sc94597 on 07 August 2025
So yeah, these world models are the next leap. Not just in games, but in general (lossy, but still semi-accurate) representations of the world which can be used with reinforcement learning or genetic algorithms to train (or "grow" rather) embodied intelligence and to essentially eliminate spatial reasoning hallucinations for LLMs and other general purposed neuro-symbolic models.
I'm starting to think the technological singularity started in 2008 with GPGPU and big-data collection via Web 2.0. Everything else has just been accelerating from then.
As for gaming, it will be interesting to see how this will be configured and combined with traditional game engines and how controllable constraints can be set.
