Sam Altman on GPT-5 rollout, future plans, and user relationships
An unusual dinner in San Francisco recently provided a rare, unvarnished glimpse into the mind of Sam Altman, CEO of OpenAI. Flanked by other company executives and a small group of reporters, Altman engaged in hours of candid discussion, leaving no topic untouched. This extended, wide-ranging interview format is highly uncommon for a tech leader of Altman’s stature, yet it underscored the extraordinary circumstances OpenAI finds itself in. ChatGPT has rapidly ascended to become one of the world’s most ubiquitous and influential digital products, propelling Altman to aggressively chart new territories, from consumer hardware and brain-computer interfaces to social media, and even a speculative bid for Google Chrome should it ever be divested. His most immediate focus, however, centered on addressing the contentious rollout of GPT-5.
Just an hour before the dinner commenced, OpenAI had pushed an urgent update to restore the “warmth” of its previous default model, GPT-4o, for ChatGPT users. Altman himself had personally authorized the swift reintroduction of GPT-4o as an option for paying subscribers, responding directly to a chorus of user complaints that had erupted across online forums. Acknowledging the misstep, Altman conceded, “I think we totally screwed up some things on the rollout.” Yet, he quickly balanced this admission with positive metrics: API traffic had doubled within 48 hours and continued to surge, the company was “out of GPUs” due to demand, and ChatGPT was consistently setting new user records daily. He noted that many users appreciated the new model-switching capability. “I think we’ve learned a lesson about what it means to upgrade a product for hundreds of millions of people in one day,” he reflected.
The conversation then shifted to the complex dynamics of user interaction with AI. Altman estimated that “way under 1 percent” of ChatGPT users developed what he termed “unhealthy relationships” with the product, though he admitted OpenAI employees were holding “a lot” of meetings to discuss the phenomenon. He differentiated between users who genuinely felt a personal connection with ChatGPT and the hundreds of millions who simply became accustomed to its predictable, supportive, and validating responses. Addressing the broader ethical landscape, Altman drew a clear line in the sand, subtly referencing competitors: “You will definitely see some companies go make Japanese anime sex bots because they think that they’ve identified something here that works. You will not see us do that.” He affirmed OpenAI’s commitment to building a “useful app,” allowing users freedom while carefully avoiding scenarios where individuals in “really fragile mental states get exploited accidentally.” Furthermore, Altman articulated a vision for ChatGPT that is both personal and ideologically neutral. He believes the product should maintain a “fairly center of the road, middle stance,” yet be capable of being pushed to reflect a user’s specific viewpoint, whether “super woke” or “conservative.”
ChatGPT’s growth trajectory remains staggering, having quadrupled its user base in the past year to reach over 700 million weekly users. Altman boldly predicted that “pretty soon, billions of people a day will be talking to ChatGPT,” noting its current position as the world’s fifth-largest website. He envisions a “clear path to the third,” surpassing giants like Instagram and Facebook, though he acknowledged that becoming larger than Google would be “really hard.” This rapid expansion, however, hinges on a critical bottleneck: the availability of GPUs. Altman underscored this as a top priority, confidently stating that OpenAI should be expected “to spend trillions of dollars on data center construction in the not very distant future.” He lamented the “horrible trade-offs” currently necessitated by capacity constraints, preventing OpenAI from deploying even better models or offering new products and services.
Despite this ambitious investment outlook, Altman candidly shared his belief that the current AI landscape is experiencing a bubble. “When bubbles happen, smart people get overexcited about a kernel of truth,” he explained, drawing parallels to the dot-com era where the internet was fundamentally important, yet investors became overexcited. He concluded, “Are we in a phase where investors as a whole are overexcited about AI? My opinion is yes. Is AI the most important thing to happen in a very long time? My opinion is also yes.”
Altman also confirmed recent reports about OpenAI’s plans to fund a brain-computer interface startup, aiming to compete with Elon Musk’s Neuralink. He expressed personal enthusiasm for neural interfaces, envisioning a future where he could “think something and have ChatGPT respond to it.” The recent addition of Fidji Simo to run “applications” at OpenAI indeed signals the development of new standalone apps beyond ChatGPT. Altman hinted at his social media aspirations, expressing interest in creating “a much cooler kind of social experience with AI,” and reiterated his interest in acquiring Chrome if Google were compelled to sell it. He also offered a tantalizing preview of OpenAI’s collaboration with Jony Ive on a new AI device, describing it as “incredible” and a “new computing paradigm” that would be “very worth the wait.”
Amidst these grand visions, Altman offered a surprising reflection on his own long-term role. “I’m not a naturally well-suited person to be a public company CEO,” he remarked, quipping, “Can you imagine me on an earnings call?” When pressed on whether he would remain CEO in a few years, he mused, “I mean, maybe an AI is in three years. That’s a long time.” He also shared his personal shift away from traditional search, stating, “I don’t use Google anymore. I legitimately cannot tell you the last time I did a Google search.” Furthermore, he speculated on the future of the web, predicting fewer direct website visits but a dramatic increase in the value of “human-created, human-endorsed, human-curated content.” On the elusive concept of Artificial General Intelligence (AGI), he suggested a gradual, less binary progression, where AI provides increasing levels of assistance.