Interview
OpenAI Sora 2 Team: How Generative Video Will Unlock Creativity and World Models
- The team intends to co-evolve society with technological breakthroughs, aiming for a future where digital copies of users interact in Sora's ether and report back to the physical world while the world adapts to new operational rules.
- A significant step-function improvement in model quality is anticipated before systems can reliably conduct scientific experiments, potentially rendering physical wet labs obsolete for biological simulations.
- Future model iterations are expected to enable feature film-length content creation by individuals at home, with the format evolving into a new medium that differs fundamentally from current cinema.
- The platform is forecasted to evolve into "Sora 2," a mini alternate reality running on phones where users can interact with digital clones of others, significantly enhancing immersion capabilities.
- A new creator economy will be established to properly monetize rights holders, allowing paid access to models for scaling and enabling specific features like a committed "pet cameo" tool.
- Long-term technical goals include unlocking a vast history of untold stories related to heroism and struggle, with shorts predicted to be the first format to achieve major industry awards.
- Video models are expected to be sufficiently advanced to support the building of video games on top of their API within another rev or two, while continuing to serve both pro-oriented creators and the general creative public.
- The team plans to structurally alter user engagement by pushing consumers out of passive consumption loops, with data suggesting 70% of creators will return to create rather than just scroll and 30% will post to the feed.
- As input bandwidth increases, models will develop deep understanding of personal relationships to function as accurate digital clones, though certain phenomena like quantum mechanics or high-speed particle collisions may remain unsimulated via video due to representation limitations.
- Potential risks and uncertainties include the inability to guarantee that adding physical agency will not degrade performance in modeling specific physical collisions, and the possibility that a wet lab will eventually be unnecessary, changing the scientific landscape.
- Future model advancements may lead to new scientific discoveries, such as improved theories regarding classical physics like turbulence or Navier-Stokes equations.
- While the team anticipates ways to scale access through paid models, they do not confirm specific pricing structures or exact timeframes for the realization of general-purpose world simulation capabilities.