Categories: World-Model Statism

16 entries

Intelligence comes from predicting the physical world; one big model should learn it from video and interaction.

Understand-Generate Unionism
Understanding and generating images are two faces of one capability and should be performed by a single unified model.
Visual Multimodalism
Aims at general multimodal intelligence but builds it on vision, treating text, documents and interfaces as things to be seen.
Robot-Foundation Imperialism
One generalist policy should control every robot body, from arms to quadrupeds to humanoids, trained on pooled cross-embodiment data.
Embodied Statism
Holds that general intelligence requires a body acting in the world, and that one large embodied model should be built centrally.
Sim-Scale Statism
Robots should learn from massively parallel simulation, compressing years of experience into hours of GPU time.
Video Scalism
Video is the next internet-scale corpus, and scaling models on it will teach them about the world the way text taught them language.
Robot Foundationism
Believes general robot skill will come from foundation models trained on large, diverse real-world robot data, gathered through a deployment flywheel.
End-to-End Driving Absolutism
Autonomous driving should be one neural network mapping camera pixels directly to steering and throttle, with no hand-built modules in between.
Humanoid Centralism
The human-built world is designed for human bodies, so general-purpose robots should be humanoids and humanoid data should be centralized.
Predictive-Coding Statism
Holds that intelligence is fundamentally prediction of incoming sensory signals, with learning driven by minimizing prediction error.
World-Model Vanguardism
Before anything else, intelligence must build an internal predictive model of the world; language and agents come after.
Visualism
Believes vision is the primary substrate of intelligence, and that language merely names what visual understanding already knows.
Bio-Hybridism
Living neurons grown on chips can be trained to compute, offering an energy-efficient substrate that silicon cannot imitate.
Video-Native Doctrine
Holds that time is the neglected modality, and that models should learn from video natively rather than from still images and text.
Grounding Doctrine
Symbols mean something only when connected to perception and action; text-only learning leaves meaning ungrounded.
Self-Supervised Vision Statism
Believes vision models should learn from raw unlabeled pixels at scale, with human labels and captions as optional extras.