Altman says energy is AI's biggest constraint, rules out handing control to machines

Summary

• OpenAI chief Sam Altman told Adipec in Abu Dhabi that humans must retain control of AI.
• He named energy the single biggest challenge facing the industry as compute demand climbs.
• The remarks followed OpenAI's agreements with Abu Dhabi spanning energy, education, investment and government services.

The latest

Humans, not AI systems, must stay in control of the future, OpenAI chief executive Sam Altman said Monday at the opening of the Adipec conference in Abu Dhabi. Speaking with The National's Editor-in-Chief Mina Al-Oraibi, Altman said there should be no handoff to AI and that the technology must serve people rather than the reverse. Asked to name his biggest challenge today, he answered in one word: energy.

Details

• The core message: Altman said the industry can debate many things, but human control is not one of them, according to The National. He described the position as a value OpenAI intends to fight for, and said he could not believe it differentiated OpenAI from other AI companies.
• The capability claim: By Altman's account, AI crossed a threshold this year, moving from trailing the world's best scientists to mostly outperforming them. He said systems can now handle a lot of parts of many white-collar jobs, but offered no breakdown of which roles or sectors he meant.
• On safety: Altman said he believed society would navigate the transition fine, but warned the industry must show it can prevent harm as systems grow more capable. He framed the open question as how companies guarantee, and give the world comfort in that guarantee, that serious safety and alignment incidents are avoided.
• The energy numbers: Altman said the power required per AI query has dropped sharply while total demand keeps climbing. Six years ago OpenAI's heaviest internal user consumed about 100,000 tokens a month; that figure is now roughly the global per-capita average, while the top user consumes about 500 billion.
• Why costs stall: Altman said costs could fall in almost every way except that atoms are still atoms, calling power the dominant factor in how much AI the world can use. He said he hoped AI would ultimately deliver a net gain for global energy availability, without specifying a timeframe or mechanism.
• The Abu Dhabi deals: The remarks followed OpenAI's agreements with Abu Dhabi covering energy, education, investment and government services, The National reported. They include an arrangement with Adnoc and technology group G42 to apply OpenAI's most advanced models to the energy industry's hardest problems.
• The maths claim: Altman said OpenAI had solved one of the Millennium Prize problems, a set of long-unsolved mathematics challenges, using 10,000 agents running an upcoming GPT model for roughly four days with no hierarchy. The National reported the claim could not immediately be verified independently.
• On the UAE: Altman praised the UAE as consistently the most forward-looking country in the world for AI adoption, saying about 80 per cent of young people there use ChatGPT. He did not identify the source of that figure or the age range it covers.

Background

Adipec is the energy industry's main annual gathering in Abu Dhabi, drawing oil and gas executives and, increasingly, technology firms. Data centre power demand has pushed AI companies into direct negotiation with energy producers and states controlling large generation capacity.

Between the lines

Altman's two headline themes are linked. A pitch built on systems that now outperform top scientists only holds if the power exists to run them, which is why he named energy as his single biggest challenge in the same appearance. Delivering that message at an oil and gas conference, days after deals with Adnoc and G42, places OpenAI's capability argument in front of the producers who control supply.

What's next

Watch for independent verification of the Millennium Prize claim, the release of the GPT model Altman said ran the work, and details on how the Adnoc and G42 agreements translate into deployed systems.

Source: The National