- The report contributes researcher interviews and technical explanation, not a new-model launch.
- Persistent world state and generated visuals are separate components in the described approach.
Keeping state separate from visuals
VAST is inviting young researchers to work on 3D-generation foundation models, three-dimensional understanding and agents, and world models through a new V·STAR talent programme, QbitAI reported on November 5. Its account includes interviews about how researchers choose and test technical approaches, rather than announcing a newly released model.
The report describes VAST's approach to interactive worlds as separating maintained state and interaction rules from generated visuals. An upstream engine tracks what has happened, while a downstream generative model produces the view. This distinction matters when a user changes viewpoint but objects, actions and rules must remain consistent.
Research choices and unfinished work
An EdenLive contributor identified under the pseudonym Zion describes internal debate about whether explicitly separating those components was the right research direction. The account says the team returned to its goal of building a world with clear rules and sustained interaction.
The report also identifies longer-term consistency, more accurate control, and trade-offs between real-time generation and computing cost as continuing research questions. The recruitment programme offers data, computing and collaboration support; those commitments do not establish that the technical challenges have been solved.
Sources & context
Go to the original material. Company claims remain attributed to their sources.
01
