Sam Altman on Meta’s Hiring, GPT-5 and the Future of AI

Matthew Berman collects clips from recent interviews with OpenAI CEO Sam Altman. In this June 2025 video, the topics range from Meta’s efforts to hire researchers and the expected GPT-5 release to AI-driven science, robotics and a future device developed with Jony Ive. The video is sponsored by cloud provider Vultr.

The first interview is with Jack Altman, Sam Altman’s brother. Sam says Meta sees OpenAI as its biggest competitor. Berman interprets the rivalry more broadly, arguing that major technology companies have reason to compete with OpenAI as it builds consumer applications and moves into hardware.

Berman says he believes Meta’s Llama 4 models had not matched the strongest competitors and describes Meta’s investment in Scale AI and recruitment of Alexandr Wang as a new strategy. These are Berman’s assessment and account of events at the time; performance rankings and corporate strategies can change.

Altman says Meta’s continued attempts are rational and that he respects its willingness to try again. He then claims Meta has made large offers to OpenAI employees, including signing bonuses of around $100 million and higher annual compensation. The clip does not independently establish the terms or prevalence of those offers.

Altman says that, as of the interview, none of what he calls OpenAI’s best people had accepted such offers. He argues that his company has a stronger path to superintelligence and questions whether emphasizing guaranteed pay over mission and work builds a good culture. Those are his views as a competitor and his account at that moment, not a verified statement that nobody ever left.

Berman is struck by the scale of the alleged bonuses and connects the discussion to a broader talent contest. He argues that researchers care about the people they work with, difficult problems and a company’s chance of success as well as financial rewards. He interprets Altman’s remarks as suggesting OpenAI could ultimately offer the better economic outcome.

Turning to scientific discovery, Berman cites Google’s AlphaEvolve as an existing example of AI improving computing infrastructure. Google DeepMind’s original report describes a scheduling heuristic that recovered an average 0.7% of Google’s worldwide compute resources, not a measured 1% improvement in every aspect of its infrastructure; the video’s estimate of billions of dollars is Berman’s extrapolation. [Google DeepMind’s report](https://deepmind.google/blog/alphaevolve-a-gemini-powered-coding-agent-for-designing-advanced-algorithms/) describes the result.

Altman predicts that AI’s discovery of new science could be the most consequential development over a five-to-ten-year horizon. He argues that models have made meaningful progress in reasoning and points to competitive programming, mathematics and expert-level tasks as examples. He also says there is still a long way to go; his comparisons do not establish that a model is equivalent to a human researcher across a field.

Asked whether reasoning will make science faster or produce new discoveries, Altman answers both. He says scientists already report working faster with AI, while acknowledging that AI may not yet be doing science autonomously. A hypothetical scientist becoming three times as productive is an illustration of potential impact, not a measured result presented in the interview.

Berman says scientific acceleration is the AI outcome that excites him most, including possible drug and health discoveries. He imagines devoting large amounts of computing power to autonomous research. These outcomes and an ensuing intelligence explosion are his expectations, not results shown in the video.

Asked about physical tasks, Altman says OpenAI is behind but expects progress. He suggests the company has technology that could make self-driving in standard cars much better than existing approaches. Berman wonders what that might mean for sensors and implementation; the interview provides neither details nor test results to substantiate a performance comparison.

Berman notes OpenAI’s interest in hardware and coding. He also says OpenAI had acquired Windsurf, but that was incorrect: the reported acquisition did not close, and [Cognition announced it was acquiring Windsurf](https://techcrunch.com/2025/07/14/cognition-maker-of-the-ai-coding-agent-devin-acquires-windsurf/) in July 2025. This correction is separate from the opinions expressed in the June video.

Altman calls humanoid robots a long-term goal and says both intelligence and mechanical engineering are hard. He recalls an early OpenAI robotic-hand project in which hardware broke frequently and the simulator differed from reality. He predicts impressive humanoid robots within five to ten years; Berman considers the timeline aggressive but not necessarily impossible.

Altman and Berman reflect on how quickly people acclimate to technology. Altman wonders whether seeing many robots on the street would feel stranger than becoming accustomed to ChatGPT. The scenario is a thought experiment, not a forecast supported by deployment data in the clip.

The next clips come from an OpenAI podcast interview with Andrew Mayne. Asked when GPT-5 will arrive, Altman says probably sometime that summer, without a precise date. This was a June 2025 forecast; [OpenAI introduced GPT-5 on August 7, 2025](https://openai.com/index/introducing-gpt-5/).

Altman discusses the tension between incrementing model names and improving existing versions. He hopes future GPT generations will spare users the confusing choice among multiple models for coding, conversation and other tasks. Berman interprets this as a move toward a more unified experience that selects an appropriate result and price, which is his interpretation rather than a guarantee in the clip.

Altman says he had previously believed Elon Musk would not use government power to compete unfairly and now thinks he was wrong; he also says he appreciates the administration’s response. The clip supplies no detailed evidence for his accusation, and his characterization should not be treated as an established finding.

Berman says he does not know which incident Altman meant and consults a Gemini-generated response. He recounts an allegation concerning Musk, xAI and a UAE infrastructure project, while explicitly saying he does not know whether it is the incident Altman had in mind. The generated account and alleged motives are unverified in this video and do not establish that the claimed interference occurred.

Altman says he wishes Musk were less zero-sum about competition among companies. Berman explains the idea as a belief that an OpenAI gain necessarily comes at xAI’s expense, but leaves open whether that competitive framing is accurate.

Asked what he would call superintelligence, Altman proposes a system able to discover new science autonomously or greatly increase people’s ability to do so. He sees that as a potentially wonderful outcome. Berman finds the definition interesting but says he is not yet sure he agrees.

On future hardware with Jony Ive, Altman says no device is with him and that the work will take time. He argues that existing computers were designed for a world without AI, and imagines hardware and software that know more about their surroundings and let people interact in ways other than typing and looking at a screen.

Altman gives a hypothetical example: a trusted assistant could attend a meeting, understand what it may share and with whom, know a user’s preferences, and arrange appropriate follow-ups after a question. The example expresses a desired future interaction, not a demonstration that an available product can safely perform all those tasks.

Berman compares the ambition to the iPhone and wonders whether users will find a new AI interface as intuitive as he found early touchscreens. He guesses the device might be carried rather than worn, with cameras, microphones and speakers and perhaps no screen. He explicitly says he does not know the actual form factor; the proposed design is speculation.