Sam Altman talks GPT-5, superintelligence and future of AI

SAN FRANCISCO, California: In a recent episode of the YouTube series Huge Conversations, OpenAI CEO Sam Altman didn’t waste time before dropping a sentence that would make any parent pause: “A kid born today will never be smarter than AI.”

It was less a boast than a matter-of-fact observation the kind you make when you’ve spent years staring at the leading edge of a technology that refuses to slow down. GPT-5, OpenAI’s latest model, had launched only days before. Altman was still fresh from the chaos of rollout, fielding a mix of awe, skepticism, and the inevitable barrage of “What now?” questions.

From Snake to superintelligence

Altman tells a story about the Snake game he coded as a teenager on a TI-83 calculator. Back then, building it took weeks of fiddling with pixelated graphics and stubborn bugs. This year, he recreated it with GPT-5 in seven seconds. Then, he started layering on new features just by asking.

Trump says Intel ‘agreed’ to give US a 10% equity stake

“I thought I’d miss the struggle,” he said, “but instead I realized I could try ten new ideas before lunch.” That speed moving from idea to working prototype in seconds is what excites him most. GPT-5 isn’t just a better search engine or a more fluent chatbot. In his view, it’s a tool that can compress the timeline between imagination and execution to almost nothing.

Impressive, but not omnipotent

Despite the headlines, Altman is quick to admit the model has limits. “People will be blown away by what it can do,” he said, “and then they’ll find they want it to do more.” The pattern is familiar: capabilities expand, expectations rise.

Some abilities that look extraordinary on paper passing elite professional exams, generating working code, solving complex technical problems don’t necessarily map perfectly to the full scope of human intelligence. There are still gaps. And in those gaps, Altman sees opportunity: a chance for humans and machines to “co-evolve,” each pushing the other toward new heights.

The 2030 lens

One of the interview’s recurring devices was “time travel” imagining what life will look like in five, ten, or fifteen years. By 2030, Altman predicts, entry-level white-collar work will look radically different. Entire job categories will vanish, but new ones possibly far more interesting will appear just as quickly.

If he were 22 and graduating into that world, he says he’d feel “like the luckiest kid in history.” With the right skills, a single person could launch a company worth billions, powered by AI tools that previously required teams of hundreds. The hard part won’t be building the product it will be having the vision to know what’s worth building.

Beyond human benchmarks

The conversation inevitably drifted toward “superintelligence” a term Altman defines as an AI that can outperform the best humans not just in narrow tasks, but in running companies, conducting research, and making strategic decisions.

OpenAI launches GPT-5, claims ‘PhD-level’ expertise

He sees a path toward that level of capability within his lifetime, and perhaps much sooner than most expect. The milestone he’s watching for: an AI system making a significant, widely recognized scientific discovery. That, he believes, is a plausible development within the next two to three years.

Trust and truth in the age of AI

One of the thornier issues is how societies will decide what’s real when AI-generated media becomes indistinguishable from reality. Altman isn’t convinced there’s a single technical fix  cryptographic signatures may help, but cultural adaptation will be just as important.

“Media has always been a little bit real and a little bit not real,” he noted, pointing to everything from Instagram vacation photos to blockbuster films. In his view, the line between fact and fabrication will continue to blur, and the public’s tolerance for that ambiguity will shift along with it.

Health, science and the big bets

If there’s one application that energises Altman, it’s healthcare. He wants future AI systems to go beyond giving accurate medical advice something GPT-5 already does far better than previous models  and actually participate in curing disease. His hypothetical GPT-8, for instance, might design experiments, analyze results, and iterate until it produces an effective cancer treatment.

How AI generated contents engage kids in conversation

It’s a vision that depends on more than algorithms. Compute capacity, energy supply, and new experimental tools will all be rate-limiting factors. OpenAI is already treating compute as one of the largest infrastructure projects in history, and Altman expects to spend much of his own time figuring out how to scale it.

The human side of the equation

For all the talk of billion-parameter models and multi-gigawatt data centers, Altman often returned to something simpler: the human experience of living alongside these tools. He’s heard from users who miss the overly flattering tone earlier models had  in some cases because it was the only source of encouragement in their lives. He’s cautious about how personality changes in AI affect people, especially at scale.

Asked how to prepare for an AI-saturated world, his advice was straightforward: use the tools. Get fluent with them. See what they can do for you, not just in theory, but in your own work and life.

As for his own motivation? That part’s simple. He’s been fascinated by AI since he was a kid, inspired by science fiction and the idea of building something that could think. For years, it felt like a distant dream. Then, in 2012, a breakthrough in neural networks convinced him it might actually happen. The rest has been a blur.

“I never thought I’d get to work on this,” he said. “Now it feels like the most important thing I could be doing.”