OpenAI bought Ive’s outfit, called io, for $6.5 billion in May, but so far the pair have been quieter than a Genius Bar during a Linux convention about what they are building.
People in the know told the Financial Times that the device is meant to be a palm-sized, screen-free gadget that listens, watches and responds to the user’s surroundings like a chatty pet rock with a PhD. But the wonder duo have yet to crack several crucial problems, from the assistant’s “personality” to the eye-watering cost of the computing muscle needed to run OpenAI’s models on consumer hardware.
A source close to Ive said: “Compute is another huge factor for the delay. Amazon has the compute for an Alexa, so does Google [for its Home device], but OpenAI is struggling to get enough compute for ChatGPT, let alone an AI device, they need to fix that first.”
Another insider from OpenAI shrugged off the concerns, saying the hiccups were normal growing pains in the product development process.
Multiple people familiar with the project said the gadget is about the size of a smartphone and uses cameras, microphones and speakers to communicate. It’s designed to sit on a desk or be carried around, and could even include several cameras.
One person said the device would be “always on”, hoovering up data all day to help build a virtual assistant with actual memory. That means it could watch and listen without waiting for a wake word, something that might make the EU’s privacy watchdogs twitch.
OpenAI, now the world’s most valuable private company with a valuation of $500 billion after leapfrogging Elon Musk’s SpaceX, sees this hardware as its next cash cow. The goal is to outdo the dull smart speakers of the past decade such as Amazon’s Echo, which mostly play music and set kitchen timers, by creating something that feels more like a “friend who’s a computer”.
“The concept is that you should have a friend who’s a computer who isn’t your weird AI girlfriend … like Siri but better,” said one person briefed on the plans.
Another insider added, “Model personality is a hard thing to balance. It can’t be too sycophantic, not too direct, helpful, but doesn’t keep talking in a feedback loop,” they said.
But designing a device that knows when to shut up is proving harder than it sounds. The team is also struggling to settle on the right “voice” and behaviour for the assistant; a tricky mix of helpful and human without tipping into creepy or clingy.
The market for this sort of gizmo is already littered with corpses. Humane’s AI Pin has been scrapped, while Friend, an AI pendant, was widely mocked for being both “creepy” and “snarky.”
OpenAI has been raiding the talent cupboards of Silicon Valley to get this thing off the ground. Ive’s io acquisition brought in more than 20 ex-Apple hardware engineers, while a dozen other Job’s Mob veterans have joined since. The start-up has also pinched staff from Meta’s Quest headset and smart-glasses team.
Manufacturing will be handled by Chinese contractor Luxshare, though the assembly might take place outside China, according to insiders.
OpenAI and Ive’s design firm LoveFrom declined to comment, which might be the only thing about this mysterious device that works as intended.