talk about that for a long time. But like it's really interesting, like you're you're you're dancing around actually both the mechanical side of and these like way more mystical, like ambiguous, almost. [53:36] ineffable side of it. [53:39] My question might actually be a little lower level, which is just like, [53:44] You're describing some kind of cool toys or some cool theories or ideas. [53:48] Synthesizers took a while to be taken seriously by musicians. Yeah. Um, [53:53] Where do you think we are with these types of ideas today? Like, what are the seeds that are actually promising? Mm-hmm. [53:59] Beyond like fun blog posts. [54:00] Yeah. [54:02] First of all, I'll comment on the [54:04] the better ways to speak of beauty rather than mechanizing the means of beauty production thing. [54:10] I think... [54:11] It's kind of mystical, but there's a concrete way to think about this, which is like, [54:16] Python is a better way to speak of software. [54:21] Mechanizing... [54:21] the means of software production would be writing a language model that's really good at writing C. [54:26] Wow. [54:27] And I think we're all better for Python existing in the world. Yeah, elegance doesn't mean... [54:32] ambiguity. [54:33] It's actually more clear, right? [54:36] It's... [54:36] more useful abstractions of a certain kind. If you're writing a database, Python is too messy. But if you're writing Instagram, Python may be right.

[54:45] And [54:47] If when I think about like [54:49] what advancement as a civilization... Maybe this is too grandiose, but when I think about advancement as a society or a civilization, like... [54:55] what would really affect an alien species landed on Earth tomorrow? [54:59] What would really impress me [55:01] is like, it would be really cool if it had a language model. That would be sick. But what would be more cool is if like, [55:09] they had really elegant physical theories or they found some more elegant way to think about music or... [55:16] They have a thing that's like the periodic table, but like even more elegant, like these things would really be impressive. [55:22] Because they are... [55:23] out of [55:24] more impressive further compressions of knowledge. [55:26] Have you read Story of Your Life or seen Arrival? [55:29] Yes. A little bit of that in there. A little bit of that. [55:32] And these things coexist, right? Like, [55:34] mechanizing the means of working with this knowledge is useful, [55:37] is useful for scaling these things. [55:39] and automating these things. [55:41] Like, [55:42] That by itself... [55:44] feels kind of empty. [55:46] aesthetically. [55:49] And then just seeds of... [55:50] Any places where you're starting to see interesting things that could soon be real in this category? [55:55] broad category. Yeah, I mean, this is what I've spent the last... [56:00] I guess, two years thinking about. I spent... [56:04] the prison project. [56:06] which is about... [56:08] Figuring out ways to read... [56:10] read interesting ideas or concepts or features out of latent spaces of models. That was really fun. That was actually also the first time I ever wrote deep learning code. It was like this research project, which is kind of wild. Also kind of hilarious project. Like a lot of its outputs are hilarious. Oh yeah, yeah, it's so funny. I remember like

[56:27] You can do things like you can put in my... [56:30] like bio from my blog and then turn off the coding feature and like turn on the culinary arts feature. And then it's like a bio that's like exactly the same style. I made a hundred dishes. Yeah, exactly. Exactly. So it's super, super fun and interesting. But like, [56:44] I think the reason that it's worth digging into why it feels interesting is... [56:48] it's a totally different way to interact with ideas. [56:52] Now, the challenge that I've spent the last two years with... [56:55] is... [56:56] Okay, you have this kind of cool toy. It can let you... [56:58] play with ideas in a really fun way. [57:01] How can I... [57:03] how can I make this more of a thing in the world? [57:06] And that's all about like... [57:07] That's partly about packaging the idea. That's also partly about like building a thing that's actually useful in some way that people want to use or people want to play with. [57:13] or people need to use because it's valuable for their business. [57:17] And that has turned out to be... [57:19] the difficult part. I think there are some [57:22] avenues that are interesting. There's a really cool company called Goodfire that is applying this in a more research domain where they're applying... [57:29] interpretability techniques to [57:31] very non-language modalities like [57:33] models that understand molecular biology and cell biology and... [57:38] other kinds of things. [57:40] But that's more in the research realm. [57:42] I'm still very interested in language. I think the most... [57:46] viable path to [57:49] production impact that I've arrived at is [57:53] Using this kind of thing to help people make sense of really large... [57:56] datasets where nuance is important.

[58:00] An example of this is like understanding people. [58:03] a big part of venture. [58:05] the business that Thrive is engaged in. [58:07] is about understanding people. [58:09] and how they work and where they are and what drives them. [58:12] And, uh, [58:14] data sets about people are really interesting because there's a lot of structure, but all of the important signal resides outside of that structure. Like, sure, I can get a list of everyone that has ever worked at OpenAI, but, like, that's not what makes them interesting. What makes them interesting is, like, [58:30] Like, what game are they playing? Like, why are they interested in this company? What are they looking for? What are they... [58:35] what really excites them. [58:37] And those things are all between the lines. Yes. [58:41] using some of this kind of [58:43] looking inside the model stuff. [58:45] we can learn to work with these data sets, not just at the level of like, where do people work? And like, what name are you? [58:50] what name do you have, but more at the level of like, [58:53] What seems to be like between these types of things that we're looking for, what seems to be the one that's like driving this person? [58:59] and and [59:00] Rather than just throwing it to this black box that's going to tell you a bunch of names, [59:05] roll it out in a sheet of paper and like... [59:07] different motivations and different intensities of motivations can correspond to, for example, different places on this chart. [59:13] And that that [59:15] There are so many reasons why I think that is much more exciting to me. The obvious one is just like... [59:19] It's more... [59:20] visual in more detail, but also [59:22] Working with people and making judgments about people is a really high stakes thing. And it's really... [59:28] It feels really important to me

[59:30] from like a principle perspective, that the people that are making those judgments are exposed to the full complexity of people that this data models. [59:37] People are such a good example because... [59:41] They're the rare category where there's incredible complexity. [59:45] I shouldn't say everyone, but many people actually have a strong degree of intuitive knowledge. [59:51] high resolution thinking about it. [59:53] but they can't always put it into work. Like there are people who are incredibly high emotional intelligence, for example. Yeah. [59:59] And they might even have, and you see this in all categories, but maybe especially people, they have this sort of high dimensional, nonverbal, [1:00:06] vibe or sense or notion that somebody's special. And we, in the words, [1:00:12] All the words that you're using are like, what do they even mean? What are they holding? Or, I don't know, investor firms, they'll say, that's a person special or good or even smart. Like, what is an insider? Yeah, or like they might use the word smart in like a really specific way or like spiky use. People say... [1:00:28] crafty or like cracked like what are these words like don't we don't we deserve better vocabulary for talking about these really important things yes but it's also cool because i there there actually is a huge light in space there that is [1:00:43] in certain people's heads. And it's like, it hasn't been extracted. Yeah, there is complexity there that... [1:00:49] I think... [1:00:49] my, I guess, the value that I want to... [1:00:52] push out into the world is that there is complexity there that we currently have very ill-suited tools for. Yes. That like, I want anyone working with these...

[1:01:01] data models of people and [1:01:04] CRMs and tools like that. I want to... [1:01:07] I think the world would be better if we forced them to contend with the complexity of what's inside people. But I also want to give them good tools to be able to do that. [1:01:16] One theme that runs across your work [1:01:18] that you actually started to maybe allude to when you were talking about warming up to instrumental interfaces, is this sort of balance between very philosophically inclined, highly principled thinking and being pragmatic and actually making stuff that's useful. I think you've done that for your own tools. You build for yourself. I'm sure you're doing that professionally. You might even be doing it ideologically. [1:01:39] And obviously... [1:01:40] that, [1:01:40] very much applies to understanding when an instrumental tool might be useful versus a more engaged one. Mm-hmm. [1:01:47] I think this is... [1:01:48] is especially relevant when it comes to building with LLMs, which is what you're focused on both in your personal and professional work. You alluded to it earlier, like LLMs are sort of this magic box, sort of. At least that's how a lot of people feel. That's what it feels like to use to a lot of people. Mm-hmm. [1:02:04] And it seems to me, and we talked a little bit about this, that you are focused on [1:02:09] how we can make these LM powered tools more robust and engineered. Sometimes that means making them more instrumental, sometimes that's [1:02:18] with engaged versions of it, but broadly moving away from this sort of mystical, ineffable, [1:02:23] hallucinatory thing to something that's more reliable and predictable. You say natural language interfaces feel like they should be super easy to use, but in practice, they can feel confusing and frustrating because they leave no room for affordances that tell you how exactly to command or control the tool. How do we fix this? Obviously, that's, as you said earlier, specifically referring to chat. Yeah.

[1:02:44] I guess at a high level, [1:02:46] before we go more into the weeds, like why, why are LM so different than most of the rest of software tooling and infrastructure? [1:02:54] at least when it comes to tactically solving problems. [1:02:59] there's a cultural aspect and then there's a [1:03:01] That's a good aspect. [1:03:03] The technical aspect can be [1:03:05] comes first, which is that [1:03:08] This thing... [1:03:10] In an academic sense, the thing that's exciting about language models is their generality. [1:03:14] But generality is actually... [1:03:16] are really undesirable. [1:03:18] property [1:03:19] for an interface to have. You want an interface to be intuitive, and intuitive means, in a lot of cases, very obvious. Generality gives you power. [1:03:27] But like, [1:03:28] there's a reason that Final Cut Pro looks very different than like, [1:03:32] an ID. [1:03:33] And in some ways, the thing that makes language models as a technology academically appealing and what is pushing the frontier of research is, [1:03:42] is antithetical to the thing that makes really great interfaces for humans to use. [1:03:46] And then there's, I think, a cultural aspect. [1:03:49] which is... [1:03:50] this like [1:03:51] academic desire for generality and simplicity of interface is applied into [1:03:57] to [1:03:59] people who are building products where there is a similar kind of desire for generality without really considering [1:04:05] what end users feel like when they're using it, which is like, [1:04:08] Okay, you have a box. [1:04:10] You were telling me this thing is useful, but I don't know what to ask. You can do anything with it. You can do anything with it.

[1:04:14] And real physical objects have less of this problem because they're like, [1:04:20] physical constraints, [1:04:22] If you're making a hammer, a hammer needs to have a thing that is like the affecting end that is like the thing that hits and then like the handle hits. [1:04:29] If you had a box of tools that was just like a bunch of squares, it would be really maybe like a terrible box of tools. [1:04:35] But... [1:04:35] I think both the kind of [1:04:37] academic culture around machine learning and the way that technology originated has influenced this. [1:04:44] Thank you. [1:04:45] I think it's worth talking about both the instrumental side and the engaged side with regard to LMS or at least kind of ends of that spectrum. [1:04:52] You say, "The interesting details are in the necessary trade-offs between how well you understand the user's intent and how cheaply, quickly, and reliably you can deliver the result." [1:05:02] I'm curious how you think about what... [1:05:05] ideal, [1:05:06] instrumental shaped LLM tools look like today. [1:05:10] beyond just like an agent. This applies a little bit to what we were talking about before, but like more practically, um, [1:05:16] if the idealized magic button [1:05:18] perfect agent is the perfect instrumental LLM tool. Like what is a good enough instrumental-ish LLM tool look like now? [1:05:30] This is kind of an oblique answer to that question, but I think [1:05:33] Got that. [1:05:34] what I want to say, which is [1:05:36] Because the technology is so general, [1:05:38] I think it's tempting to make... [1:05:40] you're as someone building something with language models because the technology is so general i think it's tempting to

[1:05:46] want your solution and your problem framing to also be really general. [1:05:50] What I mean by that is, [1:05:51] If you... [1:05:53] are using a language model to build. [1:05:57] something to help people write programs. [1:05:59] it's really tempting because the technology is so general to want to preserve that generality [1:06:04] which is maybe equivalent to power in some senses, and say, "I want to build a thing that's generally useful for anything in programming." [1:06:11] And, you know, [1:06:12] That is, [1:06:13] It's hard to resist the temptation, by the way, of the fact that LMs really are really good at being gentle. Like, it has the, like, yeah, the seeds are the potential to be gentle. [1:06:22] Although like... [1:06:24] Elements of goodness are in there for generality, but... [1:06:26] It turns out that's just really hard tactically to make a product. [1:06:31] That is so general and good at all those general things because... [1:06:37] Personally, I think when the times that I've built [1:06:40] products that feel the best are when I've been really specific and precise about what the task the model is doing is. [1:06:48] It's kind of funny because this is in the more kind of pre-language model, pre-language. [1:06:54] generality pilled. [1:06:55] classic ML world [1:06:57] The task shape is what you start with. [1:06:59] you have a bunch of inputs in the domain, and you have a bunch of outputs, and you try to like, what the model is doing is it's like learning a function that maps the inputs to outputs. And because you're building this thing, and you can't train over everything in the entire world, you have to be really precise about what the inputs are, what the exception cases are on those exception cases, what the right behavior is. - Yeah.

[1:07:18] Like... [1:07:19] You have to... [1:07:21] Really master the... [1:07:24] task domain. Like you have to have a lot of really fine opinions and, and, [1:07:29] There's complexity in the inputs and outputs that you have to have opinions about and deep knowledge of. [1:07:34] And I think. [1:07:35] that [1:07:36] principle is still true today. I think if you want to still build good... [1:07:40] products on top of language models, it really... [1:07:42] pays off to be an expert in what the right inputs are, what you want the users to type in. You should be opinionated about what the users should type in, what they should type [1:07:51] want to do with this thing, and you should also be opinionated about [1:07:55] what the right outputs for those inputs are. You should not just leave it up to... [1:07:59] The model defaults or whatever OpenAI or Anthropic Google's post-chainers want the model to say. That's the thing that feels like it's missing the most in a lot of AI products today. [1:08:07] Yeah. And again, there's this temptation of like, oh, let's just like preserve, try to preserve the generality of models, which is possible. But it just means there's no true generality. It just means that you are letting the post trainers of whatever model you're using be your product designers. [1:08:22] which I think is not [1:08:24] the best. [1:08:26] One thing that's sort of... I'm not sure where it sits... [1:08:29] on the sort of gradient from instrumental to engaged, but is like this broad frame of AI as a collaborator, which we're increasingly contending with. [1:08:39] You have a bit where you're sort of suggesting that working with [1:08:42] AIs or collaborators or a set of agents or whatever ends up being might be more like managing people or designing an organization.

[1:08:50] You say, "There may be craft to building software at a distance, with things like LM agents chugging along beneath your hands, but that feels to me like a distinct kind of joy from working directly in the medium of code." Then you go on to say, [1:09:04] And just as we shouldn't expect the same people to enjoy both coding and managing all the time, we shouldn't expect people to enjoy doing both coding by themselves and coding with a team of agents the same. And I think that's okay. I think the people like you and me who like the craft of working directly with the system will find ways to accelerate ourselves in our craft. Mm-hmm. [1:09:26] I don't know when you wrote that. I don't think it was super recent. No. Yeah. I'm curious how, like, have you found yourself doing more of both of these sides and, and, [1:09:34] Or are you more still kind of in the, like, accelerate your craft zone? I am definitely. It's funny because I think this was a while ago, as you said, and I think it's actually kind of come true. [1:09:44] One of my friends, Dan Tripper, has a company called Every and... [1:09:48] Somewhat incredulously, all of the engineers at that company... [1:09:51] on the used coding agents. [1:09:53] I think it's incredibly rare, if at all, that [1:09:56] they open up an IDE and write code, which... [1:09:59] I am somewhat skeptical of, but I believe down. [1:10:03] But me personally... [1:10:05] I really enjoy understanding and having opinions about how architect systems. [1:10:12] Looking at. [1:10:13] how things are organized. [1:10:15] And how things work under the hood and being in the weeds. And so I've... [1:10:19] I think gotten better and I've still accelerated myself. Like I'm way faster at building things than I used to be, but just in a different way.

[1:10:25] And I think it's a different... [1:10:26] kind of joy as I wrote that and [1:10:29] I think both things will hopefully continue to be accelerated. At the end of the day, I do think it's important that [1:10:34] Like, software is one of the most [1:10:37] I mean, not even one of, I think software is probably the most complex... [1:10:41] kind of machine... [1:10:43] that humans... [1:10:44] can ever make just because you can pack so much complexity into such a small amount of space. [1:10:48] And so I think it's important that as we build more software and more critical infrastructure software that... [1:10:53] Humans. [1:10:54] are forced to contend with the full complexity of the systems we're building and [1:10:59] I think there will be [1:11:00] It'll be important for... [1:11:02] that some people will. I think that's good that [1:11:04] some people will continue to get joy out of like working with that complexity including myself [1:11:10] A few... [1:11:11] bits that I think tie very closely to that and the idea of accelerating craft, uh, [1:11:17] All you. When possible, directly manipulating the underlying information or objects of concern, the domain objects minimizes cognitive load and learning curve. [1:11:25] And then a separate quote, what direct manipulation is to the graphical user interface we have yet to uncover for this new way to work with information? [1:11:33] And you go on to talk about good complexity. You say, a second type of surrogate object is focused not on showing individual attributes, but on revealing intermediate states that otherwise wouldn't have been amenable to direct manipulation because they weren't concrete. Indeed, I think it's fair to say that direct manipulation is itself merely a means to achieve this more fundamental goal. Let the user easily iterate and explore possibilities, which leads to better decisions. Okay.

[1:12:00] All of this to me, I think is like, [1:12:01] around [1:12:03] We talked about a little bit with the people stuff and the vetting stuff. It's like I think you call a data set a bag of concepts, which is great. This proximity and latent space, another framing or metaphor you use is like a brain. What if we do a brain scan of an LM? Yeah. [1:12:18] I'm curious what... [1:12:20] your early conclusions are on like maybe going back to the earlier one of those quotes is about like we have the the gooey [1:12:27] Most people today, even engineers, are either... [1:12:32] like using it to generate some amount of code, [1:12:35] tab autocomplete, whatever, full agents, like get answers. Like the broad shape of the use case for these things doesn't seem to capture a lot of this like [1:12:45] exploring latent space. [1:12:47] Idea. Yes. [1:12:49] I'm trying to figure out what the right question is. I suppose it's part why is that, part [1:12:54] Do you have any hunches of where we're going? Are you doing stuff? We've talked around some of this. I don't want to be too repetitive, but yeah, I'm curious where you sense we are on this. [1:13:05] I feel this really personally... [1:13:08] I've spent so much time [1:13:10] thinking about. [1:13:11] how to work with ideas in this more like geometric format. [1:13:14] And... [1:13:15] I'm so excited by it, and yet [1:13:18] Like, I'm... [1:13:19] Even I'm like building chatbots, you know, I, I, so like at times it feels like, [1:13:24] kind of very deep cutting personal hypocrisy at times it feels like I just maybe have it like, maybe it's like a skill issue and I just like haven't been good enough that like,

[1:13:31] injecting some of these ideas and research land into real products. I just think it'll come [1:13:35] Phil Wattler, who's a researcher who [1:13:37] contributed to Haskell. [1:13:39] has this great quote about [1:13:41] doing computer science research. [1:13:42] It's actually specifically about... [1:13:44] building a great programming language, which is, I'm going to butcher the details, but it's something to the effect of, [1:13:50] It's something to the effect of if you want to build a great programming language, first do some research and funds a language and then wait 40 years for the ideas to mature. I think. [1:14:00] surely a part of [1:14:02] What I need. [1:14:03] is more patience. [1:14:05] And I think... [1:14:07] there are ways for ideas to collide into the right space. [1:14:11] ingredients for it to become successful over time. [1:14:14] I also think it is still what I'm working on. [1:14:17] which is to find like [1:14:19] For a while, I was just doing in my head exploration. And then for some time, three years ago, two years ago, [1:14:27] I was... [1:14:29] working in a very horizontal company, Notion, and trying to find ways to make useful things with my ideas. Then now I'm in a very vertical, very concrete, [1:14:39] set of use cases and still trying to find ways this can plug in be useful. And I think I've gotten closer. Like the understanding people thing, I think, is... [1:14:46] concrete progress in where I think this can be applied, but [1:14:49] I'm [1:14:51] even more excited to actually build things in this direction and then see what breaks down. [1:14:55] And ultimately, that's how ideas get out into the world. [1:14:58] That's a great transition. Um, [1:15:00] You...

[1:15:02] work at Thrive. Um, [1:15:04] if that weren't obvious, you previously spent some time at Notion, which you just alluded to as sort of being a little bit more wide and abstracted. [1:15:12] the Thrive works more deeply. [1:15:14] kind of embedded part of that i think implicitly is just like thrive has less than 100 people most of whom you're with in a room uh or at least in a building you i think you frame this somewhere as like a much smaller tam for the things you're building yeah like 80 people yeah i've spoken to every single one of my potential users it's amazing [1:15:32] Super high level like what what drew you here and what has that? I? [1:15:36] shift in the sort of scope of your work, what has that been like? [1:15:40] Yeah, so maybe important to know that at the top, given... [1:15:45] The engineering that we do at Thrive is not [1:15:46] Super Visible is we have a team. The team is... [1:15:50] Five or six of us, including a designer and many people who are great product engineers and data folks. [1:15:56] I'm a part of that team for most of my work. [1:15:59] And, uh, [1:16:00] The team existed a while before I came, but I came in and... [1:16:04] I think kickstarted a lot of the more LLM-related parts of what we're building now. [1:16:08] They really... [1:16:10] The interesting thing that I've learned about building internal tools is that [1:16:16] When you're building anything, I think you have a kind of a fixed budget for surprise or uncertainty. [1:16:21] And when you're building... [1:16:23] at least my kind of brief experience in the world, when you're building a product to go have its own life in the world and be used by hundreds of millions of people,

[1:16:32] you spend a lot of that uncertainty on just like, [1:16:34] the problem, like the problem, different people are going to bring different problems in different use cases and contexts of use to what you build. And you have to prepare yourself for all of those things. Almost like the problem is a theory. [1:16:44] Or is this like broader zoomed out abstract thing that has to hold a whole bunch of potential people's needs? And because the uncertainty is in the market and the user is... [1:16:52] You want to... [1:16:54] I think be pretty stable and predictable and kind of, [1:16:57] mostly what you build with, where in internal products, at least in my experience, I think [1:17:02] I thrive. [1:17:03] There's a lot more like the problems are so clear. They're also kind of, I think, just barely outside the reaches of boring technologies that there is a lot more room. [1:17:13] that I found to [1:17:16] throw weirder ideas or more frontier ideas against those problems and see what sticks and see what works. And so that's been that's been really fun. [1:17:24] You can go as specific or not as you would like. [1:17:27] And you briefly alluded to it, but I think it would be helpful to... [1:17:31] hear a little bit more about what you actually work on. - Yes. - And then as a second piece, [1:17:36] I think you also kind of implied this, but like you aren't exactly the typical profile of like, [1:17:42] a [1:17:43] somebody a VC firm hires to work on internal tooling. I think that match on both sides a little unique. And a meta thing hanging over all of this would be [1:17:53] What does it mean for a... [1:17:56] investment firm to take building software very seriously. [1:18:01] Mmm.

[1:18:02] Yeah, so we... [1:18:04] It's kind of funny. I find it always really hard to... [1:18:07] I have this problem with Notion and I have this problem again at Thrive. It's really hard to explain what the thing actually is. Like if you work at... [1:18:13] a calendar app company, like I'm building a calendar. Notion is kind of a little bit of everything. So it's like, it's like you can say some people call it a note-taking app, but it's really like [1:18:21] like... [1:18:22] But the thing that people say internally is like, oh, it's like Lego bricks for software, but then like everybody's like, you know, pretentious description. [1:18:31] Here, there's a different kind of the same problem, which is the thing that we're building is kind of a suite. [1:18:36] of different tools that together [1:18:39] help a bunch of different teams inside ThriveWork. And so it's not really a calendar, it's not really a note taking app, it's not really a chatbot, it's kind of, [1:18:47] an amalgamation of all of these things to the extent that they're useful for different teams. [1:18:50] The mandate of the team is to build a kind of Iron Man suit for everyone. [1:18:54] inside Thrive and also to make Thrive's business itself much more software defined. [1:18:59] And there are three different layers to this. One layer is... [1:19:03] just to... [1:19:04] kind of [1:19:05] organize the information and data we have about the world and about [1:19:09] people and companies and investors are on Thrive to be more accessible. [1:19:13] The next layer is a kind of core product that includes [1:19:18] a thing that looks like a calendar, some kind of a news feed, a search product, a research agent tool. All of these things combined to a single coherent interface that helps people find information quickly, prepare for their day more efficiently, know what's going on, be more prepared in their job, and spend less time working on drudgery.

[1:19:36] And then there's... [1:19:37] A third pillar, which is [1:19:39] more embedded. [1:19:41] which is we have this foundation of data and automations, but there are really specific workflows where if we just, like, pick at that specific workflow and... [1:19:50] fully or mostly automated, it's going to save someone tons and tons of time. And maybe only two people of our 80 people are going to use it, but it's going to save them so much time that it's worth it. And so we build those kinds of one-off automations as well. [1:20:00] The way that different people use it differently, which is what's cool about [1:20:03] building for 80 people is that you can kind of build for every single individual person [1:20:08] But the way I use it, I use it a lot to just look up and understand different companies that we're partnered with and involved with and invested in different people around our universe. [1:20:17] Also just generally to like, [1:20:19] Our research agent is actually quite good, and so I use it sometimes as a replacement for [1:20:23] like 03, because we have some proprietary data, [1:20:27] And also to just organize my day and like prep for meetings and things like that. [1:20:32] So that's what we build. Yeah, very, very classic and former product. [1:20:36] but a lot of [1:20:37] the way they would like, [1:20:39] broken down the problem is as like a data layer, an intelligence layer, and then like a UI on top, and then automations on top. [1:20:44] which I think is a really clean separation of concerns. [1:20:47] Um, [1:20:48] And then your second question. I mean, I think inside my second question would be like, [1:20:53] Most investment firms, certainly VC firms these days, have internal tools. [1:20:59] That's true. You take building software, you personally, I know, take building software really seriously. And so my assumption here is that [1:21:05] that is to some degree to attract someone like you that is internalized here. And so, yeah, the broad question would be like,

[1:21:11] What does it mean for particularly an investment firm? [1:21:14] Um, [1:21:15] to take Facebook builds really nice internal tools too. But like, what does it mean for investment firm to take building, [1:21:21] software seriously. [1:21:22] Mm-hmm. [1:21:24] I think there are things that make it [1:21:26] personally really interesting to me as a place for me to continue to validate some of my reider ideas about [1:21:32] Information. [1:21:34] Tools? [1:21:35] And then there's also, I think, [1:21:37] ways in which this is just a really cool opportunity for [1:21:40] Thrive for any investment firm? [1:21:42] For me personally... [1:21:44] I've already spoken a little bit about how internal tools give you, in my experience, a little bit more flexibility and flexibility. [1:21:50] the methods that you apply to solve hard problems because the problems are so fixed. Hmm. [1:21:54] And because internal tool teams tend to be smaller and a little bit more... [1:21:58] Explore terrain agile. [1:22:00] I also think [1:22:02] the kinds of [1:22:04] intellectual work that [1:22:06] A lot of people do. [1:22:08] inside at least Thrive, because this is really the only firm that I know with this level of familiarity. [1:22:14] is... [1:22:15] like, [1:22:16] deeply qualitative. It's not just moving numbers around and like optimizing numbers and sheets. It's like, [1:22:21] You have to understand people. You have to understand people. [1:22:24] products and problems. You have to understand research. You have to understand news and events. [1:22:29] And these are all really qualitative things. [1:22:31] And at the same time, there are clear... [1:22:34] signals for when decisions are correct. They're also in place of [1:22:39] like raw signals of successful decisions. There's also lots of like really high quality proxies. And I thought personally, it was a really good,

[1:22:45] place to... [1:22:46] continue to build pretty high stakes. [1:22:49] tools, [1:22:50] to help people work, work on, [1:22:52] pretty deep intellectual problems and knowledge bound problems. And but then be able to flexibly explore inside that. [1:22:58] I think more generally, [1:23:00] I've been noodling on this idea of [1:23:02] for like [1:23:03] Fundamentally, venture is a services business. It's a financial services business. [1:23:08] And, [1:23:09] I think. [1:23:10] Right now is a really interesting time for... [1:23:14] Services, businesses. [1:23:15] because... [1:23:17] When you're trying to get really good at a service, [1:23:19] There are two things that are really scarce, especially in an AI world. [1:23:22] One is... [1:23:23] really, really detailed understanding of what it means to be really good at that task. [1:23:28] What does it really mean to... [1:23:29] get really good at investing or get really good at researching a market. [1:23:33] or get really good at. [1:23:35] legal contract review or whatever. There's a lot of detail in there. A lot of detail is not [1:23:39] like written about on the web. You just have to kind of like do the job and learn. And then there's also, so that's one thing that's scarce. The other thing that's scarce is, [1:23:47] like a sandbox environment or even like a real environment for you to then make decisions and learn [1:23:54] Maybe not through mistakes, but learn through feedback. [1:23:56] This is not just true of financial services. It's true of like, if you're like trying to run like an AI managed software building agency, the two things that are scarce are like, how do you, [1:24:05] do the job of like, [1:24:07] getting a client and writing the software and delivering it really well, and then also [1:24:11] just this playground or environment for writing software and then getting feedback from clients or from...

[1:24:18] compilers or what have you, the environment. [1:24:20] and [1:24:21] It feels like a really interesting... [1:24:23] technical problem given where I is at now. You're also, sorry to interrupt you, but you're always, both of the examples you gave [1:24:31] in part because they're generalized, you're in this constant state of... You have this frame of scale, X, or explore. And you're in this constant sort of... [1:24:40] that type of situation because there might, the next incremental thing might always be new. [1:24:45] if you are a consulting or an investor or something like that, like, whereas, um, [1:24:49] Inherently, big software companies have new things, but they also are like really, they're just really good at like mechanical reproduction in some sense, which is interesting. [1:24:59] Yeah. [1:25:00] Yeah, I think at the end of the day, [1:25:02] It's an environment that's very rich with really intellectually demanding tasks with a lot of detail. [1:25:07] where if you [1:25:09] built some software that either was really good at the task or helped a human get really, really good and efficient at the task. Both of those things were really interesting innovations for me. Yeah. So for it to work on it. I also think more concretely... [1:25:20] In general, Thrive is good. It's filled with really thoughtful people and [1:25:24] from my experience, at least in the last year so far, [1:25:28] we've been able to build something that's [1:25:29] really functional and valuable, but also in a really high craft way. Like one of my favorite things about our app is that when you – [1:25:35] It's one of these apps designs where like, [1:25:37] There was a ribbon of like four tabs at the bottom. [1:25:39] And when you tap [1:25:40] the tab buttons, there's a little like haptic vibration on the, and I just, it's just so nice. And I love that I can work on things like that, little animations, animations, [1:25:48] We've done a lot of work to make our chat interface itself

[1:25:51] really, really robust errors and... [1:25:53] I like that. [1:25:54] there's space afforded to [1:25:56] even if it's only for like 100 people or 80 people, to make something that is just objectively a great piece of software. And that applies to the way that we build its language models as well. Yeah. There's a sort of broader thing that your earlier comment about the Iron Man suit made me think of, which is one of the more unique things about [1:26:14] an investment firm is that investing in many ways is sort of like the most generalized job [1:26:20] Like you sort of [1:26:21] definitionally are a generalist. [1:26:24] And as a result, you're sort of building... [1:26:26] tools, [1:26:28] for this like group of high agency, fast learning people. [1:26:33] fast moving [1:26:35] generalists, which is kind of just a broadly interesting design problem. [1:26:39] I don't know if I fully agree with that. [1:26:42] I think... [1:26:44] Maybe that maybe that. [1:26:46] Especially at a generalist fund, I think maybe... [1:26:49] the topics that we want to research and understand are general, like when they will be looking to understand. [1:26:54] batteries and the next day we're looking at [1:26:57] fashion. [1:26:58] Bye. [1:26:58] The skills involved are not super general. I think the skills involved... [1:27:03] I mean, I'm not an investor, so maybe, you know, but... [1:27:06] Maybe I don't have as much detail as someone who would be, but... [1:27:10] It feels to me like the core skills are around relationship building. [1:27:14] things that resemble sales. [1:27:16] rapidly understanding and researching new topics, [1:27:20] kind of decision making under duress maybe. - You're describing something that sounds pretty generalist to me, but. - I don't know. I think like, if you really think about like what it takes to be good at like programming. - Mmm. Ah, this is a good counter. - The skills. It'll also be like, right. Like understanding really complex systems.

[1:27:36] and being able to like describe that complexity in words really well and obviously there's a concrete i don't know like the concrete skills in programming are like you gotta like [1:27:44] use an IDE. In investing, you've got to use Excel. I don't know. There are these super concrete skills, but I feel like they're a little bit [1:27:51] Two in the weeds and [1:27:53] I think the way that we describe skills, they all kind of can sound general. But I do think there are aspects of it that are... [1:27:58] I don't think investing is fundamentally more generalizable than many other skills, but I do think it's very... [1:28:04] like qualitative and like rich in detail it makes an interesting kind of task [1:28:09] On that last note... [1:28:10] How do you think about bringing an [1:28:12] engineering orientation to any kind of work or this kind of work specifically. [1:28:20] One way that I think I've grown as an engineer in the last year, which is... [1:28:25] I'll answer this question. [1:28:26] more directly later, is... [1:28:29] I think when I was really young, [1:28:30] I viewed engineering as like, okay, my job is to... [1:28:34] write some code that does the [1:28:36] solving of the problem. [1:28:38] And so the end artifact... [1:28:40] the deliverable. [1:28:41] if you will, is like the source code. [1:28:43] I grew a little bit. [1:28:44] A little more mature. [1:28:46] a little wiser. [1:28:47] And then my perspective was actually my job is not just to write the source code. [1:28:51] but to... [1:28:52] deliver a running system. [1:28:54] The system is like, I think that's derived from the source code, but it has to be [1:28:59] like reliable. There are certain kinds of like operational metrics that you have to hit. It has to be understandable and debuggable. If something goes wrong, you need to be able to root cause it

[1:29:07] and fix it. [1:29:08] If you want to make a change to it, [1:29:09] You need to be able to make that change. [1:29:11] you have to be able to make that change and be confident that it's going to be good, and so on and so forth. So there's a second layer, which is the operational aspect to the software. And then more recently, I've been... [1:29:22] thinking about [1:29:23] The first two are really... [1:29:26] They make sense in the setting of an individual coder. As a solo developer, you have to write the source code, and then you have to maintain and run the source code, the software. And then there's third layer, which is especially important if you're running, [1:29:38] working in a team. [1:29:40] where [1:29:41] The third layer they have to deliver is to build like a people system [1:29:47] or an organization that... [1:29:49] is capable of... [1:29:52] continuing to make changes to that software and evolve it as the world evolved. Yeah. [1:29:56] And in some ways, the thing that you have to deliver is not just the software, but like [1:30:02] the processes and the culture and the team itself, such that if you were removed from that equation, this whole system would be self-sustaining and resilient and robust and will continue to [1:30:12] deliver this like second tier thing of like, [1:30:14] really resilient, operationally excellent piece of software, and also the first thing that is really good source code. [1:30:19] And so – [1:30:20] Building these systems, the source code, the running system, and then the team, all three systems in a way that is easy to change and understandable and debuggable, [1:30:31] is I think what [1:30:32] good engineering is, at least the way I understand it at this point in my life. [1:30:35] And that's kind of the concrete aspect, I think.

[1:30:40] More... [1:30:41] specifically and maybe more interestingly in my specific job right now. And [1:30:46] building a tool for [1:30:48] Thrive. [1:30:49] In the near term, I think a lot of what we're building is this Iron Man suit. [1:30:54] idea of there's a ton of things that people do inside the firm that we want to [1:30:58] Make easier. [1:30:59] and give them more leverage. [1:31:01] In the long term, [1:31:02] I'm also really excited about. [1:31:05] Are there parts of [1:31:07] thrives, [1:31:08] business. [1:31:10] that we can... [1:31:12] fully turn into, like, a scaling equation where we, like, spend an incremental dollar of compute and get some incremental predictable, like, amount to return out. Hmm. Like... [1:31:21] It would be really exciting... [1:31:23] If... [1:31:24] we had some machine, some black box, where you could put in [1:31:28] like X thousand extra dollars. And like for every X thousand extra dollars you put into compute, you get like one new extra interesting founder that we discover in the world. I don't know how you would do that yet, but like that is a very, that's like a fully software defined [1:31:41] version of the business versus like a thing where you have [1:31:44] people, [1:31:45] And in that world, you would still need... [1:31:49] the people and the judgment of the people. Yes. And so it's not about... [1:31:53] automating, but it's about, like, in order to build that thing, you're going to have to have really deep understanding of people and of the job and of the decisions. And so... [1:32:02] I think it gives, especially as we want to stay as a small team as we grow the business, [1:32:07] We'll have to lift everyone up to work at a slightly higher level of abstraction. And as we lift those people up...

[1:32:11] there's going to be perhaps this underlying engine. [1:32:14] Yes. That's that's like an embodiment of of. [1:32:17] What makes. [1:32:18] people are really good at their current job. Also, just another cool example of the ways that, like, better modeling latent space might produce really interesting outcomes. Yeah, in order to build systems that are really good at these very nuanced jobs, you're going to need... [1:32:30] to give the people that are operating these tools [1:32:33] The ability to express their nuance, which goes back to what makes a good tool. Yes. Yes. [1:32:38] Has working in an investment firm [1:32:40] or broadly or Thrive specifically made you more commercial in any ways? [1:32:45] or even just like learnings about [1:32:47] that side of the world. [1:32:49] Mmm. [1:32:51] I don't think it's... [1:32:52] a consequence of working at an investment firm specifically, [1:32:56] I'm sure these are all related, but. [1:32:58] I don't think I'm more commercial now than I used to be. I think part of it is [1:33:02] the growing pains of having an idea that I was really enticed by, but then not being able to find great use cases for it. Yeah. [1:33:09] And a part of it also is like, [1:33:12] I think... [1:33:13] Thrive in particular being very generalist, [1:33:16] and doing everything from like [1:33:19] Fashion too. [1:33:20] I don't know. [1:33:21] Yeah, compute. [1:33:23] So much of... [1:33:25] So many of the things that humans are engaged in have, like, nothing to do with software or technology. I mean, everything is technology in some ways, but nothing to do with, like... [1:33:33] software and AI and like, or even like, [1:33:36] automation machines. Like, [1:33:38] Software is such a small part of the human experience.

[1:33:42] and [1:33:43] There are a lot of other kinds of businesses out there. And I think that's like a continual learning that I find really fun. [1:33:49] Hmm. [1:33:50] I want to spend our last few minutes... [1:33:53] on an idea of, [1:33:54] that I think has actually been inside a lot of what we talked about. [1:33:57] And his... [1:33:58] expressed in that final answer you gave a bit, certainly expressed in the tools you build for yourselves. [1:34:05] yourself, excuse me, including stuff. I mean, stuff you said, all great tools must be built in a serious context of use. Good tools are transparent. They let ideas through, um, [1:34:15] But zooming way out, talking about kind of technology and humanity and how they – [1:34:19] interact with each other. [1:34:21] And in this section, I'm going to have a bunch of quotes for you, so you'll have to forgive me. The first is from an essay called Create Things That Come Alive. [1:34:27] You say, building technology is fundamentally an affair by humans for other humans, and objects of technology ought to be ensconced in a romance and history and all manners of color and details and textures of life. It ought to come alive in our environment. [1:34:44] Technology is not what's shiny and boxy and delivered in metallic wraps. [1:34:49] And then you quote Ursula Le Guin, technology is the active human interface with the material world. [1:34:56] From a separate essay you wrote on Radio City, you say, I yearn to see this more irreverent and humanist relationship to technology imagined more often today. [1:35:05] when both technology and the industry backing its progress feel increasingly detached from culture and media and humanities. I dream of a humanist revival with computing systems at the center,

[1:35:16] grasped firmly in the hands of wisdom. [1:35:20] And then one final pair of quotes where you're, [1:35:22] kind of, [1:35:22] pointing at this notion that technology [1:35:26] is often thought of as like other separate alien thing that's separate from our messy humanity. You say technology exists woven into the physics and politics and romance of the world, and to disentangle it is to suck the life out of it, to sterilize it to the point of exterminating its reason for existence, to condemn it to another piece of junk. If you consider yourself a technologist, here is your imperative. Build things that are unabashedly beautifully tangled into all else in life, [1:35:56] relationships, politics, emotion and pain, understanding or the lack thereof, being alone, being together, homesickness, adventure, victory, loss. Build things that come alive and drag everything they touch into the realm of the living. And once in a while, if you are so lucky. [1:36:14] May you create not just technology, but art. [1:36:17] not only giving us life, but elevating us beyond. I kind of just honestly wanted to read all those quotes. They're amazing. My question is, how do you personally imbue the technology you build with this aliveness, with this humanity? Yeah. [1:36:32] Wow. [1:36:33] What a question. [1:36:34] Um... [1:36:36] So earlier I mentioned that [1:36:38] One way that I've grown is that I've [1:36:39] Gone from... [1:36:41] working on this stuff because I was just really excited by tools and productivity tools and the [1:36:46] Thinking of it.

[1:36:47] in the context of [1:36:49] trying to give people more power and agency over their life and over the world. [1:36:53] I think a similar parallel... [1:36:54] Maybe like growth. [1:36:57] arc that I've had. [1:36:58] is... [1:36:59] If I actually think about [1:37:01] why I like working on this stuff. [1:37:04] A part of it definitely is that I like the puzzle of writing programs. [1:37:09] And, you know, [1:37:10] It feels like I'm good at it and I want to continue doing it and be better at it because mastery itself is fun. [1:37:15] But so much of it also is like... [1:37:18] everything about the context in which I do this work. [1:37:22] I find the environment of... [1:37:26] people building companies and thrive really fun. I love the people that I get to work with and [1:37:32] get opinions and feedback from and tell them, tell them, [1:37:35] It's fun to tell them about ideas that I'm thinking about. And all of this exists within the context of, I'm a human being, there are other human beings, I'm building stuff for them, we're building with them, we're having them disagree in interesting ways. And the context of doing this work is what's fun. [1:37:50] And I'm sure I'll... [1:37:51] Over time, I'll like... [1:37:53] learn even more and [1:37:55] lean even more in this direction. [1:37:57] and [1:37:59] going back to where we started this conversation, this... [1:38:03] I think is really adjacent to this other kind of aesthetic thing that I feel of [1:38:07] The point of technology is to elevate us, to give us more [1:38:11] prosperity, [1:38:12] to... [1:38:14] let individuals [1:38:16] do more than just survive.

[1:38:17] Um, yeah, [1:38:19] I... [1:38:20] Though in some ways, [1:38:23] Not in some ways. I think [1:38:25] Most of the time, for most people, technology itself is instrumental. [1:38:28] There is obviously fun in building technology. It's fun to write programs and [1:38:32] solve these puzzles. [1:38:33] But [1:38:34] At a societal scale, technology itself is an instrumental tool. [1:38:38] And I think it's important not to forget that and the, the, [1:38:41] If it's instrumental, there has to be something... [1:38:44] That is the result that we want. Like, what is the prompt? [1:38:46] Mm-hmm. [1:38:47] And I think the most inspiring prompt is [1:38:50] whose response is technology writ large, [1:38:53] is... [1:38:55] everything that makes it really great and fun and [1:38:59] Lovely to be a human being. [1:39:01] How can we have more of it and how can everyone have more of it? [1:39:04] And how can we have it in all of the [1:39:06] variations of it that exist in the world. [1:39:10] That is the prompt. And then the answer is everything the technology has unraveled into. [1:39:15] And if we... [1:39:16] stray away from that. [1:39:18] if we develop some other [1:39:21] proxy some other benchmark that we're like optimizing technology towards that is um, [1:39:27] That's not good. Yes. And so I... [1:39:30] I think... [1:39:32] people building with technology. [1:39:34] You need to remember... [1:39:36] that these are all for other human beings and building with other human beings. [1:39:41] Another few quotes. [1:39:42] on a related theme. [1:39:45] This is a recent tweet. You say more people should create things to proliferate and aesthetic into our future, not just to solve problems. This is the quality that every artist and engineer I respect shares most universally. Without this, you are doomed to churning out slop.

[1:40:01] What values do you create to spread? [1:40:04] What image do you dream about? [1:40:06] What is the feeling of a tomorrow you want to give form to? Have a position, stand for something, don't just create value. [1:40:14] and then you're a separate idea about dreams. You say too many tools for thinking, [1:40:18] Not enough tools for dreaming. [1:40:21] But really, aren't some of our best ideas found in dreams? [1:40:25] And then finally, you're riffing on this idea of neural media. [1:40:28] And you say, could a generative image model, never having seen Voyager 1's pale blue dot, have been used to create such beauty? Can a neural generative model imagine beyond its own world model? Yeah. [1:40:41] How can we build models that can help imagine new worlds, not just permutations of the one we know? [1:40:50] Imagery is this... [1:40:52] pattern or this theme across all of those ideas. Mm-hmm. [1:40:55] aesthetics, but maybe even more so dreams. Dreams are so image based, at least in my experience. How does imagery underpin your creativity? [1:41:06] Your work. [1:41:07] And I realize you're sort of talking about imagery in the literal sense and maybe in this sort of broader perspective. [1:41:12] sense too. [1:41:13] I do think in a lot of these cases, imagery and dreamer [1:41:17] interchangeable or at least deeply entangled. [1:41:20] And dreams are maybe the more fundamental thing, although imagery is the way in which it comes across. [1:41:27] There's a thing that I feel like... [1:41:30] A lot of... [1:41:31] what I read around me.

[1:41:32] is engaged in these days, which I promise this is a roundabout way to answer the question. There's a thing that a lot of what I read around me is engaged in, I feel like, these days, which is [1:41:42] to pretend that [1:41:44] technology has its own arc and its own destiny, and it's going to, like... [1:41:47] go somewhere by itself and it's like [1:41:50] There's some like manifest destiny version of technology that is where like there is some final form that we're trying to achieve. [1:41:56] and [1:41:57] Going back to like humans build technology for other humans thing. [1:42:02] Sure. I understand that. [1:42:05] The reason that this is sometimes true, like, [1:42:07] there are kind of really powerful structural forces that [1:42:10] let certain technology really proliferate really easily and sort of have more evolutionary pressure and tailwind against it. [1:42:16] And so in some sense, it's true. But I also think this is often used as a way for people to either [1:42:21] not be intentional about the direction in which they are building the technology or or. [1:42:28] to sort of forget that [1:42:30] They have... [1:42:32] the agency and the responsibility to [1:42:35] try to push against it or try to shape the direction in small ways at all. [1:42:39] It goes back to the first thing you said at the top of the conversation. Yeah. [1:42:42] Again, there is a default path for technology like roll down the hill and [1:42:46] it's still really hard to predict. [1:42:49] Like... [1:42:50] This like kind of fetishization of prediction is maybe – [1:42:55] Most obvious in this question that you get in San Francisco all the time, like, what is your timeline? Or is this a vision like correct answer to this question?

[1:43:03] And my kind of. [1:43:05] pp responds to this always it's like what do you mean the timeline like you're the ones that are like [1:43:09] It's sort of like if you ask me, like, oh, like, let's make a... [1:43:13] Let's make a prediction, like when you're going to wake up tomorrow. Okay, I understand that there is the spirit behind the question, which is like, okay, there's going to be some probabilistic distribution over when I'm likely to wake up. [1:43:24] More likely you wake up around like 8 or 9 than like 5 a.m. Sure. There are some... [1:43:29] likelihood distribution that we're talking about. But also, this is a thing where [1:43:34] like, [1:43:35] If I really wanted to, I could wake up whatever, you know. We have will. There is some will that... [1:43:40] And not only do we have some will in the case of maybe, you know, waking up, when do I wake up tomorrow is an inane question, but like in the in the case of building technology. [1:43:49] I think... [1:43:51] people should be building technology intentionally to express who they want, [1:43:56] to be and who they want, what they want the world to look like. [1:43:58] And again, there's a default course. [1:44:01] Bye. [1:44:02] I think the fun in technology is to like, [1:44:04] Use it. [1:44:06] to... [1:44:07] try to deviate the world from the default course in whatever way you think is interesting and useful and good for you and good for the people around you. [1:44:15] And so – [1:44:16] It has, of course, but... [1:44:18] Like... [1:44:19] The course is [1:44:20] Ultimately... [1:44:22] the sum vector of like every, every little person that's like, [1:44:25] pushing on technology in different ways. [1:44:29] One of the other ways in which I think I've grown over time is – [1:44:31] It used to be that the only way that I would influence...

[1:44:35] Where... [1:44:36] technology was going around me was by like building stuff. And then I think over time, I've [1:44:40] ended up in this really privileged position where [1:44:43] a bunch of other people care what I think about stuff. [1:44:45] And so now I have this like other lever to push on technology, which is I can build technology on my own or I can like try to convince lots of other people that there are other directions for technology to go in and like. [1:44:55] get their help in pushing things where I want to go, which is kind of, I guess, what I'm doing now. [1:44:59] But there are all these different ways that you can... [1:45:02] tug or nudge technology to go in the direction that you want to see the world [1:45:06] look more like... [1:45:07] And, [1:45:08] If you're building technology out of this power, you should take advantage of it. [1:45:11] It's incumbent to money to take advantage of it. [1:45:15] I think... [1:45:17] aesthetics and dreams and all these things too are... [1:45:21] Highly rational, ambitious, smart, whatever people can maybe underrate their – [1:45:27] Their weight... [1:45:29] or their persuasion or their impact on what we do too, which is cool. I think also this is, maybe I'm just a romantic, but it's like, [1:45:39] kind of the [1:45:40] point of it all like [1:45:41] When you're like... [1:45:43] enjoying a really delicious meal. [1:45:45] There's like the functional point, which is to sustain yourself. And then there's the takes. [1:45:50] And I think it's totally valid to be like, I'm eating this thing because it tastes amazing. And sure, you can make whatever meal, but like... [1:45:58] I think it's much better to think about trying to make something that tastes really good. [1:46:02] And the taste of the meal in my head is like the value and the principles behind what you make.

[1:46:09] I have just one final question, but I do have some more quotes before we get there. [1:46:14] This is you. In these visions, I fell in love with the idea that there was some quality other than truthiness that ought to guide our search for knowing more about the universe and about living. [1:46:26] This ineffable quality I've come to call by many names. Among them are words like novelty, surprise, and wonder. [1:46:34] So, [1:46:34] Also you. [1:46:36] Productivity is about the industrialization of creation. Wonder, in contrast, defies systemization because it gets its power from uncertainty and surprise by nature. You can't optimize wonder because to optimize requires knowing the output and the process. [1:46:53] Wonder is the discovery of new outputs and new ways of getting there. [1:46:58] And one final quote from [1:47:00] The beginning of a book I love by Lawrence Wechsler is the biography of Robert Irwin. I'm in the middle of this one right now. Amazing. He opens the book. [1:47:08] with an anecdote about Irwin. [1:47:10] During the early 60s, when Robert Irwin was on the road a lot, visiting art schools and chatting with students, he was preferred an honorary doctorate by the San Francisco Art Institute. The school's graduation ceremony that year took place in an outdoor courtyard on a sunny, breezy afternoon, sparkling clear. Irwin approached the podium and began, I wasn't going to accept this degree, except it occurred to me that unless I did, I wasn't going to be able to say that. He paused, waiting as the mild laughter eddied. [1:47:39] All I want to say, he continued, is that the wonder is still there.

[1:47:43] whereupon he simply walked away. [1:47:47] My final question is, what does it feel like to be lost to wonder? [1:47:53] I think at the root of... [1:47:55] everything that I find really... [1:47:58] fun and invigorating, [1:48:00] is some... [1:48:02] useful... [1:48:04] substantive sense of novelty, which is, I think, if I had to really clinically analyze it, what wonder feels like to me. [1:48:10] This is a part of solving our programming problems. This is a part of [1:48:14] trying to come up with new abstractions, [1:48:16] a part of [1:48:17] Spending time with new people is like behind all of those things. There's something new that I feel like I could understand if I just put more effort into it and spent more time with it. [1:48:24] And [1:48:25] There's something like fundamentally very satisfying about coming upon a... [1:48:30] thing that feels... [1:48:31] New and mystical... [1:48:33] And... [1:48:35] than kind of figuring out a model for how to understand it and how to model it. [1:48:40] And I feel like... [1:48:42] times in my life where I feel like I'm having... [1:48:45] making the most of my... [1:48:46] life. [1:48:47] are when I'm in a repeated cycle or a flow of coming up on something new, understanding it, coming up on something new, understanding it. [1:48:55] And [1:48:56] to do that continually over and over is to be lost in wonder. [1:49:00] That's all I got. Minus, thank you. [1:49:01] Thank you very much.