New AI framework adds beliefs and intentions to world models

Current world models like Sora or Genie simulate physics but ignore what people think, want, or feel. A new 'Mental World Modeling' framework adds mental-state variables such as beliefs and intentions, and even weaker language models using this approach reportedly outperform stronger models that lack it. The biggest remaining bottleneck is predicting how physical and mental states change together.

Why it matters: If borne out, this suggests architecture and training approach can matter more than raw model scale for tasks requiring an understanding of human intent — directly relevant to building AI agents that act in environments with people, from robotics to assistants. It also echoes other recent findings, like agent harness quality mattering more than model quality, that point toward smarter task framing outperforming brute-force scale.