OpenAI CEO Sam Altman used a recent talk at Stanford to mount a forceful defense of the strategy that built ChatGPT: simply making large language models bigger, trained on more data and more computing power.
According to The Decoder, Altman argued that "a whole generation of researchers" actually slowed the field down by underestimating what this scaling approach could achieve. In other words, he suggests progress could have come faster if more people in AI had believed that scale alone would keep producing better results.
As evidence, The Decoder reports that Altman pointed to OpenAI's recent disproof of a mathematical conjecture — a case he offered to show that scaled-up models can do more than skeptics expected.
The comments are notable because the "will scaling keep working?" debate sits at the center of AI's biggest bets. Companies including OpenAI have poured enormous sums into ever-larger models and the data centers that run them, on the wager that bigger keeps meaning better. Many researchers have questioned whether that curve eventually flattens, making smarter methods — not just more horsepower — the real path forward.
Altman is, of course, not a neutral observer. As the head of the company most identified with the scaling bet, he has a direct stake in the argument that scaling was underrated rather than overhyped. His framing turns a technical dispute into a pointed critique of his own field's caution.
Why it matters: how this debate resolves shapes where billions in investment, talent, and energy flow next — and whether the AI industry keeps building bigger or starts betting on different ideas.