Sam Altman reportedly foresees three ultimate outcomes for AI: a global regulator, consolidation of top labs into one entity, or one company breaking through first and gaining the ability to disable rivals' hardware.
The first is the establishment of a global regulatory body, the second is the consolidation of top labs into a single entity, and the third is that one company will take the lead in breaking through key technologies, even gaining the capability to paralyze the hardware of its competitors.
Sam Altman argued that AI water usage is greatly exaggerated, comparing 38,000 ChatGPT queries to the water needed to produce a single California almond.
For every 38,000 ChatGPT queries, that is the same amount of water used in the production of one single almond in California.
Sam Altman reportedly told OpenAI staff there are three possible endgames for AI, one of which is a single company dominating so completely that it effectively "melts everyone else's GPUs."
melts everyone else’s GPUs
Altman says he is personally accountable for which moral framework ChatGPT applies and has not slept well since launch.
The person who must be held responsible for the decisions about which moral framework is applied is me. I do not sleep well at night. I have not slept well a single night since ChatGPT was launched,
Altman says the small decisions about model behavior, multiplied across huge usage, carry big impact and worry him.
What keeps me up at night are the small decisions we make about how a model behaves. The impact is big,
Altman raises the dilemma of whether ChatGPT should provide bioweapon instructions even for research purposes, highlighting the tension with user freedom.
Should ChatGPT teach you how to build a biological weapon even if your aims are merely research-related?
Altman disagreed that AGI had already arrived but said OpenAI is close and expects to reach AGI by year-end.
OpenAI's Sam Altman also didn't agree with Huang's assessment, saying that he thought his company was almost there and expected to reach AGI by the end of the year.
Sam Altman promises that AI technology will bring shared prosperity on a scale that seems unimaginable today.
shared prosperity to a degree that seems unimaginable today
Altman reportedly told OpenAI staff he rejects the 'Eliezer plan' endgame in which one developer achieves self-improving AI and uses its power to 'melt everyone else's GPUs' and shut out rivals.
Obviously I think that one’s terrible.
Altman predicts AI will bring disease cures and universal extreme wealth for everyone.
universal extreme wealth
Altman reportedly told staff the superintelligence race could end in an IAEA-style regulator, a CERN-like lab merger, or one firm dominating via recursive self-improvement.
Sam Altman reportedly told OpenAI employees there are three plausible endings to the race to build superintelligence: an International Atomic Energy Agency-style inspector that would audit frontier labs, a merger of the leading labs into a single CERN-like entity, or one firm pulling so far ahead through recursive self-improvement that it could 'melt everyone else's GPUs.'
Altman reportedly called the scenario of a single firm using recursive self-improvement to disable rivals' hardware 'terrible'.
Altman called that outcome 'terrible,' according to the reporting.
Altman reportedly told staff that the "Eliezer plan", in which one lab wins recursive self-improvement and starves rivals of compute, is a terrible outcome included only for completeness.
In this scenario, one developer reaches recursive self-improvement first, then uses its runaway economic lead to starve every rival of compute — or, in Altman’s blunt phrasing, “melt everyone else’s GPUs.” It’s a winner-take-all outcome dressed up as a thought experiment, and Altman reportedly included it “only for completeness.”
Altman told staff the current ship-after-safety-tests approach is "pacing 1.0" and that the industry must choose a path for safe frontier AI within about a year.
According to The Information, Sam Altman walked OpenAI staff through three possible “endgames” for safely developing frontier AI in an internal talk last week. He reportedly called the current playbook — shipping models once they pass safety tests and flagging any concerns — “pacing 1.0.” The decision, he told employees, has to be made within roughly a year.
Altman reportedly favors an IAEA-style global watchdog that audits frontier labs, compute and energy use.
Per the report, Altman appeared to favor this option — and it’s not new. The IAEA-style idea appears in OpenAI’s own May 2023 post “Governance of superintelligence,” co-authored by Altman, Greg Brockman and Ilya Sutskever.
Altman reportedly described the AI safety choice as "pacing 2.0," urging a decision within a year and a speed limit or stopping point for self-improving AI.
Altman described the choice as “pacing 2.0” and called for a decision within a year. He said the approach must include a “speed limit” or stopping point for AI that improves itself.
Altman reportedly said the IAEA-style and CERN-style plans require heavy coordination and OpenAI cannot afford to cede ground now.
Altman reportedly told staff that the first two plans demand heavy coordination and that OpenAI cannot afford to cede ground now, with both the startup and Anthropic moving toward public offerings.
Sam Altman said OpenAI accepts that some things will go wrong and warned that a tidal wave of cybersecurity problems will come from open-source models.
We are willing to accept some things going wrong,” OpenAI CEO Sam Altman told Vanity Fair’s global editorial director, Mark Guiducci, last month, adding that “there is a coming tidal wave of cybersecurity problems that are going to come from these open-source models.
Sam Altman warned that losing control of the future to AI is unacceptable, stressing that OpenAI is on Team Humanity and AI must always serve people.
There are two ways AI progress could go very badly and that we must avoid. First, we could lose control of the future to AI. This is unacceptable; we are unapologetically on Team Humanity, and AI must always serve people.
Altman said he supports independent evaluators having employee-like access to AI labs.
OpenAI CEO Sam Altman said at the time that he also supported the idea of independent evaluators with employee-like access.
Sam Altman said in September that the timing was wrong for OpenAI to go public.
“Right now would be an ill-advised moment to go public,”
Sam Altman said OpenAI will not go public in 2026 because more AI safety work is needed.
CEO Sam Altman said in September that the company would not go public in 2026, citing the work still needed on AI safety.
Altman says people's inability to intuit exponential growth is a blind spot that gives founders who recognize AI's trajectory a huge advantage.
The world does not understand how to intuit exponentials
Altman recalls that when others dismissed deep learning, OpenAI felt it held the world's biggest secret, and being underestimated was an incredible gift.
For years at OpenAI it felt like we knew the biggest secret in the world
Sam Altman warned Congress in 2023 that AI could cause serious harm if it goes wrong.
“If this technology goes wrong, it can go quite wrong.”
Altman praised Qi Lu as one of the most impressive technologists he knows when announcing him as head of YC China.
I am delighted to announce that Qi Lu is joining YC to launch and run YC China. I have been trying to recruit Qi for many years — he is one of the most impressive technologists I know,