Self-play and autonomous learning are the critical pathways for AI to transcend the ceiling of human data.

In any domain where we can engineer a controllable, cost-effective, and high-fidelity environment for trial and error, AI possesses the potential for continuous evolution—ultimately surpassing human capabilities.

The true determinant of AGI will not be model scale, but our ability to construct a general World Model capable of representing physical laws, causal structures, and long-term dynamics. This is essential to enable AI to simulate, predict, and plan, achieving open-ended intelligence growth through self-feedback.

Yet, the journey has only just begun…