Photo-Illustration: Intelligencer; Photo: Getty Images

Last year, OpenAI announced it would be working with iPhone designer Jony Ive on a range of physical devices. Rumors swirled about various surveillant doodads: a pen, an earbud, a necklace, a pocket shard. Now, a few weeks out from the unveiling of Ive’s other big post-Apple design collaboration — the Ferrari Luce, the intensely polarizing look of which led the company’s former boss to say he hopes “they remove the prancing horse from that car” — reporter Mark Gurman has leaked information on OpenAI’s first device, a smart speaker that’s “mobile, screen-free,” and intended to function like an AI “home computer.”

The product — still under development — is meant to serve as a humanlike AI companion that lives in the home, said the people, who asked not to be identified because the project hasn’t been announced. It will help control smart-home appliances, play media, answer questions, respond to messages and tap into the range of capabilities offered by OpenAI’s ChatGPT, they said.

This sounds like a bid to create a slightly new category of hardware. Smart speakers that live in your house tend to stay in one spot or work in unison across different rooms. Movable smart speakers, like the Sonos Move, or the countless bluetooth speakers that include little-used Alexa features are marketed as portable and are mostly dedicated to music. In addition to the devices being built around ChatGPT, the suggestion here is that, in order for them to be constantly available — and able to listen — users will be expected to take them along inside the home.

Okay! We’ll see. Smart speakers have sold well over the years and have gradually and naturally morphed into a few subcategories. You’ve got the classic early Echo-type devices, mostly used for playing music, setting timers, and, particularly within the past five years, operating increasingly common smart-home configurations. More recently, companies like Amazon shifted toward countertop tablets, which likewise fulfill some old “home computer” needs, doubling as TVs, calendars, and picture frames. In the 12 years since Amazon opened up the category, though, the “smart” part of these devices has been a consistent letdown. In 2014, the Echo launched with the promise that, the more you use it, “the more it adapts to your speech patterns, vocabulary, and personal preferences.” As with Siri, which, years into the LLM era, is just barely exceeding the capabilities it was first marketed with in the 2010s, the smart-speaker concept has only just broken out of set-timer purgatory, and not completely: Alexa+, the post-GPT revision of Amazon’s platform, can now chat with you at length and is better at taking notes and updating calendars, but it still feels like using Alexa. 

Companies such as Amazon, Google, and Apple have had years to integrate their speakers with numerous smartphone protocols, and all have the benefit of existing within larger ecosystems. Whatever OpenAI builds will need that stuff to work pretty well right away. All three of those tech giants are also in the process of further building out LLM-based features across those ecosystems, so a device that talks like ChatGPT won’t have enormous novelty by default (Apple’s new Siri is built on top of Google Gemini). For OpenAI’s part, attempts to assistantize ChatGPT with daily digests and proactive communication haven’t yet taken.

There are signs that Apple is, if not worried, sort of annoyed: The company is currently suing OpenAI for what it characterizes as fairly brazen attempts at IP theft. But what this report mainly suggests is that OpenAI (which, in the years following the release of ChatGPT, frequently used the the movie Her as an aspirational touchstone — at one point releasing a demo so on the nose that Scarlett Johansson posted an angry letter about it) remains, in its alleged era of avoiding “side quests” and doubling down on paying enterprise customers to compete with Anthropic, fixated on the idea of creating an intimate personal assistant.

her

— Sam Altman (@sama) May 13, 2024

This isn’t a bet any of the other big labs are making so explicitly, but it’s one with which OpenAI has since gained useful, though complicated, insights. The model generation it was demonstrating with the Johansson-ish voice was ChatGPT 4o, which would later be withdrawn for being too sycophantic but which was also linked, more clearly than any other model to date, with AI psychosis, attachment, and incitements to harm. In one way, for OpenAI, this was a disaster and a scandal. In another, it was proof of concept: They really can get people attached to their AI assistants. It just might drive some of them insane.

Sign Up for John Herrman column alerts

Get an email alert as soon as a new article publishes.

Vox Media, LLC Terms and Privacy Notice