“Ahhh, this thing is so frustrating!”

Those words are from my brother, who was trying to ask the Rabbit R1 a simple question while I was driving: “What's the nearest coffee shop?” It's a question that's so easy for a smartphones to answer, but the R1, with all its artificial intelligence prowess, left us in silence. It didn't even want to give us an audible cue that it was thinking—just silence.

On that cool spring evening, when we asked a few other questions, the R1 often took so long to respond that my brother decided to reup the query right when the R1 spat out an answer; we had to start all over again. It was funny, annoying, and reminiscent of the earliest days of Alexa and Google Assistant, where it was common to have these kinds of mix-ups with the burgeoning technology. This particular instance seemed to be a bug because most of my R1 interactions until then have had an audio cue with much more reasonable response times.

Bug or not, the experience largely has been the same during my time with it. I have found the Rabbit R1 fairly useless over the past week I've been testing it. There are moments when it spits out a response to my query that leaves me nodding in surprise at its accuracy (or perhaps the fact that it gave me a helpful answer). But the biggest issue I have with the R1 is finding a use for it. Not only do I now have to carry a second device everywhere I go, but more often than not I end up pulling out my phones to finish the task the R1 can't complete. This red-orange gadget isn't a personal assistant, it's dead weight.

Follow the White Rabbit

Rabbit announced the R1 at CES 2024. An AI gadget launching at CES? Get in line. But Rabbit's founder Jesse Lyu did have some standing. He sold his first startup, Raven Tech—a Y Combinator-backed company that was building a mobiles operating system called Flow—to Chinese tech giant Baidu in 2017, which he later joined as a hardware general manager. It was, at the time, described as “China's answer to Alexa.” However, the tech never seemed to take off.

At CES, Lyu didn't have any demos of the R1, yet the company said it would be shipping around Easter. It was an exciting announcement. This cute, retro, orange-red gadget was designed by acclaimed Swedish design firm Teenage Engineering, the same company that has helped design products from Carl Pei's Nothing. (TE's founder is now Rabbit's chief design officer.)

I can't deny its looks. It's a bit plasticky, but the color is eye-catching. There's a small, vertical, 2.88-inch, left-aligned touchscreen, though you don't use touch to interact with it (only when you need to use the keyboard for things like inputting a Wi-Fi password). Instead, there's a large scroll wheel on the right and a button on the edge. There's a camera in the corner, and it can rotate front and back (and to the side when it's not in use for privacy). It recharges via USB-C.

Keep in mind that after you buy one, you will need to cough up additional dough to supply it with a data plan—there's a SIM slot where you can add a SIM card, and the R1 supports only 4G LTE. It can otherwise connect to Wi-Fi (or you can tether from your smartphones). There's Bluetooth support too, so you can pair wireless earbuds.

The promise was simple. Speak into the device and it'll complete tasks for you thanks to Rabbit's “large action models”—call an Uber, reserve dinner plans via OpenTable, play a song through Spotify, or order some food on DoorDash. Just speak and it will handle it, just like if you handed your smartphones to a personal assistant and asked them to do something for you.

Naturally, the core function of the gadget is also push-to-talk access to large language models. You can ask it anything and get an answer like if you typed something into Google (it primarily uses Perplexity's AI model). There's a camera too, so you can point it to subjects and get answers to visual queries. It made several Best of CES lists from the tech press (though not ours).

Hare-Brained

The reality of the R1 is that it promises to do so much but does very little. First, I need to mention the setup process, which is one of the sketchier things I have done in the nearly 10 years I've spent testing personal devices. You connect to and set up your device on the Rabbithole, a bespoke web portal that's the only way to access all the information and third-party integrations with your R1.

The homepage is where you'll see all the things you've asked your R1, very much like the Humane AI Pin's web portal, though the R1 has large gaps of data where it didn't record the things I asked it. Then there's the Connections tab. This is where you connect your accounts for currently supported services, which just include Uber, DoorDash, Midjourney, and Spotify.

To connect one of these services, Rabbit uses a Virtual Network Computing system—you know, the type of software an IT team will ask you to install so they can remotely access your computer. This is indicated in the URL and the fact that the entire experience is slow. You'll sign into a service with your login details, more or less handing Rabbit your authorization credentials (as indicated in its privacy policy). The company says it doesn't store any user credentials for these third-party services, but knowing that doesn't make me feel any less queasy.

These apps don't have any official relationship with Rabbit. You're using little bots that have spent hundreds of hours learning how to use Uber's interface, and Rabbit has to make a custom interface for its gadget to show you details on its tiny screen. I'm curious to see whether a company like Uber will try and block these “Rabbits,” since this convoluted interaction scheme means Uber is no longer in control of how its customer perceives its user interface. Just remember that Rabbit is now acting as a layer that can collect all the data on what you're doing with these third-party services (much like Google and what you do on its androids mobiles operating system).

There's also been a lot of blowback on social spaces about Rabbit's approach, the privacy issues it creates, along with characterizations of Lyu as a "grifter." Are so-called LAMs actually using artificial intelligence? Rabbit's system runs androids, but the company claims all of the AI computing happens on the cloud. So why can't this be an app?

Either way, the cloud-powered interactions are janky. Uber did not work for me—“There was an issue with the Uber service” is what it frequently spat onto the screen. I don't use Spotify or Midjourney (it requires a paid account), and Rabbit does not recommend making a new account to connect them with the Rabbithole. “To ensure the best experience and compatibility with our services, connected accounts should not be brand new,” the company says in its instructions. “We strongly encourage you to use accounts that have been active for a substantial period of time.”

So that leaves DoorDash, and … well. You can see this in my X post below. I asked to order some burritos, and it served me three to four restaurants in my area. I chose one, and to my surprise, I was able to see a total of only six items on the menu via the R1's screen. Two of them were burritos (not the one I wanted), and the rest were drinks and churros. Great. I ended up pulling out my phones.

When the R1 connects to these third-party services, it warns you that it might take some time to load. Waiting for DoorDash to open only to then scroll through a measly number of options is just plain ridiculous. Just use your smartphones! Lyu's whole pitch with the R1 was that instead of using apps, which he finds annoying to use, you can just use your voice with the R1. But if this is the experience, I'll choose my apps and Big Screen smartphones, thanks.

By the way, you can't connect these services to your R1 if you access the Rabbithole on your smartphones. It currently works on desktop only. How very mobiles.

Hopping Mad

The primary way to interact with everything on the R1 is the scroll wheel, and then a push of the side button for selection. Unfortunately, the scrolling function is inconsistent. It doesn't scroll as quickly as you'd expect, there's no haptic feedback, and you have to scroll way too much to move to the next selection. How Rabbit couldn't get this right on a device with such limited functionality is mind-boggling. Even changing the volume is a task—and asking the R1 to lower the volume results in, “I cannot adjust the volume on the R1 device.”

Looking at my Humane Ai Pin review, the issues with large language models are also present here—chiefly inaccurate information. On April 24, I asked the Rabbit R1 when the next total solar eclipse was, and it unhelpfully told me it's on April 8, 2024. I finally got it to answer what the nearest coffee shop was, but it mentioned a place two and a half miles away—I have multiple coffee shops blocks away.

I will admit, it is impressive when you do get a correct answer. I asked whether an auto shop I was going to offered inspection services, and it accurately said yes, after looking through dozens of user reviews to find the answer. The problem is some of these answers are so lengthy, and the R1 sometimes doesn't lead the response with the information you're looking for. Just say yes or no!

You can use the Vision feature to ask the R1's camera to look at what's in front of you and give you information about it. This works pretty well—it described my dog, called my wife a young woman, which she was happy about, and me a “middle-aged man.” She laughed a little too hard at that one.

At the R1's launch event in New York City, Lyu demoed an example of having the R1 look at a paper with a printed spreadsheet on it. He asked the R1 to swap two columns, and then send the result to his email. I didn't have a spreadsheet on paper, but I did have an auto-inspection report that I wanted to send to my email. I asked the R1 and … it said it didn't have my email address. (I set up my Rabbit account with my email information.) I asked the company about this, and I was told the R1 didn't support documents other than spreadsheets yet. Great. So I printed a spreadsheet, asked it to swap two columns, and sent it via email, and it sort of did this. It swapped the two columns, but for some reason, it didn't include several other columns that were on the paper.

I picked up my copy of Kazuo Ishiguro's Klara and the Sun and asked the R1 whether it could look at it and tell me what it's about. The R1 instead just described the cover and said it's “likely” a work of fiction. If it could read out the name, why couldn't it research it at the same time and give me a synopsis? Even the Humane Ai Pin could do this.

You can also have the R1 take notes, and edit these notes in the Rabbithole, but there's no reminder functionality. I also find it annoying that the Rabbithole keeps logging me out after some time, so whenever I want to check a note, I might have to log in first. There are also voice recordings, and the R1 plays a nice tape recorder animation when it's working. Too bad the recording itself is low-quality and muffled. It does summarize the contents of the recording though, and you can download the WAV file.

The translation capabilities, much like the Humane Ai Pin, are good. Just ask it to translate a specific language, and you can now have a back-and-forth conversation. The R1 will automatically change the translation language, so when I speak English, it changes it to Spanish. When the person across from me speaks Spanish, it swaps to English.

Hop to It

You know what else does all of this stuff pretty well? Smartphoness! This is also the question I receive repeatedly whenever I show someone the R1. “Why can't it just be an app?”

I posed this question to David Widder, a postdoctoral fellow at Cornell Tech studying open source artificial intelligence. “Hardware is cool—there's increasing frustration from app developers on having to give so much money to Apple and Google. I think there's a little bit of, ‘We want to do our own thing and not be beholden to them.’”

That's fair, but the R1 is just not ready yet. I considered skipping this review and writing a more experiential story, but this is a product anyone can buy right now. A company is charging you $200 to be its beta tester, and while Rabbit has a roadmap of features and services—including a Teach Mode that lets you train the R1 to do specific tasks—I don't see a reason to buy it now. Revisit it when it's more feature-rich and genuinely useful, and buy it then if you want.

At the very least, I haven't had the battery issues plaguing other reviewers. The R1 recharged quickly for me and doesn't deplete juice too fast in standby mode. When you do use it, the battery drops fairly quickly though.

In the end, the biggest issue boils down to the fact that I now have to carry two devices. I'm WIRED's resident smartphones reviewer and I hate carrying two phoness—it's why I always put my personal SIM into each new device I test. Over this past week, I forced myself to use the R1 but often ended up using my phones instead. (Weirdly, the Humane Ai Pin was better in this regard, as it is wearable and I don't have to carry it in a pocket or hold it.)

Rabbit was clear in saying that the R1 will not replace your phones, but if I can do all of the same tasks and so much more on my smartphones (Google's Gemini has given me identical if not better results than the R1), I have no reason to use it. At least it looks pretty. I'll add it to my growing collection of AI-powered paperweights.