YouSaid · the spoken record

Jesse Liu

lines on the record
92
first
2024-10-07
most recent
2024-10-07
sittings or episodes
1
sources
podcast

Every line below is reproduced as it was said and linked to the record it came from. Nothing here is summarised or generated. Directory · Search · Corrections

  1. Yeah, yeah. I mean, I don't get legal documents to my door. Maybe I will get one, but maybe they think we are too small. But yeah, we do things in our way. I guess that's what I want to say where we're really done to the ground team. Like, that's my style

    2024-10-07 · Decoder with Nilay Patel · Rabbit CEO Jesse Lyu isn't thinking too far ahead · IDENTIFIED FROM THE TRANSCRIPT

  2. We're going to keep releasing all the stuff. And in fact, we pushed 17 OTA within five months. The other company pushed like what? Three, two, three, four, five OTAs. So I think I really hope people can see us as, you know, we're a bunch of underdogs. Our solution isn't perfect. But it is David versus Gaudi on day one because it's a reality. And don't expect perfect stuff from us because we are not perfect, right? We raise very little amount of money and we're a small team, but we move fast. What we can guarantee is that when Rabbit shows you something, you probably couldn't even find somewhere else. Just like the hardware, just like the playground, or even the very janky day one version of LAM, we are the first company that has Apple Music can be streamed to our device.

    2024-10-07 · Decoder with Nilay Patel · Rabbit CEO Jesse Lyu isn't thinking too far ahead · IDENTIFIED FROM THE TRANSCRIPT

  3. Very unusual way that they have this new phone, but all the things are still not there. We're making a very small dent, right? But that even doesn't matter. I think for us, we care about our customers. One thing I want to say is that yes, there are a lot of misinformation, there are hates, there are all that feedbacks, criticisms. If you talk to the R1 user, they're happy. That's what I care. That's what I care. Otherwise, there will be a lot of returns. There will be a lot of refunds. Have license on 5% return, put that term in any consumer market, electronics device. It's a good benchmark.

    2024-10-07 · Decoder with Nilay Patel · Rabbit CEO Jesse Lyu isn't thinking too far ahead · IDENTIFIED FROM THE TRANSCRIPT

  4. But there are a lot of people that they say, oh, I'm a founder, I'm cool, no, I've grown enough to get rid of that, probably the same way as if I'm 21, 22, but now I'm like 34. Startups are really tough. It's a war. It's about survival, right? It's really, really tough. It doesn't really matter if others want to do something like whatever. You have to be survived. Just survive by your own is tough. in any sense. So that's why a lot of people ask me, I got asked a lot, like, okay, what if they do this? What if they do that? Well, end of day, there's nothing you can do. You have to do your thing. And they will react to it. I think it's fair to say that with rabbit other startups like us, biggest company like Apple, they react to us. They react to us in a very hustle way, right?

    2024-10-07 · Decoder with Nilay Patel · Rabbit CEO Jesse Lyu isn't thinking too far ahead · IDENTIFIED FROM THE TRANSCRIPT

  5. What ifs, and I make short, clear, concise decisions based on current fact. And in fact, if you do the recap for what we launched back in the CES, it was probably the best timing. The price is probably just right, the color probably just right, right? And the decisions of not negotiating, spend six months negotiating with T-Mobile is probably just right. You know, like I make on current decisions and that's my style. And I talk to people, everyone talked to me. I told my team, you know, everyone in my team, they can find me anytime, talk to me anytime. I spend a lot of people, a lot of time talking to my people. And it's just, we are in general just a very real team like down to earth. And I really don't like some of the other type of startups that they kind of like spend too much time enjoy the feeling if you understand.

    2024-10-07 · Decoder with Nilay Patel · Rabbit CEO Jesse Lyu isn't thinking too far ahead · IDENTIFIED FROM THE TRANSCRIPT

  6. Predictions is that I don't like to make predicts. Like what happens if this happens, if that happened, what do you think? Like, I think when I manage my team, I tell people we make decisions based on current fact. And we find the best solutions to it. If you spend too much time, at least if I spend too much time think about what if Apple knocks on your door, what you're going to do, and what if this A happened, then B happened, then C happened, what you're going to do. Most likely you're going to get a different strategy, right? Because if you think about if B is a solution to A, when A happens, you just do B. But there are other type of people that are like, hold on, have you ever thought of when A happens, then D happens, then E happens, then F happens? Are you still going to do B? If you think it that way, probably not, right? So I just choose not to predict a lot of.

    2024-10-07 · Decoder with Nilay Patel · Rabbit CEO Jesse Lyu isn't thinking too far ahead · IDENTIFIED FROM THE TRANSCRIPT

  7. I'm a very intuitive person. I like to trust my intuition on big directions like what's going to happen in the long run. But meanwhile, I'm quite conservative that I hate to predict things. So I think people when people replay this episode, they will hear probably I got really tricked by some of your questions. It's just my brain couldn't work for like.

    2024-10-07 · Decoder with Nilay Patel · Rabbit CEO Jesse Lyu isn't thinking too far ahead · IDENTIFIED FROM THE TRANSCRIPT

  8. But I'm really, I can't wrap my head around on have to put a frame that doesn't fit. Yeah, we'll see. I think even the current smartphone, I think, is perfect. I really like the state of a glass or a screen phone factor. But the real problem here is not about the phone factor. The problem is about the apps, right? Because now we see all this agent technology, AI stuff. They're doing things that app are doing and they're doing things that apps can't do. So I think the problem is with apps.

    2024-10-07 · Decoder with Nilay Patel · Rabbit CEO Jesse Lyu isn't thinking too far ahead · IDENTIFIED FROM THE TRANSCRIPT

  9. So, I think I'm very, very conservative on the hardware form factor. Talking about a glass, that's a different story. I think your skull actually grow to fit the frame, not the other way around. Because I used to wear like prescription frames. I know the pain. Like your scroll is growing to faded glass frame. Now they are the way around. So I think there is really no generic fit on the glass frame. I was having fun with my design team joking. I'm like, maybe if we do the glass, we probably do the dragon ball style, like the power. Or whatever that is, but

    2024-10-07 · Decoder with Nilay Patel · Rabbit CEO Jesse Lyu isn't thinking too far ahead · IDENTIFIED FROM THE TRANSCRIPT

  10. Know you're gonna push, you know, a wheel probably can scroll. There's a screen you can look at things. So the R1 phone factor is very conservative in the sense that it derisks the software. It's just like people haven't figured out how to interact in a virtual world. And all of a sudden, back in 2016, there's like 200 different companies making goggles, right? And they all fell.

    2024-10-07 · Decoder with Nilay Patel · Rabbit CEO Jesse Lyu isn't thinking too far ahead · IDENTIFIED FROM THE TRANSCRIPT

  11. Yeah, I'm not against any phone factors. In fact, I really think that There will be a lot of fun factors, but We were trying to design our one, the reason is that we know it's not going to be a smartphone because we know people are going to do a lot of other things on smartphone, which the current AI cannot do. So we deliberately avoided smartphone phone factor. Talking about pins with lasers and glasses, I have different comments for each phone factor because there's no universal rules here because let's talk about pins, right? So like I think my general pushback for making it as a pain with a laser like humane, I think, first of all, I think it's really cool. But I think it's too risky. You are trying to offer a new way of utilizing your technology to have user use software. And that's already new to them. And you don't want to just introduce a sci-fi kind of type of gear. So two new things stack together. That's too risky. So if you look at R1, it's like very familiar design. You know, there's a button.

    2024-10-07 · Decoder with Nilay Patel · Rabbit CEO Jesse Lyu isn't thinking too far ahead · IDENTIFIED FROM THE TRANSCRIPT

  12. I would say that I'm very optimistic in the three years term. But I think, like I said, right now and starting of next year is everyone trying different approaches and we'll see which one works. But I think we're confident on the approach that we're taking right now.

    2024-10-07 · Decoder with Nilay Patel · Rabbit CEO Jesse Lyu isn't thinking too far ahead · IDENTIFIED FROM THE TRANSCRIPT

  13. Think the AI model will get very smart very fast. But I think we're talking about a generational shift. I think obviously we don't want a 2024 piece of technology operating on eBay's website, which is basically designed back in 1990, right? So I think a lot of the infra needs to be refreshed. And the biggest gap as I can see here is productionized. So I think in our roadmap, we think that it's very likely that we can get all this separate piece of technology we have like land playground teach mode and rabbit OS at some point maybe next year kind of like merge into a new RabbitOS 2.0 and that actually will push a huge step forward towards this generic goal but my general take is that AI model is smart enough But the action part is a lot of infrastructure. There's a huge gap between research and production ISO. That's what we learned.

    2024-10-07 · Decoder with Nilay Patel · Rabbit CEO Jesse Lyu isn't thinking too far ahead · IDENTIFIED FROM THE TRANSCRIPT

  14. They're actually giving a hand and legs for the AI to do things. I think it is an entire human's effort and a lot of the resources can be shared instead of each company has to go race for this amount of money and take that amount of time to achieve the same goal. So it's really hard to say, but we know we need more money and resources, that's for sure. But I think you've seen how efficient this team has been performing. From seven people, 17 people till today, like we raised obviously much less than humane or any big companies out there. I think it's actually one of our advantages that we can do things in a relatively cost-efficient way and fast.

    2024-10-07 · Decoder with Nilay Patel · Rabbit CEO Jesse Lyu isn't thinking too far ahead · IDENTIFIED FROM THE TRANSCRIPT

  15. We have this very limited budget is that how can we convert the latest technology and research and build to a product that we can ship early and collect feedbacks and learn from it? So, a lot of people have a different definition of AGI. I don't really talk about this term because I think so many people have so many different definitions for it. But I do think that AI understand what you say and help you do things. And maybe, you know, here we're talking about virtually help you click buttons and stuff. There are a lot of companies doing human Android.

    2024-10-07 · Decoder with Nilay Patel · Rabbit CEO Jesse Lyu isn't thinking too far ahead · IDENTIFIED FROM THE TRANSCRIPT

  16. I might have a different opinion here. I think Foundation models like OpenAIs, you know, obviously they're raising for a crazy amount of money. I think we take benefit from what they've been worked on, right? Because their primary services is selling their models as APIs, which saves a lot of money. We don't want to recreate retraining like an OM. It might not be as scary as a lot of people might think. I think there's a huge gap between converting the latest technology into a piece of product versus pushing for a more advanced technology. Obviously, I'm very proposed to do like high-end research. We want to have a research house here set up at the same scale as OpenAI and deepmind, even though they already far, far behind. But I think what we're trying to do right now at this current scale, because here's the money we have, right? We don't have like $1 billion. We don't have like $2 billion.

    2024-10-07 · Decoder with Nilay Patel · Rabbit CEO Jesse Lyu isn't thinking too far ahead · IDENTIFIED FROM THE TRANSCRIPT

  17. We'll take a look and we'll see how we adapt. But the agent is out there yet already. There's a lot of agent running already. So I think there's no turning back and it's very unlikely they say, hey, we're going to stop agent using our services. That's not going to happen.

    2024-10-07 · Decoder with Nilay Patel · Rabbit CEO Jesse Lyu isn't thinking too far ahead · IDENTIFIED FROM THE TRANSCRIPT

  18. Yeah, that's exactly what I'm talking about. I don't think that they are not willing to change their terms. And I think unlikely they're going to put terms like it has to be a human. It cannot be. There's a lot of automation tools out there already, right? So like where there's no turning back. I think what they would like to work with any companies, including us, is that when they see a popular demand from this new kind of agent technology, they want to charge for it. We ask our user and us to pay for them. And that's a business deal. That's more like a money terms. That's what I can see. But yeah. For now, we're not breaking any of their terms and agreements, right? And if they change the term and agreements tomorrow,

    2024-10-07 · Decoder with Nilay Patel · Rabbit CEO Jesse Lyu isn't thinking too far ahead · IDENTIFIED FROM THE TRANSCRIPT

  19. No, at the moment, we don't have the resources to fight that fight. And at the moment, that's not a real threat to us because they said we're too small.

    2024-10-07 · Decoder with Nilay Patel · Rabbit CEO Jesse Lyu isn't thinking too far ahead · IDENTIFIED FROM THE TRANSCRIPT

  20. There's a YouTube training video can be used by this or that. There's all sorts of things like this. Yeah. But I think it's not just the builder are adapting, the industry are going to adapt to the builder too. At some point, there's going to be a conclusion that, okay, this is a new policy. This is a new kind of like terms that we need to use.

    2024-10-07 · Decoder with Nilay Patel · Rabbit CEO Jesse Lyu isn't thinking too far ahead · IDENTIFIED FROM THE TRANSCRIPT

  21. I don't think anyone in AI in the industry can give you a very definitive answer, be like, hey, if you just do this, here's the structure. This is going to guarantee you the best result in the long run. I think that's not a good way to think of it. But yeah, I agree. Everyone in the industry are experimenting something new and a lot of companies that we saw are going to, like you said, run into some sort of legal problems. There's music generation platforms.

    2024-10-07 · Decoder with Nilay Patel · Rabbit CEO Jesse Lyu isn't thinking too far ahead · IDENTIFIED FROM THE TRANSCRIPT

  22. We are confident that this technology is the current technology route that it will work. And I haven't yet to see another approach that actually make any generic agent system work in any other manner. That doesn't mean that we're locked in to one technical path. If you talk to any company, it's probably not a smart idea to say, hey, we just bed this for next 10 years. The technology changed so fast, you have to adapt. Right now, I think we're off the good start, we launched a concept with playground, with free off-charge that you can explore so that we understand how this system can be improved. In fact, I believe the But we're not here to say, hey, we stuck into this. We do have patterns about this, but we're not saying, hey, we think this is a correct path to go.

    2024-10-07 · Decoder with Nilay Patel · Rabbit CEO Jesse Lyu isn't thinking too far ahead · IDENTIFIED FROM THE TRANSCRIPT

  23. Something will happen with Samsung. As that universe of competitors expands, it feels like the core technology you're betting on is being able to automate a VMC with a large action model, right? You're going to open up user sessions for people in the cloud, and then your LAM is going to go click around on the web for them, and that will get you out of the challenges of needing to strike API deals with various companies, with other kinds of deals, copyright deals with various companies, whatever you might need. Is that durable, right? The idea that this will keep rabbit away from needing all of the deals, the big companies will just go pay and get. Because that's the thing that I think about the most. I can think of 10 companies that came up with a technical solution to a legal problem. And even if the technical solution was amazing, the legal problem eventually caught up with them.

    2024-10-07 · Decoder with Nilay Patel · Rabbit CEO Jesse Lyu isn't thinking too far ahead · IDENTIFIED FROM THE TRANSCRIPT

  24. Then there's an article by the Verge just use that title 5000, which is wrong. I can tell you that's wrong. That's very wrong. That's misinformation right there.

    2024-10-07 · Decoder with Nilay Patel · Rabbit CEO Jesse Lyu isn't thinking too far ahead · IDENTIFIED FROM THE TRANSCRIPT

  25. We really are hoping for people to discover more use cases, but unfortunately we have like four or seven apps on the connections, right? That's one of the bottleneck. So if you check for the total query, most of the cases you ask a question, you forget about it, right? So it's not about how many times you ask R1. It's about what kind of task you ask R1 and is R1 actually going to help you. Guess, yeah, very unfortunate. It seems that that's a misinterpretation. So, what's that number

    2024-10-07 · Decoder with Nilay Patel · Rabbit CEO Jesse Lyu isn't thinking too far ahead · IDENTIFIED FROM THE TRANSCRIPT

  26. I said it can be misinterpreted. Okay. Yeah, first of all, I think we saw a very steady growth of all the people interacting with R1 and each time with new features there's going to be more people using it. I give you some numbers that I want to throw to you, and maybe I can share a very detailed usage sometimes in the future. First of all, there are about 5% people. They have their R1, they're not happy that return less than 5%. Which is a very good number. And I think the top features that people are using are asking questions and visions and all that.

    2024-10-07 · Decoder with Nilay Patel · Rabbit CEO Jesse Lyu isn't thinking too far ahead · IDENTIFIED FROM THE TRANSCRIPT

  27. I said there can be misinterpreted. What I said is that if you go look at the data doc right now, you probably will find 5,000 people Using R1, at least five. I'm just quoting.

    2024-10-07 · Decoder with Nilay Patel · Rabbit CEO Jesse Lyu isn't thinking too far ahead · IDENTIFIED FROM THE TRANSCRIPT

  28. Monthly fee, but I really don't find any reason to upgrade that because people are talking about Rabbit being launched too early. Now you have the company like Apple, if you go to Sunset Boulevard in Los Angeles where it's close to here, or I guess Mission Street in Santa Monica in San Francisco, you go to any major city, you see this gigantic posters, billboards that Apple puts there, right? iPhone 16, iPhone 16 Pro. What are the other lines underneath? It says Apple intelligence. Is it ready? Is it out?

    2024-10-07 · Decoder with Nilay Patel · Rabbit CEO Jesse Lyu isn't thinking too far ahead · IDENTIFIED FROM THE TRANSCRIPT

  29. But obviously, I also believe when big companies want to kill startup, they have a million ways to kill you. That's just the reality. I think people keep talking to me and asking questions. What happens if the risk is too high? What happens if the company dies? I really don't think that all these questions matters because we're on this course. We're going to see the end, whether it's a good end or bad end. And I don't think any answer to this question will change our course, to be honest. I can go here and tell you and be a crybaby like this is super hard. This is impossible. Everyone in the answer can tell us easy or the YouTube reviewer can kill us by posting a video. It doesn't change the course because we are doing things. We're launching, we're shaping things, right? We're moving forward. So it would be interesting to see what Apple came from. I was on the Apple iPhone upgrade program. So I automatically get a new iPhone every year by paying the same.

    2024-10-07 · Decoder with Nilay Patel · Rabbit CEO Jesse Lyu isn't thinking too far ahead · IDENTIFIED FROM THE TRANSCRIPT

  30. We have a spectacular launch. We are the most selled dedicated hardware yet. And we have make good profit. We fix all the day one problems and the company actually controlled the size. So we're growing, we're moving fast. And now we drop this. I think, like you said, put a marker between today and yesterday. I think today I can say a lot of things that you can do on R1. You cannot do it on the iPhone, right? I believe eventually everyone will be able to come to the same solution that all the device can do same similar stuff. But I firmly believe at least this remaining half the year or the Q4 of 2024 and probably the Q1 2025, it is still a game of you have something that they don't have versus you have you guys all have the similar stuff with done better. So I think relatively we have a good six to eight months ahead of start. Like we have our little room here.

    2024-10-07 · Decoder with Nilay Patel · Rabbit CEO Jesse Lyu isn't thinking too far ahead · IDENTIFIED FROM THE TRANSCRIPT

  31. That can become a definitive recipe, including Apple. I think the benefit for Apple to do that is that, yes, they understand user better, much, much better than any complaints out there. And they have infinite money, theoretically infinite money, and they have a very close ecosystem. The way that they're rolling this out is that they have this SDK called App Intent, right? Companies or app developers need to choose to enroll or not enroll with that to have the new serial control stuff. I guess my relative advantage as a small group as rabbit is that we move fast. We move fast and we keep growing. I think if we put all the cars on our table

    2024-10-07 · Decoder with Nilay Patel · Rabbit CEO Jesse Lyu isn't thinking too far ahead · IDENTIFIED FROM THE TRANSCRIPT

  32. So I said as if right now, I don't think we have serious competitors from startup. And then when we talk about competitors, obviously there's Apple, there's every big companies out there, including OpenAI. First of all, I think this is good for us because it validates our direction, it's absolutely correct. And I also are curious about what are going to be the definitive route for the generic agent technology because different people in the industry might have different ideas, right? There are still debatable states. There is no evil for agent systems yet. There's no very good evil yet. And you can see a lot of different research houses and companies trying different routes. Obviously, there's API routes like GPTs, which doesn't really take off. There's pure neurosymbolic routes. There's Hebrew routes. There's all this multimodality. So we're still in the phase of everyone trying their own recipe and hopefully.

    2024-10-07 · Decoder with Nilay Patel · Rabbit CEO Jesse Lyu isn't thinking too far ahead · IDENTIFIED FROM THE TRANSCRIPT

  33. Of all speaking for myself, I've sold my company before when I was 25. I don't want to build another app. I should chase my same dream because I really think that the grand vision that I have and our team was working on is actually the current direction everyone's chasing. And it just feels so bad if you don't chase the same dream, no matter how hard it is, really. And in reality, we feel blessed and happy to say the exact situation because we don't have any serious competitors from startups, to be honest.

    2024-10-07 · Decoder with Nilay Patel · Rabbit CEO Jesse Lyu isn't thinking too far ahead · IDENTIFIED FROM THE TRANSCRIPT

  34. To use this feature, we don't care about how to do it. In fact, because we know that you don't care how that's been done, I don't want to spend six months, eight months sued up to talk to Spotify people and Uber people. And one by one, let's do that, right? So.

    2024-10-07 · Decoder with Nilay Patel · Rabbit CEO Jesse Lyu isn't thinking too far ahead · IDENTIFIED FROM THE TRANSCRIPT

  35. And you have to chase for the services people want. We want to offer this music feature. And we think Spotify is the best experience overall. And we want to chase for this partnership. And we're still chasing for this partnership. But to talk from technical perspective, why I said I don't like APIs because think about Alexa. Alexa speaker are all using APIs. And you literally have to go there and negotiate because like I said, today not everyone is opening APIs. A lot of the traditional services don't have API. And then startups, for startup, it's impossible. When you go talk to them, they think you're too small, right? We did that. We just did that to everyone. They think we're too small. They don't care. So we can't get API. Does that mean that we're not going to figure out the alternative way to make it work? No. Hell no. We're going to make it work. And this is exactly how we make it work. So we care about you.

    2024-10-07 · Decoder with Nilay Patel · Rabbit CEO Jesse Lyu isn't thinking too far ahead · IDENTIFIED FROM THE TRANSCRIPT

  36. Don't need to care. We need to care. We need to care. And we need to care because we checked what are the good APIs we can use. Don't get me wrong. Perplexity API has been great. OpenAI break every day or two on the Very detailed breaks per day is more than, I guess, more than 10 on average that ChatGPT API breaks or instable, whatever it takes. We have a notifier. So API, first of all, API is not stable. It is not stable.

    2024-10-07 · Decoder with Nilay Patel · Rabbit CEO Jesse Lyu isn't thinking too far ahead · IDENTIFIED FROM THE TRANSCRIPT

  37. That's my second part. Yes, API is 100%, but you are relying on they gave you the API that's stable, that works. I'm the user

    2024-10-07 · Decoder with Nilay Patel · Rabbit CEO Jesse Lyu isn't thinking too far ahead · IDENTIFIED FROM THE TRANSCRIPT

  38. I really want hard. Two things. I disagree That the Spotify is not working good. Spotify has been working amazing. Maybe two times we put on a maintenance, a total amount of time we put on a minus probably under one hour. You can ask any R1 users. That's not through API, which is impressive. That's through agent.

    2024-10-07 · Decoder with Nilay Patel · Rabbit CEO Jesse Lyu isn't thinking too far ahead · IDENTIFIED FROM THE TRANSCRIPT

  39. I'll tell you, Spotify changed interface all the time. And I think in the past six months, five months since the first lamp was adding to Spotify with the connection since launch, I think we probably put the Spotify under maintenance for like maybe two times one hour in total. A very hard proof, yeah.

    2024-10-07 · Decoder with Nilay Patel · Rabbit CEO Jesse Lyu isn't thinking too far ahead · IDENTIFIED FROM THE TRANSCRIPT

  40. I'm not sharing my credentials to my body, right? I'm not telling my phone password. I'm not telling my DoorDash password. I'm not even sharing my credit card info. All he do is just add to the card and click confirm. That's it. So this guy is equivalent of the first generation of LAM, which is unfortunately we don't like it. So that's why we work so hard. Now we have playground, which is more generic technology.

    2024-10-07 · Decoder with Nilay Patel · Rabbit CEO Jesse Lyu isn't thinking too far ahead · IDENTIFIED FROM THE TRANSCRIPT

  41. Which is very aggressive. That is taking screenshots of your local computer. This is what I thought was happening in AI in the early days. There's going to be a lot of different takes and tries. And eventually people will reconcile and agree on single piece of terms and agreements. But if you check how we automate the website to their interface, most important part is we don't create fake user. We don't create spam user. We don't log in on your behalf. And you are you the way I help you to do things is by helping you click the buttons on mouse. It's equivalent of if I want my body to help me Give you an example. So if I'm busy, I'm about to head into the meeting. I want my buddy to help me order a burger from DoorDash. All I need to do is I unlock my phone, I pass my phone to my guy, and my guy helped me click that. And in this process,

    2024-10-07 · Decoder with Nilay Patel · Rabbit CEO Jesse Lyu isn't thinking too far ahead · IDENTIFIED FROM THE TRANSCRIPT

  42. Don't care until everyone's building smart speakers, they're like, okay, we have to resell the whole copyrights for this particular phone factor. I guess end of day is about money, right? They want to sell the same copyrights to as many phone factors as they want if there's the popular one. So we're okay to have this kind of negotiations, but certainly like you said, there's bigger companies that are doing similar things or even more advanced things that needs to be addressed. I give you another example, like, you know, Siri and Microsoft, there's a feature called Microsoft Recall, right? Which they pull back that feature now. And I think they relaunch it.

    2024-10-07 · Decoder with Nilay Patel · Rabbit CEO Jesse Lyu isn't thinking too far ahead · IDENTIFIED FROM THE TRANSCRIPT

  43. Have other Asian companies like Motion, even the GPTs are doing this, right? So this is like a new way emerging for all this old services that they have to think about. But I can tell you my personal experience dealing with scenarios like this, like when we first start building one of the first smart speakers back in like 2013, you know, like all this music label, they don't care.

    2024-10-07 · Decoder with Nilay Patel · Rabbit CEO Jesse Lyu isn't thinking too far ahead · IDENTIFIED FROM THE TRANSCRIPT

  44. I guess there's always a way that you can detect, but I think the question is, this is actually a very good topic that we're talking here, is that, you know, think about captures. Lamb playground or any capable AI models now can go there and solve text-based captures. So their old system to prevent automated systems like this are currently fouling. And this is an industry effort to push everyone in the industry to rethink about now with this AI, now with all this agent, how their business is going to reform or how their business, how all these policies need to be changed. I do agree this is a very complicated topic. But what I can see is that this is not rapid doing some really fancy magics here. Like every company is doing this.

    2024-10-07 · Decoder with Nilay Patel · Rabbit CEO Jesse Lyu isn't thinking too far ahead · IDENTIFIED FROM THE TRANSCRIPT

  45. It is a little bit more technical to achieve that, which we are working on the other platforms. I think I showed very specific group of people of a working prototype that LAM is operating on the desktop OS, such as Linux with all that local apps. So we're definitely heading to that direction.

    2024-10-07 · Decoder with Nilay Patel · Rabbit CEO Jesse Lyu isn't thinking too far ahead · IDENTIFIED FROM THE TRANSCRIPT

  46. It's not a problem for now. We'll see how this problem evolves. But I remember When Apple is relatively not so big, I mean, not as big as today. When I read Steve Joss book, there's one chapter, he said, okay, go talk to Sony from tomorrow, 99 cents per track, right? Remember that moment. So at some point, this level of negotiation needs to be happened. I'm not sure if we're leading this or someone else is leading this, but this is the working proof that we're not using API. And I don't think the services are not building API just because they're trying to prevent people from automating the company, just because API to them is not making money. And they for sure will love to set up a negotiation in some phase later when we grow bigger. But I guess, you know, we try to reach out to Uber with it.

    2024-10-07 · Decoder with Nilay Patel · Rabbit CEO Jesse Lyu isn't thinking too far ahead · IDENTIFIED FROM THE TRANSCRIPT

  47. I see where, first of all, I see where you're going. Before you go deeper, I just want to say, first of all, we're not using API. Second of all, to say LAM is not there, That's false Because we have all these services. If you really pay attention to their documentation, there is no API for DoorDash. There is no API for Uber.

    2024-10-07 · Decoder with Nilay Patel · Rabbit CEO Jesse Lyu isn't thinking too far ahead · IDENTIFIED FROM THE TRANSCRIPT

  48. We are basically gave everyone a virtual machine, which is a VNC. Which is 100% within policy. And you have the rights to access that VNC. And on that VNC, we basically work directly on website, just like today's land playground. So we're not getting the audio from server, from spotted firmware somewhere else. We're basically going to Spotify website and play and do the things for you and play that song for you.

    2024-10-07 · Decoder with Nilay Patel · Rabbit CEO Jesse Lyu isn't thinking too far ahead · IDENTIFIED FROM THE TRANSCRIPT

  49. I've always been very curious about this. I've been dying tasking these questions. So I ask my R1 to play a song somewhere in AWS, a virtual machine fires up, opens a web browser, opens Spotify, logs into my Spotify account using my credentials, clicks around on Spotify, pushes a button to play a song, and then you capture that audio and restream it to me.

    2024-10-07 · Decoder with Nilay Patel · Rabbit CEO Jesse Lyu isn't thinking too far ahead · IDENTIFIED FROM THE TRANSCRIPT

  50. Have a conversation, they realize this is agent behavior. And we said, look, we ask user to log in on your website. And they are 100% legitimate user and they're a paid user. And when we do the trick, we help them click the button.

    2024-10-07 · Decoder with Nilay Patel · Rabbit CEO Jesse Lyu isn't thinking too far ahead · IDENTIFIED FROM THE TRANSCRIPT