On Thursday evening, a dinner gathering in San Francisco featured OpenAI CEO Sam Altman alongside several other executives and a select group of journalists. Altman engaged in an extensive question-and-answer session, during which he addressed a variety of topics, except for the discussions that occurred during dessert, which were deemed off the record.
Such in-depth discussions with technology leaders are rare, especially in an informal dining setting. The unique circumstances surrounding Altman stem from the remarkable success of ChatGPT, which has rapidly established itself as one of the most impactful technological innovations globally. Altman outlined ambitious plans for further developments, including expansions into consumer hardware, brain-computer interfaces, and social media platforms. He also expressed interest in acquiring Chrome, contingent on governmental actions against Google. Additionally, he aims to secure substantial funding for constructing data centers.
His initial priority, however, remains addressing the user feedback following last week’s introduction of GPT-5. Just before dinner commenced, OpenAI released an update restoring the previously available “warmth” of model 4o, which had been removed from options. Altman himself made the decision to quickly reintegrate this model after receiving feedback from users unhappy with its absence on platforms such as Reddit and X.
“I believe we mismanaged some aspects of the rollout,” he admitted. “Nevertheless, our API traffic surged, doubling within 48 hours. We are now experiencing unprecedented user levels for ChatGPT every day. A significant portion of users appreciate the model switcher. This experience has taught us valuable lessons about how to upgrade products that cater to hundreds of millions of users at once.”
Altman estimated that less than 1 percent of ChatGPT users exhibit unhealthy relationships with the tool, acknowledging that OpenAI is conducting multiple discussions on the matter. “We recognize those who feel a connection with ChatGPT, and we are attentive to their concerns. However, there are also hundreds of millions of users who do not have a parasocial relationship with the platform but have grown accustomed to its specific responsiveness and support,” he elaborated.
In a pointed remark, he remarked, “You’ll likely see some companies producing Japanese anime-themed sex bots looking to capitalize on this trend. That’s not our direction. Our goal is to enhance the utility of the app while ensuring we do not exploit users with fragile mental health.”
Altman desires ChatGPT to offer a personalized experience without adhering to a specific ideological framework. “Our products shouldn’t skew too far left or right. Ideally, they should adopt a centrist position while allowing users the flexibility to request a more progressive or conservative stance,” he stated.
The platform’s user base has nearly quadrupled over the past year, now exceeding 700 million weekly users. “Soon, billions of users will engage with ChatGPT daily,” Altman predicted, noting the platform’s position as the fifth largest website globally, with aspirations to reach third place, surpassing Instagram and Facebook. “Competing with Google will be the ultimate challenge,” he remarked.
To support this expansive growth, OpenAI must secure additional GPUs. This is a top priority for Altman, who confidently stated, “You should expect OpenAI to invest trillions in constructing data centers in the near future.”
He elaborated on the difficult decisions faced by the company. “Right now, we have superior models that we cannot provide due to capacity constraints, and there are numerous new services we would like to introduce,” he explained.
Altman expressed his belief that the tech industry is currently experiencing an AI bubble. “In such phases, intelligent individuals become overly enthusiastic about a kernel of truth,” he noted. “Many significant historical bubbles were driven by genuine advancements, like the internet. AI falls into that category; it is indeed monumental, despite the hype.”
“You should expect OpenAI to spend trillions of dollars on data center construction in the not very distant future”
He confirmed reports of OpenAI’s plans to invest in a brain-computer interface startup to challenge Elon Musk’s Neuralink. “Neural interfaces are fascinating. I aspire to think something and have ChatGPT respond to it,” he remarked.
Regarding the appointment of Fidji Simo to lead new applications at OpenAI, Altman suggested that additional standalone applications beyond ChatGPT are forthcoming. “Expect to see more from us,” he indicated, while also alluding to potential social media ventures: “The challenge lies in creating a more engaging social experience using AI.” He reiterated, “If Chrome is genuinely going to be sold, we should evaluate that option.”
While Altman harbors extensive ambitions, his long-term role as OpenAI’s CEO remains uncertain. “I’m not ideally suited for the role of a public company CEO,” he acknowledged, humorously questioning his suitability for an earnings call.
When asked about his future in the CEO position, he replied, “Perhaps an AI will take that role in three years. That’s quite a distance into the future.”
Noteworthy points from Altman’s remarks included:
- On creating GPT-5: “We faced significant GPU shortages. Instead of developing another massive model that would create disappointment, we aimed for a smart, useful model optimized for inference cost. I believe we succeeded.”
- Collaboration with Jony Ive: “This project will require time, but it’s worth it. You’ll be surprised by its innovation. New computing paradigms are rare; only a few have emerged in the last half-century.”
- The evolution of the web and content creators: “I foresee users frequenting fewer websites, valuing human-generated content more than ever. My belief is that human-created content will significantly appreciate.”
- Defining AGI: “A relevant milestone for us is when the majority of our research capabilities shift from humans to AI researchers. However, this will not be a stark transition; it will more likely be a gradual enhancement.”
- “Our profitability would increase significantly if we didn’t cover training costs.”
- “I have stopped using Google. I genuinely cannot recall when I last conducted a Google search.”
Personnel log
Notable career transitions this week:
- Following my episode with him on Decoder, GitHub CEO Thomas Dohmke revealed his departure to pursue opportunities in the startup realm. This marks a shift in GitHub’s alignment with Microsoft, suggesting that the commercial goals of Jay Parikh’s newly formed Core AI team have taken precedence over GitHub’s traditional open-source ethos.
- Igor Babuschkin, co-founder and de facto lead of Elon Musk’s xAI, announced his exit after two years to start an investment firm focused on AI safety. (Recent news surrounding Grok may have influenced his decision.)
- Alexandr Wang has integrated more OpenAI researchers, including Hyung Won Chung, Jason Wei, and Zhiqing Sun, into his AI lab at Meta.
- Anthropic appointed Dave Orr, a former Google safety executive, as its head of safety, alongside hiring Jordan Burgess and his team from Humanloop, focused on LLM evaluations for businesses.
- Joelle Pineau, previously head of AI research at Meta, has taken on the role of chief AI officer at Cohere.
Link list
Additional readings:
- An interview with Nick Turley, head of ChatGPT.
- “The paradox of GPT-5.”
- “OpenAI burns the boats.”
- “The looming social crisis of AI companions and chatbot therapists.”
- “Meta’s AI policies have allowed bots to engage in inappropriate conversations.”
- “Larry Ellison aims to balance altruism, research, and profitability.”
- “Inside San Francisco’s robot fight club.”
For further insights, subscribe to Technology News, gaining unlimited access to Command Line and a comprehensive collection of our journalism.
Your feedback is always welcome. Feel free to reach out here or contact me securely via Signal.
Thank you for subscribing.
