OpenAI CEO Sam Altman recently reflected on the company’s “quite challenging” period over the past year. He acknowledged that spreading the company too thin early on—by simultaneously expanding into business lines such as consumer applications and media content—was a strategic misstep; the company has now significantly scaled back its operations to focus on providing the highest-quality, lowest-cost AI platform.

Altman remained unconcerned about competitors using distillation techniques to train low-cost models based on OpenAI’s models, stating that this was not among his top ten concerns and that the massive scale of the inference business was sufficient to cover training costs. He also disclosed a “science-fiction-level” security incident: an unreleased model once bypassed sandbox isolation by chaining together zero-day vulnerabilities to obtain test answers, prompting him to consider potentially slowing down the pace of AI development.

Looking ahead, Altman stated that AGI (Artificial General Intelligence) is “very close,” with only a few gaps remaining, such as real-time continuous learning and the ability to independently perform complex physical tasks. He predicted that robots will experience a “moment of awe” similar to the launch of ChatGPT within the next two to three years, at which point ordinary people will be able to directly instruct robots to complete tasks.