Rabbit R1 Before Release: Questions About Cost, Actions and Privacy

Matthew Berman says he ordered a Rabbit R1 on announcement day and is eager to review it when it arrives. In this February 2024 video, he responds to questions from viewers about the proposed voice-first device. Rabbit contacted him and sponsored the video, a relationship he explicitly discloses.

One viewer questions whether a $200 device without a subscription can sustainably cover cloud costs. Berman says Rabbit told him it had no subscription planned at that time, while acknowledging that the company could change its plans. The video does not independently establish the economics or future pricing.

Berman explains the proposed distinction between a large language model and Rabbit's large action model: language interprets the request, while an action system is intended to operate an interface on the user's behalf. He describes efficiency and cost advantages, but these are pre-release claims rather than independently measured results.

In an interview excerpt, Rabbit founder Jesse Lyu says the company is not building its own large language model and plans to work with language-model providers. He describes its action system as neuro-symbolic and says it was trained using recordings of people interacting with software. The excerpt does not provide enough detail to verify the technical or performance claims.

Lyu argues that relying solely on APIs can limit a startup's access to services or features. Rabbit's proposed alternative is to interact with application interfaces. Whether that generalizes reliably across changing interfaces, authentication and service terms remains a practical question.

A common viewer objection is that the product could have been a phone app. Berman favors the experiment of dedicated hardware and a voice-first interaction, even if it means carrying another device. He considers competition with existing phone assistants valuable; his enthusiasm is a product opinion, not a demonstration that a separate device is technically necessary.

In another interview excerpt, Lyu describes reaching for a dedicated device to search or play music by pressing a button and speaking. He argues that the interaction can be faster and more intuitive for him than navigating a phone. These are his examples and expectations before Berman's hands-on review.

Berman then reads Rabbit's privacy statements. The company described a physically blockable camera, push-to-talk rather than an always-listening mode, and a face-down privacy mode. He likes the opt-in interaction, but the statements are the company's account of its design rather than an independent verification.

Rabbit's published policy, as quoted in the video, described controls in a web portal and restrictions on use or sharing of personal data. Berman presents those commitments as reassuring. Such statements are time-specific and do not alone establish how data are handled in every deployment or how security controls perform.

The company also described a dedicated cloud environment for each user's agent, linked third-party accounts and confirmation for sensitive actions such as payments. It said it did not store service passwords. Berman treats the approach as an explanation of the planned system; users should distinguish these claims from verified security guarantees.

Jason Calacanis asks Lyu how an order would work after a user links a service. Lyu describes selecting services in a web portal, authenticating with the provider and then allowing a cloud-hosted action system to interact with the chosen service and return a result to the device. This is a description of the intended design, not a live independent test.

Berman closes by saying he expects his device soon and plans to make hands-on reviews. He predicts a future with more natural-language interaction and fewer conventional applications, but those are his forecasts. At recording time he had not yet assessed the shipped R1.