This story was supposed to have a different beginning. Earlier this week, I attended the glitzy unveiling of Rabbit R1, a new AI gadget, in New York City and then stood on the windy curb outside the venue to hail an Uber home. device. Instead, he probably set it up and messed around for an hour or so, then failed to connect.
The R1 is a bright orange device with a camera, microphone, and small screen. Press and hold a button, ask a question or give a command using your voice, and the cute bouncing bunny on the screen will listen and talk to you. In theory, it's like communicating with ChatGPT via a walkie-talkie. You can ask them to identify specific flowers through the camera, or play a song based on half-memorized lyrics. You could call for an Uber, but your phone might hang up on the last step and you'd be stranded in Queens.
When I finally got back to my hotel room, I turned on the R1's camera and held it up to a cold slice of pizza. “What am I looking at?” I asked. “You're looking at a slice of pizza,” the voice said to me. (Correct!) “It looks freshly baked and delicious.” (Well, no.) I decided to try something else. “What is Top 10…” I stumbled and let go of the button. I thought again. “What are the top 10 best use cases for AI for the average person?” The device, perhaps confused by our previous interaction, began listing pizza toppings from his number two. sausage. 3. Mushrooms. 4. Extra cheese. ”
Until now, consumer AI has been primarily defined by software: chatbots like ChatGPT and powerful autocorrect on the iPhone. Now we are experiencing something like this: materialization: Companies are launching and manufacturing physical parts of metal and plastic entirely dedicated to AI functionality. These devices differ from previous AI gadgets such as the Amazon Echo in that they incorporate more advanced generative AI technologies that are trending these days, allowing users to interact more naturally. There are also pins, pendants, and all-new smart glasses.
But despite its promise, this new era is not going very well. Take Humane, a Rabbit competitor that launched the wearable AI Pin earlier this month. The device is positioned as an alternative to smartphones and is priced accordingly. It costs $699 and requires a $24 monthly subscription fee. Reviewers criticized the pin for being slow, overheating, and having difficulty answering basic questions. “It's hard to pick one thing that really stands out,” The Verge wrote.
R1, by comparison, is small enough for its purpose and (relatively) affordable ($199, no subscription). The device itself is fun and retro-chic. Rabbit founder and CEO Jesse Lyu reportedly bought each member of his team a Tamagotchi for inspiration. And to be fair, R1 does do some interesting things. On stage, Mr. Lyu demonstrated how the device interprets a handwritten table and converts it into a working digital spreadsheet. When I asked, he managed to outline the handwritten page, but with only about 65% accuracy. The gadget allowed him to order acai bowls on DoorDash, but it didn't allow for customization. (I wanted peanut butter.) And I couldn't get an Uber to work. (At one point, however, the device notified me that the request had failed, when in fact it hadn't, and I was forced to take a $9 ride that I didn't even take.)
One of the big selling points of R1 is probably its ability to do what we would call high capacity. action Model, or LAM – a play on the word large language model, which is the technology that powers modern chatbots. ChatGPT can answer questions and compose mediocre essays, while R1 can theoretically perform actions you might perform in another app, such as giving your friend $20 of her money. . Rabbit says that if you teach a device, it can learn from any app. Liu likened the technology to Tesla. While on Autopilot, a Tesla can theoretically recognize a stop sign not because engineers have taught it what a stop sign looks like, but because it has been trained on countless hours of footage to recognize the sign's physical attributes. Because there is. Similarly, R1 allows you to perform tasks on your phone without having to teach each app.
The problem is, none of this is actually real. At least not yet. Like many AI products, R1 is supported by hype rather than a compelling use case. (After all, much of its functionality can be performed on a smartphone.) Back in February, Lyu said Rabbit was training its models on 800 apps. This week, Spotify, DoorDash, Uber, and Midjourney (a popular AI art generator) were released with the ability to use just four of them. The company says LAM is in “very early stages.”
On stage Tuesday night in front of an audience of reporters and Rabbitohs fans, Liu appeared nervous at times, even encouraging people to laugh to ease the tension. Prior to the event, a user posted on GitHub accusing Rabbit of misrepresenting its technology. “For anyone with a technical background, it is painfully obvious that artificial intelligence and action models at scale do not exist,” said the anonymous post, which has since been deleted. About X, Liu characterized the post as “all false claims.” Lyu promised to fix any bugs that may occur on R1 devices. Before demoing DoorDash on stage, he admitted that the feature still isn't working as fast as he hoped. “But I want to show you guys, and I want to be honest with you.”
But Lyu also envisions high-concept systems that will one day merge the physical and digital, allowing people to point at various smart items in their homes and control them through Rabbit's AI. We breathlessly announced many new initiatives, including . (Never mind that R1 launched without many of the promised features.) Towards the end of the presentation, there were these words: be humble It appeared on the huge screen behind me. “We're a really, really humble team,” Liu told the crowd. A few minutes later, the curtains on either side of the stage lowered dramatically to reveal a conveyor belt loaded with boxes of R1, but the words were still visible. Music blared and people started lining up to get their hands on it.
R1 is a reminder of the disconnect, for better or worse, between Silicon Valley's culture, which often prioritizes speed over quality, and consumers' high expectations for the products they use. And to be fair, expectations have been raised, at least in part, because of the extraordinary products that have come out of the same culture of competition and repetition over the years.
Once the party was over, news of the first bug arrived. There was no way to change the time zone on the devices, and many were programmed to the West Coast by default. In the future it turns out that he is 3 hours late.