OpenAI’s rumored ‘always on’ AI device sounds terrifying – but Sora 2 shows it doesn’t care about boundaries

OpenAI may be struggling with its still-in-development AI device. According to a new report, OpenAI CEO Sam Altman and his lead designer, and former Apple^[1] aesthetic genius Jony Ive, are wrestling not so much with the form factor but more crucial elements like local compute power and personality for the always-on device.

If the Financial Times^[2] is right, this is a level of care and concern not typically on display from OpenAI AI, and, to be honest, I’m not buying it.

In OpenAI’s recent consumer-facing AI past, there is a long history of moving fast, breaking things, and mopping up later.

We have a couple of solid and quite recent examples, including the GPT-5^[3] model release. While there’s little doubt this ChatGPT^[4] model is smarter, better at avoiding hallucinations by slowing down and not leaping to fill in knowledge gaps with nonsense, its personality was a bit cold. At least, compared to previous models^[5], which, in a rather shocking move, were suddenly unavailable.

Surely, Altman and company must have known it would be a bad idea to do that, but they did it anyway and then apologized later^[6], promising to bring back GPT-4o^[7] access.

When OpenAI released image-generation tools that allowed any prompt to deliver images in styles that resembled, for instance, Studio Ghibli art, there was a huge outcry^[8]. Altman’s response, though, was a mixture of “empathy” and, as he told Chris Anderson earlier this year^[9], “I think it would be cool to figure out a new model where if you say I want to do it in the name of this artist and they opt in, there’s a revenue model there…”

That sounded good, but then a few months later, Sora 2^[10] arrived. If you’re not familiar with it, I’m sure you’ve seen the videos.

This powerful, new generative video tool lives in a new, invite-only Sora 2 Social Media app, and it’s already letting people generate extremely believable videos featuring proprietary characters and actors, both living and dead. In fact, I’ve been appalled to stumble on a series of Sora 2 videos featuring Robin Williams^[11]. The late actor has been dead for over a decade, and, honestly, these videos just fill me with enormous sadness.

Altman, of course, is somewhat apologetic and promises in a post on his blog^[12], “We are going to try sharing some of this revenue with rightsholders who want their characters generated by users.”

The comments echo what he told Anderson in April and follow Altman and OpenAI’s well-worn pattern of rolling out new features, seeing how quickly and badly things break, and then mopping up. The good news in this case is that Sora 2 will soon give the rightsholders control, with opt-ins for use of characters and likenesses.

Launch fast and break things

I’m glad OpenAI and Alman are addressing this fiasco, but it gives me little comfort when it comes to OpenAI’s secretive AI device.

As described in this and other reports^[13], this gadget would feature at least one camera, a microphone, and a speaker. It would live on the desk or on your person. Other reports say it’ll be palm-sized and wearable, with a source telling the Financial Times that it will “take audio and visual cues from the physical environment and respond to users’ requests”. They added that “the concept is that you should have a friend who’s a computer who isn’t your weird AI girlfriend”.

As an always-on AI device that builds an ambient AI relationship without you having to prompt it (until you want to), this could be either a very powerful consumer wearable or a disaster.

The struggle within OpenAI seems to revolve around compute power (how to have enough local and, especially, cloud-based processing to support instant generative AI response for potentially millions of customers), and, perhaps more importantly, the personality.

As we can see from the above, OpenAI still struggles to get this basic component right. After launching ChatGPT in 2022 with a human-pleasing, prompt-response interface, OpenAI has constantly adjusted ChatGPT’s personality, often in response to concerns that it was too sycophantic^[14]. The problem with going in the other direction was that it turned out many people liked that AI’s somewhat cloying personality.

No AI weirdos

But an always-listening AI life partner that is maybe too friendly could be construed as creepy. OpenAI, probably with Jony Ive’s considerable influence (after all, no one ever accused Apple’s hardware or software of being “creepy”), is surely desperate to avoid the AI creep factor.

That does not mean it will. In fact, if past is prologue, OpenAI’s secret AI device will launch with exactly too much power, too much AI nosiness, and considerable amounts of creepiness.

That will launch a wave of stories, complaints, and, perhaps as OpenAI likes it, attention. Sam Altman will, within a week or so post-launch, announce that he’s heard the complaints and that OpenAI is at work on a fix so the new gadget is watchful and helpful without being weird.

This assumes anyone actually wants such a device. But if and when it arrives, mark my words, it will be far from perfect, it will break things, and we’ll simply rinse and repeat, AI style.

Follow TechRadar on Google News and add us as a preferred source to get our expert news, reviews, and opinion in your feeds. Make sure to click the Follow button!^[15][16]

And of course you can also follow TechRadar on TikTok for news, reviews, unboxings in video form, and get regular updates from us on WhatsApp too.^[17][18]

You might also like