OpenAI CEO Sam Altman says AI now surpasses the smartest scientists

A leap in cognitive capability

During a speech at the Abu Dhabi International Petroleum Exhibition and Conference, OpenAI CEO Sam Altman described a rapid transformation in machine intelligence. He noted that at the beginning of this year, artificial intelligence was generally not as smart as the world's most talented scientists. Now, he claims the situation has flipped, with models mostly exceeding that level of intelligence.

This shift represents a significant milestone in the development of large language models. These systems are now capable of performing substantial portions of white-collar work that were previously thought to require human expertise. Altman's observation suggests that the barrier between human-level reasoning and machine output is narrowing quickly.

The speed of this progress has been a central theme of his recent remarks. By highlighting the dramatic leap in capabilities over just twelve months, he underscores how quickly the technology is maturing. This evolution moves AI from a simple assistant to a sophisticated entity capable of complex problem-solving.

The human role in AI governance

Despite the increasing power of these models, Altman stressed that human oversight remains a non-negotiable requirement for the future. He argued that while AI can perform many tasks, human judgment, creativity, and accountability must stay at the heart of the technology. He warned against a scenario where people gradually hand over decision-making power to machines.

Altman expressed a clear philosophy on the relationship between people and software. He stated that AI needs to be a tool built for people and needs to serve people, not the other way around. This perspective aims to reassure users that the goal is augmentation rather than total replacement of human agency.

The CEO emphasized that there should not be a handoff to AI where it decides what is best for humanity. By keeping humans in control, the goal is to ensure that the disruptive impacts on global economies and employment are managed responsibly. This balance between high-performance capabilities and human-led direction is a key pillar of his vision.

Energy as the primary bottleneck

As AI models grow more complex, the physical requirements to run them are becoming a major logistical challenge. Altman identified the physical power supply as the single biggest bottleneck for the entire sector. He explained that while many other technical hurdles can be overcome, the limitations of the physical world remain a constant factor.

The CEO noted that atoms are still atoms, meaning the infrastructure required to power these systems is the dominant factor in determining cost and abundance. Because computing requires massive amounts of electricity, the availability of energy will dictate how fast the industry can scale. This moves the conversation from software limitations to hardware and infrastructure constraints.

Addressing these energy demands is a priority for the companies building the next generation of models. Without a steady and scalable power supply, the progress of AI may be slowed by the realities of the physical world. This focus on energy highlights the growing intersection between the technology sector and the global energy market.

Global adoption and partnerships

The rapid adoption of these tools is already visible in regions like the United Arab Emirates. Altman pointed out that approximately 80% of young people in the UAE are using ChatGPT on a weekly basis. This high rate of usage demonstrates how quickly the technology is becoming integrated into the daily lives of the younger generation.

To further integrate these models into industrial sectors, OpenAI announced new partnerships with the Emirati technology group G42 and the state oil producer Abu Dhabi National Oil Company (ADNOC). These deals represent a significant milestone for the company's reach into the energy industry. They mark the first global deployment of OpenAI's frontier models by an energy company.

These partnerships signal a move toward applying high-level AI to heavy industry and infrastructure. By working with major state entities and technology groups, OpenAI is positioning its models to handle large-scale industrial applications. This transition from consumer tools to industrial partners shows the broadening scope of the technology's utility.

Trends in token consumption

The scale of AI usage can be measured by the consumption of tokens, which are the basic units of text processed by these models. Altman reported that global per capita token consumption has surged from near zero just six years ago. This growth shows a massive increase in how much people are interacting with AI on a regular basis.

Currently, the average person is consuming roughly 100,000 tokens a month. This represents a significant jump from the baseline of a few years ago. The data suggests that AI is becoming a standard part of the digital experience for a growing number of people worldwide.

The scale is even larger for professional and institutional users who interact with the models at a high volume. Altman noted that top institutional users are now consuming about 500 billion tokens every month. This massive volume of data processing highlights the intensity of the current AI boom and the demand for high-capacity computing.