Sam Altman predicts AGI will arrive in 2025

Ask most people what they are looking forward to in the coming year, and you will hear about promotions, marathons, or family trips. Ask Sam Altman, and the answer is ushering in artificial general intelligence.

During a recent conversation with Y Combinator head Garry Tan, Altman was asked what excites him most about 2025. His immediate answer was AGI. Right after that, he casually dropped another massive life update: he and his husband, Oliver Mulherin, are expecting a baby, which he called the most exciting thing in his personal life. Putting the singularity and fatherhood on the exact same timeline is quite the flex, even for Altman.

It is not just a one-off comment either. Days earlier on the 20VC podcast with Harry Stebbings, Altman claimed that within five years, tech progress will move so fast that people might feel the AGI moment simply came and went, all while society itself changes surprisingly little.

Clear path or wishful thinking?

Altman told Tan that the path to AGI is basically clear, framing the remaining hurdle as an engineering challenge rather than a need for brand-new scientific breakthroughs.

That optimism lands in the middle of a messy debate about whether large language models are hitting a wall. Rumors around OpenAI’s Orion model suggest its jump over GPT-4 is much smaller than previous generational leaps, particularly in coding. Meanwhile, high-quality public training data is drying up, reportedly prompting OpenAI to create a dedicated Foundations Team just to tackle fundamental data bottlenecks.

Still, researchers inside OpenAI like Noam Brown and Clive Chan are backing Altman’s timeline. Their conviction hinges on the o1 reasoning architecture, arguing that test-time compute unlocks an entirely new scaling dimension when traditional pre-training hits friction.

At the same time, safety alarms keep ringing. OpenAI’s former AGI readiness lead Miles Brundage left recently, openly stating on his Substack that neither the company nor society is ready for what is coming. Shortly after, OpenAI dissolved the entire AGI readiness team.

Why it matters

OpenAI’s internal framework currently places its tech at Level 2 out of 5 (reasoners). Jumping from there to human-level versatility within a single calendar year sounds almost detached from reality.

We think this timeline is classic Altman narrative management. With whispers growing louder that pre-training scaling laws are stalling out, publicly doubling down on a 2025 milestone reassures investors and keeps the pressure squarely on Google and Anthropic. It tells Wall Street that even if pre-training hits diminishing returns, o1-style reasoning will carry the torch across the finish line.

For developers and enterprises, there is no need to redesign roadmaps around a doomsday or utopia scenario next year. AGI or not, the practical shift in 2025 will likely be better reasoning and cheaper execution on complex multi-step workflows, not a rogue digital god taking over your operations.