#37: A fresh look at Rabbit

All right, this is going to be a fun one. You know, almost a year ago I had Jesse who started this little rabbit company that makes his little AI computer on my show and it kicked off a whole bunch of stuff.

Yeah, it's a roller coaster ride for sure.

But it's been a crazy year for you, I'm sure. You know, I heard you're a prod. I've seen you ship agents. I mean, there's a lot of craziness in your life.

Yeah, yeah, absolutely. I've been enjoying so far. You know, I think. Absolutely. You know, the AI hardware is a big hype for this year since human and rabbit and there's a lot of conversations going on. Right. So I think for us, we were thrilled and honored to be able to sell and ship more than 100,000 units, which is way above our estimation. I think the first time we chatted.

We said I was hoping we would sell 5,000.

Exactly.

Yeah.

I told you before we even started shipping, I said, hey, I'll be happy if it's like three to five thousand. You remember that, right? So, yeah, it was crazy. So yeah, finally to be able to see you again. So ask whatever question you have. I think it'll be so I know.

A lot of these people have them as desk plates or in a drawer somewhere. You know, the bottom, maybe they haven't seen them for a while. Why should they pull it out of the drawer and try it again?

Well, I think, you know, it was certainly the case that for the first version which was shipped around April 23rd, there was a. There's a lot of bugs and a lot of people. I totally understand a lot of people like they tried it, they don't like it. Put it somewhere eating dust. But, but the team is being really, really has done. We actually push more than 20 different OTAs over the past couple months. And if you, if you face them now, it's almost like a one week brotier. So obviously, you know, me personally, I actually watched all the YouTuber reviews and you know, all the media reviews and all that. I actually had a notion file.

Did you have to go?

Oh, no, I mean, you know, I, you know, it's, it's kind of like, it kind of like a weird, I think, you know, this kind of information or conversation, you really don't guess, don't, don't pass out to the external context. Right. So because me as a CEO, I'm only, obviously I'm embarrassed by the day one experience. But low key, I'm happy because I can Use that to push my engineers be like, hey, everyone is talking about this. Go fix it right away. And that's exactly what we do. And I think you know, to build and fix stuff is the nature of startup. I don't think this is absolutely anything out of auto normal. I think that's, that's basically our food and water every day. So we have done build all the stuff. First of all, to answer your question directly, you know the. Even though there's a huge wave of hate and you know like misunderstandings, misinformation and the critical reviews in the early days, the total return number for this product is still remain less than 5%, believe it or not. So that's, that's very, very impressive from anything Gen1 hardware. Like even you know, mature products could.

See Apple Widget probably had more than that.

Well, well, absolutely. Yeah, you know, like I think 5% less return rate is a very strong number. And obviously our goal walking towards, you know, let's fix all the stuff, let's trying to make people happy. Right. Because they pay for this. And we basically fix all the day one, you know, bugs right away. Like you know, I think MKBHD was mentioning about the battery. We basically put a battery over three days within like two days after that. A lot of things is there's again the word I like to use is misinformation. A lot of people think, oh you're just a new team, you don't know what's going on and you probably don't have a QA to test it all out. And for instance the battery was actually caused by a long websocket link to a Spotify. It was nothing to do with the hardware. So it's a lot of stuff because the entire cloud infrastructure we have to build from scratch. So long story short, I think most of bots benefits very, very quickly and then we've been focusing on lam obviously you know, like I think the beginning of the year when talk about Asia there's like just so many definitions and now I'm sure you saw, I know you have this crazy X that was like a million followers and you saw.

Them in the same lately man, the world is speeding up. There's like new things every two minutes.

Exactly, absolutely. You know I admire you have that meta set up. You know, like that was really, really amazing.

I need some agents to read my list so I could go to the beach.

Absolutely, yeah, absolutely. And hopefully next time you have some sort of agent from Rabbit to summarize what's going on. Right. And Only give you the alert. But anyway, when we talk about the agent in the early days, I think a lot of people have this like a kind of like very blurry imagination of what an agent should be. And I think for us, when we first introduced the land first version, we support 4 apps only we talk about this and I explained, hey, we have to use the recordings to make this work. But very quickly we actually move towards what we call Lamp Playground now, which is this complex system with different components to be able to have your R1 do whatever task you want as long as the destination is a website. We basically stay up to our promises and we release all the stuff based on our roadmap. I think we see a very good increase in engagement. There's a couple of waves. The first wave is obviously all the bugs being taken care of most of the day. One bugs being fixed. And the second wave is since October 1st, which we dropped the LAN playground. I think you pick it up from my Twitter as well. I think the reason for that is obviously people in the early days asking why it's not an app. Is there anything that you can do that my phone cannot? Well, guess what, with Apple's iPhone or Android, you cannot just talk to their AI, be like, hey, go to Amazon, add a 12 pack die code until my house tomorrow. Right? So there's, there's, it's a very. Right, I mean you can go to the app and do it for yourself, but that's not the concept of the agent. So I think people slowly getting to understand what the agent should represent. I think, you know, we, we obviously we've been building out in, in public for sure. You know, like I post every latest update, I talk to the Discord members all the time. But yeah, overall it's been enjoying it. I think, you know, for us a story kept holding and we catch up with our own speed. Really. Just the thing is we don't have any serious competitor from startup and that's very painful. And you know that from early days of, you know, for example, like Tesla, right. People compare the first version of Tesla with Pris saying that, oh, you have less mileage, you're not even a car, you're like modified Lotus and you have shittier interior design than Maserati or whatever. So I think we're running the same challenges. Like people compare this to the previous generation phone, which is, yeah, give me.

Three reasons to pull this out of my drawer, load it up with the latest software and try using it again.

Yeah. First of all, if you go to the new rabbit hole. What you'll see is that you will see a. Actually I can share my screen because we're in Google chat because why not, right? So first of all we put a lot of new stuff here right off the bat. What you're going to see is that we have this land playground, the LAN playground is where this new generation of LAN exists which is a no pre training general agent that can perform any task on our website. You can use from the rat hole to type if the prompt is super long, but you can just grab your R1 and ask about it. You can literally just grab your R1 and say hey, go to Amazon, do this, go to somewhere else and do that. I'll tell you some of my use cases. I basically manage all my subscriptions from R1 now because the Amazon killed the dash button. If you remember the dash button from the early days, that was really, really neat. Just press one button, you got another typos delivered to your door.

Ryan, the PR guy is showing up on my screen. Ryan, can I have you leave the call please because you're taking up the space.

Thank you. Yeah, So I use R1 to do automated tasks from website. Actually this is quite smart and a good example is that we put a sample prompt in here. It's like you can stack all the queries that for example, yeah, I just need to refresh my page. You can be like, hey, go to Redis, search for people's recommendation of this year's best TV and then go to Best Buy and see if that's in stock and if not then go somewhere else. You can actually stack infinite amount of information. For example, if I'm just running this as a sample query and you see both of my hands, so that means I'm not operating this and it's pretty fast. I will just do things for yourself. Obviously anything that I demonstrated here you cannot achieve by the current version of Siri or Gemini. It will just based on your prompt. This is really handy. This is the next generation LAN and all of a sudden your R1 can do more than four apps. You can do anything. If it's a website, people be using this to do simple things. As simple as I have a hot yoga place, I go and they have this very obscure website that I have to manually schedule for the next, you know, class and just use R1, talk to it and find a slot and that will be it. And there's also, yeah, look things up.

It can buy things, it can play.

Wordle, it can play games.

That's happening in some of the bigger models.

Exactly, exactly. But yeah, we dropped this back in September, which, you know, it's pretty, pretty popular. Another thing that I want to emphasize is obviously the login for this. We introduced this cookie jar which is really neat because we don't save the credentials. You just go here and you want to log into a new website, we'll just save the cookies. This is universal Crossrabbit os. Once I logged in with my Amazon or Google here, I can just say, hey, send an email to Robot using my Gmail. It will just put a cookie. You don't need to log in again, which is automated task. The second reason for you is obviously the long way to teach moc. I think this is very exciting and a lot of people, to be honest, they told me that they bought our one purely because of this. We did actually finish the Teach mode development quite early. I think somewhere around April to May we have our code ready for this. But the reason that it takes so long to release to the general public, we actually just rolled out this to every R1 user around like 3 weeks ago. 2. 3 weeks ago. The reason is that we started with like a small RPHA group that they had to sign an NDA and they work with our engineer directly because we realized that these two sometimes can be more powerful than you thought. Right? Because the idea of the Teach mode is that you train a new lesson and you just demonstrate how you do things once with natural language, zero code needed. Which is amazing, but it's very easy to think of if there's a bad actor, if someone trains a malicious agent. Right. You know, for example, we actually were very scared of it. So that's why we decided to kind of figure out what exactly we should enforce to.

So I could teach somebody how to fix a washing machine or code something or something like that. What kinds of things are people teaching people right now?

So again, this current version, I'll give you some examples. All of a sudden if you go to. Because you know Google Home Automation has, or nas, it has this website and you can just go there and all of a sudden your R1 can control the entire Google Home system within 5 minutes. Basically it's very simple, right? So if I go to the website and I will just type something here, I'll just give you a quick example. I actually managed my community gate system where now I just use my R1 to say, hey, Robert is visiting me. Add robot to my gate and that would be it. The idea is that you go to any website for Example, if I go to Amazon.com and if I go here and it's very simple, it's just going to load a VNC here for the destination website and think about this can be like a smart home system or whatever. All you need to do is just basically do this and it goes here and I'll just do. I log in and it goes to teach session. And here very simple. You just, you just need to put three informations right. Lesson names like the file name. You can, you can take whatever name. You can be like Amazon, Amazon one or whatever. And task solution is where it's very interesting. So for example, you instead of writing a code, if you think about developing an app for iOS instead of writing tens of thousands of lines of code, now you can just describing natural language. So I can do something be like hey, I want search for e.g. xYZ specific band of toilet paper. And that's, that's the, that's the, that's the merchandise I want or anything in particular you want and add to my cart or, or check out or set up the subscription for three months or four months and you just describe this and you start teach. When you start teaching and everything here will just waiting for you to input. All you need to do is just to demonstrate the sequence. Once that's how it works, then you said hey, I'm done process. It takes two minutes. All of a sudden if you go back here, you have your own lessons. You can also see a lot of the community lessons that we curated. These are all from our own users. The importance of this feature is that for me, this is my vision towards the future App Store. But it's for agent. If you have to think about what makes iOS so successful is because of the App Store, we have to think about what does it take for us to develop a legitimate app. Either it's on Android OS or iOS, you have to know some coding, you have to work on the UI and stuff. And that's certainly not for everyone. Even with the tragedy GPTs I think they rolled out last year, you still have to work with some sort of API, you have to write some sort of a Python code. But with this all you need to do is just demonstrate once zero code and the process time takes two to five minutes. And then we have some crazy numbers to share. We roll out this to general public around two, three weeks ago. And now I believe we have more than 2,700 agents being created.

Wow.

So you have, you have to think about a velocity, right? So people using this to make your own customized agent within two to five minutes. And eventually we're working on a business model to where if robot, if you build an agent, a million other people would like to use it, you should be making money from it. Yeah, exactly.

I thought AI was going to take our jobs. You're creating new jobs for us.

I think this is just an evolution on what we call traditional programmer or app developer instead of. We significantly lower the barrier that as long as you know what you're doing from a computer and person perspective, you don't need to figure out the code because Teach mode can take care of it. And then the next question is that what's the difference between Teach mode and the land playground? And you can think about Lamp playground. Is this. No, per training, more generic agent that can do most of the tasks, but you're giving a lot of the options for AI to guess which to pick. But Teach mode is repeating exactly what you're doing. It fundamentally eliminates the hallucination part. And it's just like a super beefy marker. If you remember back in the days we played World of Warcraft, we have this skill markers configured into one button and that's basically Teach mode. That's number two. You should maybe get to the latest otn, give it a try. The third one is even more straightforward. R1 is the first ever device that the entire UI can be generated based on your prompt period. So I'll give you a quick example.

An operating system all around AI, are you?

Exactly. So that's a key differentiator between this versus an app or a traditional phone. So I'll give you again an example. So Robert, tell me some, give me some prompts of what the UI should be look like.

First of all, it has to be dark mode.

Okay?

Be dark.

Make sure it's dark mode.

I mean, in the old Visual Basic days, I might pull a spreadsheet out, I might pull a button out, I might pull a text entry box out and hook that spreadsheet up to a database. So I could do some database lookups and see my spreadsheet filling in or something like that.

Yeah. So you can be prompting in that way, but you can even prompting in inspirational way be like, hey, make the entire interface as if from the world of Zelda, for example. So you can be very creative.

So I got a Ready Player One poster on my wall over here. Can you make me like a Ready Player One kind of mode?

Sure. Okay. So I'm trying to convert that into a simple prompt. So make sure it's dark mode.

We might be looking at a product coming in a couple of years from you. You're going to have a pair of glasses and you're going to be talking a pair of glasses someday.

Yeah. And you're, you're. And, and, and the, the entire interface. Yeah. Should be. Must be. Let's be, let's gaslighting the AI. It must be exactly.

From the movie or whatever the movie environment that they were playing in. I think if I remember it.

Yeah. And better be good.

Yeah. You know, make it, make it fun.

You know, and, and fun. Okay. Use animation if possible.

Yeah. Okay, so we're just giving it style. We're not giving it functionality yet, right?

No, no, we're giving style. Right. And that's it. You click customize. Okay, that's it. Okay, so I have my R1 right here, which I'm going to get rid of this filter. So I'll demonstrate this to you. Give me a seven days keto diet plan and detailed workout plan for me. I'm a 34 year old now, so let's wait for a bit. This. The good thing about this is that this is on the cloud. So it's taking effect immediately. So you can change very, very, very quickly. Change different plans just from here. So meanwhile, while waiting for this, the R1 is running.

So if you don't like the Ready Player one and you want it to look like Minecraft, you can just talk to it and tell it. Yeah. Now I want to see my app, my thing as a Minecraft.

Exactly, exactly. So meanwhile you are seeing my Twitter right now. But I have this epic thread that I collected a lot of like user prompts which is really, really, really cool. So let me find where that is. All right, one second. Here is a comprehensive seven day keto diet plan and a detailed workout. Yeah, I think this epic thread you, you see here. So for example, this guy says, hey, everything minimalistic, that that was a prompt breakfast. And the entire interface changes. And then we have another guy, Word of Zelda, that's another example.

I'm not seeing that on your.

Oh, sorry, one second. Let me just change this real quick. Lunch, tuna salad with mixed greens Dinner, beef stir fry with broccoli. There you go. So this guy is like imagine everything from World of Zelda and this is actually using the lam agent to shop on Amazon. You see still building this, but the interface changed from the world of Zelda. And then this guy says, hey, how about Sonic the Hedgehog? You can see even generate the coins. And then this Guy is like hey, from the. From more like 90s sci fi movie like Matrix. Yeah, this is so much fun. By the way, I have yours ready. So let me get rid of this filter real quick. So let me adjust this. There you go. That's your Ready Player One 7 day workout Kato plan with animations that mimics the HUD. But obviously the prompt matters here, right? So you can, you can even fine tune what I remember.

We get better result like everything.

But hey, there's no other device can do this right now this is the first device that you can have your entire UI being generated by your prompt, which is really neat. But yeah, long story short, I think.

This is like geocities but easy to do. I lost about half the audience. What the hell is geo?

Yeah, exactly. But those three things is what I would recommend to every user. Take it out, give to the update and give it another try. And really the importance of the agent I cannot emphasize enough because everyone now is racing to that. We still believe that what we have is the most advanced system and also plug and play. Instead of another company, I think Anthropic released the computer use. There's multiple other companies doing their own version. But you have to think about agent. It's really not just an API. It's like you have to do something that for every user to understand, they can just talk to it and get a result. That means that you have to the model itself, the vision model, the LLMs and you have to have this entire cloud infrastructure to handling the logins and the cookies as well as bulk detection captures and all that stuff. But we have all the systems ready and they see this 190 nano subscription box right away.

That's awesome. When you first were talking about large language models with me a year ago, you were laying out that, you know, someday the software is going to be able to look at an Uber app and figure out how it works and get you an Uber ride or look at, you know, Excel spreadsheet and go, I know how to do, you know, all sorts of things with Excel. How much of that vision has arrived and what's still to be done.

So most of the most of the early visions has been achieved. There are some challenges, which is not too rabid, but rather to the entire industry, which is obviously what I mentioned. Captcha and bug detection. So we're currently working on a new universal driver that basically makes our agent to be identified as human as possible. I mean we have most part of it. In fact, one of the reasons that R1 is very handy is that it's using because the entire agent runs on cloud, right and everyone make the agent runs on cloud and the R1 is using your local proxy. So to Amazon they're not seeing some random Oregon IPs where our AWS service hosted for instance they're seeing me as Jesse in Santa Monica. So that's one of the challenges. I think some websites are kind of very generous about the policies on this Some of the website not necessarily saying they're less secure or more secure but their system is just very old way of looking at agents as 100% bought which I think it's going to change right because everyone now is working on agent, the Captcha system, the cloudflare stuff they have to change at some point so to simply go somewhere and do something that's being accomplished. What we're working on right now is we're expanding the land to cross platform and I think what I can do is I can give you a private demo after this because we haven't announced yet but what we can do is we can give you idea and we're going to announce something very very soon that we have all these agent pieces here and there. We have the TEACH model, we have the playground but we also have a LAN that works directly on Windows and Linux and we have a LAN working on Android system as well. So our next biggest focus is that really just trying to get all these components to become one operating system that's more like a movie her internally we've been achieved some example be like hey, you just wake up and you have four things to do. I wake up today I have four things to do. I will just say the four things and that's it for user. You don't need to worry about oh this renewal DMV license. You have to go to their website and to write a document you have to go to Microsoft World and then the agent will be smart enough to parallel run all the tasks for you and for you getting the result and more importantly a result in generative UI with only information you need, nothing else. That's our next big focus and we've been working on that since the second half of this year.

My friend Andy built his own agent to call businesses to make check with a restaurant if there's a reservation available. I need a haircut. I ask chat GPT or the Rabbit. You know hey, can you give me a haircut appointment at Super Cuts? Yeah, well there's an API at Supercuts for that right? And there or there's at least a website where you can sign up for, you know, one o' clock this afternoon kind of. How close are we to having a rabbit that can do that? They can call a business and talk to the person at the business, get an answer and bring it back and, and then say yeah, you got a haircut at one o' clock or you got a, you know, restaurant reservation or, or your laundry is ready to be picked up or something like that.

So, so I think just to make an appointment, change appointments through a website. We already can do that with Playground and that's, that's already been there in the, in the system that doesn't have.

It all hooked up on the website. You know, it has a, answers the phones, you know.

Yeah, yeah, that, that can be handled pretty well. That can be handled pretty well. So the, the challenge being obvious to making phone calls. Right. I think, I think Google has a internal project they cancel called Duplex.

Yeah. A bunch of them come and there's a bunch of companies that do voice, you know.

Yeah. And then there's, there's obviously after Google D, there's multiple other third party companies working on that. The good thing is it's theoretically possible already because all you need to do is just hook that to each mode to that destination. Right. So we have all these components ready for user to configure their own routines. Sometimes you don't necessarily want to give all the options for AI. I'll give you a quick example. If you say hey, just order me a brand new iPhone. Well the AI will go there and pick a color of AI's choice. And then if you don't say here's my name, here's my address saved in the cookie jar in a safe spot. What the AI is going to do is the agent's going to basically just put John Dong and ABC administrator. He's just going to start making things. But still the capability is there to start proceeding to the final purchase. So I think what you mentioned is already possible. To be honest, what's not possible is a very clear, easy to understand front end user product so that they don't need to go get repos from 20 different GitHubs and working from here and there.

You can do that because he's a nerd, but I'm not a nerd.

Right, exactly. That's something that we're enabling on this device which just push the talk button. I think by the end of this year you will see the continuous velocity of tmode. We obviously start working on a business Model part of it, the app store idea part of it. How can we make the developers beneficial from this system and how can we kind of like cross share this agent? Because the fact is if you train an agent to control Google homes, the entire system learns that. Right. So, so it's not necessary for second people to do the same thing and we can just offer that as a capacity or capability to every R1 user. So we're going to figure out that part as well as most important thing for us is to get every cross platform agent in one space and presented in a very simple way. So that's basically what I'm being personally on our team's been focusing on.

One thing I tried with ChatGPT at Thanksgiving dinner was to have multiple people talking to the phone at the same time about something, arguing over where should we go for dinner or whatever, just talking. And it, it failed miserably because, you know, I would give it an instruction and then somebody else would start talking and while it was giving the answer, it would stop. Yeah, not answer me, you know, and not be very human with it.

Yeah.

Where is your voice to text technology? You know, the system that's listening, how can it handle any of that kind of thing? Or is it still a single voice or.

Yeah, I think it's very interesting that you mentioned that because I've been working on smart home speakers 10 years ago. I think that was the debate back then, the wake up wars and the way that we handle dictation. I don't think it has anything to do with the speech to text engine. Speech to text engine is pretty accurate. What's happening on your scenario is that you just simply have too much background noise and the algorithm to detect when to stop and send that query is messed up. And the algorithm is very simple. Basically it waits for the decibels to drop to a certain layer.

It's not just noise, it's multiple people talking, multiple people talking, time, having an argument or whatever. In the perfect world, they would track each of those people, understand what they're each saying, and then maybe batch an answer to each of us and say, hey Robert, you said this. Here's my answer. Alex, here's your answer and Mary, here's your answer.

Right, yeah. So that part, what we call internal, we have a code name called qmone. We can handle that with multiple queries. Stack questions, you can ask different, completely different related questions. But for our hardware design, I was super, super cautious about wake up words or always listening. So that's why I started thinking about the most challenging, most demanding scenarios or where human needs to use voice, which is obviously military and EMS and all that. Firefighters, all those are push to talk button. Right. They're not open mic and that's for a reason. Because we don't want to get the AI system to be messed up.

Well, a lot of people say, well, you just need to mute it after you talk to it. But they're not, they can't really listen to a group talking around the device. Right. And we want that at some level.

If you say that, then that's effectively push to talk. You have to click that button to mute it. Right. You actually have to click twice because you have to make it listening again. So that's even worse. Yeah, I'm just not a big fan of always listening at this point. You know, there's, there's various reasons for that and I really, really hate wake up words because, because, you know, the least thing you want to do is like we have this amazing podcast, we are both logged in and then there's Alexa speaker somewhere. Be like what I can do for you? That happens all the time. So I don't like that. Yeah. But anyway, I think this use case.

Because half my family is Iranian and they talk sometimes they're talking Farsi back and forth in the kitchen. I'm talking when my brother is in English and I want.

Right.

A device that's telling me what the other people are talking about, which is a little weird, you know, privacy and all that. But you know, my family's used to me recording them and turning on devices and playing with, you know, weird ass things. So they're, they're in that way. But, but it just doesn't work very well for now even a single user talking to that them, you know, it like talking to somebody who only speaks Farsi. It does re. It does translate, but it also often breaks and it doesn't, it's not fluid like a human translator would be, who's standing right there and translating it for you and telling you what's important that the other person said that kind of thing.

Right, right. Exactly. Yeah. My personal experience on the hardware being that, you know, the microphone that we pick for something that can handle that, which is a microphone, far field microphone field versus the standard two piece or one piece microphone. I think the performance is like vastly different. You will, you will need a microphone array, you know, a six mic array or like four mic array at least to handle that from, you know, different locations even which a phone is.

So it has a mic Array. Right. It has four mic. Three or four microphones in an iPhone.

Right.

We have one microphone.

No, we, we do have two and our two microphones. Yeah, you have to make sure it's a far field mic because the range they pick up is quite different.

I'm even just holding the thing right, right in front of them, you know, because I'm a nerd. I don't care about shoving something in your face. Right. Or keeping it on the counter right in front of us or that kind of thing. It does. It's fine at picking up your voice. It just gets confused by all this.

Exactly.

Noise and multi party because it loses track of who's talking and then it interrupts you or it gets interrupted. You know, it's not perfect and it just fails.

Yeah.

But you know, I'm watching.

But they're gonna announce 12 different things starting tomorrow. So we'll see. I mean I'm super excited for the competitors and industries.

Yeah, I mean we'll try that Christmas dinner again. In fact, there's a voice to text company I'm interviewing who does multi voice and keeps track of who's talking around the phone and stuff like that. And it's pretty good.

Yeah.

There's a lot of developers who are working on breakthroughs to get us to the promised land of having a device that always works when you talk to it.

Right, right, right.

And that's. I love your little device, you know, because it's something that I can just leave on the counter and talk to it when I, you know, I don't need to look for my phone.

Well, speaking of that, you know, like one thing that completely caught me off guard is, you know, we have a very high percentage of the user that they are actually professional American truck drivers. They love R1 because they, I guess the modern system that our company put some cameras to detect if you're driving using your phones and that's not okay. But they have the ham radio and this was considered as a radio which push talk. So a lot of truck drivers bought R1 and they are driving eight hours per day every day and they use this to listen to podcasts, you know, check the weather for next destination, you know, maybe even book a motel, whatever. So I think different people treat this device quite differently. I think our main customer is still remain as a teenagers and we have a whole lot of teenagers and they're not just average teenagers, they're like hackers, they're geeks and they love teach mode, they love to take on the UI and they Love to work on the rabbit. So yeah, it's quite amazing to see you will get a lot of meaningful data once you reach 100,000 users instead of 10,000 users yourself.

So you know, you sort of mentioned there navigating around a new city, having this thing in your hand. Right. You know, the large language models really know a lot of things. Like everything. They know everything in the Smithsonian, everything in every restaurant I've been in, everything in the Disneyland, everything around the world. I've been traveling a lot this year and testing out these large language models. Whether it's chat, GPP or anthropic, I don't care. They all do a similar thing now, right?

Yeah, yeah.

But they're not hooked up to transactions. You know, like, like if you're a trucker and you're going into Chicago, you need a hotel.

Yeah.

You know, you, you can find the hotel by talking to a large language model. And it's pretty accurate at that, right? No videos. You can tell it, you know, hey we, I need a hotel. It's about 300 bucks and it, your thing is pretty good. But it doesn't let me transact. It doesn't let me say, oh, go get me that hotel. I'm, I'm.

So what we can do right now.

Get shopping hooked up.

But yeah, yeah. So, so for instance, like with a cookie jar that, that you know, for a lot of services like Uber DoorDash, Uber Eats or even Amazon, you have your pre approved saved credit card info. Right. If that's the case, it depends on merchants. If that's the case, it will just go through. But for you know, purchasing new merchandise from somewhere else or booking random hotel. Yeah, that's not there yet because for us we have to literally apply for a special license to be able to handle direct transaction. We don't have the short term plan yet, but that's certainly something we can do. But overall I would say, you know, us is basically very far behind. If, if you, if you compare the countries like Japan or China. Right. Like everything, everything's QR code. Like no one carries a wallet. Well, unfortunately like the system here, like there's a lot of things that still just stuck with the classic like credit.

Card but you know, phone call on the back end. Right. Because yeah, talk to and say I need a hotel. Hotel.

Yeah.

Hook up to the website and you know, pay for it. Yeah, you can call the hotel and have a little conversation. I have a user on the, you know, who's outside, they want the King suite and you know, and have a conversation with the other person on the, at the hotel. And the, and the person says, oh, you know, how about a credit card? Oh, here's my credit card. Right?

Yeah, yeah, exactly. Yeah. You know, that's one of the reason I probably would never switch to Android because everything is. I just love Apple Pay. You know, it's one of the things, you know, Apple Pay imessage.

It works everywhere, man, particularly Europe. Over in Asia, right. New York City, if you have an Apple watch, you can get. Use it on the subway. You just tap in and go into the subway and it pays for. Everything is great.

Yeah. But I want to share like one particular memory that I had. It was roughly back in 2015 or 2016, where I was going to Guitar Center, Sunset Boulevard, Los Angeles. And it was, I'm sure the Apple Pay was there already probably the two years, three years before that. And I didn't bring my wallet. I grabbed some cables and say, hey, you know, do you guys take Apple Pay? And the Guitar center says, no, but we only work with Samsung Pay. I was like, who the fuck use Samsung Pay? And this year I think everyone works with Apple Pay. But it was just recent years. I mean, before that it was just all credit cards. But when you compare. Again, when you compare like countries like Japan or China and, and it's crazy. Like everything, everything's.

I mean, everybody has WeChat there, right?

Exactly. Yeah. Everything's like QR code and give me a dollar.

Your rent, R and B. Give me an R and B.

Yeah, I mean, I mean, certainly, yeah, that, that, that part. We, we are still a bit behind, but yeah, for the device itself, but it's happening.

I mean, more and more places in my life are taking Apple Pay. Over in Europe, everybody, even the street vendor takes Apple Pay, right. Or Google Pay.

Right.

Contactless payment. Right. You know, you tap your phone on a device and it pays for it, right?

Yeah, absolutely. Yeah. But again, I think the big question behind this is that when we will make a phone, right? Because there's a lot of like the thing you mentioned about phone calls or payment and all that. You know, I think for R1, we deliberately separate this.

When you're going to make a robot, the phone.

We talk about that. Right, we talk about that. I think on the. One of the spaces you hosted. But yeah, we had a lot of.

Right, soon we're going to get glasses. In fact, at ces there's a whole bunch of AI driven classes that are coming that have a display, have a headphone, have a Microphone.

Right.

And an AI computer, a large language model somewhere running so you can talk to it.

Yeah.

And you would be perfect for fitting into there because you're already thinking about operating system with this and looking at things and being able to understand them in more detail and you know, and, and you have a community of people who are building all sorts of cool apps to do all sorts of cool shit, right?

Yeah, yeah, exactly. Absolutely. We have to do the hardware design ourselves and we want to do it differently. And speaking of glass, obviously it's gravity is a lot of interest. My opinion was that, is that I like the fact that you can have a screen right in front of your eye. I like the fact that you have a POE camera that can document a unique angle that traditional handheld device cannot. My biggest debate is still the frame is because there's prescription people, there's non.

Prescription people gonna do one and Google's gonna do one. You know, you're not gonna compete with those people. Yeah.

Your scroll grows to fit the frame, not the other way around. Yeah, I know that because I did my LASIK surgery a couple of years ago and I used to wear a glass and, and you wouldn't, you wouldn't convince me to, to, to try for a slightly better frame because it's going to be uncomfortable for a couple. And that's the biggest challenge because then you want to do a consumer electronics and you want to make sure that's a universal fit. But I'm afraid there's no universal fit, so. It's interesting though. I know Meta sold so many glasses, but I never in my real life and maybe I'm just too, too much about autism, but I never in my life walk into someone that introduced the Google Meta. Sorry, the Meta glass, the Ray Bans while, while wearing it.

Yeah.

You know, they always take it out from their pocket, be like, hey, have you seen this? But I never, I never seen a.

One like actually more and more people wearing them. But you know, I'm in Silicon Valley where everybody's.

Yeah.

And very willing to try out new things.

Yeah. Yeah. But I think it's very interesting. I was joking with you Kansas City.

Very often for sure.

I was joking with my, with my, with my team. I said hey, if I were trying to design that, maybe we do the Dragon Ball power meter instead of a traditional glass. But I like the fact that you can have something right in front of your screen. So that's one of the fun factor, worth actively thinking. So it just, it's going to be.

Interesting to see how you evolve it as this stuff comes on the market, because it's coming at LG and Samsung and BOE showed off the projectors. They're right the size of a sugar cube. They fit right on the edge of a waveguide, which looks like that, right?

Yeah.

And it puts an image in your eye. It's really, really amazing. Right. So, you know, normal people, they're, they're not gonna really consider this for 2, 3, 4, 5 years. Right. As they get better and as they get cheaper and all that. And as people like you innovate on them and come out with really amazing operating systems.

Yeah.

You know, the AI does a whole bunch of stuff with it. Hey, built me a game or hey, go buy me a. You know.

Yeah, I think really, you know, it's a similar, similar story to where back in the 2016, I guess when the all of a sudden 200 companies start making VR goggles. But none of those figure out what the internal operating system user interface should be. Right. So same with computer graphic user interface. The mouse was in fact invented before the graphic user interface. Right. And then in the early stories, I'm not sure, I'm sure you must have.

Heard it like it was 20 years ahead.

Exactly. And no one gives a shit about mouse and everyone be confused, like, why do I need this? So it's always a good idea to like work on the software, make sure the software makes sense. And then for us, we want to build a software to be able to put into any form factor. And then once we do that and we can release like a variety of, of different hardware projects. So we do have some, some orange.

Rabbit in our hand. You know, we're going to be walking around any time.

Yeah.

You know, I don't carry it out in the real world because I don't want to carry two devices. And this is not yet a replacement for your iPhone.

Right, Right.

I need the iPhone all the time. This I leave at home on my office and I have another one on the kitchen just to pick up and fool around a little bit. It's a fun little conversation device, that kind of thing. And now that it's starting to become really useful because your community, you know, I'm using it a little bit more and more. Right?

Yeah, yeah. It's never about how many hours you stick with your rabbit R1. It's about you have something that you want to deal with, deal with it, and then forget about it, move on. So that's, that's the whole concept. So obviously it was not replace Your phone yet? Yeah, yeah.

So you can do things like hey, you know, like even. I talked to Grok about the weather. Is it raining outside? I could see. Yeah, it's not raining because I live in California. It's nice and sunny. But it gives me an accurate answer. Right?

Yeah.

And people do all sorts of weird stuff like that. Hey, it was, you know, is certain movie playing tonight at. Or is there tickets for a movie at Lincoln center tonight or something like that?

Yeah. There is one particular use case that I feel really amazing. It was actually ar one user sent me a DM send me a whole letter. This guy used R1 in a very creative way. This guy said hey Jesse, I want to share a story. His wife will handwritten all the menus for their entire family for. For to cook for a week. Right. So this guy, this guy basically used the R1 with camera and point to his wife's writings of the. Of the, you know, daily menus. And. And basically he. He asked around. He said hey, this is, this is my family's meal menu for this week that my wife put together. Analyze the immigrants and use agents to go to the grocery website to get. No, to go to like I think it's once or Ross, you know the grocery website to go. To go to grocery. Grocery website and purchase all this immigrants for my wife ready to cook. And then he showed me a photo where all those being delivered to the door and that was like so leaving.

This on the kitchen counter. Right. You know, live your life a little bit. Makes life more fun. And.

I like this case in particular because it really utilizes a lot of the different features. Right. It has the AI vision. It takes the image from the handwriting and then they extract the context from the menu. Extract the context into immigrants and then use the agent to understand where to purchase from all this and then have the awareness of the user safe logging which has this shipping address and everything. Just deliver the door. Which is really amazing. Yeah. We start seeing a lot of like good use cases.

I told normal people when I meet them at a bar or I don't go to bars much anymore, but at a restaurant or something like that. Oh, did you know that AI can read the Bible to you? It's true, right?

Yeah. Yeah.

Guess what? It can read a machine learning book to you.

Yeah.

Picture of the page you go.

Yeah.

I'm confused by this. You know, dumb it down to a nine year level, you know, or something like that.

Right?

It does.

It's.

It's amazing. And it can do the same in the Bible. And people are like, whatever, absolutely.

Yeah, yeah, I, I think, you know, I, I'll be very excited to see how actual human develop on the agent instead of just have this model to guess where to go. So, you know, the teach mode is really something that I really feel proud of. And there's a lot of.

Sends you a toy from Germany. It's all in German. You take a bit. Hey, can you translate this to English? And it's.

Yep, yep, yep. Yeah. So again, you know, despite of the early days of a crazy roller coaster, I want to define on this product a podcast with you that still, as of today, and I believe till the CES in the coming months, R1 is still the best all in one handheld, dedicated AI device that can do most of the things.

Like there's pendants and pins. Oh, you're gonna be like a rap star soon. You're gonna have all sorts of hanging off your neck.

Right? Well, well, but, but that's, that's the, that's a different scenario, right? That's for recording your conversation and extract that into conversation. So if you talk about something like a Swiss knife, right? Or. Or Leatherman multi tool and there's rabbit R1. That's it. Yeah.

Some of us are listening to business meetings. You take them in, which is sort of weird use case. I mean, that's not a common use case yet, but you know, I know Avi has one of these pin or these pendants, right. When he takes it to business meetings or listens to him pitching VCs and in his ear it tells them. Oh, you just that up.

Yeah, yeah, yeah. Exactly.

Why do you make your pitch this way? You. You forgot to bring up this key feature or this key thing. You forgot to ask for the sale or whatever, right?

Yeah, yeah. I, I like all these guys. I like all these guys. Little project and I really enjoy them. Ordered all of them. You know, I think last. Last AI. Not AI, but last gadget I ordered was the. The daylight, which is which the high fresh ebook. So, yeah, I, I think, you know, really, really happy to see that people start breaking the walls and making variety of funky colorful gadgets again. That brings us back to, you know, 90s.

I think the orange is the best thing you ever did, man. I just love looking at this. Like I don't even turn it on. And it's like, you know, it's like, I mean, even if you're going to use it as a paperweight, you know, it's a little expensive for a paperweight, but it's fun to just have around and it always causes conversations. What the fuck is that?

Yeah, yeah. And that was, that was a good choice by us. But yes, stay tuned to our next project. We have a couple of the amazing, amazing ideas going, but main focus was.

Really that we should be watching for any new updates that are top secret in your lab. Yeah.

I think the very next thing you should be watching for is this new Rabbit OS 2.0 where we get everything together. And that's basically. I believe, I personally believe we'll get released that before everyone else. It's just a very brand new concept. Sometime next year, very early next year. Sometime next year. Yeah. And the, the concept is really just. You have, you have X amount of things to do. Right. Just, just talk to that system and the system take care of it. So that's, that's the next big thing. And meanwhile we have, you know, two independent hardware project that's in preparation, exploring new form factors. But like I said, you hook it.

Up to an AI art system. I mean there's so many much innovation, shipping and 3D, you know, pictured video, video 3D world. I mean there's all sorts of stuff I'm seeing all day long.

Yeah.

Because this has an affordance of fun, right. That's the kind of thing I, I would love to use it a lot for is like just make memes show up on the screen like oh, look at this, you know, and show the family that kind of thing.

Yeah, absolutely. I think, you know, one thing I think didn't get emphasized from the industry enough is that there's a huge difference between you being a geek getting GitHubs all over the place to our computer, versus you have something by your hand and press a button. Just talk to it. Right. So for us, you know, we even, even though we're working on legitimate like you know, serious agent models and systems, but bottom of the bottom of the rabbit is a consumer product company. You know, we want to do something that's easy, simple to use and fun at the same time.

I mean, can I push a button and say, hey, give me a picture of Santa in a cybertruck. That kind of thing. And if it did and it did a good job that Christmas dinner. Hey look, it's Santa has a cybertruck.

Yeah, yeah, absolutely. Yeah, absolutely. You can do that. Yeah, sure.

Play some music with some words. You know, give it some words. Hey, you know, here, here's 20 words. Make me a song, make me some beats.

Yeah. One of the most amazing features we roll out is magic camera. I'm not sure if you saw that, but that was quite some days ago, basically. Now with the R1, you can take a photo. And if you take a photo, it will convert into any style you want. By default, we have our secret prompt, which right now we haven't updated that yet because people love that too much. Originally we were thinking about update our secret prompt for every. Maybe like every week or every month. But people say, hey, please don't just keep this. So anyway, if you just go to double tap, I'll do that right now. So I'll try to take a photo from this podcast. So if I just double tap this, it takes to the camera and yeah. Three, two, one. There you go. All right. Yeah, I have a photo of you. So what happens is that you can press and hold exactly like you said. Hey, take a photo and transform this into whatever style, whatever prompts you want. Or you can just take a photo. And what's going on is that if you say a particular style, it's going to transform, but if you just do the default.

Make Scoville look dystopian or make Scoville look like candy canes, something like that. Right.

I have you right here on the rabbit hole that you can download. And we're actually gonna make the R1 be able to get this photo directly. So this is without a prompt. You just convert it into this kind of like podcast guy with PixelReady style. But it's all over the place and you can create very creative photos. This is like pixelated zombie rabbit I created. But this was just like my dining table, actually counter table to the bed. I just take that. Yeah. So this is one of the fun features that we added probably a couple of months ago and we generated I guess, millions of magic camera photos already. Yeah.

It's crazy because it has an affordance for being fun, right?

Yes, yes.

Grok has this fun mode in it.

Yeah.

And you start, you know, I was talking about Santa Claus and I. I said, Santa Claus fired Rudolph. I heard and bought a cyber. And they went on. Oh yeah, the North Pole is modernizing.

Yeah.

You know, they've made some changes up to the North Pole, right?

Yeah, yeah, yeah. I. I use grog. I use grock one time because they have a roast me mode. Right. I said, hey, roast me. You know, it was quite funny. And he said, it looks like you and your miserable little rabbit is already getting roasted all over the place.

I was on an X Audio space last night and there was an AI host and that's what it was doing. You would talk to it and it would roast you or call you names or, you know, have fun with you and be very on point, you know, this is crazy.

Yeah, yeah, absolutely. Make. Make it fun is actually harder than make it useful to be on.

You gotta have a sense of humor to make it fun.

Exactly. And yeah, I think we did a very good job that people treat R1 as like a, you know, not very serious device. It's more like a Tomagotchi or walkie talkie, Pokedex ish toy, which is exactly where we want to be for the first generation. But I hope that people see the hard works that we put behind for the actual agent part that R1 is.

Indeed a toy that actually does something useful. Oh, my.

Yeah, exactly. Yeah, exactly.

We haven't had a toy in our life like that before. You know, I've had a lot of toys in my life, but I'm not. Not one that can read a spreadsheet.

Exactly like what you said. Very fun, very fun way. Yeah. This year I'm not going to do that boring 45 minutes keynote, but we're gonna do something very, very fun. But yeah, I think again this year, main focus, fix our stuff, get people happy and keep our promises. Keep our, keep our goal of getting the agent right. Yeah.

And.

And end of this year, early next year, putting all the agents in one place, introducing this, this new system. Yeah. And then we follow up by a couple of other form factors. That's kind of like the big pictures.

Thank you so much.

Thank you.

Rob, where should people go to learn more about a rabbit? It's 200 bucks, right?

Yeah. So best, best way, again, I would argue X is really the best platform to get the news. Follow. Follow our official Twitter account or X account. Go to Rabbit tag. We have all the information there. But more importantly, just reach out to me directly. Yeah.

I have 5,800 companies on X, all in AI.

Right.

What's going down?

But probably you have the most orange ones called rabbit. But anyway, thanks, Robert.

Thanks so much.

Thank you.