SoftBank's Masayoshi Son: OpenAI Using AI to Design AI Models, Predicts ASI Will Arrive Within 2 Years

In a recent interview with CNBC, SoftBank CEO Masayoshi Son revealed that OpenAI is actively employing artificial intelligence to design and develop subsequent AI models, accelerating the timeline for Artificial Super Intelligence (ASI) from a decade to just two years.

Son, who speaks regularly with OpenAI CEO Sam Altman and the company's engineering team, stated that AI models are now capable of designing future iterations of themselves. This paradigm shift toward "AI designing AI" marks a fundamental change in development methodology, one that Son believes will expand across the industry until human engineers can no longer independently design more powerful models.

The SoftBank chief previously defined ASI as AI systems possessing intelligence 10,000 times greater than that of humans. While he initially predicted a ten-year timeline for such capabilities in 2024, Son now admits that forecast was conservative. "The real expectation has changed to within two years," he told CNBC.

OpenAI appears to be laying the groundwork for this transition already. In February, the company disclosed that GPT-5.3-Codex represents its first model capable of "participating in creating itself." The Codex team reportedly utilized early versions of the model to debug training processes, manage deployment logistics, and diagnose evaluation results.

Son practices what he predicts, noting that he personally uses ChatGPT for two to three hours daily. He acknowledges the system already surpasses his own intelligence in most subjects and anticipates AI will exceed human capabilities in 70 to 80 percent of academic disciplines within the coming years.

The acceleration of recursive self-improvement in AI systems represents a pivotal moment in the technology's evolution, with major technology leaders now publicly forecasting the imminent arrival of machine intelligence that vastly eclipses human cognitive abilities.