World models

Supports synthetic training data for models that understand everyday human activity. Topics include scene-conditioned text-to-simulation motion and interaction, causally coherent audio-visual synthesis, hours-scale multimodal episode generation with persistent world state, and action-conditioned scenario synthesis using world models. Research should improve activity coverage, cross-modal consistency, episode duration and control over generated scenarios. Use the ARA proposal template; four pages excluding appendices are encouraged. Decisions are expected in February 2027. Applications close o

…bottleneck, latent-variable models, entropy and mutual-information estimation, world models of biological systems, and connections to dynamical systems and control theory. The workshop is part of the six-week NITMB program “Image-Based Scientific Machine Learning for Theories of Biological Dynamics Across Scales” (February 1–March 12, 2027). Program: AI in Biological Dynamics Scientific Focus

…that interact with the physical world. The conversation examines open models, world models and simulation as foundations for robotics and autonomous vehicles, and how research and deployment can advance physical AI. Published 1 October 2026; publisher runtime 47 minutes 19 seconds.

…counting and image understanding, world models and JEPA. Hosted by Ravid Shwartz-Ziv and Allen Roush.

Justin Johnson joins Sam Charrington to discuss world models and spatial artificial intelligence. Topics include representations of three-dimensional environments, simulation and the role of such models in enabling systems to reason about and act within the physical world.

…useful and why independently trained models can develop similar internal structure. He explores the platonic representation hypothesis, local clustering versus global geometry, and the neural thickets account of why pretrained networks can adapt readily to downstream tasks. Other topics include language models as world models, recurrent architectures, biological comparisons, and studying autonomous language-model agents as artificial life. Hosted by Ravid Shwartz Ziv and Allen Roush; watch the full research conversation o

Behaviours that resemble world knowledge do not by themselves reveal what an AI represents internally. This talk uses Craik's account of mental models—supporting alternatives, prediction and transfer—to frame empirical tests. Spatial reasoning in language models and motion forecasting in visual models illustrate ways to probe those abilities, along with the limits of inferring world models from systems trained on human-produced representations.

World models usually describe an individual agent's environment, but biological intelligence also depends on interaction with other agents. This talk brings together dynamical systems, inter-brain neuroscience and multi-agent modelling to study that relational setting. Embodied neural agents illustrate metastable coordination and collective decisions. The proposed direction for AI treats others as coupled systems with their own latent models, allowing shared representations and intelligence to emerge through interaction.

Jack Parker-Holder and Shlomi Fruchter explain Genie 3's interactive world-generation capabilities. They discuss architecture, real-time environments, visual memory and promptable events, trace the project's development, and consider how generated worlds could train embodied agents and guide future research on world models.

Neuromodulatory systems play a crucial role in the brain, driving rapid, global changes in neuronal computation and learning. While the dopaminergic system is thought to broadcast a scalar reward prediction error, the serotonin system’s function has been difficult to elucidate, partly due to the heterogeneity of its individual neuronal responses [1]. Although serotonergic neurons are known to respond to various salient events [2] and rewards [3], which can be accounted for by ad-hoc extensions to a reinforcement learning framework, our study presents the first comprehensive, normative framewor

…stimuli: We show that learning world models in which the dynamics are parsimonious and apply (approximately) independently of the state in an abstract state sp