Rabbit CEO Jesse Lyu on launching the R1, future of AI hardware, and going viral at CES | E1885

Episode Summary

Executive Summary: Rabbit founder Jesse Liu explains the R1 as a deliberately low-risk first device for a software-led AI interface, not hardware built for its own sake. He distinguishes Rabbit’s neurosymbolic Large Action Model (LAM) from an LLM: it learns to operate apps through recorded human interactions and can perform tasks in a cloud environment. The discussion covers early demand, product limits, possible revenue through teach mode, and the company’s longer-term ambition to make software interaction more intuitive.

Main Topics: R1’s software-first rationale and hardware design: Liu says the R1’s retro, compact form factor was chosen to make a new and uncertain interaction model approachable. He describes hardware as a necessity to support software, rather than an end in itself, and credits Teenage Engineering with the product’s industrial design. How Rabbit’s Large Action Model works: Liu emphasizes that Rabbit is not building its own LLM. Its LAM uses neurosymbolic methods and examples of people using apps to learn how to carry out actions through interfaces, rather than relying solely on APIs or fixed screen coordinates. Authentication, cloud execution, and task confirmation: Users connect services through a web portal and authenticate directly with providers. Rabbit says tasks are executed in a cloud-hosted virtual environment and results are rendered back to the device. The founder acknowledges that complex or consequential tasks need clearer review and confirmation flows. Use cases, latency, and current limitations: Liu highlights quick requests, music, search, and visual summaries as useful early functions. He says direct LAM actions can be fast, while up-to-date search and image understanding take longer. The R1 is not intended to replace a phone’s social messaging or content-consumption features yet. Adoption, economics, and potential business model: The R1 costs about $200, and Liu says the company makes money on the hardware while recognizing that hardware margins are generally low. He describes a future teach mode in which users demonstrate workflows, with the possibility of sharing or selling those taught capabilities. Long-term device and industry direction: The conversation explores possible future formats such as standalone headphones and watches, alongside cloud-based AI and the constraints of chips and energy. Liu frames the R1 as an initial experiment rather than an immediate attempt to replace smartphones.

Key Arguments: Liu argues that new hardware should be justified by software that works better on a dedicated device; choosing a familiar, compact form factor is intended to reduce adoption risk for a novel AI experience. Rabbit’s central technical bet is that a model can operate many services through their interfaces, avoiding dependence on vendors providing comprehensive APIs. Liu says the LAM learns from human interaction recordings and uses symbolic reasoning rather than relying only on fixed screen coordinates. The founder presents the cloud as a practical way to run the system’s tasks, while acknowledging that privacy, latency, and infrastructure costs remain considerations. He says the LAM’s direct actions are fast, but search and vision are slower. The R1 is positioned as a task-focused companion, not a phone replacement. Liu cites music and direct requests as appealing uses, while conceding that users still need their phones for messaging, social features, and richer media consumption. Liu sees user-led teach mode as a possible ecosystem and monetization path: people could demonstrate app workflows and potentially share or sell the resulting capabilities. He characterizes this as a future direction, not an established business. The interview argues that incumbents may struggle to shift from app-centered operating systems because of existing business incentives, while a small company can experiment with new interaction models and industrial design.

Data Points: R1 sales reported by the host: Over 60,000 units in five days - The host cites this early demand when introducing the device; Liu says the scale of the response exceeded his expectations. Sales pace reported by the host: About 10,000 units per day - Early sales pace cited in the introduction. Founder’s initial sales expectations: About 500 units on day one and 3,000 units for the early-adopter market - Liu says he expected much lower initial demand and had prepared a plan to increase production if needed. R1 price: About $199–$200 - The host and founder discuss the device’s roughly $200 price point; Liu says the company makes money on the hardware. R1 weight: 110 grams - Liu uses the weight to argue that carrying the R1 alongside a phone is manageable. Company headcount: About 17 people - Liu’s estimate of the team size at the time of the interview; he says the company is hiring. Fundraising: $10 million - The host references the company’s funding from Vinod Khosla. LAM development and research: About two and a half years of data collection; an idea dating back roughly nine years - Liu says Rabbit began collecting human app-interaction data around two and a half years earlier and describes the broader idea as a much longer-running effort. Reported task latency: About 500 milliseconds for some generic questions; roughly 2–3 seconds for current-information search; about 8–10 seconds for vision - Liu gives approximate examples and says vision and search are the main sources of delay; he expects continued optimization. Planned early Perplexity offer: First 100,000 units to include $200 of credit - Liu describes a partnership offer for R1 buyers.

Pivotal Quotes: "Hardware is never a choice by preference. You don't build a hardware because you want to build a cool hardware." — Jesse Liu: His explanation of why the R1’s form factor serves the software concept rather than being an end in itself. "We're not making any large language model. The large action model is neurosymbolic." — Jesse Liu: A key clarification about Rabbit’s technology and the distinction between its LAM and commercial LLMs. "Technology is not reinventing new demands. Technology is solving the same problem over and over again, but each generation in a much more intuitive way." — Jesse Liu: His broader view of how new devices and interfaces create value.

Implications: Rabbit’s bet is that AI can shift interaction from navigating apps to stating tasks, but its success depends on reliable app operation, faster responses, user trust, and a viable ecosystem. The R1 is an early experiment, not yet a phone substitute; cloud dependence and incumbent platforms remain major constraints.