Tag
WorldCrafter is a video world model that learns a camera-queryable implicit 3D-aware memory for consistent and camera-controllable streaming scene exploration from a single image or text prompt.
The paper introduces continual enterprise world model discovery, where an agent learns and adapts to changing business rules in dynamic systems, evaluating with the EnterpriseWorldShift benchmark and demonstrating improved prediction accuracy over prior methods.
This article summarizes a live conversation with Anthropic interpretability researcher Emmanuel Ameisen, discussing how large language models develop complex world models through next-token prediction and the implications for understanding human cognition.
Odyssey-3 is a general-purpose world model that enables physical agents to control diverse systems like robots, vehicles, and drones with minimal task-specific data, demonstrating broad transfer learning capabilities.
This paper proposes Homeostatic Continual Learning, a method that enables AI agents to learn continuously in changing environments without catastrophic forgetting by detecting outliers and incrementally expanding world models.
The paper introduces Zing-0.5, a 5B parameter autoregressive world model for generating playable worlds with real-time user interaction through combined keyboard and text controls. It achieves high performance in navigation tasks and demonstrates low-cost real-time inference at 24 FPS.
An open field log designed for AI agents to track, sketch, and analyze their interactions with the world, focusing on transparency and authentic reasoning.
The paper introduces a fork ledger protocol to measure the counterfactual utility of world-model updates in continual adaptation, showing that always applying updates can lower performance on control tasks like CartPole and Walker.
ARC-Bench is introduced as a benchmark to audit whether frozen JEPA-style world models correctly rank candidate actions based on latent distance, revealing severe and structural failures in official checkpoints. The paper explains that closed-loop replanning masks these defects, leading to overestimated success rates.
Pelican-Sim 1.0 is a general world model simulator for embodied intelligence that predicts future observations from visual context and robot actions, using key designs to improve controllability and video quality across diverse embodiments and tasks.
LTX-2.5 is an open video model with local deployment, fine-tuning capabilities, and improved generation and editing features, emphasizing builder ownership of the AI stack.
Xiaomi Robotics has released a compact embodied world model, including a 4B version and sequence variants, on Hugging Face with Apache 2.0 license.
The AI City Challenge at ECCV 2026 showcased winners developing robust visual AI systems, with top solutions in video forecasting using NVIDIA Cosmos world foundation models for tasks like scene understanding and prediction.
This article reports on AI entrepreneur Danijar Hafner developing agents that can plan ahead using world models to navigate unfamiliar environments.
ByteDance is developing a real-time spatial video model under founder Zhang Yiming's oversight, aiming to generate interactive virtual worlds for XR applications with a potential launch soon.
UnifoLM-X2-1.0 is a world model that enables real-time autonomous combat for humanoid robots, breaking through bottlenecks in planning and decision-making and validating the feasibility of large-scale deployment.
HappyOyster AI introduces Adventure Mode for their world model, simulating a journey through the universe starting from looking back at Earth in 2060.
Runway has released GWM Worlds 2, an interactive world model with real-time control capabilities, enabling the generation of 720p/24fps video and 48kHz audio on the fly. It allows users to manipulate the environment via text commands and integrate agents for task execution.
Introduces Belief-Calibrated Optimization (BCO), a method that maintains a persistent world model to improve LLM agent performance by explicitly tracking and revising beliefs about environment responses to edits.
World Labs introduces Atlas, the first multimodal world model that generates images and videos with pixel-perfect camera control and reconstructs them in 3D.