AI's chat interface problem and Lobe's imaginary seed round

Studio (00:02.774) All right, episode two, episode one had two parts. We went really long on that one, so we broke it into two parts. Episode two, and this one's gonna be a short episode, which... By short we mean we're skipping two of our normal segments, which are mental models and tool time, you know, where we dig into a tool. And we just talk about a subject. So kind of a... Episode two, episode one had two parts. went really long on that one, so we broke it into two parts. Episode two, and this one's gonna be a short episode. By short we mean. By short we mean we're skipping two of our normal segments, which are mental models. In tool time, you we dig into a tool. And we just talk about a subject. So kind of a one-shot hot take on a topic. And I chose the topic this time. They'll choose it next time. So... Well, if you remember how the topic came about, it was... We were testing recording and I was showing you a tool. kind of a one-shot hot take on a topic. And I chose the topic this time. You'll choose it next time. So. Studio (00:42.904) Wait, how did it come about? Studio (00:46.862) Mmm. John's mic (00:51.798) Rube. yes. And then all of a sudden you just had a lot to say. I had a lot to say. the chat interface. Yes. So today's short is about the chat interface for AI and its ubiquity, its current ubiquity and my very strong feelings about the problems with it. And so we're going to dig into chat interface but first you wanted to follow up on the startup that we yes, yes. Studio (00:57.664) I had a lot to say because I had written a blog post about this. Yeah. So today's short is about the chat interface for AI and its ubiquity, its current ubiquity and my very strong feelings about the problems with it. and so we're going to dig into chat interface, but first you wanted to follow up on the startup that we created in episode one, which is lobe. episode one, is low. So, the, so there's a prompt and we'll, we'll see, I think we're going to do this, do this a few times, but so I gave Eric a prompt for the first part of episode one to a start an AI startup with a constraint of it has to be a screenless startup. Screenless device. Yeah. Screenless device. And then the other constraint. strike there has to be a screenless Screenless device. was he had to name it first and then come up with the product. Do you want to give a 30 seconds on what you came up with? So give us the elevator pitch then. Studio (01:51.374) He had to name it first. And then back into the product. Do you want to give a 30 second song? Yeah, I've been working on the pitch because I've been... So give us the elevator pitch. Okay, the elevator pitch. Studio (02:09.026) The deepest insights about your life come from your dreams. The deepest insights about your... Studio (02:16.171) But remembering dreams is difficult and when you wake up, the dream is often fragmented and is forgotten. Lobe helps you discover the deepest parts of who you are by recording biometric data as you sleep, especially during REM sleep, and then reconstructing your dreams after you recount what you remember. But remembering dreams is difficult and when you wake up, the dream is often fragmented and is forgotten. Lobe helps you discover the deepest parts of who you are by recording biometric data as you sleep, especially during REM sleep, and then reconstructing your dreams after you recount. Okay. you go. And find the deepest parts of you. Yeah. And a hardware device that sits by your bed. Right. Awesome. That's just a, that's just a detail, right? If you want to, if you want to the deepest parts of you, know, fact that it's a device that sits by your bed. Just a detail. Yep. Okay. So congratulations. You've just received a $3 million seed round. Loeb, Loeb, find the deepest parts of you. Studio (02:51.758) Hardware device that sits by your bed, right? But that's just a detail, right? If you want to find the deepest parts of you, the fact that it's a device that sits by your bed is just a detail. So congratulations. You've just received a $3 million seed round to pursue your dream. to pursue your dream, pun intended. Okay, so $3 million seed round. Right. So what do you do first? Or what's the go-to-market plan? Okay, three million dollar seed round. So. Okay, Well, this is tricky because at least part of that is, at least part of the money needs to go to producing the devices, right? I mean, there's a... Yeah, I there's a hardware. There's a hardware aspect, right? And people have high standards, a lot of times high standards for hardware these days, although this can be very simple. Agreed. So... John's mic (03:32.44) Well, this is tricky because at least part of that is, at least part of the money needs to go to producing the devices, right? mean, there's a... Yeah, I mean, there's a hardware. That's a real... There's aspect, right? And people have high standards. A lot of times, high standards for hardware these days. Although, this can be very simple. Agreed. So... So a decent bit of the money is actually going to go towards battery technology so you can actually, so that you don't have to charge it very often. You can just leave it sitting there for a long time. because it generally only records during REM sleep, it has a pretty long battery life. But we're going to have to either acquire that or develop it. So a decent bit of the money is actually going to go towards battery technology so you can actually, so that you don't have to charge it very often, right? You can just leave it sitting there for a long time. because it generally only records during REM sleep, it has a pretty long battery life. But, you know, that's, we're going to have to either acquire that or develop it, probably acquire that technology initially. So super long battery life. Okay. So these are really, really designed for really mobile because I was imagining an alarm clock where you just plug it into the wall. But, but yours is like you're traveling all the time. You know, yeah, whatever. So you got to have the battery part. Okay. Absolutely. Yeah. Make sense. I think it looks inductive charging some sort. Right. And then you don't, but the whole point is you don't have to charge. Hmm Studio (04:38.862) Sure. Yeah. Studio (04:43.274) Absolutely. Yeah. And it just, you just set it next to your, you know, I think it looks nice. Right. And then you don't, but the whole point is you don't have to charge it very often. So let's just say a third of the funds are going to go towards the hardware side. Right. So that's research, you know, sourcing, manufacturing, et cetera. so, all right, we have 2 million. Right, right, right. Let's just say a third of the funds are going to go towards the hardware side. Okay, so that's research, know, sourcing, Right, you have two million dollars left. This is great. I don't know, is a million enough for that? I don't know. We'll find out. Yeah. This is great. I don't know, is a million enough for that? We'll find out. Well, gosh, the Kickstarter stories. So the initial go-to-market is going to be through partnerships. So I think it'll be much more difficult to crack the We can always go to Kickstarter if we need some more funds. the initial go-to-market is going to be through partnerships. I think it'll be much more difficult to crack the biometric, like the big biometric trackers, know, so, ORA Ring and Fitbit and of Apple Watch. biometric, like the big biometric trackers, so Aura Ring and Fitbit and of course Apple Watch. John's mic (05:50.798) Right. So if you can't beat them, join them. Well, no. Absolutely not. But a company like 8 Sleep that focuses specifically on sleep. Right. I it's a really interesting channel. There are the apps. I Sleep Score was the app that I mentioned. So there are several. Maybe there's an acquisition there. Just go ahead and. I mean, you're definitely partnering with. Well no, absolutely not. But a company like 8 Sleep that focuses specifically on sleep, I think is a really interesting channel. There are the apps, think Sleep Score was the app that I mentioned. So there are several. Totally, mean you're definitely partnering with people who are potential suitors. Right. Oh, I just don't know how well that app's doing anymore. Maybe you should buy it. Buy their technology. Because, I mean, that's one of the best strategies for a seed-round company is to acquire another company. Sorry, sometimes that is a good strategy. Right. I mean, they work. yeah, that is super interesting. Studio (06:35.054) Well, think that the Yeah, so a million dollars is we're gonna go buy the sleep score app, you know For a lot of years on that, I was joking, but that actually might be a decent idea. actually is not a bad idea. Okay, we're gonna explore that. Okay. All right. But my initial thinking was that that was a distribution plane. Could be both. Yeah. Yeah. Because they have kind of a freemium model, but the combination there... And no hardware. Yeah, no hardware, right? But you sleep better if you don't look at your phone right Exactly. Yeah. So they need the hardware. They need the hardware. It actually is not a bad idea. Okay, we're gonna explore that. We're gonna explore &A with that. But my initial thinking was that that was a distribution play. Because you're just buying their user base, right? Because they have kind of a freemium model, but you know, the combination there. That's interesting. You had no hardware, right? But you sleep better if you don't look at your phone right before bed. we're... They need the hardware. This is actually... This is... They need... They need us. They need love. They need love. To help people sleep. Right. Because if you have your phone next to your bed, you're going to at it before you go bed. Exactly. Yeah. But love is really... It's screenless. exactly. Okay, so that's the second million is that we're going to try to acquire a sleep score. Okay. And then... We've got a million left. Okay, we've got a million left. Well, that's just going to be like runway for... Yeah. like, the objects, know? Yeah. they need Lobe to help people sleep, you because if you have your phone next to your bed, you're going to look at it before you go to bed, you know? But Lobe is really... Yes, exactly. Okay, so that's the second million is that we're going to try to acquire a sleep score. And then, okay, we got a million left. Well, that's just going to be like runway for, for like Opex, you know? But the, okay, go to market. So partnerships. John's mic (07:47.886) But OK, go to market. So partnerships, we're going to try to acquire sleep score to just have immediate distribution. then. we're gonna try to acquire sleep score to just have immediate distribution. And then... This is gonna sound old school, but I think physical mattress stores, actually. This is gonna sound old school, but I think physical mattress stores, actually. I love that. I love that idea. You know? And actually, think what you could do is the... You know, like you just, this is, actually I think what you could do is the, you know, you could basically give it away for free with a mattress purchase. Cause people don't purchase mattresses that often. Right. Right. And so you put, you know, it's, that's a thing. You get your mattress, you go take the sheets off of your bed, pull your old mattress off, find a dumpster because you don't want to dispose of it. John's mic (08:18.286) You could basically give it away for free with a mattress purchase. You wouldn't want purchase mattresses that often, right? Sure, agreed. That's a thing. You get your mattress, you go take the sheets off of your bed, get your old mattress off, find a dumpster because you don't want to dispose of it. Yeah, I actually have a story about that. So there you go. I think that's going to be a big one. Because if you buy loads So there you go, I think that's gonna be a big one. Because if buy Loeb, if Loeb comes with your mattress, there's a mental association there, right? If lobe comes with your mattress, there's a mental association there, right? Yeah. So this is funny because the moment you mentioned that, I imagine us hiring an army of teenagers twirling signs that say lobe in front of mattress doors. Exactly. Is that where you're going? Totally. And then the, what's the guy, like it has some, the, like, I don't even they load in front of mattress doors. Exactly. Exactly. totally. And then the, what's the guy, like it has some, like. the right word is. It's the thing that's like, I don't know, eight feet tall that like blows around the character. What is that thing called? In our family we call that a silly guy. Okay. Because that's what one of our children said, look it's a silly guy. So there's the silly guy here and the master store in the background. With the blower, yeah, the blower with the character. Sort of waving around. Yeah, Exactly. So you've got that here. You've got the teenager. Studio (09:17.76) It's the thing that's like, I don't know, eight feet tall that like blows around. yeah. What is that thing called? In our family, we call that a silly guy. Okay. Because that's what, you know, that's what one of our children said. Look, it's a silly guy. But it's a silly guy here. It's a blower and it's like, yeah, there's a blower on it. And so he like, you know, sort of waving around and exactly. You've got that here. You've got the teenager twirling like the lower side. twirling like the lobe sign and then the mattress door in the background. I think that's the right visual. Okay. Surprisingly well. Right. Most of which will go to our marketing strategy around. Yeah. Yeah. Yeah. Yeah. Exactly. That's right. That's our, that's a big part of our go to market. All right, how did I do? So I spent $2 million and I've got a million dollars of runway left. Go to our marketing. Exactly. That's right. As it should. Yeah. As it should. There we go. Loeb, go to market. All Can we talk about AI chat stuff? Because of course, I have really strong feelings about this. I wanted to actually, I'm going to share my screen. Do we need to? All right. All right. Can we talk about AI chat stuff? Yes. Because of course I have really strong feelings about this. I wanted to actually, I'm going to share my screen. Do we need to start with the one site that this all came from? Or do you just have a, do you want to just back up and give your- of the not let here. Yeah. And I'm just going to show you. Studio (10:32.142) I'm actually going to run the gauntlet here and I'm just going to show you. Let me see. Let me pull this up. Let me pull this up. Studio (10:43.682) Okay, I'm just gonna run through a bunch of tools here. OK, I'm just going run through a bunch of tools here. Nice. So this is a tool called PoE, and it's essentially a chat aggregator. think this is from the people who So this is a tool called PoE, and it's essentially a chat aggregator. think this is from the people who... The people who did Quora. Quora. Yep. So this is like a... They were early. They were very early actually. Yep. You can do group chats and all that sort of stuff, right? But let's just look at the general UX here. This is Bolt. So AI Builder. Stunning apps and websites by chatting with AI. people who did Quora. Quora, yep. They were early. Like early on. getting early, actually. You can do group chats and all that sort of stuff, right? But let's just look at the general UX here. This is Bolt. So AI Builder. John's mic (11:20.17) stunning apps and websites by chatting with Replet. Studio (11:30.87) Lovable. create anything, which I love to come. Create Anything, which I love the company name, Anything. Pretty awesome. I don't know. But I think this is, yeah, create anything dot com. Rube, this is the one that, this is the one that. Yep. It is a cool name. I wonder how much anything.com cost. John's mic (11:46.636) Rube, this is the one that, this is the one that, yeah, that sparked the whole discussion. And then, of course, close to my heart, it is V0. But what do you see about all of these, what do you see about all of these that seems like a common theme? And then of course close to my heart is v0 But what do you see about all of these? What do you see about all of these that? Seems like a common theme John's mic (12:11.628) I feel like this is a setup, but I'm going to say the chat interface for 100. And actually here, this is Notion. So I'm logged into Notion, and this is Notion AI actually. So this is a logged in state. So. Thanks. And actually here, this is Notion. So I'm logged into Notion. This is Notion AI, actually. So this is a logged in state. Studio (12:32.04) I have issues with the seeming ubiquity of the AI chat interface because I think there are several major problems with it. And I'm just going to run through my list of problems and then we can either talk about it. You can confirm or deny that you think it's a problem and then let me know if I'm missing anything. I have issues with the seeming ubiquity of the AI chat interface. I think there are several major problems with it. And I'm just going to run through my list of problems, and then we can either talk about it. You can confirm or deny that you think it's a problem. And then. Let me know if I'm missing anything. I do have an interesting defense of the chat interface. I don't think it's strong, but it's worth noting. Okay. Wait, do you want to start with Yeah, I do kind of. Okay. Okay. Wait, you don't want me to try and tear it down and then you defend it? No, I think it'll be more fun this way. Okay. Yeah, this is short because it is one point with about four or five points on the other side. But the one point is imagine you created an app and say it was around 2020. Okay. Okay, wait, do you want to start with that? Okay. Wait, you don't want me to try and tear it down and then you defend it? Great, okay, go for it. John's mic (13:28.984) Your app grew faster than any other app has ever grown. You have 800 million users after a couple of users. It's a certain interface. And a bunch of people copy the interface like, sure, that checks out. Makes sense. So that's my defense of, well, there are 800 million people using this interface. Maybe we shouldn't try to reinvent the wheel. Sounds familiar. Studio (13:41.694) out yeah yeah yeah Using this. maybe we shouldn't try to read them with the whip. Totally. Well, we'll talk about this. Okay. Great. We'll talk about this. In full disclosure, I thought a huge amount about this topic over the summer and wrote a blog post about this. Yes. A pretty long blog post, like 4,000 words or something like that. So I'm just going to shamelessly steal all of my thoughts, my previous thoughts from the blog post. We'll talk about this. In full disclosure, I thought a huge amount about this topic over the summer and wrote a blog post about this. Yes. A pretty long blog post, like 4,000 words or something like that. And so I'm just going to shamelessly steal all of my thoughts from my previous thoughts from the blog post. I did not write a blog post about this. I did not have as many thoughts about it. OK. Okay, I'm going to, we'll put this in the show notes. I'm actually gonna start by reading a couple of quotes because my strong feelings about this are actually not, I'm not alone in this. I guess I actually could. I guess I actually could. Here, I can share, let me pull this up. I'll just share this so people can read along if you're on video. John's mic (14:22.326) to look with us in the show notes. I'm actually going to start by reading a couple of quotes because my strong feelings about this are actually not, I'm not alone in this. Okay, so you're not going to quote yourself from your blog post. I guess I actually could. I guess I actually could. Here, can share. Let me pull this up. I'll just share this so people can read it. and while you do that, Anything.com is available for $2 million. Maybe Loeb should consider a branch. Okay, this first quote is from Naveen Rao, VP of AI at Databricks. There's some nice colorful language in here that I'll skip just in case you're listening in the car with children like me. consider a brand? Spend two-thirds of the money on anything. Okay, this first quote is from Naveen Rao, VP of AI at Databricks. There's some nice colorful just in case you're listening in the car with children like me. So he said he was talking about, we were actually at this conference and we saw this talk. Right, we did, yeah were both there. Which was really cool. And so he said there's all these hard problems at the sort of leading edge of the challenge of the next big fundamental breakthrough in AI technology as sort of one major area. And the other major area is UI innovation. Studio (15:13.198) So he said, he was talking about, we were actually at this conference and we saw this talk live. Yep, which was really cool. And so he said, there's all these hard problems at the sort of bleeding edge of, you know, the, the challenge of the next big fundamental breakthrough in AI technology as sort of one major area. And the other major area is UI innovation. said, a lot of the stuff people are delivering as a chat bot, which we just saw. So a lot of the stuff people are delivering is a chatbot, we just saw. It's the worst interface I've ever seen for most applications. I'm so tired of seeing chatbots. Please fix this. Give people insights, intelligence to the right user at the right time. This is why IDEs and coding interfaces have taken off, because they deliver value where I need it. They don't make me cut and paste something into a chatbot. It's the worst interface I've ever seen for most applications. I'm so tired of seeing chat bots. Please fix this. Give people insights, intelligence to the right user at the right time. This is why IDEs and coding interfaces have taken off because they deliver value where I need it. They don't make me cut and paste something into a chat bot. That's actually a side note, very interesting based on our discussion about Zoho and file systems, right? Because it's just grepping. You don't have to like cut and paste and dump a bunch of stuff into the context window. It's actually signed up very interesting based on our discussion about Zoho and the health systems, right? Because it's just grepping. You don't have to cut and paste and dump a bunch of stuff into the context window. I love this. This is so great. He clearly has strong feelings about this, too. This is the stupidest interface I've ever seen. Here's another interesting one. This is Benedict Evans, who... I like, this is so great. He just, clearly has strong feelings about this too. This is the stupidest interface I've ever seen. Okay, that's one. Here's another interesting one. This is Benedict Evans, you know, who I think he was at, he was at a venture firm for a while, but has been a tech analyst for a very long time. And I can't remember, I want to say it was Excel or one of the big name brands. John's mic (16:29.646) I think he was at a venture firm for a while, but has been a tech analyst for a very long time. I can't remember, I want to say it was Excel or one of the big name brands. Has written for a very long time, I mean really just a wonderful thinker and writer. And he wrote an interesting post about the adoption metrics for these AI tools. Interfirm's, has written for a very long time. I mean, really just a wonderful thinker and writer. And he wrote an interesting post about the adoption metrics for these AI tools. And one of the things that he points out, I'll just read here, it might also be that the chatbot as chatbot is the right UX. And one of the things that he points out, I'll just read here, it might also be that the Chatbot as Chatbot is the right UX only for some people and some use cases. And most people will experience this technology as features and capabilities wrapped inside other things. I don't think we can know that. only for some people and some use cases and most people will experience this technology as features and capabilities wrapped inside other things. I don't think we can know that. Hence the real question as I've hinted a couple times is how much LLMs will be used mostly as actual user facing general purpose chat bots at all or whether they will mostly be embedded inside other things in which case trying to measure their use at all will be like trying to measure machine learning or SQL. Hence, the real question, as I've hinted a couple times, is how much LLMs will be used mostly as actual user-facing general purpose chatbots at all, or whether they will mostly be embedded inside other things, in which case trying to measure their use at all will be like trying to measure machine learning or SQL. How many times a day do you use a database? Who cares? And this is in the context of a post where he talks about adoption metrics. So you mentioned there's 800 million Studio (17:34.968) how many times a day do you use a database, who cares, right? And this is in the context of a post where he talks about adoption metrics. So you mentioned, okay, there's 800 million people using GPT, chat GPT, but he makes the point that in the numbers, it's hard to get at the actual underlying data, but. people using GPT, chat GPT, but he makes the point that the numbers, they're still, it's hard to at the actual underlying data, but apparently the daily active numbers are very low. A huge percentage of users use it maybe a couple times a week. So here's a thought. Think about, Apparently, the daily active numbers are very low. A huge percentage of users use it maybe a couple times a week. John's mic (18:17.28) I think the mobile phone is a great, I don't know, analogy or whatever that is called to this where mobile phone comes out and you can only call then call and text essentially. And obviously when the iPhone comes out, it enables a form factor that works with all these other things. So it enables a form factor that works with all these other things. Yep. So, and it seemed dumb. And I even remember and it seemed dumb. And I even remember like, but I like a physical keyboard. You know what I mean? A lot of people were like, but I like a physical keyboard. And if, and if you were just only texting, the physical keyboard was still better, especially than the original, you know, keyboards. So I wonder if there's an analogy here where like, if what you're using it for is just quote, You know what mean? A lot of people are like, but I like the physical keyboard. The palm. And if you were just only texting, the physical keyboard was still better. Especially than the original kind of keyboard. So I wonder if there's an analogy here where if what you're using it for is just, quote, texting. I mean, if you use ChatGBT, you can talk to it and you can text it. mean, it creates images, so you need a visual piece for that. But what's the big shift that makes it so it's just so obvious that this is a limiting form factor? I that's my question. John's mic (19:15.448) Texting and I mean those like if you use chat GBT like you can talk to it and you can text it The I mean it creates images so you need a visual piece for that but but what's the like big shift? That makes it so It's just so obvious that like this is a limiting form factor. I think that's my question Okay. Studio (19:41.304) to say that my own words, like why is it problematic? Like why is the form factor of the chat interface problematic? Okay, we'll get there. I wanna do one more quote. to say that in my own words, like why. John's mic (19:48.222) Yeah, what is it limit? What is it? Yeah, sure. Studio (19:56.236) This is really interesting. Jacob Nielsen's a, he's a UX expert, has been writing for years, I mean decades about UX. I thought this was really interesting. I doubt the current set of generative AI tools like ChatGBT, BARD, remember that name? The original name for Google's. Yeah. Right. Things are moving quickly. This post isn't that old. I doubt these are representative. This is really interesting. Jacob Nielsen, he's a UX expert, has been writing for decades about UX. I thought this was really interesting. I doubt the current set of generative AI tools like ChatGBT, BARD, remember that name? The original name for Google's. that's right. I'd forgotten. Right? Things are moving quickly. Yeah. This post isn't that I doubt these are representative of the UIs we'll be using in a few years because they have deep-rooted usability problems. Their problems led to the development of a of the UIs we'll be using in a few years because they have deep-rooted usability problems. Their problems led to the development of a new role, the prompt engineer. This new role reminds me of how we used to need specially trained query specialists to search through extensive databases of medical research or legal cases. Then Google came along and anybody could search the same level of usability leapfrogging is needed with these new tools. Better usability of AI should be a significant competitive advantage. This new role reminds me of how we used to need specially trained query specialists to search through extensive databases of medical research or legal cases. Then Google came along and anybody could search the same level of usability leapfrogging is needed with these new rules. Better usability of AI should be a significant competitive advantage. okay, so we can, here I'll stop, I'll stop screen sharing here. Studio (20:57.742) Okay, so we can, here I'll stop, I'll stop screen sharing there. I thought that was really interesting, right? The signal of like a prompt engineer, prompt engineering as a new skill set for how to wield these tools is a really interesting symptom of a potential UX problem. Okay. I thought that was really interesting, right? Yeah. The signal of like a prompt engineer, prompt engineering as a new skill set for how to wield these tools is a really interesting symptom of a potential UX problem. Okay. Studio (21:28.812) Back to your question. So restate the question. Yeah, restate the question because I have some answers for this. Back to your question. So restate the question. Okay, restate the question. Yeah, yeah. So, so their original analogy was early mobile phone, text and call. Then iPhone comes out and people are like, excited. This is cool. Like it has a web browser now, but like I had a web browser on my other phone and it wasn't that good. And I want a physical keyboard. I'm John's mic (21:56.598) at least for a while and then eventually like people were like, and then, and they kind of understood the new form factor. Then you had an app store and like the whole thing. So I guess my question is current state. Let's just use chat GPT. For example, you've got voice and people do talk to it. I think that's a fairly common interaction. The most interaction common is, is text back and forth. We have a little bit of text input with like an image output. That's a thing. current state, let's just use chatgpt for example. and people do talk to Studio (22:19.33) Yep. And then, yeah, so essentially you change the form factor. what do you think we're limiting ourselves with the current form factor? Is it a limiting thing? Like with the iPhone, it's like, now that I can collapse the keyword, like I can watch movies and it's a great experience, you know? So there's a, are the, what's the limiters like current state? And then the other component, I think to the question is. Yeah, so essentially, you change the form factor. What do think we're limiting ourselves with? Is it a limiting thing like with the iPhone like, now that It's a great experience. So there's a limiters like current state. John's mic (22:55.246) I guess I've heard a lot of people say this, like, cool, so what do you recommend? You know, there's that, there's that like, yeah, it's fun to be edgy and complain about the chat interface, but like, what do you think we should do? And often people- people don't have an answer and they just love to be, you know, contrarian. And then it's like, cool, what are you recommending? And then you hear crickets. I guess I've heard a lot of people say this, like, cool, so what do you recommend? know, there's that, yeah, it's fun to be edgy and complain about the chat interface, but what do you think we should do? You're gonna make me actually have a chat. But people don't have an answer. They just love to be, you know, contrarian. Sure. And then it's like, cool. Totally. Yes, I can answer that. have some. So let's talk about the limiting the limitations. I think that there are three main limitations and we can dig as deep as you want into any of these. The first is a writing barrier. So. I can answer that. have some... I know. I know you do. That's why I'm bringing it up. John's mic (23:33.281) Yes. John's mic (23:37.678) Green England. John's mic (23:43.234) The first is a writing barrier. So... yes. Okay. This is something I've thought a lot about because, mean especially because I took a job as a writer. I think about writing a lot. This is something I've thought a lot about because, I mean especially because I took a job as a writer, so I think about writing a lot. It's actually hard to find data about writing skill generally. The closest thing that I found was a... proficiency, right? Yeah, so how good are people at writing? And because the, I think, clear articulation... It's actually hard to find data about writing skill generally. The closest thing that I found was a... Like writing proficiency. proficiency. humans. So how good are people at writing? And because the, I think, clear articulation... people who I know who are good writers over index for the value that they get out of using the chat interface. John's mic (24:26.2) people who I know who are good writers over index for the value that they get out of using the chat interface. interesting. OK. Something that I've noticed. Sure. Yep. And I mean, you can call that prompt engineering. You can call it whatever you want. But I actually think that just being able to very, very clearly articulate what your is something that I've noticed, right? And I mean, you can call that prompt engineering, you can call it whatever you want, but I actually think that just being able to like very, very clearly articulate what your intention is, what you're trying to do, instructions, all that sort of stuff, right? intention is what you're trying to do instructions all that stuff right yeah it's a superpower that just got that already was a superpower that just got like massively magnified exactly yeah exactly right and i think about it in terms of Exactly, exactly. And I think about it in terms of... You know, we talked about prototyping actually in episode one with vZero, like in the context of product, right? So you can, you know, prototype something that's really high fidelity and that's minimizes, you know, it can minimize the need to be overly verbose in a product requirements document or something like that, right? But you still have to use words to build the prototype, you know, with AI. John's mic (25:08.3) You know, we talked about prototyping actually in episode one with v0 like in the context of the product, right? So you can, you know, prototype something that's really high fidelity and that minimizes, you know, it can minimize the need to be overly verbose in a product requirements document or something like that, right? But you still have to use words to build the prototype, you know, with AI. So the best proxy that I found for writing skill was essentially a composite of reading and writing, which is literacy. So reading, comprehension, and writing. Yeah, yeah, yeah, yeah. And so literacy, so this is a very widely cited study. And I think they do this worldwide, but I just looked at the numbers for the United States. So the best proxy that I found for writing skill was essentially a composite of reading and writing, is literacy. Okay. So reading, comprehension. So you're doing research. You're trying to find numbers. Yeah, So this is a very widely cited study. I think they do this worldwide, but I just looked at the numbers from the United States. Okay, they rank literacy on a five level scale, so five being the most proficient and one being the least proficient. Okay, they rank literacy on a five level scale, so five being the most proficient and one being the least proficient. Does that correspond? Because I feel like I've seen the scores with like high school level reading, collegiate level reading. You know, they kind of did it by grade standard. Do you think that corresponds? I don't know. Because they rate people on that standard and then they rate like books and Studio (26:21.794) seen the scores with like high school level Studio (26:27.502) they kind of did it that standard? That's a good question. don't know. I don't know. Because they people on that standard. and stuff on that. really? Yeah, yeah. third grade level or whatever. yeah. I'm not sure what that was. I'm right. Yeah, yeah. This is like a middle... Yeah, exactly. I'm not sure what that was. I'm sure we could look it up. One to five. And over 50 % of Americans rank level two or lower. So over half of Americans are below average in terms of literacy proficiency. So one to five. The, and over 50 % of Americans rank level two or lower. Okay. So over half of Americans are not, are below average. Okay. In terms of literacy. John's mic (27:06.04) proficiency, which is reading comprehension and writing. Which, I mean, the reading comprehension thing actually is a component here as well, because generally what is returned is text. I mean, there's an image component as well, but not usually. So I think that's problem number one, which is the writing skill generally, I think, which is reading comprehension and writing. Which, I mean, the reading comprehension thing actually is a component here as well because generally what is returned is text. I mean, there's an image component as well, obviously, but a huge amount of text. So I think that's problem number one, which is the writing skill generally, I think, means you get more value out of the LLM. And so you sort of limit, there's a fundamental limitation on how much value people are gonna get out of this because of a fundamental deficiency in writing skill. Now, this is not a new problem in that Google had to figure out how to take fragments of phrases and translate them into user intent to return. means you get more value out of the LLM. And so you sort of limit, there's a fundamental limitation on how much value people are gonna get out of this because of a fundamental deficiency in writing skill. Now, this is not a new problem in that Google had to figure out how to take fragments of phrases and translate them into user intent to return high. highly relevant search results. And the LLMs are doing the same thing. They're performing web searches based on what they interpret as user intent, et cetera. Definitely better. Definitely better. And this leads into the second problem. I think the challenge is with Google Search, it's returning a bunch of links that are really relevant. John's mic (28:04.778) search results, right? And the LLMs are doing the same thing, right? They're performing web searches based on what they interpret as user intent, et cetera. And arguably better. Definitely better. Definitely better. But I think what is, and this leads into the second problem, I think the challenge is with Google search, it's returning a bunch of links that are really relevant. Yes. The actual. Studio (28:42.7) Yeah, the actual task that the user is trying to perform is very specific. I'm going to a search engine and I am looking for something and I want a bunch of, I'm gonna get, the search engine's gonna return a bunch of links. Yeah, the actual task that the user is trying to perform is very specific. I'm going to a search engine, and I am looking for something, and I want a bunch of, I'm going to get, the search engine's going to return a bunch of links. And then the transaction's over, and they're browsing another site, and Google's done at that With an LLM, it's... Exactly, right. With an LLM, it's a blank page, right? And so, in fact, being more articulate is more helpful because you need to understand and articulate your own intent of what you're trying to do, right? And so it is interesting. it's a blank page. And so in fact, being more articulate is more helpful because you need to understand and articulate your own intent of what you're trying to do. And so it is interesting. Studio (29:29.506) on an animal. Studio (00:03.682) So. So I think that's what's interesting is that in Google search, it's a known problem in Google search. I think that's what's interesting is that in Google search, it's a known problem. In Google search, the intent was very clear and the terms of the transaction were very clear. This is completely open-ended, which I would argue actually means that writing skill is even more important, like I said before. The other thing that's interesting, you brought up the idea of texting with a keyboard and then that goes away. Again, the terms of that transaction. the intent was very clear and the terms of the transaction were very clear. This is completely open-ended, which I would argue actually means that writing skill is even more important, like I said before. And the other thing that's interesting, you brought up the idea of texting with a keyboard and then that goes away. But again, the terms of that transaction were very clear. were very clear. And this is very open-ended, right? I mean, the name of the company that we just looked at is Create Anything, right? And so, okay, so we have that sort of input problem, which, you know, the prompt engineering thing, the major providers are certainly, like, using LLMs to understand user intent and optimize. John's mic (00:41.921) this is very open-ended, right? I mean, the name of the company that we just looked at is Create Anything, right? And so, okay, so we have that sort of input problem, which, you know, the prompt engineering thing, the major providers are certainly, like, using LLMs to understand user intent and optimize. Sure. Optimize against that limitation. I still think it's a fundamental limitation. The main interface for this is you articulating your intent in a completely, essentially open-ended environment. Right. And so if writing skill is low, then there's, I think, a fundamental value ceiling. So that's one. Two is the blank page problem, which I think, this is really interesting to me. optimize against that limitation, but I still think it's a fundamental limitation, right? Like the main interface for this is you articulating your intent in a completely, essentially open-ended environment. And so if writing skill is low, then there's, I think, a fundamental value ceiling. So that's one. Two is the blank page problem, which I think, this is really interesting to me because because it reminds me of the Amazon Alexa challenge. Right. Which we've talked about. have. So you could essentially run your entire home as a smart home. It reminds me of the Amazon Alexa challenge, which we've talked about, right? So you could essentially run your entire home as a smart home by speaking to Alexa or Google Home or whatever, right? These smart home devices, Exactly. John's mic (01:50.052) by speaking to Alexa or Google or whatever, Smart appliances and light bulbs and TV and everything. And the research shows that overwhelmingly people just use those things for very, very basic things, for playing music, creating a grocery list, checking the weather, checking sports scores, or doing performing search. I have something that I to know. Hey, Alexa, you know. And the research shows that overwhelmingly people just use those things for very, very basic things, for playing music, creating a grocery list, checking the weather, checking sports scores, or doing performing search. I have something that I want to know. Hey, Alexa, what exact year did the Berlin Wall come down? what exact year did the Berlin Wall come down? Which I think I've done research on this before I read an article. And I believe the top things. You want to guess what the top two are? weather in search? The top two, I believe, are set a timer. Studio (02:39.698) Mm-hmm. and maybe weather. I think those are the top two. Just hilarious. anyways, yeah. That's the blank page. That, I think, is actually a really interesting limitation. It's very surprising to me that we have seemed to struggle to move beyond that because of what Okay. Just hilarious. anyways, yeah. That's the blank page. That I think is actually a really interesting limitation. It's very surprising to me that we have seemed to struggle to move beyond that because of the breadth of what these LLMs can actually do. It's amazing. because of the breadth of what these LLMs can actually do, it's amazing. You can create anything, right? That is super interesting. But people really struggle with the blank page problem, right? If you tell me I can create anything, what do I actually do? And there are certainly little suggestions and tiles that say, create a website or... you can create anything, right? That is super interesting. But people really struggle with the blank page problem, right? If you tell me I can create anything, like what do I actually do? And my, and I mean there are certainly little suggestions and tiles that say like, you know, create a website or, you know, optimize your calendar or other things. And so there's certainly UX hints that are, you know, being added to this and. John's mic (03:32.824) you know, optimize your calendar or other things. And so there's certainly UX hints that are, you know, being added to this and, know, I'm sure, I'm sure we'll get better. So funny timing for this conversation because the chat GPT store like just launched. yes. Do you think that, I mean, I think that helps, but how much do you think it helps? Cause I think it might help with both of your, your two, you know, I'm sure it will get better over time. Studio (03:50.387) yes. John's mic (04:02.775) problems. I think it'll go a long way. But I also think that the, I think it'll go a long way because it is a way for users to more clearly articulate their intent by finding something that is related to a specific task where, you know, per Google search, per texting on a physical keyboard, the terms of the transaction are much clearer, right? So I think that's actually just a way to more efficiently is it is a way for users to more clearly articulate their intent by finding something that is related to a specific task where per Google search, per texting on a physical keyboard, the terms of the transaction are much clearer, right? So I think that's actually just a way to more efficiently uncover user intent. That's a way to more efficiently uncover user intent. Right, without a page. Yeah, because they clicked on the Spotify app and then used AI to create a playlist or whatever. Exactly. So that's interesting. And then the last problem, is I think more specific probably to knowledge work, that to leverage an LLM, Studio (04:43.735) Exactly. Right, exactly, exactly. So that's interesting. And then the last problem, which is I think more specific probably to... you know, more specific to knowledge work is that they're, you know, to leverage an LLM, leveraging an LLM to sort of, let's say it's full value often requires some pre-existing knowledge or skill in the subject that you are, that you're engaging the LLM for. an example would be, well, two examples, right? One that's very obvious is writing software, right? So, Leveraging an LLM to sort of, let's say, its full value often requires some pre-existing knowledge or skill in the subject that you are, that you're engaging the LLM for. an example would be, well, two examples, right? One that's very obvious is writing software, right? So more and more agents are writing software, but I'm very upfront about the fact that I don't have formal training in software engineering. more and more agents are writing software, but I'm very upfront about the fact that I don't have formal training in software engineering. And so I have built multiple pieces of software, know, fully end-to-end software with, you know, a database and, know, sort of a fully functioning front and back and all that sort of stuff. But you don't want me as, you know, an engineering lead putting something into production for millions of users because I don't actually know how to fix it if something goes wrong, you know? John's mic (05:36.9) And so I have built multiple pieces of software, fully end-to-end software with a database and sort of a fully functioning front and back and all that sort of stuff. But you don't want me as an engineering lead putting something into production for millions of users because I don't actually know how to fix it if something goes wrong. I would say writing actually is similar. You can... I would say writing actually is similar. You can articulate what you want to communicate with GBT and it can generate a really nice email for you, but it won't give you the core skill set of a writer. And you will help litter the world with dashes. you can articulate what you want to communicate with GBT and it can generate a really nice email for you, but it won't give you the core skill set of a writer. John's mic (06:24.031) That's so true and and when we're reading on episode one the inner ring speech has a lot of M dashes in it from 1940 something and I was looking at that and clearly has nothing to do with LLMs, but like forever more somebody's just gonna assume So I do think It does. Totally. It's really sad. I am sad about that actually. Studio (06:43.519) I know. I know. Isn't that sad? So I do think the chat interface is really good for certain things. Search being one of them, which is why I think so many people are using it for search because it's familiar. It actually is a, it's a phenomenal step forward in search interface, right? I mean, there's some challenges to overcome, right? Where the chat interface is really good for certain things, search being one of them. Which is why I think so many people are using it for search, because it's familiar. It actually is a phenomenal step forward in search interface. I mean, there some challenges to overcome, which I think is happening, where you're sort of getting receipts and links out to the information that the LLM is. which I think is happening, you know, where you're sort of getting receipts and links out to, the, information that the LLM is, is aggregating for you as opposed to just it returning some information and it's non-deterministic and you don't know, you know, so I think that is getting better, exploring a topic it's way better, you know, it's way more useful than Google search for a lot of things. And so it's a great form factor for that. But my problem with it is that we're applying that form factor to lots of. is aggregating for you as opposed to just returning some information that's non-deterministic and you don't know. So I think that is getting better, but exploring a topic, it's way better. For sure. It's way more useful than Google search for a lot of things. And so it's a great form factor for that. But my problem with it is that we're applying that form factor to lots of other things. Yes. That makes sense. OK. you know, other, lots of other things. Okay. I actually think, so the other thing I wanted to do was starting there. So it's a good form factor for search, right? For performing search. And then I think actually as things become more high context and the terms of the transaction become more clear, the chat interface becomes more useful. Okay. John's mic (07:49.206) I actually think, so the other thing I wanted to do was starting there, so it's a good form factor for search, right? For performance search. And then I think actually as things become more high context and the terms of the transaction become more clear, the chat interface becomes more useful. I'm gonna share my screen again and just show a couple examples of, you So I'm going share my screen again and just show a couple examples of what I mean by that. Yep. And I did look up the top Alexa use cases. I missed one. Music playback. So that's the top one that looks like then timers and alarms. then like queries, like news queries or. Yep. Yep. really, okay. John's mic (08:36.395) So interestingly, if you click on Notion AI, this is just a free Notion account. So interestingly, if you click on Notion AI, this is just a free Notion account. But I think a lot of them, they give you some usage. so here, I'm going to say this is mid-tier. So we have completely open-ended. some. know, some usage. Okay, so here, I'm gonna say this is like mid-tier. So we have like completely open-ended. Studio (09:03.619) You have completely open-ended GPT chat, which I think is used very heavily for search and other form factors where it makes sense and sort of overcomes the blank screen problem and the writing problem. Then you have, would say, this is in context. And so this is just a very simple example. I was trying to pull together a bike tour for a trip to Amsterdam, you know, to save some money. Yeah, exactly. So putting together my own bike tour, which I used GPT very heavily to do all the search and planning, which was super cool. You have completely open-ended GPT chat, I think is used very heavily for search and other form factors where it makes sense and sort of overcomes the blank screen problem and the writing problem. Then you have, this is in context, and so this is just a very simple example. I was trying to pull together a bike tour for a trip to Amsterdam to save some money. Yeah, bike one does. Yeah, exactly. So putting together my own bike tour, which I used GPT very heavily to do all the search and planning and everything. And it worked out pretty well? It worked out great. Nice. And so this is actually interesting, right? So here, you are within a document that has a bunch of really specific information. And the chat interface, think, is very high context here, right? So maybe we could say, turn this into a bullet. It worked out great. And so this is actually interesting, right? So here you are within a document that has a bunch of really specific information, and the chat interface I think is very high context here, right? This is, you know, so maybe we could say like turn this into a bullet point. itinerary. Studio (10:02.179) with a timeline or something like that. Okay, this is super high context, right? And so I think it gets better the deeper you go into high context, which is the quote from Naveen Rao said that about LLMs as well. something like that. Okay, this is super high context, right? And so I think it gets better the deeper you go into high context, which is the quote from Naveen Rao, you know, said that about Alalens. So you think the chat interface gets better the deeper you go into high context. high context and you overcome the blank page problem and it's much easier to articulate what your intent is. The other thing actually that's really interesting, that's cool. That is cool. Right, because you're super high context and you overcome the blank page problem and it's much easier to articulate what your intent is. The other thing actually is really interesting. That's cool. Okay, so see it's, you know. This is a common problem I have with the interface. I wanted what was there before, all of the, if you want to call this like a middle context, or maybe just high context, all of the notion-esque, I've used several tools that are like this, they seem to be all pretty aggressive about... I know. Studio (10:52.033) Yep. Replacing what's there. Right, right. I know. Exactly. But actually I think that that sort of reinforces my point about the terms of the transaction not being clear. It's very unclear to me. This is actually, this is very cool, but it overwrites what I did and I don't know that that's going to happen. Right? There are just some cues there that are, you know, challenging. Like replacing what I already have. And sometimes it's what I want, sometimes it's not what I want. Exactly. But actually I think that that sort of reinforces my point about the terms of the transaction not being clear. It's very unclear to me. This is actually, this is very cool, but it's overrides what I did and I don't know that that's going to happen. Right? Yeah. Right. Yeah. mean, yeah, for sure. And it's not just most of the ones here. It's unclear. Even from my intent, I'm not even blaming the LLM. It's unclear from my intent. Are we trying to turn this document into this new form factor, or do want me to create a new form factor from this document? Exactly. So when I say mid-tier, this is just an example because I don't really have anything in this personal notion other than me doing this document. But when I say mid-tier, what's interesting is a lot of companies use notion for, you know, Even from a rain. Studio (11:40.832) Exactly. So, and when I say mid-tier, this is just an example because it's, I don't really have anything in this personal Notion other than me dumping this, you know, me doing this document. But when I say mid-tier, what's interesting is a lot of companies use Notion for, you know, running their business, right? So content calendar, knowledge base, internal documentation, all that sort of stuff, And so that's very high context for the business. And if Notion's AI tool can sort of, you know, running their business. knowledge base and internal documentation, all that sort of stuff. And so that's very high context for the business. And if Notion's AI tool can sort of cross all those different pieces of knowledge within the same closed system. Glean is another example of this within a business context where they sort of aggregate, have Slack and Google Drive and Notion that sort of solves the fragmentation problem. know, cross all those different pieces of knowledge within the same closed system, Glean is another example of this, you know, within a business context where they sort of aggregate, you know, Slack and Google Drive and Notion, it sort of solves the fragmentation problem, right? So that's mid-tier. And then I would say, you know, the most high context would be something like an IDE, you know, like cursor. And here the intent is, Right. So that's mid-tier. And then I would say the most high context would be something like an IDE, like Cursor. here the intent is very, very specific. I'm within a specific code base, and I have very specific intent about what I want. I'm optimizing it, I'm building a feature, et cetera. And so I think the utility is greater the higher context that you get. Which again, I'm just reiterating something that is Yeah. But if you tracked. very, very specific. I'm within a specific code base and I have very specific intent about what I want to do with that. I'm optimizing it, I'm building a feature, et cetera, right? And so I think the utility is greater the higher context that you get, which again, I'm just reiterating something that Navin Rao said. John's mic (12:56.675) spend, think token spent in 2025 generating code would be really, high. Super high. Super high. For sure. Because the form factor makes a lot of sense. Now, I think the other interesting thing that we're seeing now is a couple things. So we're seeing things go more multimodal. tokens spent in 2025 generating code will be really, really high. Super high, super high, right? Because the form factor makes a lot of sense. Now I think the other interesting thing that we're seeing now is a couple things. So we're seeing things go more multimodal. Right. So like v0 is a great example of this, right? They have a design mode, right? Right. V0 is a great example of this, right? They have a design mode, right? And you can actually click around and change text and other things like that so that you're actually interacting with the website itself. And then Cursor actually rolled this out as well. So this is just a little project. Cursor has a browser. So you can see here I'm running this. and you can actually click around and change text and other things like that, so that you're actually interacting with the website itself. And then Cursor actually rolled this out as well. So this is just a little project. Cursor has a browser, so you can see here I'm running this app. This is the browser inside of Cursor. you can see that I can actually So this is the browser inside of cursor and You can see that I can actually you know interact with So this is non AI So this is this is like you could pick the word feedback there and change the color by clicking on it and picking a new color for example, right? Studio (13:57.699) interact with the different components here. this has become- Studio (14:12.117) Yes, yes, right? you know, here's the color spectrum, right? Yep. Which, which I bet they used AI for the like selection, reading all the objects in selection. But then you have a non AI part where you are clicking and physically changing the code, which is a really interesting, I think a great example of non chat interface that, but, it's still AI powered, but, but invisible. And I'm just guessing, I assume that's how they did it. like reading all the objects in selection. then you have a non-AI. Studio (14:32.867) That's a really interesting, I think a great example. Studio (14:38.467) but it's still AI powered. Totally. Yeah, yeah, yeah. Yeah, it's interesting. So I think the biggest opportunity is with the general, like the very general blank page, you can sort of do anything. Because I think that's where the most challenges are. Right? And I think it's really cool to see tools like cursor and V zero go multimodal within a very specific context to overcome some of the limitations of just having, you know, chat. so So I think the biggest opportunity is with the very general blank page, you can sort of do anything. Because I think that's where the most challenges are. And I think it's really cool to see tools like Cursor and V0 go multimodal within a very specific context to overcome some of the limitations of just having chat. So here's my question for you. Here's my question for you. Why, if there are these very clear problems, and we're not the only people who think that these are problems, why is it so ubiquitous? I think you hit on one thing, which is this is a form factor that actually, I mean, it's a large language model. And so it actually was really brilliant, I think, of OpenAI to use this interface. Also, well, we can get there, but I think it's If there are these very clear problems, and we're not the only people who think that these are problems, why is it so ubiquitous? I think you hit on one thing, which is this is a form factor that actually, I mean, it's a large language model. And so it actually was really brilliant, I think, of OpenAI to use this interface. John's mic (15:41.518) Well, we get there. But I think it's brilliant. You're interacting with a large language model, so you use a language to interact. It sort of makes common sense. That took off. There's 800 million people who have used this. And so why would someone try to do something dramatically different? Yeah. Yeah. That, for sure. Brilliant, you're interacting with a large language model, so you use a language to interact with it. It sort of makes common sense. That took off. There's 800 million people who have used this. And so why would someone try to do something dramatically different? John's mic (16:05.955) And then of course there's talk of these dramatically different form factors, which we talked here. I can show this. Remember Rabbit? We'll pull that up real quick here. Rabbit R1. Let me show this to everybody. So this came up a while ago. then of course there's... Studio (16:11.789) Yep. Studio (16:18.169) yeah. Yeah. Studio (16:30.093) This came up a while ago. John's mic (16:36.503) So this is Rabbit R1, and it's a physical device. Scroll down so you can kind of get an idea of the form factor. It's got a scroll wheel. It's such a nice package. Yeah. A physical button. I'm going to reach out to them and see if we can use the same supply chain for Lobe. Yes, perfect. But yeah, it's a really. So this is RabbitR1, and it's a physical device. Scroll down so you can kind of get an idea of the form. It looks super. It's such a nice package. I'm going to ask, I'm going to reach out to them and see if we can use the the same supply chain for Loeb. Yeah, it's perfect. But yeah, it's a really, and I think they're going down the app, where they're going to have apps. Yep. And obviously there's a voice. And I think they're going down kind of the app route where they're going to have apps. you know, obviously there's a voice component here. They've got their magic camera, which to be honest, I don't fully understand what that does. But yeah, so you kind of get an idea for the form factor. You can talk to it. You can interact with it. It's got some, some basic apps that it ships with. think there's going to be, or is an app store. But yeah, so you kind of get an idea for the form factor. You can talk to it, can interact with it. It's got some basic apps that it ships with. I think there's going to be, or is an app store. John's mic (17:35.344) and and Studio (17:40.515) Yeah, I don't know. at launch, yeah, so this is a rabbit last two. At launch, they launched really early. Which there's kind of a first mover thing there that sometimes is an advantage. Yep. They got really pretty roasted. Hmm. A few people especially. Yeah, I don't know. at launch, yeah, so this is a rabbit to us too. At launch, they launched really early, which there's kind of a first mover thing there that sometimes is an advantage. They got really pretty roasted by a few people, especially, I think it was MKBHD, who's a really big YouTuber, kind of like the product guy. is a really big YouTuber, kind of like the product guy, reviews every iPhone and... reviews every iPhone and Android phone that comes out and is kind of the abroad guy. And kind of got roasted for like, hey, you released this too early, it's not good. Type of thing. But then they've kept going and they're on version two and yeah, this would be a cool redemption story. Hope they can get it in his hands again. But this is one thing I thought of, somebody pushing hard to break the chat only interface. But it's perfect. Totally. Right. Yeah, exactly. John's mic (18:39.883) But it's the exact thing that, you know, they released early, they pushed hard, kind of got roasted on the original version and everybody else saw. What do you think? Everybody's like, yeah, we'll just stick with chat. So there's just a high risk thing, I think. It is high risk and there's also, there are also major, a couple of really major advantages. So one is that, and this gets to, I do want to actually try to answer your question of like, if not chat. version and everybody else saw, everybody was like, yeah, we'll just take a chat. Yeah, totally, totally. So there's just a high risk thing, think. It is high risk and there's also, there are also major, a couple of really major advantages. So one is that, and this gets to, I do want to actually try to answer your question of like, if not chat, know, then what? Lobe. If not chat, then lobe. So. Right, what? Lobe. mean, clearly. So, what are the advantages? I mean, this is brilliant, right? So, number one, it's really expensive to run these AI tools at scale. Yes. And so, when you start making decisions about running jobs or performing tasks on behalf of your user without... What are the advantages? I mean, this is brilliant, right? So number one, it's really expensive to run these AI tools at scale. And so when you start making decisions about running jobs or performing tasks on behalf of your user when you're assuming their intent, the bill gets really high. And so we know this from by when you're assuming their intent, the bill gets really high. And so we know this from building products, right? There's things that you don't know until you get something in a user's hands. It's like, OK, you have all these assumptions. Even if you do research, you actually get something in the user's hands. And that's when you really learn how they're going to use it. And so the cost of that experimentation cycle is extremely. Studio (19:46.456) building products, right? They're just things that you don't know until you get something in a user's hands. You know, it's like, okay, you have all these assumptions, even if you do research, you actually get something in a user's hands, and that's when you really learn how they're going to use it. And so the cost of that experimentation cycle is extremely high if you start doing things on behalf of your user and assuming, right? if you start doing things on behalf of your user and assuming, right? Well, and we're being told, which I think is accurate, that the bottleneck, if you listen to episode one, global bottleneck for all of the AI advancement is the data centers. Yes. So what are you going to spend that precious resource on? Making the foundational model better? Exactly. Application layer experience. Like, I could list other things, but OK. Yes. Studio (20:29.631) exactly. John's mic (20:34.753) making the foundational model better, think is so far what we've decided. But there probably will be a point where people, where there's maybe some kind of diminishing return and then like the money goes into like more money goes into that and resources go in the application layer. Yeah, yeah, exactly. I think that that will happen, right? But right now it makes total sense for the user to explicitly declare their intent before you start, you know, before you start. Right. Which is, it's just a fundamental, like we were talking about with Zoho. Exactly. Studio (20:46.371) Exactly. yeah, exactly. think that that will happen, right? But right now, it makes total sense for the user to explicitly declare their intent before you start incurring costs, right? John's mic (21:03.811) computer, you're to do this at 800 million person scale, even if they only use it a couple of times a week, you want that kind of serverless type model where somebody interacts, something spins up really fast, somebody interacts, it does the thing, and then it's off for the however many hundred hours that they don't use it. Like the math, that's how the math works. Right. So the user requiring the user to initiate the chat makes a lot of sense. Also, you can just roll out new features without having if you're going to do this at 800 million person scale, even if they only use it a couple times a week. serverless type model where somebody interacts. Something spins up really fast, somebody interacts, it does the thing, and then it's off for however many hundred hours that they don't use it. Exactly. That's how the math works. Exactly. requiring the user to initiate the chat makes a lot of sense. Also, you can just roll out new features without having any interface work. True. interface work. True. mean, that's, that is not, should not be undersold. It is amazing how much you can do a GPT. Now I, again, like because of like page thing, think there's a limitation. can schedule stuff. I don't think a lot of people know that. you know, like at a, at a super broad scale, right. Because it's It is amazing how much you can do with GPT. Now, again, because of the blank page thing, I think there's a limitation. You can schedule stuff. I don't think a lot of people know that at a super broad scale because it's not very clear in the UX that you can do that because you can do anything as a blank page problem, right? John's mic (21:55.588) It's not very clear in the UX that you can do that because you can do anything as a problem, Which, here's another one that you might not even know exists. On the $200 a month plan, which I had for a month and then canceled, there is a, I think it's called Pulse. Oh yes. And it's that, it's that thing that we were just talking about where they do have this like proactive, hey every day we're gonna kind of like. Yeah John's mic (22:19.403) go through your chats and your integrations and tell you useful things. And it's pretty good. It's kind of cool. Not worth 200 bucks to me, personally. Well, mean, which is interesting, right? And I think that goes back to the cost thing. And the reason it's on the $200 a month plan is because it's so expensive for them to do this proactive thing. So I do think that there are some really, really smart things about leveraging the chat interfaces that primarily Yep. Yep. Totally. Well, I mean, that's which is interesting, right? And I think that goes back to the cost thing. That has to be really valuable. It's so expensive for them to it. So I do think that there are some really, really smart things about leveraging the chat interface as the primary interface that overcomes some very real structural business problems that enable velocity of rolling out features. There are just a lot of interesting things about it, but for the general use case, think it has a lot of challenges. So I think a couple things. What is this going to look like in the future? that overcomes some very real structural business problems that enable the last few rolling out features. There are just a lot of interesting things about it, but for the general use case, think it has a lot of challenges. So I think a couple things. What is this going to look like in the future? So one, as the cost component and the economics of the whole thing, Studio (23:07.293) One, as the cost component and the economics of the whole thing, once the dust begins to settle on that, I think we will start to see more proactive things happening autonomously. And even if you think about cursor or any of the other IDE tools. you know, once the dust begins to settle on that, I think we will start to see more proactive things happening autonomously. Right. And even if you think about cursor, you know, or any of the other IDE tools, If you repeatedly use cursor to work on a code base, they're collecting a huge amount of knowledge about If you repeatedly use Cursor to work on a code base, they're collecting a huge amount of knowledge about the software, about your intention as a user, about your role, like what you generally do on it, right? And so I think we'll start to see like really good recommendations about what should be done next or changes that should be made or other things like that, right? the software about your intention as a user, your role, like what you generally do on it, right? And so I think we'll start to see like really good recommendations about what should be done next or changes that should be made or other things like that, right? So this will be a good topic for another episode, sleep time compute, which is a fun topic where the, I mean, high level idea. Hmmmm John's mic (24:06.467) is human sleep every night and we know that like fundamental to like making memories and storing memories like there's a whole process that has to do with that. And the idea for sleep time compute is mirroring that and then taking advantage of that time when active chat isn't happening with an agent to compress memories and to like it's a very interesting like it's super interesting space that would be fun to talk about. Humans sleep every night. Studio (24:15.971) Yeah, for sleep time. Yep. shot, isn't it? Studio (24:29.718) It's super interesting. Okay, sleep time compute. We're putting it on the books for a future episode. So I think we'll see more proactivity. And then I think one thing that I'm excited about is, and we're already starting to see this on a much larger scale, is that AI disappears. Yes. I think we'll see more proactivity. then I think one thing that I'm excited about is, and we're already starting to see this on a much larger scale, is that AI disappears into the background and software just gets way better. Actually, an interesting example is Riverside, the tool that we're using for this. Yeah, sure. into the background and software just gets way better. Actually, an interesting example is Riverside, the tool that we're using to record this. I mean, you know, the terminology that they use isn't, you I think it's very marketing-y, right? But you record an episode and then they will process it and automatically generate some assets for you to use, which is really, really helpful, right? And they sort of call it AI magic or whatever terminology they use. But it's like, okay, well that is... the terminology that they use isn't, I think it's very marketing-y. But you record an episode and then they will process it and automatically generate some assets for you to use, which is really, really helpful. And they can call it AI magic or whatever terminology they use. But it's like, OK, well, that is awesome. That is awesome. I'm recording this episode, I save it, and then in the post-processing, they just do a lot of things that would be really useful, very, very manual to do otherwise, that make a ton of sense. It's like, oh, this software is awesome. It's just making this thing that I wanna do phenomenally easier, right? And it just happens to be AI under the hood. So I think we'll see a lot of that. And then... John's mic (25:30.339) episode I save it and then in the post-processing they just do a lot of things that would be really useful very very manual to do otherwise right that make a ton of sense it's like oh this is this software is awesome it's just making this thing that I want to do phenomenally easier right and it just happens to be AI under the right so I think we'll see a lot of that and then I don't know, it'll be interesting to see what happens with the generic chat. I I think it's there to stay as a form factor for search. But as we've seen with vZero and Cursor, more multimodal, high context experiences I think are going to be more and more common. And so how is that gonna change? know, I think it's gonna be a matter of you have the store in GBT, I think it's gonna be figuring out ways to get to user intent faster. It'll be interesting to see what happens with the generic chat. I mean, I think it's there to stay as a form factor for search sure But as we've seen with v0 and cursor more multimodal high context, you know experiences I think are going to be more and more common. And so how is that going to change? I think it's going to be a matter of you have the store, know, in GBT. I think it's going to be figuring out ways to get to user intent faster. I agree. All right. Awesome. Our first short, we did it in an hour. Yeah. All right. Our first short, we did it in... Studio (26:31.427) Alright, we'll catch you on the next one. We'll do mental models and tool time on the next episode. Great. All right, we'll catch you on the next one. We'll do mental models and tool time on the next episode. All right, see you.