Join us in this insightful episode of Beyond the Prompt as we explore the multifaceted world of AI with Dan Shipper, Co-founder and CEO of Every. Dan, a pioneer at the intersection of business and technology, shares his unique perspective on using AI as a powerful tool for enhancing both personal life and professional ventures. Discover how custom AI instructions can revolutionize decision-making and streamline processes, leading to impactful changes in daily routines. Tune in for an intriguing discussion on integrating AI into our lives, beyond the hype and into practical, everyday applications. Don't miss this thought-provoking conversation that could change the way you think about and use AI
Check Every on https://every.to/ and Dan on https://twitter.com/danshipper
Disclosure: Henrik is an angel investor in Dan Shipper's company.
📜 Read the transcript for this episode: Transcript of Dan Shipper, Every: "Use AI to make you and your business better" |
[00:00:00] Henrik Werdelin: Welcome everyone to the latest episode of beyond the prompt, where we dive into the fascinating intersection of technology and business. I'm your host Henrik. Wertland alongside my cohost, Jeremy oddly. Today, we're joined by Dan shipper. The founder is CEO of every, dennis here because he's an incredible finger, but also one of the people I know, who knows the most about how to use AI, not just for your professional life, but also to make your personal life as a little bit easier. So with that, let's dive in.
[00:00:26] Dan Shipper: Hi I'm Dan Shipper. I'm the Co-founder and CEO of Evry. Evry is a daily newsletter. We publish one long form essay a day at the intersection of business and technology. We've been around for about three years. We have about 90, 95,000 subscribers.
And for every, I write specifically every Friday, a column called Chain of Thought where I talk about AI as a creative tool and as a tool for thinking. And I'm excited to be here.
~Jeremy and I've been working for a while now on a bunch of different things. and we both been fascinated about ──── gen and the. The rapid growth.~
Now Jeremy and I have been collaborating on a lot of projects over time and we share this deep fascination for chatty beauty and AI in general.
[00:01:01] Henrik Werdelin: And I think one of the things that we're increasingly interesting in is how~ how~ fast it is happening and how few people that are using it. And so that Delta is really interesting to us. And so we've taken upon ourselves to make a podcast where we basically talk to people who know how to use it, in a very practical manner to really share some of the best practices and stories that are out there, ~so we're I~
~looking to talk to people who know what it is and have the ability to talk about the way that they use it ~ jeremy, was that kind of like the, and a non shitty explanation?
[00:01:28] Jeremy Utley: Perfect. Better than~ than~ Chad GPT could have done it.
[00:01:30] Henrik Werdelin: Great. I love it. .
[00:01:33] Dan Shipper: I'm actually doing something similar, like a Twitter series like this.
I just interviewed Sahil Avinjia about how he used to chat GVT. I have this thought that if you give people super powered alien technology and just drop it in their laps, they'd figure out how to use it.
But , the reality is that no one does. And there are so many people who are just going through their day being like, Hey, I know I should use this, but I have no idea how, and I don't have the time to figure it out. And I think stuff like this where people get concrete use cases is perfect.
[00:02:00] Jeremy Utley: ~I don't know, Dan. ~One of the things that's fascinating to us is personal versus professional. And I think because. So much hype around market cap, multiple expansion, big tech overtones folks immediately think in terms of the professional, but I find that actually then there's there's too much activation energy to even getting started.
And one of the things I love is your old post about your kind of your daily journaling practice. And I wonder whether there are some call it. AI Habits that people can form in their personal life that actually reduce kind of the friction to trying and provide them on ramp.
So then they go, Oh, wait what if I did that at work? It's not starting at work, but actually starting in their personal life. What do you think about starting from a, from the personal life versus an organizational diagnostic or assessment,
[00:02:54] Dan Shipper: for example? ~Yeah, of course. Definitely.~
One is just my custom instructions, I think are like very, detailed.
[00:02:59] Jeremy Utley: Yeah, custom instructions are a great kind of tangible starting point. So why, when we talked about personal AI habits, did you, did your mind immediately go to custom instructions as like a starting point?
[00:03:10] Dan Shipper: The reason I I thought of custom instructions first is because it's this sort of secret power up that makes everything about the way that you interact with chat.
You'd be much better. And I can explain how and why. But basically, I the way that I explained custom instructions. For people who don't know what they are, is when you interact with chat, GPT currently, it's acts like drew Barrymore on 51st dates. Like it just doesn't know who you are every time you interact with it.
And you have to like, re explain all the context. And you may not even notice that re explaining context is like annoying. But it makes a huge difference . When you can do that and it has a little bit of background on who you are and what you care about and how it should respond to you.
And it will make a bunch of connections for you that you don't even know you needed to make if it has that information there. So it takes a lot of the thinking off your plate. If you do it correctly, the problem is that it's hard to know what to put in there. So here's some starting points of what you might want to put in there.
So I just put, my name is Dan blah, blah, blah. I'm a writer and entrepreneur. Tell it a little bit about every, and just having that in it is super helpful because it knows my name. I have enough writing online that it knows who I am
and it will just, respond. When I say, Hey, I'm working on something new for every, it knows what that is. And that is really helpful and important. ANd then I just talk about things that I like books and writing and technology and programming. I love, I tell it about the kind of writing that I like. And I talk about, I like lyrical writing.
I like when it's accessible. I like when it's emotionally vulnerable. And then I have a bunch of people who I think are great as examples. And this is quite good. Like I'll often ask Chachi Bt for, I don't know, something like I'm writing something and I really need an example or a story or an anecdote and it can help pull from these people automatically without me asking. That's which is like really helpful. Or, I'm often looking for recommendations and it'll be like, Hey here's something that maybe Scott Alexander wrote that you didn't know about.
And that's really great I use chat GPT a lot for kind of working through interpersonal decision making, or just decision making in general. And then one big category is decision making around complicated situations with people.
I use chat GPT to help me bring awareness, bring my awareness to things that I. ordinarily might get wrong or might miss in my decision making. And it helps me work through areas that are difficult for me. So for example, , one of the key things in my life right now is I found that in interpersonal situations, , I index a, a little too high on the agree, agreeableness scale.
I don't really like saying no I'm like very afraid of the feeling of guilt that I might get if I said no to someone. And it's this like interesting, complicated thing for me because on the one hand.
I don't like saying no to people. And on the other hand, I really have a strong desire to collaborate and connect with people. Like I love people. And so in any case, there's a great rich tension there. Yeah. And so in any sort of interpersonal situation, it's complicated because I have to pull apart.
What am I motivated by here? Am I motivated to do this thing or hang out with you guys on this podcast, because I really want to connect and and hang out. And if that's the case, I should do it. Or am I on if I am I doing this podcast, for example, because I'm too afraid of saying no, because I, I might feel guilty about saying no.
And obviously I want to hang out with you guys. And I want to connect. So I'm here. But there are a lot of situations in my life where I may determine I actually don't really want to do this, but I'm just really afraid of saying no. And that would be ~a that's like~ an insight that you might have, going about your daily life or in therapy or whatever.
But it's often really hard to like, operationalize that or get it into your head day to day and get it into the decisions that you make. And what I love about chat GPT is I can say this stuff, I have a desire to connect with people. I'm also really afraid of rejecting other people and the resulting like shame or guilt that I might feel.
And I really want to be a little bit less agreeable. Can you help me with that?
[00:07:10] Jeremy Utley: So can I ask two questions on that right there, just because this is a fascinating thing. So for folks who are listening, Dan's highlighting this text says, I have a desire to connect with people and to collaborate.
I'm afraid of rejecting people because of the results in guilt or shame. I'm working on being less agreeable and a little more shame resistant and clearly setting a communicating goals to my team. I'm a little too opportunistic and I want to be more strategic. So two questions for you. How did you discover that's actually the thing that you need to put in custom instructions?
And then second. When you get an email like you did from us, Hey, want to talk to us on our podcast? Yeah. How does this custom instruction then interface with an
[00:07:50] Dan Shipper: inquiry like this?
Yeah. So to your first, question, how did I decide to put this out of anything else into chat, GPT? And the answer is really I just experiment, like I'm curious. I have a lot of different ways that I'm, I feel like I chat, GPT could be really helpful.
And~ I know. That~ I know already that I like use it for decisions. ~And so the sort of obvious thing is ~ maybe there's like a section of this that I could use to help me with decision making and in order to do that, I need to give it my, the patterns , the things that I'm constantly working on and trying to get better about.
And I'm constantly modifying and changing things because right now custom instructions is it's like, there's 1500 characters for the, what would you like Chachibita to know about you? And then what would you, how would you like Chachibita to respond?
So there's actually like pretty limited space. I think this is going to change significantly in the next couple months. It would be a really easy thing for open AI to. To raise the character limit and do better memory here. And I think it will be like a true game changer when they do. And I think they will.
But for now I just experiment, I move things in and out and see how it changes the responses. And then I think your second question was how do I. Use it? Like, how do I get into talking to chat GPT? Is that's the question? Yeah. Yeah.
[00:09:04] Jeremy Utley: Also, you said I use it for decisions. What does it mean? I think sound like an
[00:09:08] Dan Shipper: alien statement to some. Yeah. Yeah. Let's talk about it. So to be clear I have this section here where I'm telling people like, here's what I'm working on.
And then in this how would you like to respond? I have if you see an opportunity to help me be less agreeable or prone to shame, take it same for me, bringing me back to the big vision rather than being opportunistic in my decision making don't do this unless it's very relevant. So I'm giving chat to BT some specific instructions, help me in these situations. If you see the opportunity, take it. And then I have a couple of different frameworks here these are like past insights that I want to bring back up for me. So integrate past insights into our conversations where relevant expectation versus reality.
I often brace for negative outcomes, especially in social situations that seldom materialize. Team trust. My team is usually very supportive of me. Proclivity for distractions. I can be drawn away from my core objective objectives by new opportunities or by compromising with other people. Make sure I'm not losing sight of my objectives or compromising with others in ways that undermine my goals.
The no guilt question. If I'm in a complex interpersonal situation, ask what would I do if I wasn't afraid of feeling guilty? Help me balance the line between genuine connection and over accommodation. Help me contact anger if it's there. So
[00:10:20] Jeremy Utley: again this, these comments reveal more self knowledge than I think most people
[00:10:24] Dan Shipper: possess.
[00:10:25] Jeremy Utley: So it's just,
[00:10:28] Henrik Werdelin: it might be. tWo things. One, I think it's first good to just point out that the custom instructions are available only for dbt plus subscribers as far as I know. And I think you tweeted about that at one point. Like basically if you've only used a chat dbt in the free version, then you basically.
Have to do a do
[00:10:49] Dan Shipper: over or something like that. Yeah. It's it's pretty necessary to use chat. You can keep plus I don't think that you can get a good read on how useful this technology is unless you're paying for it right now.
[00:10:59] Henrik Werdelin: And the second thing, which I thought was interesting and I just discovered yesterday is this research that come out that actually.
Maybe counter to what most people think it actually helped to be pretty emotional with these models Like if you write in an emotional way, it seems that they respond better. Was that did
[00:11:19] Dan Shipper: I get that? Yeah if you say stuff like take a deep breath and think before you respond or I'll lose my job if you get this wrong, like all that kind of stuff will actually make them make the responses higher quality, which is like pretty wild.
[00:11:35] Henrik Werdelin: Crazy the two things that we talked about now, which I think most people don't do is to pay for the. The version that gives you these extra features and now bundle both the ability to make pictures and manipulate kind of Excel files and do these kind of custom instructions.
They don't do that. They don't pay for it. And two, how different your instructions are I was comparing them to the ones that I have and they're mine are much more utilitarian, right? Be highly organized like I think you have that mistake of real trust, consider new technologies
it's just like a list of things, almost like I was being prescripted to an MBA in turn. I'm definitely going to experiment trying to put them like these more kind of rounded parameters of how I like to be treated as a person.
[00:12:20] Dan Shipper: Yeah. Yeah.
[00:12:22] Jeremy Utley: So here's maybe one way to think about just wrapping kind of Henrik's observation there.
Say somebody goes, I'm trying to figure out whether I should pay for chat GPT. I don't know that 20 bucks a month is really worth it. What would you say what are the three to five first things they do that you believe would demonstrate maximum value where they go, you know what, I shouldn't
[00:12:44] Dan Shipper: cancel this subscription. Yeah. I, that's a really good question. I really, it's so dependent on who they are.
And so I can talk about what I use it most for. And I can even go through a little bit of an example with you about deciding to be on this podcast, like whether I should do it and we can see how these things help chat GPT respond. But~ I ~ the way to think about it is chat GPT is like a incredibly smart.
intern who never sleeps and sometimes is dumb. ANd like the question that you should ask yourself is if someone offered you that, like a human who would be around you 24 seven, who would follow your instructions, who would sometimes get things wrong, but is also like really smart and would help you see around corners and see your blind spots and help you get work done.
Would you pay 20 a month for that? And if the answer is yes, then you should it's probably a no brainer. And the answer is no. And I think for some people, the answer is no, because it's like I don't want to have to manage a fricking intern. Then don't do it. I think the technology is in this place right now where it is like intern level.
It will get to a place where you can be a lot more confident that like everything that it does is really working. And in that case I think for some people, they're just a little further along on the adoption curve and they don't want to deal with the, like the hiccups.
I love the hiccups that the hiccups make are so fun for me and interesting. And I feel, I just feel very curious and excited to understand how the technology works, but some people are not like that. And I think it'll, it will be for those people in a couple of years. And that's totally fine.
[00:14:19] Henrik Werdelin: Maybe a bit of a detour, but I noticed that I already started to use different of these models for different things. So for writing or for understanding data and then suggesting insights, I use Cloud more than GPT for making, let's say, background images for PowerPoint slides, mid journey, makes more like.
iLlustrations that are more visual compelling than Dali, but Dali specifically in the new version it's much easier helping you figure out how to prompt it properly. Do you, a, do you already also use different models for different things? And two, do you think we'll see a world where these models will increasingly specialize?
Where pi. ai is going to be your personal friend and, OpenAI is your Microsoft, or do you think it all going to basically get mushed together? And one, one
[00:15:16] Dan Shipper: winner takes them all. So I do definitely use. different models for different things sometimes. So I think that Claude gives some sometimes can give slightly better writing help.
anD Claude until yesterday had a much bigger context window. So if I needed to Ask a question about a really big piece of text. Sometimes cloud was a little bit easier to do that with. That changed yesterday. GPT for turbo has 128 K context window now, so you can feed it 300 pages of text, which is more text than anyone probably needs to feed it.
For a long time. I don't know how that performs, honestly. And my experience is that with long context windows. The attention mechanism for the GBT is it's, it gets fuzzy, it's harder to , get it to remember things that happened like at the beginning of the context and stuff like that.
So I, I would say I've experimented with a lot of them and I've definitely personally have collapsed on mostly using chat GBT. And I think my. theory is that in general people have room for one or two chatbots in their life. And but I want to make a really clear distinction between chatbot and AI.
I don't know, the predictive text on your iPhone is like, And so as an AI, right? And I think there will be all of these different places where generative models are like a little bit more seamlessly integrated that are not going to a specific place like chat to BT and like having a conversation with it.
But I do think like in the having a conversation with it. There's going to be like, each person is going to have one or two that is like dominant in their lives. And I think it's a really interesting question , to try to understand like, oh, is it going to unbundle? And are we going to, is there going to be a sort of winner take all?
Or is there going to be like this effect where there's lots of different models for different people? On the kind of like winner take all front, I think. Just for people. If you're not using it every day, you're going to forget about it. And so people are just going to collapse on one thing that generally is useful for them.
And what's really interesting is, Chachi BT is a mixture of experts model, which means that when you interact with Chachi BT, you're not just interacting with one model. There's basically a bunch of different models under the hood that are specialized for different things.
And the way that chat GPT responds to you, obviously, as you can see from my custom instructions is custom. My experience with chat GPT is going to be different from your experience, which means it'll be able to like work with a broad variety of people for different use cases, and it will just get better at that over time.
And so I think those are. Points in the camp of there's going to be one model for or one, chat bot for that, that sort of takes the market. And I think chat GP is like clearly the leader here. On the other hand, I think one of the really interesting effects of this technology is the more popular the model, the more popular the chat bot
the more diplomatic it has to be, the more it becomes the CEO of the fortune 500 company or the presidential candidate, or becomes like reading Wikipedia. And that's necessary, as they get more successful the reason that they get more diplomatic is because they have more to lose.
And the effects of what they say are much higher. And and I think that the same thing is true for these kinds of technologies, like weirdly the more that open AI scales, they have a hundred million users now, and I think they'll have more soon. The more careful they have to be about how the chatbot what it will do and what it won't do because their risk of service area is much higher.
And I think that opens up opportunities and you can even see this right now for Elon launching his grok model for people who are are a little bit further back in the race to be like I'm going to carve out a little niche of being able to take risks in a different way.
So one way to take risks is just to be . a little bit more inflammatory or whatever, but that's not the only risk that you can take. You can take risks around I'm not going to be the big consumer model, but I'll give you legal advice. Which at GBT, like probably we'll try to refuse to do, or I'll give you medical advice or whatever.
And so I think there's going to be a lot of opportunity for chatbots that are willing to take risks. And those opportunities will get bigger as large models get larger and need to serve more people. And , I think the effect is similar to what in media where large media companies, the New York Times or, whatever.
Not always, but in general, the larger you get, the blander you have to be. And then there's like lots of room for small media companies that are gonna, talk about the world in a way that's more idiosyncratic to them and really voice and that kind of stuff.
So that's my general take. And I think yeah, there's going to probably be like one or two, like really big consumer scaled chatbots. And then there would probably be a whole vortex of smaller ones that are more specific for different sub communities of people.
[00:20:14] Jeremy Utley: One thing I thought would be cool is if you want to play through a recent decision where you feel like chat GPT provided helpful guidance
[00:20:23] Henrik Werdelin: totally.
[00:20:24] Dan Shipper: I can walk you through a hypothetical so , I can show you how it would respond to me trying to talk about the podcast ,
[00:20:31] Henrik Werdelin: I think that's fine. So you get an email from us saying, Hey,
[00:20:35] Dan Shipper: we'd love you to have, let's see , I would just be , Hey I got an email from a friend asking me to be , on his podcast. I'm not sure if I can do it if I should do it.
Can you help me think it through?
So even something like this, where it just knows your name is just like super helpful. So first of all, it asked me to consider like goal alignment, right? And this seems like a small thing, but I think it's I think it's quite important for me as a person, one of the things that I feel like I'm less good at is, especially in an interpersonal situation, stepping back and be like what is my actual goal here?
reGardless of what people want or, what I think is expected of me, how does it align with like my overall goal? And I've set some overall goals for every and for myself in custom instructions. And I can ask and ask me like, Hey how does this align with your goals? It could boost being on podcast could boost my visibility.
It might also drive traffic to every. So like one of the things we're going for is a hundred thousand page views a week. So like maybe it could help with that, but it's at least worth considering. And even as you
[00:21:49] Jeremy Utley: say it, it's amazing how often we interface with a request without considering our goals.
[00:21:54] Dan Shipper: Exactly. It's
[00:21:55] Jeremy Utley: it's so obvious, but it's like, how many times do you interface with a request considering your burrito or considering your sleep or there's other dominant considerations other than remember what you say you want.
[00:22:07] Dan Shipper: Yeah, exactly. And that's the thing. It's never going to do.
Sometimes it says something that you're like, holy shit. Like I would never have thought of this in a million years, but a lot of the ways that it's useful is for just reminding you of the brain dead, simple stuff. That's just really hard to remember. And that's really valuable.
So next is this opportunity to GJK opportunistic does the podcast align with your target readership? Will this Appearance position you in every in a way that you aspire to. So like similar to the first reflect on the time commitment versus the potential return.
So is it worth it? And then the last one, which is the, which is really important. Consider if any hesitation stems from fear of guilt or shame, perhaps the worry of not living up to expectations, their podcast, not going well, remember the no guilt question. What would you do if you weren't afraid of feeling guilty?
So that is like a really important question for me, which is like, Am I considering doing this because I'm might be afraid of feeling guilty if I said no. I think chat GP is also worried that I might be afraid of going on it for fear of it not going well, which is not usually the thing for me, but like in general, the, like the no guilt question, what would you do if you weren't feeling guilty?
Yeah. What do you do next?
[00:23:12] Jeremy Utley: Cause to me, just as I interface with this, it's actually a lot to consider. And I almost wonder do you take it a piece at a time? Yeah.
[00:23:19] Dan Shipper: How do you, what happens next? This is a good one. So I actually have in my custom instructions please ask me one question at a time.
And it doesn't do that yet. So I'll often happens to be all the time. I'm like, I said one question. Yeah. Yeah. Yeah. So I'll often be like, Hey, this is a lot. Can you ask me one question at a time? So I don't feel overwhelmed. And another really important thing is I sometimes just put it in this prompt too.
And that just makes it a little easier. So I don't look at a wall of text. I think this will get better pretty soon, does the audience align with the target readership you have for every the answer is yes, I think so. They definitely seem to cover topics relevant to tech entrepreneurs and people interested in
[00:23:59] Henrik Werdelin: AI.
~That's besides the, this is maybe not part of an interview so much. Every is really an outlet that is very useful. I don't think just as an entrepreneur, but for anybody who's thinking about how to innovate within our organization in many ways. Maybe even more and so I was curious why you see the readership as being as tech entrepreneurs, like it's besides the point, but it's fascinating when you ~
[00:24:01] Dan Shipper: ~talk about it.~
~Yeah. I think I think for any, like when I think about who we want to serve. I think the probably like larger word is like founders or builders, people who want to make stuff. And I think like in the middle of that dot are people who are starting tech companies because that's the world that we come from and I can like talk to those people pretty easily.~
~And that group of people I think is aspirational for many people who want to make stuff. And if we. If we drive our ship toward that, I think we'll bring along all of the people who who are not just aren't just in that small dot, but are want to be like that or want to make stuff in general.~
~But it's, I think it's a useful framing for me to think about are we serving the core of the audience that we want to have. ~And
[00:24:01] Jeremy Utley: this interaction is making it obvious why you started with custom instructions, because the nature of your interaction is so bespoke to you and the way you are asking for advice or asking for input on a decision.
Of course, you don't want generic advice, right? Yeah. So to me. We almost have to circle back to custom instructions to understand, like what one tactic that I like to use is, interview me or something like that. Yeah. Could you, if somebody doesn't have any custom instructions, they go, I have no idea what to put in that, how to talk to me.
Is it as simple as saying, interview me about the custom instructions. I should
[00:24:39] Dan Shipper: set, I love that idea there. I think that's so good. And just while I'm responding to that let me just show you what this chat is like if I don't have custom instructions enabled, because it'll be way worse.
If I just do this same thing. Okay. Dan just
[00:24:55] Jeremy Utley: turned off custom instructions. I
[00:24:57] Dan Shipper: turned off custom instructions. So this is actually, this is interesting. So it does get who's the target audience? It does get Stuff about the time commitment, but what it doesn't get is friendship dynamics.
This is interesting. So it's going for some of the same categories of things, but it does, it's not going to ask me the specific questions or the, it doesn't say how does this align with every target readership, which is like tech entrepreneurs or and stuff like that?
How does this align with your goals which are to do xyz and what is the what would you do? If you weren't afraid of feeling guilty, so it's not going to do that stuff and I think that those like little things are actually make it make a big difference it feels like it knows you in this like really interesting way
[00:25:41] Jeremy Utley: , it still requires your translation.
If I think about the cognitive load, as I'm reading this, I still have to define target audience. Whereas the prior example, it defined target audience. And that's all of a sudden now, like it's just one extra step of translation. You don't have
[00:25:56] Dan Shipper: to make, , it's a totally different ball game.
It makes it much easier. And I think, yeah, your question is really apt. It's like. Where do I start? Should I have chatty? Would he interview me about what should be my custom instructions? I feel like that's a really common solve for a lot of AI you don't know what to do, ask the AI and it will help you know what to do,,~ that's really helpful. I've not had it. Do that, but~ definitely ask Chachi to interview you. ~I have some articles about this. I can send you all for all of your listeners that go through some of the basics of what to put in there. ~And I think also , one thing to think about is, when we're going through our days and we're talking to friends, families, coworkers, or we're reading books or whatever, there are always things that pop out and you're like, wow, It would be great if I could X I don't know I wrote an article , a couple of years ago about this CEO, Andrew Wilkinson, who he doesn't take any meetings at all.
And he runs a public company. And yeah, and he's, the article is titled the CEO of no. And that's an article where I might read that and be like I want to be a little bit more like that. And the minute I have that thought, I can just be like, okay, I'm going to just throw into custom instructions, like that little insight of I want to just try to take no meetings or whatever.
And just see how it goes. And you'll over time, you build up this library of things that are the kind of thing that you might normally write down in your note taking app that maybe instead of doing that, you just put it into chat GPT, because it's going to help you actually operationalize it rather than just have it be listed somewhere in a note.
On
[00:27:13] Henrik Werdelin: that note, a huge hack that I started to use is the, that they have the Whisper feature implemented in their app. I hope, may I? Yeah. So most of the time when I get a thought, obviously Danish being my first language and me being somewhat on the philosophical side, some of my kind of inklings, sometimes a little bit of philosophical in nature.
And so they can be tough for people to compute if it's not somebody I worked with for a long time. However, I found that. The chat GPT app is incredibly useful for that. So I will wake up and have a thought that I think is brilliant. And I'll basically babble into the chat app for a good three or four minutes and basically say, could you just write it up in a more structured way?
, even for myself, it's a much better kind of tools to like just take this very unstructured kind of like concept and then try to figure out what is the actual insight there. And then often I then want to share it fast with somebody. And obviously I can just call them and try to get a half an hour .
On their calendar, which is tough, but here I can just copy paste that thing, probably adapt it a little bit and then send it
[00:28:18] Dan Shipper: over. Yeah. One, I'm curious how that works for you. Cause one of the things that I found, like I tried to do the same thing, but I end up pausing a lot when I'm thinking through something.
And if I pause, it thinks that I'm done responding and then it responds back to me and I'm like, wait, I wasn't done. I almost use filler text. I'll go, can I yeah. Is that's a secret. Yeah,
[00:28:39] Henrik Werdelin: no, ~this is maybe I, maybe for a, ~this is like detailed feature stuff. There's two ways you can talk to chat dbt.
You can basically go into voice mode with it, but they also have five small lines next to the text input field. And if you press that, it basically records until you press the button again. Okay.
[00:28:55] Dan Shipper: That's, I should try that. And that's what
[00:28:57] Jeremy Utley: you that's what you called whisper, right? Just to be clear for a listener.
Whisper is like the little whiskers by the text box.
[00:29:05] Henrik Werdelin: Yeah. Dan will explain better what whisper
[00:29:07] Dan Shipper: is than I. Oh yeah. Whisper is the the open source library, ~right? I think it's open source, but it's, yeah, it is. It is open source. The open source library that trans~ that takes audio and turns it into text.
Open and I release it. It's like really good transcription software that is built into chat GPT, but any developer can take it and use it for their own applications.
[00:29:23] Henrik Werdelin: What's really cool for a non English native speaker is that it speaks all other languages, and so it's the only voice to text kind of system I've ever tried that actually understand Danish to the point where you don't have to rewrite
[00:29:36] Dan Shipper: everything.
Yeah, it's pretty crazy. They have ChatGPT, they have all this crazy stuff, and then, oh, by the way, they they like solved transcription and it's better than human level. And it's a problem that people have been working on for decades and it's a little side project for them.
~Have ~
[00:29:48] Henrik Werdelin: ~you ever played with~ speaking of language? Have you ever played with languages? And ask, because I was playing with. This thing called 11 laps, which are very good at doing text to voice for customers. And they had a new feature where they, you basically put in a YouTube video of yourself or somebody else.
And then you say, please translate the language and it uses the same voice and change the genetic and it's pretty remarkable as a media outlet. Is that something where you have been thinking maybe I should. Try to put this in German or
[00:30:18] Dan Shipper: whatever. nOt yet. Maybe I should.
I think we're, we still have a lot of English speakers to cover before we get to other places, but I think eventually we'll probably get there. That
[00:30:29] Jeremy Utley: was one of Mr. Beast's big insights, right? I, if I remember correctly, he's like ~only, whatever the number is, right?~
There's only 800 million English speakers in the world. whatever the number is, right? We've got to make all my videos in Spanish and Portuguese. And he's got, like hundreds of millions of followers who follow his language specialization kind of sub
[00:30:45] Dan Shipper: channels, right? And now you can
[00:30:47] Jeremy Utley: do that weight. He uses voice actors, but it sounds like you can do it in a fraction.
Hey, ~I want to do,~ I want to take a hard left just to see if it goes anywhere. Tell us about Lex and the decision to spin Lex out and Henrik and I, in another kind of area of our lives are fascinated by spinning out technologies and things like that, but tell us how you developed it and why you decided
[00:31:07] Dan Shipper: to spin it out.
Yeah, it's a great question. So Lex is it's an AI writing app that we incubated at every, you can think of it google Docs with with AI built in. So as you're writing, you can press command, enter and it will start to complete what you're writing about.
And we launched it about a year ago. And when we had originally built it we had intended , to run it inside of every, but when we launched it about a year ago, it just went like crazy viral. , it took off only that neither me nor my co founder, Nathan, who's the one that built the original version had expected.
And I think like when you're running a business and then you start another business inside of that business that has a very different shape from the business that you're running. There, there's always a bit of a question of like, how do we do the best thing for both businesses to succeed? And it.
It felt pretty clear that like Lex could have a venture backed software shape. And, every as a media company we raise a little bit of money, but I think we've always been pretty wary of. Putting the media company on the venture path. It was something that was like was a thing that I think, especially in the beginning, we were like maybe, but also the history of media companies that have raised lots of venture money is that it doesn't really work out very well.
And what's more important to me is to make really great stuff for a really long time than it is to scale to a billion dollars in five years with the newsletter. aNd so when you're looking at the media company, which has like a very profitable cashflow business shape, and then the the kind of AI software company that has it could have the profitable cashflow shape, but it also could have a sort of a venture back trajectory.
It doesn't really make sense to keep them together. So what we did is we we spun it out. We mirror the cap table. So All of our investors they got their serve, of stake in Lex. And then Lex went on and raised like a 3 million seed round from true.
So everyone got markups. And that's actually a very cool business model for media companies. I think we'll do more of those. We've, since that time we've launched a couple of other like little tools and I can, I. They're always good subscription moments, like subscribers really like them.
And I can see us , as we launch more of those and to the extent some of them succeed, see us spinning them out as well. One of the things I've learned over the last four years of running a media company is you're spending all of your time building up the audience and the reputation and putting out great stuff, and then you have to monetize it as.
In as many ways as you possibly can and so we do all the the basic stuff like subscriptions and ads and courses and all that kind of stuff. But I think software is this thing that we're actually like pretty well positioned to do because, we have a background in building software companies and is also, software companies are great businesses.
And I think especially with all the AI stuff right now. There's a lot of opportunity to build interesting new tools for people. I'm excited about that as a way forward.
[00:34:02] Jeremy Utley: We'll just from the totally from the outside, I've seen what y'all have done and I not understanding kind of all the rationale that you just laid out.
I admire the singular focus because it's easy to say, another company could have easily said, now let's become this other thing. It's Whoa, we're getting pulled. Like you said, this crazy viral moment. Yeah. It's like maybe, there are, there, there's an archetype of founder who says we thought it was this, it's actually this other thing.
And it's easy to rationalize. How did you personally think about staying true? To your original vision. Because I'm sure that there are investors who are like, dude, are you crazy? Let's go all in on this other
[00:34:38] Dan Shipper: thing. We definitely had that. And my co founder Nathan is now running that.
So I think that was definitely his feeling was like, this is actually the thing that he wanted to spend most of his time on. I think for me. Maybe I'm just like revealing my like craziness, but I just love getting to write every day. I love getting to run a business that's built on writing.
I love finding writers and helping them be great. And I think there's a really limited supply of the kind of writing in the world that I think is really good. And we have an opportunity to make that. And I didn't want to throw that away. I think it's it's a very rare thing.
I'm doing the thing I want to do. Did you
[00:35:14] Jeremy Utley: leverage, did you ask Chad GPT for guidance in your own thinking there?
[00:35:19] Dan Shipper: I'm sure I did. It was a long, it was a long process. I had a lot of different conversations with a lot of different people about what to do. Cause you know, you never want to like, in a high pressure situation like that, you never want to like, Make the wrong decision.
So I'm sure that there was some guidance from chat GBT. But yeah, I don't have any specific chat GBT memories from that
[00:35:36] Henrik Werdelin: time. wHat's your thought on this potential for creating new software on top on? Specifically when you are somebody who is serving a customer, for example, entrepreneurs, obviously, as you say, every has a lot of opportunity to do that, but I would imagine anybody who's listened to this working within a specific domain.
And if they're curious about AI now, they're honestly probably early. And so I think a lot of probably said anything, Hey, it'll be cool to build like. A custom, some layer on top, do you have a mental model for how to think about that? Or are you just thinking, , it's going to be much we democratized distribution through the internet and increasingly like code development with these models.
[00:36:23] Dan Shipper: What's next? It's a really good question. We're obviously. sTill in the process of figuring that out. I don't think anyone, really knows. And I think open AI is trying to figure it out too. Like you can see yesterday they launched the ability to create custom GBTs so anyone can build a GBT chatbot with access to private knowledge and its own personality and access to tools without having to code.
And they've also created this whole like app store where you can. List your GBT and people can download it and pay for it and all that kind of stuff. And I think that's their next like foray into trying to enable a sort of platformy type experience. I think that, that sort of gets back into are we going to have a one chatbot to rule them all situation?
Are you going to use separate chatbots for different use cases? And~ Yeah, I think they're, I think they're really,~ I think they're really trying to figure that they're in this like weird position where they started off as an API company wanting to enable developers, and then they accidentally created the like, most popular fastest growing consumer app of all time.
And have right. Yeah. And the really interesting thing is they don't train on API users. They only train on chat GPT. And so they need chat GPT to keep improving their models at pace. And then that kind of creates this like weird Strategy question around if you're going to make decisions, who do you prioritize?
Do you prioritize ChatGPT and it's 100 million consumer users, or do you prioritize the developers that you're trying to enable to build new businesses and all that kind of stuff? They don't say this, but I think they have to prioritize ChatGPT. So I think that's a really important thing for, Developers who are looking to build businesses in this area is you should be aware that if you are in their way in the way of chat, GBT, you're going to get.
Replaced, basically. And so it's a dangerous place to be building. Even yesterday, chat, GPT, launched all this stuff they added retrieval into their API. Retrieval is a process by which for any sort of chat bot when you ask it a question the chat bot can go and find chunks of knowledge from a internal knowledge base, and then use that to construct its response.
And there are all these developer libraries like Lang chain and Lama index, where I'm an investor in Lama index and a bunch of other ones. And , that are built to do retrieval and it doesn't mean that those libraries aren't useful anymore, but definitely opening. I just took a chunk of their functionality and just integrated it right into the API.
And building at that sort of infrastructure layer, I think is a little bit dangerous. And then, yeah, building any sort of, very general chat bot, I think is a little dangerous. And then, but there are places around the margins where I think I think I think there are opportunities.
And I think open AI is going to be incentivized to. Create more of these sort of app store things. They want people to build stuff. They just don't know how to do it. And so this sort of like GBT thing is another test of that so that they previously built plugins, which didn't really work~ now they're going, it's ~
[00:39:17] Henrik Werdelin: ~also fill with people who try that, right?~
I remember. Facebook platform being a big thing for a long time until it was not. And, it definitely seems that a lot of these companies are trying to figure it out in real time and don't always
[00:39:29] Dan Shipper: get it right. Yeah, exactly. The app store is the best example and everyone's trying to build their own, but ~like how well that actually,~ how well it actually works it's all about the sort of the dynamics of the products and the needs of the users and all that kind of stuff.
While I am skeptical of there being a need for more than one chatbot. I do think that there probably will be a form factor that OpenAI can figure out that upgrades your GPT like your chat GPT with a specific set of knowledge and specific set of instructions that you want.
In the same way that I might go to the bookstore to buy a book for myself. I might want to go to the app store to buy a book for chat GPT, to give it some skills and some knowledge that it might not ordinarily have push it in a way that is like more custom for me where
I don't have to explain it. I don't have to write the custom instructions. Henrik has put together ~his, ~ all of his experiences from all the things he's built. And I can just press a button and load it, ~it's ~ like in, in the matrix when like Neo downloads, the thing
[00:40:23] Jeremy Utley: we're getting dangerously close.
~That's I'm like, ~ I'm imagining. ~Imagine~ a world where you don't need to read books anymore. You just have your
[00:40:28] Dan Shipper: AI read the book. That's I really think that's that's a thing that will probably end up happening like it's much easier to load all the knowledge and information to your eye than into your brain.
It doesn't mean that books are not useful anymore, but I do think that's a new content format where people are probably going to specifically write. Long pieces of content for AI. So it's meant to be read by AI and that's its own thing. That's wild.
[00:40:54] Jeremy Utley: A whole different readership. Is there a collapsing of the two conversations we've had between call it custom instructions and then retrieval where, both Henrik and I are authors, right?
Why do our custom instructions not include our books? AS when you're responding to me about decisions, feel free to leverage the frameworks I've already put into the world, right? To me that's On Brandand, right? ,
[00:41:17] Dan Shipper: yeah.
[00:41:17] Jeremy Utley: What better way of feeding customer
[00:41:19] Dan Shipper: instructions than my own work.
Yeah, I know. You should and I think that's a I think that's coming. And you might be able to do it today, if, especially if the content that's in your book some of it is also online, like Chachi b team might know about it already. It's worth trying.
[00:41:31] Jeremy Utley: Dan, last question, just to get personal. ~Cause I also I used to blog every day that. Became a little onerous and I do kinda once a week, I'd love as one writer, from one writer to another.~
~And our audience, of course, they're interested in mine and Henry's writing as well.~ Why write ? And especially just given your answer to the whole question about Lex and spending it out to me, you clearly have a passion and I'm sure you've got blog posts about it. Yeah. But if somebody only had a minute to hear you wax poetic about why writing is so integral Yeah.
To your life and worldview. Perhaps maybe a better question is in the age of
[00:41:55] Dan Shipper: AI, why bother writing? It's a great question. And I think it's the answer to that is similar to like, why start companies? It's like you do it cause you can't not do it.
And if you can do anything else, do something else. Cause it's it's really. ~Fucking~ hard. But I think in both those cases for writing and for starting companies I just love it. I can't not it's what my mind does when I have nothing else going on, I'm like, oh, I want to, I have an idea for something I want to write, or I have an idea for a company or a product or whatever.
That is just what I do. It's it's basically breathing. And in an age of AI, it's like, it's a really good question. I don't know what writing is going to be like in five or 10 years, but I know I'm going to try to figure it out. And that, I think that's the fun thing is I get to spend all day being curious about new ideas and writing about what I think about them.
And that's like a, that's what more could I ask for is really how I think about it. It's a great existence.
[00:42:42] Jeremy Utley: Thanks for joining us today. This has been inspiring and illuminating, and I'm sure that our audience is going to get a ton out of your insights.
[00:42:48] Dan Shipper: So thanks. Thanks for having me. It's great to hang out with you guys.