Inside the Rise of Jesse Lyu and the Rabbit R1

Everyone seems entranced by Jesse Lyu, the new AI prophet.

Clad head to toe in black, Lyu stands on stage and fires off live demo after live demo using a bright orange-colored AI gadget called the R1. He speaks to the pocket-sized device and the AI voice assistant responds back.

At first, Lyu starts with some easy requests like looking up the weather and basic search questions. Next, he demonstrates the R1’s computer vision for identifying and describing what’s in front of him and then dives into more advanced functions. He uses the R1’s camera to convert a hand-written chart into a spreadsheet that’s then emailed to him. It happens instantly over a 4G LTE connection.

The crowd of several hundred early adopter techies explodes with cheers at the speed and perfect conversion from the AI. Lyu streams music from Spotify, orders McDonald’s via DoorDash, calls an Uber, generates images from Midjourney — all using his accented English (no wake words like “Hey Siri” or “Alexa”) which could typically trip up other voice assistants.

The 34-year-old founder and CEO of Rabbit Inc. had just wrapped up a splashy launch party for the R1 the night before at the TWA Hotel adjacent to John F. Kennedy Airport in New York City. Many of the attendees had flown in from across the country to meet Lyu and pick up their pre-ordered R1 devices.

When we meet up at a WeWork in midtown Manhattan he’s looking a little tired but tries to hide it underneath his baseball cap, which appropriately has the words “Out of Office” stitched on its front. He tells me he only got a few hours of sleep and I believe it because he and Rabbit’s chief technology officer, Peiyuan Liao, were up past 3 a.m. helping me troubleshoot my R1 setup issues. They were also busy responding to the feedback and out-of-box problems customers who picked up their orange-colored devices were sharing to social media platforms like X.

“I know exactly what to improve.”

“In my mind, I’ll be very happy if the R1 gets a 6/10 review score, today, in the first 24 hours [of it being released],” Lyu tells Inverse. “I know exactly what to improve. We’re going to fix the battery [drain issues] and fix the clock [displaying wrong time zones].”

Lyu’s profile began to rocket at the start of the year when Rabbit “won” CES with the introduction of the R1. The AI gadget, which I likened to a real-life Pokédex (the gadget that fictional Pokémon trainers use to identify new monsters), stole attention from iterative versions of devices we already have such as TVs, laptops, and speakers. That a startup nobody had ever heard of could come out of left field with a device that — at least on paper — seemed to make artificial intelligence seem desirable with a new handheld device told Lyu he was on the right track.

While Rabbit and the R1 are fresh to the consumer tech world, Lyu is not. Prior to forming Rabbit in 2021, and from the moment he left his hometown in Xian, China to go to university in the UK, he began working his way through the industry. He’s worked on apps, which led to startup entrepreneurship, which led to his last company, Raven. In 2017, Lyu launched a robotic smart speaker co-designed by Teenage Engineering called the Raven H in partnership with Baidu, in China. The Swedish consumer electronics and industrial design company, Teenage Engineering, doesn’t do design work with everyone. It has a certain aesthetic that can be described as retro and technical-looking, but funky and playful, and a love for tactile inputs like buttons, knobs, and switches. Devices like the OP-1 synthesizer and sampler tool, and most recently, its work with Nothing for transparent phones and wireless earbuds, exemplify non-conformist design and utility. To summarize: Lyu’s done hardware and software before, and he knows the intricacies of doing both as they now collide with AI.

In hindsight, the dots all connected for Lyu to become a figure in AI consumer tech — using AI, specifically natural language processing via voice input and computer vision — to enhance our curiosities about the world, and maybe even give us back time, whether that’s saving time working or not getting lost in apps.

Nobody, not even Lyu, can be certain that the R1 will succeed (though 100,000 pre-ordered units since January is a great start) or whether smartphones, which are starting to get similar AI capabilities, will render it and other AI gadgets like Humane’s Ai Pin obsolete before they can even gain traction. But Lyu isn’t fixated on taking down Apple or Samsung or any other big consumer tech company. His full attention, now that the R1 is in the hands of real customers, is delivering on the promises of the device and its AI. That means shipping the new features that were announced at the launch event and listening to feedback for improving the R1 — and moving fast. Judging at how engaged Lyu is sliding into the replies on X, Rabbit is moving at the speed of light.

Lyu’s Pathway to Tech

Like any kid growing up in the ‘90s, Lyu loved technology and gadgets. His first contact was video games — specifically, arcade machines. Back then, China was not yet the technology powerhouse it is today. The country lagged years, if not a decade, behind the rest of the world with computers and IT infrastructure. But even though his access, especially in Xian, which was not yet urbanized, was limited, he found a way. He’d beg his parents to allow him to stay at his uncle’s place during Lunar New Year (a week when work effectively shuts down and everyone in China goes back to their hometowns to spend time with their family) just so he could visit the local arcade.

“When I was five or six years old, that was the first time I got my hands on an actual computer,” Lyu says while slowly sipping his Diet Coke, kicked back in shorts, and comfy Air Jordan 36 Lows. “It was pretty early in China and it was an Intel Pentium 486 chip running on Microsoft DOS with no mouse. There were a couple of DOS games on a floppy disc. I played that and I was like, ‘Holy shit, this is way different than Street Fighter at the arcade.’”

Video games and computers weren’t just a hobby to Lyu — they were part of his life and his identity, and they always found a way to intertwine. By his early adulthood, he was a semi-professional Warcraft 3 player and ranked 8th at the World Cyber Games in 2010.

“I just like to try all the gadgets,” Lyu says. “I saved up my own money for a Game Boy. Not the color one, the original one.”

With the Game Boy and computers piquing his technology curiosity, the natural next step was to learn how to program. Lyu says because he was a minor and there was no access to online courses or YouTube tutorials back in the ‘90s, he again had to find alternative means — begging his older sister to register for a workshop teaching adults the C programming language so that he could tag along.

“That’s when I started programming,” he says. “I was just building Flash games and some random small projects.”

By 2008, he graduated high school and applied for university, and it was soccer, not technology, that would free him from China. A big fan of Liverpool FC, he sprung at the opportunity to attend the University of Liverpool, where he majored in mathematics.

“In the remaining year of the workshop, the teacher showed up with a mouse,” he says, and it was a revelation. “The teacher was like, ‘From today, whatever you’ve learned is obsolete’ because they started installing Windows and I was like, ‘I shouldn’t start computer science because then if I graduate, everything in the next four years is obsolete’ so I went with something more fundamental like mathematics.”

In the second semester of Lyu’s first year at university, he got a random email from one of his professors asking if anyone had an iPad and knew how to use the Objective-C programming language to create tablet apps. Lucky for Lyu, his mom had just gifted him one, so he replied, and together with his professor and a few other students, he got his first taste of app development trying to create digital and interactive textbooks. The project ultimately failed when the professor funding the project ran out of cash.

Lyu doesn’t seem to fear messing up.

That failure, however, made Lyu realize how much he liked working with a collective of people all trying to build something — essentially startup culture. He’d work on several more projects with three of his university mates by graduation time, including a “social calendar network” called Time Meet that allowed groups of friends to organize plans based on the free time on their calendars. Lyu’s first official company was financed by money borrowed from each founding members’ parents. “Each of us borrowed £3,000 and then we spent 75 percent of that just to pay the lawyer to register the company. We realized then it was not gonna work.”

Curiosity and determination have always been guiding forces for Lyu. He didn’t have the Silicon Valley connections handed to him, but he had the grit to figure it out. Lyu doesn’t seem to fear messing up. He didn’t fear losing the first term sheet that he got from an early venture capitalist when his co-founders decided to drop out of the agreement. And he doesn’t fear the critical feedback that early R1 customers are pointing out about the device and its AI. Despite calling it the most “down period of his life,” Lyu says he didn’t dwell on it for long. He picked himself back up to figure out what he wanted to do after college.

Birth of An AI Computer

After spending some time bumping around (talking to VCs and such), TEDx invited him to give a presentation. It was making that keynote when Lyu found his true calling.

“I wanted to build a J.A.R.V.I.S.,” he tells me referring to the fictional AI assistant that aids Tony Stark in Marvel’s Iron Man movies. “I wanted to build something from a sci-fi movie, something for everyone, so I pivoted quickly. That was the early idea of building a computer using natural language.”

An application to Y Combinator, a VC and startup accelerator that invests and incubates startups, moved Lyu and a few friends to Palo Alto, California where they formed Raven.

“We were one of the first companies really focused on conversational AI, which was before LLMs” Lyu says, referring to the large language models that give programs like ChatGPT the ability to replicate human conversation. “We were really working hard on natural language processing.”

Lyu says the natural language processing that Raven was working on predated Amazon’s Alexa voice assistant (which launched in the first Echo smart speaker in November 2014). “We didn’t know each other.”

Had it not been for his investors telling him that hardware was really hard to do, Lyu says he could have released a smart speaker before Amazon.

“The only decision I regret is starting with a music-based phone app. [The investors] convinced me that it wasn’t a good time and that I didn’t know how to do hardware, and then Alexa came out. That felt really bad for me. I had the idea [before them] not knowing that there was such a thing as a smart speaker.”

Raven continued working on its natural language processing. Then, in 2016, Baidu (the Google or Yahoo! equivalent in China) reached out to Raven because “search was dying” and “they weren’t doing anything on mobile,” according to Lyu. The company wanted to invest in and pivot to AI. Baidu originally wanted to buy Raven, but Lyu wasn’t interested, so they landed on a partnership instead, which led to the creation and launch of the AI-powered Raven H smart speaker in 2017 and Raven R smart robot (which was never released). It was Baidu’s deep pockets that allowed Lyu to collaborate with Teenage Engineering on the two Raven smart devices.

“I wanted to build a J.A.R.V.I.S. I wanted to build something from a sci-fi movie, something for everyone...”

Baidu’s decision to pull the plug on the Raven R can be viewed as a failure, but it did connect Lyu with the tech company and designer he worshiped in his youth — a partnership that would later extend to Teenage Engineering designing the R1 hardware. So even though his time working on hardware with Baidu was short-lived, he knew that if he were to ever do hardware again (Lyu says it was always a when not if), he’d tap Teenage Engineering for the job without hesitation.

Somehow, Lyu’s always had a little bit of luck. When the deal with Baidu ended in 2018 and he relocated from Beijing back to the San Francisco Bay Area, Sam Altman rang him up. Altman, who catapulted into the spotlight last year with the release of ChatGPT in late 2022 and has become the face of AI, showed him what OpenAI was developing. Lyu says he was intrigued, but didn’t feel it went far enough. He wanted to see humans manipulate AI to truly understand a user’s intention and actions when they’re using a computer.

“It was roughly around 2020, during hard Covid that I got access to the early version of GPT-3 — and it was mind-blowing,” he says. “I talked to my team and said ‘Let’s not try to recreate the wheel,’ AI understanding a user’s intentions is 100 percent solved. Transformers, too. Let’s focus on the next step. Can this language model understand [a user’s actions with a computer]?”

In order for an AI to understand a user’s computer actions, it needs to first be trained on how people use computers. How do they control a mouse? What buttons do they click? What do certain buttons do?

“We took a step back to think about how humans learn from nature. We observe, we imitate, we practice. Can we find something in common? Because we don't want to just build a Chrome plugin that just does one thing. We don’t want to build an agent or action model for solving [commands] on one platform. We wanted to find a universal solution.”

That common factor ended up being modern user interfaces — software that’s designed for humans to easily understand and operate. For example, everyone knows that a gear-shaped icon usually refers to a settings menu with options for adjusting different preferences. With that commonality identified, they’d just need to train an AI (neural-symbolic AI) with tons and tons of videos of people interacting with modern UIs on computers and mobile devices. Three years of nonstop training later, all that AI learning morphed into Rabbit’s Large Action Model, or LAM.

It was at that inflection point after meeting with Altman that Raven became Rabbit. Why Rabbit? Lyu says there’s “no special meaning” but shares a few unofficial inspirations. One: He has a thing for animal names (Raven and Rabbit) and didn’t want a techie-sounding company. Two: He likes The Matrix and in the movie, Neo follows a white rabbit to scratch his curiosity. Three: The R1’s LAM and its Teach Mode (allows users to teach the AI how to use apps, services, and other computer actions) and how it can “spawn different instances” reminds him of the rabbit population at the beginning of the animated film, Zootopia. And four: “There are two syllables, A and I, in Rabbit, which could be interpreted as a nod to AI.” Lyu also adds that a rabbit is “so easy to visualize” compared to a name like Humane, which is more of a feeling, than a thing.

“We started with 150 [of the most popular apps] like Expedia, Spotify, DoorDash, and Uber,” says Lyu. “Those were the first.. and we quickly learned there were similarities [among app designs], like McDonald’s actually uses the same [app] vendor as Taco Bell.”

By the end of 2023, Rabbit had trained its AI on 800 apps and it was time to think about hardware to put it into. I was expecting Lyu to tell me how challenging it was to nail down a design or procure components and manufacturing, but he didn’t. (When manufacturing hardware, factories request a Minimum Order Quantity, or MOQ, which is a certain number of units set by suppliers and manufacturers in order to make it worth their time and profit to produce devices.) If a startup cannot meet a supplier’s MOQ, then they can’t build the product in volume or they may be forced to work with lower-tier supply chains which usually means access to second or third-rate components. Instead, what he told me next made my jaw drop — all of it.

“I knew exactly what I wanted to build: a Tamagotchi-Pokédex-walkie-talkie.”

“We started the R1 design phase in October 2023,” he says, only three months before Rabbit introduced the device at CES 2024. “Hardware is an elimination game. You don’t just put a lot of stuff all over the place and it [becomes] bulky. All you have to do is minus. [For R1] I thought about what would absolutely be necessary. A camera is absolutely necessary. The screen is necessary. A SIM card is necessary.”

I ask how that’s even possible considering hardware design and production usually take at least a year (if not years) of planning. Because Lyu had worked with Baidu to produce the Raven H, he knew how the supply chain worked and was confident he could get what would become the R1 together within six months or less. He bought a Tamagotchi off Amazon in March 2023 and gave the Rabbit team one single instruction: “AI is scary, but let’s make it fun and cute.”

“I knew exactly what I wanted to build: a Tamagotchi-Pokédex-walkie-talkie.”

With that vague concept in mind, and Teenage Engineering ready to roll (Lyu sits on their board), Rabbit landed on the R1’s square design, which Lyu claims “took us 10 minutes to design.” He says they started with an iPhone 15 Pro Max and cut it in half, penciled in essentials like the display, the camera, the analog scroll wheel, the Push-to-Talk (PTT) button, the speaker, the microphones, the SIM card, and the USB-C port.

“I basically drew [Jesper Kouthoofd, co-founder, CEO, and head of design at Teenage Engineering] a box and said somewhere there needs to be a big button, and I want the menu to be controllable with a scroll wheel.” He says it only took 10 minutes of emails to bounce ideas like making the camera rotate and selecting a color. Kouthoofd actually suggested making the R1 a dark red, a color picked from RAL (a German color standard) instead of the Pantone Color System, which is what many designers select from. Lyu didn’t like it and chose the bright and neon orange-red paint job shipping on final units. It was the first and only time that he vetoed Kouthoofd on a design decision.

“[The R1] took us 10 minutes to design.”

Shortly after I got my R1 up and running, I dove into the device’s settings menu and noticed that it has 128GB of internal storage. I wondered if that was a mistake, perhaps in the software mislabeling it. Lyu tells me it’s not, there’s really 128GB of storage in the R1. A miscommunication between Rabbit and the flash storage manufacturer led to devices shipping with four times the storage instead of the originally intended 32GB the startup had wanted, so Rabbit just decided to roll with it. The company didn’t get the extra storage for free, though. It still had to pay for the extra storage per device (only an additional 6 to 8 cents, according to Lyu). However, don’t be surprised if future versions of the R1 have less storage. Because with almost all of the device’s features processed in the cloud, there’s really no need for so much.

Storage miscommunication with the supplier aside, Lyu says: “There’s a broad misunderstanding that we don’t know how to do hardware. This is actually our strongest part. My head of hardware has been in the industry for over 30 years and he knows everyone that knows everyone back in Shenzhen.”

Sizing Up the Competition

The Rabbit R1 launched right on the heels of Humane’s Ai Pin. Both “AI gadgets” are alike in that they use voice-based AI to answer questions, control supported apps and services; become a “second brain” for offloading notes and such; and both have a camera for identifying objects and scenes.

They’re also quite different. The Ai Pin starts at $700, is worn on clothing, has a Laser Ink Display to project a mono-colored interface onto the palm of your hand, requires a separate phone number for calls and messaging, and needs a $24/month subscription to function. On the other hand, the R1 starts at $199, comes with a color display for viewing information and what the camera sees, doesn’t support calls or messaging, and doesn’t require a subscription fee (though it does have an unlocked nano-SIM card slot if you want to connect to LTE cellular instead of only Wi-Fi).

Whereas the Ai Pin hopes to one day replace your smartphone, returning you back to a life that is more present as opposed to glued to a display and apps, the R1 has no such ambitions to kill the touchscreen slab. “We never said this is gonna replace my phone.”

“We never said this is gonna replace my phone.”

“I'm not chasing them or checking them. Their philosophy is quite different. When I started Rabbit, on day one, I knew I was a new startup. The number one mistake I see from them is that — yes, they have over three decades of experience at Apple — but you’re starting as a new company. You’re not Apple, you shouldn’t do things the Apple way. You cannot customize your first-gen product. It’s just plain wrong.”

Unlike Humane, Lyu says that he told everyone at Rabbit that hardware needed to be “de-risk.” That’s why the R1 doesn’t include wireless charging or even a USB-C cable for charging up the device. “Why would we offer wireless charging if we don’t know how the thermals perform? Why would we offer a charging cable if we haven’t tested if it bends and could break at the dining table?”

“Once I saw them trying to replace the phone, I had a huge relief, and I knew we were heading in separate directions. I never worried [after that].” Humane’s original pitch was that the Ai Pin would replace the phone, but the company has since walked back on that stance and is repositioning the device as a phone companion instead.

With over 100,000 devices sold so far, the R1 is the first AI gadget hit. The price is 3.5x cheaper and initial impressions from myself and first-order customers seem to favor the AI speed and capabilities of the R1 over the Ai Pin. Yet, there’s still immense skepticism about whether the R1 needs to exist when it’s very likely that many of its AI features will be baked into new smartphones.

“I never believed that my first-generation product was going to convince everyone [to buy it],” Lyu says. “That’s just very unrealistic thinking. But we know where all your pockets are on your pants and they’re basically molded to [fit phones]. So we just cut [a phone in half] and made it as light as possible, and gave it a color that’s so unrealistic that you just want to have it right here and stare at it. We make all this fidgeting analog stuff for you just to play around with it.”

Lyu believes if phone makers don’t add AI capabilities and change, they’ll die because “application systems are dying.” I don’t believe many people share that belief — think about how phones are now cameras, entertainment portals, banking and payment systems, health and fitness trackers, and more, and it’s hard to see them just falling out of style when they’re so central to living.

It’s at this point of our conversation where it feels like the R1 raison d’être overlaps with the Ai Pin. Lyu says that the R1 is designed to help you quickly compute and then move on. There are no addictive apps like TikTok or Instagram to get lost in, like on a phone.

“You go [to your phone] to waste your time or to kill your time,” he tells me while waving his iPhone 15 Pro Max, covered with a purple Neon Genesis Evangelion case (his favorite anime) in the air. “But you go [to the R1] to save your time.”

“You go [to your phone] to waste your time or to kill your time. But you go [to the R1] to save your time.”

The Ai Pin is the R1’s most direct competition right now, but a wave of AI gadgets is popping up, too. I ask him if he’s concerned about other action-based AI gadgets, such as Open Interpreter’s 01 Light eating the R1’s lunch. The stone-shaped AI gadget can control your computer and perform tasks on your behalf, and costs $99 pre-built or you can DIY one yourself with its open-source design and code.

“They basically said they just want to build an open-source Rabbit R1,” he says. “I’m actually supportive of that. We have the same goal and we’re chasing the same dreams. But what I’ve learned is that if you really want to build the best end-user experience, you have to be a little bit closed. There needs to be polish on both sides [hardware and software].”

He raises a hypothetical scenario, where there are five different brands of an AI computer, and they all use different microphones to listen to a user’s voice command. They may run the same software, but the AI experience will be “all over the place” because of the different microphones each brand will use.”[With the R1] we know exactly the driver chip, the microphone of each component — we polish to them. But that cannot be achieved if everyone just downloads the same GitHub code and deploys it on their local environment. We saw that from Android.”

At this time, there are no plans to license the R1’s software, rabbit OS, to other brands to make their own devices. It’s the opposite of Humane, which has said licensing its Cosmos AI operating system is part of its long-term growth strategy.

“We care about the experience,” Lyu says.“We care about design. We don't want any trash designs to run our OS. It just doesn’t work.”

The Real Work Begins Now

By the hour mark into Lyu’s R1 launch party keynote, the nerds — including myself — are getting restless standing around, but excited at the possibilities of what the R1 can do. Lyu starts wrapping up his keynote and throws up a slide showing a list of features that are coming by the summer and features in development for future software releases.

The one thing that seems to be bugging everyone is why the R1 doesn’t have a way to set alarms (it’s coming in the summer update); the Ai Pin doesn’t have it either (it’s also getting it soon). “Do the founders not use alarms or calendars?” I tweeted during the keynote. So I ask Lyu and his answer is simple: They didn’t have time to add it.

“Our team is too small. We have a lot of other priorities. That’s it. We know it needs an alarm, but we didn’t have time to do it.” Rabbit, after all, is really small — only 45 people (including engineering and marketing) at the moment. The company is scaling up (not too fast, though, Lyu says) with more engineers to further develop AI and add new features. Because AI is always evolving, it’s not something that will ever be “finished” so in a way, any AI gadget will be never-ending, so long as the hardware can handle the updated AI.

The period after a product launch — when customers are using it in the real world — is make or break. Lyu is well aware of this and has said it during his keynote, on X, and again to me, that Rabbit is monitoring all feedback (on Discord, on social media, on tech blogs and YouTube channels, etc.). Without the community, the R1 has no future. So Rabbit is taking the job of listening very seriously. That means delivering the features that it’s promised and continuously adding new utility.

We wrap up our interview (over 40 minutes past the originally allotted one hour) on AI trust, privacy, and security. For AI and AI gadgets to take off, the answers that the AI responds back with need to be correct. Generative AI such as ChatGPT has improved at a breakneck pace since it was released in November 2022, but AI still hallucinates — there’s a 50-50 chance you’ll get an answer or a correct one.

“We’re super early and we move fast.”

“Any common facts, up-to-date search information — I 100 percent trust,” Lyu says. “That’s a good sense of where the AI that we’ve created is at. Anything legal-related? Hell no. Unless you train [the AI] on dedicated data and models, which is happening. This trust phase is going to be relatively short. I will say that in the next couple of months, this will completely become a legacy issue. The only way to move forward is to build it. We accept the current stage of AI, and that’s why we have a disclaimer in the settings [menu]. For example, don’t use [the R1] to check your health.”

Jumping off trust, Lyu says even though Rabbit is small, security and privacy are top priorities. There were rumors that R1 user information might have been stored on a server in China, given his past relationship working with Baidu. Lyu debunked that during a media Q&A session at the R1 launch event, confirming that all Rabbit data is stored on Amazon Web Services, or AWS. His Chinese-U.S. relationship is more personal than business.

“If you know… just by speed, there’s no way that [the speed of R1’s AI] can be hosted in China.” He says there would be even greater latency if it was. “I came here to chase the American dream and work hard for my passport. I’m not even a Chinese citizen [anymore]. Next year, I’ll be an American citizen. Unfortunately, I was born in China… I don’t need to lie about that. But I left Baidu and all that in 2018.”

As for security, Lyu reminds everyone that “any system can be hacked” but that doesn’t mean they’re taking it lightly, either. “We’re running infiltration tests, every single day. We’ve intercepted numerous attacks. It happens all the time. We’re getting blackmailed. A lot of people say they have our code and if we don’t pay $2 million in Bitcoin, they’ll share it to Humane. It’s just the nature of running any software company. You have to deal with it. Security is a high priority for us.”

Lyu and I walk out of the WeWork into gorgeous, 70-something-degree weather. The sun is just beaming warmth everywhere. It may be difficult for new apps to change the world the same way they did over a decade ago, but they’re not going anywhere, even with the new boom in AI.

“We just need to run fast.” Like any startup, Rabbit is smaller and scrappier, and as a result, leadership can make decisions much faster than at a large company like Apple or Google. “We’re super early and we move fast. Our approach into [using AI to control] third-party apps — there’s no other route and we have some pretty strong patents.”