Sam Altman · OpenAI · 2025-06-10
OpenAI's CEO argues that humanity has already crossed the threshold into superintelligence — but the transition feels gradual because, in his framing, "wonders become routine, then table stakes." There's no thunderclap. There are agents writing real production code in 2025, novel insights produced in 2026, and downstream of that, a recursive loop where AI accelerates AI research itself.
The optimistic case rests on two compounding mechanisms: research acceleration (AI does more of the science, faster) and infrastructure scaling (cheap robots build the next data center). Altman keeps the framing emotional rather than technical: "wonders become routine" is the load-bearing claim, and the implicit ask is that society treats this as a manageable adaptation problem rather than an apocalyptic one. Whether that framing holds is itself the question.
- 2025: agents that do "real cognitive work" — writing computer code that ships - 2026: systems that "figure out novel insights" - Energy: 0.34 watt-hours per query — "about what an oven would use in a little over one second" - Economics: the world getting "so much richer so quickly we'll be able to seriously entertain new policy ideas we never could before"
> Wonders become routine, and then table stakes. This is how the singularity goes.
The bet is that recursive self-improvement (AI helping build smarter AI) and automated robotics (cheap physical buildout) are the levers. If both work, the curve is gentle because each generation absorbs the prior one's gains before the next arrives. If only the first works, you get cognitive abundance with physical scarcity — a different scenario entirely.