Sam Altman, over bread rolls, explores life after GPT-5

I’m looking at Alcatraz Island from a Mediterranean restaurant in San Francisco with hundred-dollar fish entrées on the menu. As I make small discuss with different reporters, OpenAI CEO Sam Altman jumps by the door on my left. Altman’s trying down at his naked iPhone to indicate us all one thing, and an intrusive thought slips out of my mouth: “No telephone case is a daring alternative.”

In fact, I instantly notice that the billionaire CEO of OpenAI, who employs Apple veteran Jony Ive, cares extra about preserving the iPhone’s unique design than the $1,000 it prices to switch one.

“Hear, we’re going to ship a tool that’s going to be so stunning,” says Altman, referring to OpenAI and Ive’s forthcoming AI device. “In the event you put a case over it, I’ll personally hunt you down,” he jokes.

Altman has gathered roughly a dozen tech reporters to hitch him and different OpenAI executives for an on-the-record dinner (and off-the-record dessert). The evening raises extra questions than it solutions.

As an illustration, why is Nick Turley, the VP of ChatGPT, kindly passing me a lamb skewer only a week after launching GPT-5? Is that this to encourage me to write down good issues about OpenAI’s largest AI mannequin launch but, which was comparatively disappointing given the years of hype round it?

In contrast to GPT-4, which far outpaced rivals and challenged expectations of what AI can do, GPT-5 performs roughly on par with fashions from Google and Anthropic. OpenAI even introduced back GPT-4o and ChatGPT’s model picker, after a number of customers expressed issues over GPT-5’s tone and its mannequin router.

However all through the evening, it turns into clear to me that this dinner is about OpenAI’s future past GPT-5. OpenAI’s executives give the impression that AI mannequin launches are much less necessary than they had been when GPT-4 launched in 2023. In spite of everything, OpenAI is a really totally different firm now, centered on upending legacy gamers in search, client {hardware}, and enterprise software program.

OpenAI shares some new particulars about these efforts.

Altman says OpenAI’s incoming CEO of functions, Fidji Simo, will oversee a number of client apps exterior of ChatGPT — ones OpenAI has but to launch. Simo is slated to start out work at OpenAI in just some weeks, and she or he would possibly find yourself overseeing the launch of an AI-powered browser that OpenAI is reportedly growing to compete with Chrome.

Altman suggests OpenAI would even contemplate shopping for Chrome — probably a suggestion that may be taken extra severely than Perplexity’s bid — ought to it turn out to be obtainable. “If Chrome is actually going to promote, we must always check out it,” he says earlier than all of us and asking: “Is it truly going to promote? I assumed it wasn’t gonna occur.”

Simo additionally would possibly find yourself working an AI-powered social media app — one thing the OpenAI CEO has mentioned he’s considering exploring. The truth is, Altman says there’s “nothing” inspiring to him about the way in which AI is used on social media as we speak, including that he’s considering “whether or not or not it’s potential to construct a a lot cooler form of social expertise with AI.”

Whereas Turley and Brad Lightcap, OpenAI’s COO, largely give the ground to Altman, ingesting wine alongside the opposite seated company, Altman additionally confirms reviews that OpenAI plans to again a brain-computer interface startup, Merge Labs, to compete with Elon Musk’s Neuralink. (“We’ve not executed that deal but; I would really like us to.”)

How intertwined that firm can be with OpenAI’s fashions and gadgets stays to be seen. Altman describes it solely as a “an organization that we’d put money into.”

For all of the discuss of browsers and mind chips, although, the elephant within the room stays GPT-5’s tough reception. Ultimately, the dialog circles again to the mannequin that has prompted our group dinner within the first place.

Turley and Altman say they’ve discovered rather a lot from the expertise.

“I legitimately simply thought we screwed that up,” says Altman on deprecating GPT-4o with out telling customers. Altman says OpenAI will give customers a extra clear “transition interval” when deprecating AI fashions sooner or later.

Turley additionally says OpenAI is already rolling out a brand new replace to make GPT-5’s responses “hotter,” however not sycophantic, such that it received’t reinforce unfavourable behaviors in customers.

“GPT-5 was simply very to the purpose. I like that. I take advantage of the robotic persona — I’m German, you understand, no matter,” says Turley. “However many individuals don’t, they usually actually like the truth that ChatGPT would truly test in with you.”

It’s a fragile stability for OpenAI to strike, particularly on condition that some customers have developed dependencies on ChatGPT. Altman says OpenAI believes that lower than 1% of ChatGPT customers have unhealthy relationships with the chatbot — which might nonetheless be tens of tens of millions of individuals.

Turley says OpenAI has labored with psychological well being specialists to develop a rubric to guage GPT-5’s solutions, guaranteeing that the AI mannequin will push again on unhealthy behaviors.

That mentioned, evidently GPT-5 hasn’t harm OpenAI’s enterprise. The truth is, Altman says OpenAI’s API visitors doubled inside 48 hours of GPT-5’s launch, and the corporate is successfully “out of GPUs” due to all of the demand. Cursor and different AI coding assistants have since made GPT-5 their default AI fashions.

In some ways, the evening’s contradictions — disappointing launches, record-breaking utilization — mirror OpenAI’s unusual actuality proper now.

Given OpenAI’s bets — and others the corporate is making round information facilities, robotics, and power — Altman clearly has ambitions of working a a lot larger firm than simply the ChatGPT maker. The ultimate type might look one thing like Google’s father or mother Alphabet, however even perhaps broader.

Because the evening winds down, it turns into clear we aren’t gathered to mirror on GPT-5 in any respect. We’re being pitched on an organization that’s desirous to outgrow its well-known and controversial product.

It appears probably that OpenAI will go public to fulfill its huge capital calls for as a part of that image. In preparation, I believe Altman desires to hone his relationship with the media. However he additionally desires OpenAI to get to a spot the place it’s not outlined by its greatest AI mannequin.