Rabbit CEO Jess Lyu on the high-risk, highly-hyped r1 AI companion | StrictlyVC LA

Uh Devin coldway again my esteemed Colleague from Tech crunch guys thank You so Much thank You you good yeah all right uh Jesse Thank you so much for coming out uh we Met briefly at CES uh but this is your First like event since then right yeah Um we're moving extremely fast you know Spend every second building and this is The only Event uh that I attend yeah I it's Really it's really awesome thank you for Uh thank you for making it out so I Wanted to like right at the start here I Want to if there's anybody out there who Thinks this thing doesn't exist you got One in your pocket right now Right I think So there it is yeah there's the thing um You know one thing I want to say is that The the the digital camera is really Hard to get the colors right you have to See it um but we we chat it about after The the the the speakers yeah so we'll Uh yeah we're going to we're going to Play around with this backstage but then Like you can go and circulate later um Don't don't rush the stage uh it's you Know it only responds to his voice Anyway uh so uh tell me uh why make this Device this is the sort of question Every is like Connie said like why not Make an app why why not do something we

All have powerful Gadgets in our pocket Why make this device and why make it Now well I'm I'm not sure just that's Question alone I can finish answer it Within like 18 minutes but uh I'll keep Yeah I'll keep it short I think first of All the application based operating System um is dying there's no question About that um from user perspective User experience to be exact that's Where that's the start of the thoughts Right because we Realized how challenging and how Difficult even for uh younger generation To you know learn how to use different Apps and you know solve a single problem Using four or five different apps and Jump back and forth um it just it's just Not intuitive enough and uh you know I Had this vision many many many years ago Actually 10 years ago um but the Technology won't ready I think with the Generative AI especially with you know Major breakthroughs like Transformers And all this key technology uh this is The first time in history that devices Like this is actually possible uh now Speaking of the R1 in specific uh Because we know that this is a brand new Territory and this is a new way of Interaction we want to get get it right I mean there is natural language um Applications like Siri and Alexa and Google many many years ago and we don't

Think that they get the experience right So we're extremely cautious to make sure That we you know we want to think Backwards we want to say hey can we Reach the bar of the user experience if We can do that then let's lower the risk Because you know it's really really hard To convince people get extra devices You're going to have your phone Regardless um so like I said before R1 Is a result of drisking we want to say Hey let's don't make a um ,000 device Let's don't make a $500 device uh with Subscriptions let's make a $200 device With no subscriptions uh we want to loow Not a $100 device or a $50 device I mean You it feels like you kind of just named A number and maybe that was was it was It unrealistic do you think no uh I mean Hardware you put all the parts together You have a bomb cost right so uh lower Than this is become really really Fragile uh and trust me on that I I I Want to offer you guys a $5 device but That's just not possible I think you Know it's the hardware is a compromise Between design functionality build Quality and user experience I think Ry Is a very good balance so and you're You're shipping the plan is to ship in The end end of March beginning of March So we're really close to 100,000 orders Um but uh us uh first batch us orders Starships on uh March 31st which is in

About a month and 100,000 pre-orders and You were expecting what were you Expecting like 10,000 20,000 I don't Know we I mean two days before the KE Note I I told my team it'll be really Nice if we can sell 500 units on day one Uh but we sold $18,000 on uh1 18,000 units on day one So um we immediately ramped up the stock So we're not going to have any delay and We ramped up the stock and we're ready For 100,000 units and the the design Certainly I want to talk about the the Software but I Al but since we're Talking about the hardware right now Uh Teenage engineering which you're on The board of right uh what is it like Working with them to build something Like this because I mean I'm familiar With them from their audio devices and Interfaces uh but this is still like This is sort of outside of their their Sort of their normal bounds well first Of all team engineering is my hero Company I I I collect vintage sensor Sizers uh for past 15 years and they are My hero company because ever since I Bought the op1 you know I become I'm a Huge fan of them uh so just to clear the Record teenage engineering is not a Design agency right the company doesn't Generate Revenue by you know taking Design contract they're pretty much Logged in in their own territory which

Is audio um and portable synthesizers Which they're extremely good at um but Uh you know they rarely do some of the Collaborations you know we we had the First collaboration s years ago um but To work with uh Teenage is just very Very intuitive I mean we don't have Email we don't have you know slack we Have a secret Instagram account that we We we we post uh sketches and renders And someone find this account find the Secret Instagram account and send it to Me Devon Tech crunch.com or Jesse you Can just give it to me later yeah I Think once we ship our one to customer I Think I might be able to share some of The screenshots but to work with them is Really really magic confirmed it's Coming to me uh watch techcrunch.com for That uh so I can't remember if I Interrupted you or not so we're just Going to move on to the software uh so The the large action model this is your The sort of the centerpiece of the whole Thing is this what you call large action Model and we've heard we've heard a Little bit about it and we've heard you Say basically that is only possible as Of you know the last year year or so so Tell us a little bit about it is it Transformer based is it does it what's What does the training data look like Give us a little of the secret sauce Here yeah so uh uh first of all large

Action model is not a branding it's not A fancy name we just put together um we Got access for GPT apis in very very Early days uh I think it's actually Around 4 years now uh so we got a first Access I saw the first version of the GPT back in 2019 uh at open Ai and uh um I immediately thought okay because if You trying to do something like this There's two part there's intention There's action and I started my previous Company I work really hard on what we Call NLP natural langage processing back In the days um to solve the intention But I think it's roughly around early 2020 I got convinced that the Transformer will solve 100% intention Like chbt understand exactly what you're Trying to talk about right um then we're Immediately trying to use super promps To force this language model to do Things and the result is very miserable Because um yes in theory you can you can Force the language model to you know Work on the tokens and kind of like get To a point that you know there's a demo From another company be like use l to go To Mr B's latest YouTube video and leave A comment uh yes in theory language Model can do that but at what cost you Have to literally watch your screen Doing that step by step and it takes Roughly around 2 three minutes to finish One task like that uh we just don't

Think that how that can convert into a Good end user experience so then we're Like okay hold on let's take a step back Like we should do a AI generic AI that Just focusing solving on actions instead Of of intentions because intentions is What we call large language model and Transformer and when we trying to solve Actions there's couple of options There's one way you can do like gpts or Plugins you can work on with apis um but As a startup that's obviously not the Best way to move forward then we started To look at neuros symbolic we generally Designed a method to where we collect Real human working on most frequent apps With our data collector uh vendors so we Work with third party companies back in 2020 uh we start asking data collectors To help us gather uh real people Interacting with Spotify Uber xedia you Name it so we have a top 800 uh highest Frequent apps and then we colnect those Recordings of real human interacting With different interfaces and then we Set up this uh neuros symbolic Network And as this a AI which now we call large Action model to review those clips but Frame by frame so in General the idea is that symbolically The AI will be eventually smart enough To extract all the buttons all the Elements using symbolic method and then We can basically build a logic to

Automate Up and a good ide a good example uh Which is not quite as similar I'm trying To get less damic here uh but if you Think about Tesla uh camera based Autonomous driving I mean Tesla is not Hiring Engineers to hard program what is A stop sign right they set up I think They get someone else to do all the stop Sign circling so they have Ai they they have their AI to basically Train based on Millions hours of Road Dash cam footage and the AI will be Eventually understand oh when I see this This absolutely 100% a stop sign and Then once the AI learns what is a stop Sign it really doesn't matter whether The stop sign is in 12th Street or or in Hollywood Boulevard a stop sign is a Stop sign so large action model is kind Of like similar approach in in in in Idea wise not necessarily in code based A little little more reliable maybe yeah So we our assumption is if this is a Music player and the music player has a UI I mean all the modern software has a UI and the UI was was purposefully built For human eyes to to see and human brain To understand that in orderers for you To know how to use so it really doesn't Matter let's say if Spotify or apple Music they change their UI they re ramp Their entire stuff there's going to be a Big fat play button somewhere right cuz

Cuz cuz that's that's our assumption and Then we set up this neuros symbolic to Kind of like take them down uh from Frame by frame on what are the real Humans doing on all the s sort of uh Interfaces that's how we trained large Action model and to make it work the Best example is that if I have a Assistant and my system name is Tom for Example I want to order a pizza how but I'm busy I don't have time to do it so On my phone all I need to do is I unlock My phone I pass my phone to Tom I gave Tom I unlock the iPhone and I said hey Tom go get me a piece on door Dash right And Tom does not need to know my Password because I authenticated to him And Tom knows how to use door Dash to Get the pizza so all I need to do is I Talk to Tom Tom gave me a pizza through Door Dash right so in this case Tom is Large action model that's kind of like The idea gotcha and so I wanted to so I Want to talk about that because this is Where people in security and privacy are Like and for instance Meredith Whitaker Who was just up here and just backstage I was like oh yeah and I'm talking with Jesse from rabbit and she was like wait Tell me how it works again and then when I said so she was like it does what and How can they guarantee the SEC this Sounds like a nightmare uh so can you Explain why it's not a nightmare to just

Imagine Meredith is listening cuz she Will watch this later well there is a Lot of misunderstandings to was H Lam You know access on behalf of you uh we Do have a full privacy page on our Website explaining in detail but I Actually kind of like already explained It when you get a device if you choose Not to log in with any of your third Party application account this is by Default the best GPT device out there But if you go to what we call rabbit Hole it's like a website you go to Rabbit hole and you start connecting With your Uber account with your Spotify Account with your door Dash account etc Etc what happens is that we don't save Your credential we never save your Credential because we don't need to and We don't want to and we after you hit Connect Spotify it redirects you to Spotify login page and you you do all The loging process on each of the third Party system not on here rabid OS will Recognize oh this account has been Connected with that particular R1 and You only need to to do that at once so It remembers it connects with it so so Does it keep a does it keep like a Virtual session open in like a you know Web browser somewhere in the cloud or Whatever so all the lar model is running In Cloud but uh we're not doing a Virtual instant for each user we have a

Super uh Cloud host and has a super Advanced Dome we actually had a pattern Around that uh that basically handles All the traffics but um just like what I Described I'm giving my phone to Tom and Tom is ordering a a pizza from door Dash Tom does not need to know my credit card Info or iPhone password yeah I'm worried About like I'm worried about the guy Standing next to Tom who's like looking And being like okay Spotify huh looks Good what was the what did you what was That you just played yeah so like I said We don't even handle payment here Because you don't you don't you don't Bind your credit card info here what Happens is that when you log in for for Instance when you log in with Uber you Get a car and on your Uber settings you Have that saved in their server and it Just charged by your connected card we Don't handle that yeah speaking of uh Speaking of credit cards how are you Going to make money it feels like 200 Bucks uh no subscription and we didn't We talk about this yet but you do use Third- party LMS like perplexity and What have you for for the intent part That you mentioned right there's there There's no no need to reinvent wheel Just do it API to whoever yeah like it It feels like aren't we it feels like You're throwing there's a lot of money Involved here and 200 bucks doesn't

Cover the bill yeah I Mean I think investor ask about the same Question before they decide to invest us Right and they're not dumb So um so first of all um we're not Losing money by selling R1 which is a Very very very significant achievement Especially for new startup on gen one I I gave all the credit to my Hardware Team amazing amazing guys being able to Basically negotiate down to the parts And on the bomb cost so we're making Money selling R1 And we're not looking for a subscription Model because we don't think that's a Good way to go in general um we have a Teach mode right the idea is that with Teach mode using lamb anyone can go There show the lamb how you do to how You how you work on any software for one For one take and the teach mode will be Able to generate a new rabbit for you And then you can choose to publish that Rabbit and then it's like classic App Store logic you can sell your create Creation and others will buy it and then We can take a portion through it but That's in the long run that's not Definitely not day one um but overall You know I I Think yeah we're not going to be Bankrupt by selling more units and I'm Not worried about that and in fact Praxity is not costing us it's

Benefiting us because all the all the R1 Or the first 100,000 R1 they offer $200 uh credit on their uh app version So it's a it's a free onee credit that They offer to us but all the praxity on R1 is free is Grandfathered so if you buy a R1 if your First 100,000 orders the device is Effectively free because you you got Free proxity here forever and then you Got $200 credit so I think what everyone Is trying to do here is really genius Yeah well it sounds good uh except there Are enormous multinational billion Dollar companies trillion dollar Companies that probably want to snap you Up uh Google and Microsoft we talked About this are you know fighting this Proxy battle with open Ai and anthropic And stuff and so I think they're kind of Occupied with each other but Tim Cook Just said like yesterday said that apple Is going to break new ground that's me Making apostrophes uh in AI this year That sounds like to me my suspicion is Siri with an action model that can Operate iPhone Ms and other stuff Otherwise I don't know why they even Bother making a model what are you do You feel threatened by any of these uh These Titans well first of All I'm not delusional to where that we Think that we're not a startup we are a

Startup I mean um one the first lesson I Ever learned from y combinator 10 years Ago is that 99% of startup will die That's the very first lesson I learned So as a startup you're betting the odds Like no question about it um then the Second part is I don't Care like they're going to do what they Going to do I'm going to do what I'm Going to do right like if your mentality As an entrepreneur is that oh I have a Genius idea and I can guarantee this Will work no matter whether all this big Type company whatever they try this is Just fundamental a better idea I mean You're dilutional there's no such thing Like that reality is that startup is a Survival game and you better spend your Time focusing on your own stuff other Than concerning about this this and that And that uh there are got to be some Founders when they heard apple is doing Apple cars they stopped right there got To be some guys out there now what they Just canceled right so I think it's it's It's good to have this level of Competition that only going to help us Grow faster or that Faster which is the nature of startups But what was That it's either or I don't know yet I'm Trying my best to be to like I said it's A survival game yeah live fast die fast The life of a startup well thanks Jesse

Uh we're out of time thanks very much For coming Out


Coinbase is a popular cryptocurrency exchange. It makes it easy to buy, sell, and exchange cryptocurrencies like Bitcoin. Coinbase also has a brokerage service that makes it easy to buy Bitcoin as easily as buying stocks through an online broker. However, Coinbase can be expensive due to the fees it charges and its poor customer service.

Leave a Comment

    • bitcoinBitcoin (BTC) $ 58,690.00 0.68%
    • ethereumEthereum (ETH) $ 3,152.02 0.03%
    • tetherTether (USDT) $ 1.00 0.09%
    • bnbBNB (BNB) $ 530.91 0.81%
    • solanaSolana (SOL) $ 139.66 0.64%
    • usd-coinUSDC (USDC) $ 1.00 0.1%
    • staked-etherLido Staked Ether (STETH) $ 3,156.58 0.14%
    • xrpXRP (XRP) $ 0.534357 14.34%
    • the-open-networkToncoin (TON) $ 7.32 0.18%
    • dogecoinDogecoin (DOGE) $ 0.110230 2.28%