Humane Unveils the Ai Pin, Its Large Language Model-Linked Wearable — with Some Embarrassing Errors

Self-described "experience company" Humane has unveiled its artificially-intelligent wearable, the Ai Pin, which taps into large language models (LLMs) like OpenAI's ChatGPT to "reimagine interaction" — though the promotional video for which highlights a clear shortcoming in the concept.

"Ai Pin is the embodiment of our vision to integrate AI [Artificial Intelligence] into the fabric of daily life, enhancing our capabilities without overshadowing our humanity," claim Humane co-founders Imran Chaudhri and Bethany Bongiorno of their company's creation. "We are proud to finally unveil what we and the team at Humane have been working on for the past four years. For us, Ai Pin is just the beginning."

The Ai Pin itself is a compact, square wearable which attaches to the user's clothing with a magnetic clasp. Built around an eight-core Qualcomm Snapdragon processor running at 2.1GHz and with 4GB of RAM and 32GB of eMMC storage — putting its specifications towards the budget end of the smartphone market, a comparison invited by its inclusion of a cellular modem and mandatory line plan — the Ai Pin lacks a screen, very deliberately. Its primary mode of interaction is voice — though it also includes a laser projector which can beam interactive menus onto the user's palm.

The pin also includes a range of sensors, including accelerometer, gyroscope, magnetometer, and an indirect time of flight (IToF) depth-sensing camera driving the gesture-based control system, a touchpad along its front face, a GPS receiver, and a 13 megapixel camera — though the company has not yet shared its video resolution. Everything is powered by an internal battery, and while Humane has yet to go on record with a battery life estimation the inclusion of two wireless-charging "battery booster" backs in the box as standard is telling.

As the name implies, though, the Ai Pin is for more than hands-free calls. Following a manual interaction not wholly dissimilar to the activation of a Star Trek Combadge, the microphones begin to listen to your query — and it does more than just answer back with the output of ChatGPT or another large language model: Humane claims it can summarize your inbox, provide foreign language translation, "craft messages in your tone of voice," and even recognize objects to the point of being able to advise you on the nutritional value of the food you eat.

There's a catch, however, and it's one Humane itself inadvertently highlighted in its slick announcement video. While demonstrating the Ai Pin's ability to answer questions, the company had it detail the next total solar eclipse — and, as spotted by David Friedman and others, suggest exactly the wrong half of the world in which to see it. Later in the same video, the Ai Pin recognizes a handful of almonds and confidently overstates their protein content almost threefold.

These, presumably, are cherry-picked examples of the Ai Pin performing at its best — but its failings should come as no surprise, as they're rooted in the very technology of a large language model. LLMs are, in effect, text-completion engines: given an input, they generate the most likely output — with no care as to whether the response is factually accurate. Responses from the Ai Pin, then, should be taken with a pinch of salt — and with no screen, you'll have to pull out your smartphone in order to check the truth of anything it tells you.

The company is to launch the AI Pin next week, priced at $699 for the pin, two of the wireless "battery boosters," a charging pad, a charging case, and a power supply — but all purchases require a mandatory $24-a-month T-Mobile line plan, which provides the Ai Pin with unlimited calls, text, and data, plus "full access to Humane’s growing suite of Ai-powered services with no limits on the number of queries you can pose," the company says.

More information is available on the Humane website.