On July 17, during the 2026 World Artificial Intelligence Conference (WAIC), Jiang Xu, founder and CEO of Liangyuan Innovation and a core contributor to ChatGPT, outlined three key paradigms for embodied AI development: scalable pre-training, scalable alignment, and scalable deployment. He noted that while the path to training large models is becoming clearer, truly identifying the right product forms and commercial scenarios that match model capabilities is no easier than training the models themselves.
Jiang Xu pointed out that before ChatGPT’s success, OpenAI went through extensive product exploration, many of which didn’t pan out. For embodied AI companies, a more realistic strategy is to keep an open mind and continuously explore diverse application scenarios, rather than prematurely assuming the one correct commercialization direction.