AI gadgets like Humane’s Ai Pin and Rabbit’s R1 are never “finished”

What do we actually know about good AI gadgets? And is it “good” when you can theoretically improve something forever? These are two questions that Ware addresses but leaves largely unanswered.

The Humane Ai Pin was launched with the intention of being an open-minded review at best, and issues with the device’s overheating and slow response times suggest that a conversational wearable (not worn on the wrist) might be a better idea. It often exceeded my expectations. While the company promises fixes and new features and says it’s willing to listen to feedback, it appears to be taking little steps to address the fact that the product it shipped was incomplete.

The Rabbit R1 gained a lot of goodwill upfront by being significantly cheaper. Teenage Engineering’s design was also fine. But based on the short time I spent with my own R1, even though it looks and feels pleasing, it doesn’t perform all the features featured, and what it can do isn’t all that detailed or rich. Rabbit, like Humane, promises future updates and addresses customer issues on social media, but overall the company seems happy with what’s been shipped.

This is kind of confusing for anyone who reads or cares about product reviews. But it also reflects the weirdness of AI hardware in general. These products can go from providing the wrong answer to providing the right answer in a matter of hours with seemingly small gains. It can be updated quickly and permanently, so it’s never actually finished.

into the clouds

Neither the Rabbit R1 nor the Humane Ai Pin are technically “headless” in the sense that they don’t have a graphical user interface to interact with, but they prioritize the use of voice and keep you informed of what’s actually happening on the device. Make it vague. The intent here is pure.that should It feels more natural to ask for something verbally than to fiddle with the menu. But the problem, as many have pointed out, is that monologues aren’t appropriate in all situations.

Ai Pin gets around this problem by using the R1 with a projector and a small touchscreen, but most of the time you’re still just making a voice request and waiting for a response. This magic happens on the backend with AI frameworks, large-scale language models, and large-scale action models (LAMs) that interpret what the user wants, connect to information from the appropriate services, and act on the user’s behalf. and interact with those services. and provide the answers you are looking for.

But there is no way to know what the device is actually doing to produce the answers it provides. Only the answer itself is meaningful or not, right or wrong. Despite producing attractive shells at high or affordable price points, the products Humane and Rabbit sell are more of a software experience than a hardware experience. It’s also software that can be tweaked remotely in a way that would normally require downloading an entirely new software update.

can always be improved

Both Humane and Rabbit are planning summer updates, which they claim will introduce features promised at launch, as well as new features suited to their different form factors. Both devices have been tinkered with ever since they were in customers’ hands. On the day of the launch, my colleague Raymond Wong asked Rabbit R1 what time it was, and not only was the wrong time shown on the screen, but verbally he was given a time that was 16 minutes off. . The issue has since been fixed, with Rabbit CEO Jesse Lyu claiming that future over-the-air updates will improve battery life and “further reduce hallucinations.”

Humane currently maintains a public changelog of adjustments made to the backend and Ai Pin’s own Cosmos operating system. According to updates published on its blog, the company’s current focus is on “improving battery life and thermal management” for Ai Pin, reducing the time it takes to respond to requests, and “improving accuracy for a smoother experience. “Fine adjustments”.

It feels like a bolt out of the blue in terms of what these products can accomplish, if both companies can actually make the gadgets work perfectly as they’re supposed to. By relying on models created by other companies (in Rabbit’s case, an app that he can train his LAM to use), Ai Pin or R1’s capabilities can theoretically expand indefinitely as long as the AI ​​models continue to improve. may be.

Already reached the ceiling

But from another perspective, the fact that these devices shipped unfinished shows that we are already honing in on the limits of what AI can actually do on its own. There may be. You will realize that hallucinations can only be “reduced”, not eliminated. Features are not released in a functional state; they are “improved” and “tweaked.” Language models and the nebulous cloud of AI capabilities that depend on them are so unpredictable that they can never be perfect.

Sam Altman, CEO of OpenAI, which provides GPT models that both Humane and Rabbit use in some form, said that to address the hardware bottleneck OpenAI faces in developing future models, He was reportedly seeking trillions of dollars. This would theoretically cost Apple or Microsoft’s worth to release his GPT-6 or 7. At what point does it cost more than it’s worth to create a better version of the AI ​​chatbot we’ve developed?

Chips aren’t the only problem. The data used to train AI models is running out. new york times Companies report that they will run out of high-quality web data to train models by 2026. Companies developing AI models are already considering other sources, such as acquiring entire publishers or using AI-generated text for training. Is it all because of somewhat natural language and frequently inaccurate responses?

When something unknown becomes a product

New ideas and new products are naturally exciting. Especially if it offers a different experience than the old Siris or Alexa. This sense of possibility is the best part of the current wave of AI hardware. Maybe we really are going somewhere new, and the journey to get there doesn’t have to end.

However, the open-ended nature of current products like the Humane Ai Pin and Rabbit R1 also easily cover what is currently a frequently frustrating experience. And it’s a sense of frustration that neither company is committed to eliminating. Because they can’t do it. They sell hope for the future in a way that few consumer products that actually work have ever been able to do. And it’s not even a very detailed future. In my short time with the R1, I never felt like it had a purpose. I don’t think it should necessarily be the customer’s job to figure it out, even if it’s fun. None of these AI gadgets are finished. That’s because that unknown feeling, and all the good and bad points contained in it, are products.