According to CoinMeta, former ByteDance intern Tian Keyu revealed his startup plan for a world model, which aims to enable AI to learn about the movement and changes of objects in the real world by designing a vocabulary containing 200,000 visual symbols, with the intention of applying this knowledge to fields such as video generation and robotics. His team has raised nearly $30 million in funding from institutions including Wuyuan Capital and IDG Capital, and the post-investment valuation is $200 million. The team currently consists of about 10 people and has not yet released a product publicly. Tian Keyu stated that this method has the potential to reduce the cost of generating videos per second to one-tenth of the current level or even lower. The team plans to train the model with 100 million hours of video data and is preparing to release the complete model in 2027. Additionally, Tian Keyu admitted to shutting down another intern's program; ByteDance dismissed that intern in 2024 and later sued for compensation of 8 million yuan.