Topic
world model
New Critique of World Models Proposes Generative Latent Prediction Architecture for AGI
The arXiv paper 'Critique of World Model' defines the primary goal of world models as simulating all actionable possibilities for purposeful reasoning and acting. It examines key design dimensions—data, representation, architecture, learning objective, usage—and proposes a new Generative Latent Prediction (GLP) architecture for a general-purpose world model.
New Unified Definition of AI Hallucination Pins It on Inaccurate World Modeling
A new arXiv paper by Liu et al. proposes a unified definition of hallucination in large language models, defining it as inaccurate internal world modeling observable to the user. The framework subsumes prior definitions and distinguishes true hallucinations from planning or reward errors, and introduces the HalluWorld benchmark for stress-testing models.
Kairos Stack Promises Native World Models for Physical AI Across Heterogeneous Experience
Researchers have introduced Kairos, a world model stack designed for Physical AI. It features a Native Pre-training Paradigm using a cross-embodiment data curriculum, a Native Unified Architecture with hybrid linear temporal attention, and a Deployment-Aware System Co-Design for real-time performance. Kairos achieves top-level results on embodied world-model, long-horizon, and action-policy benchmarks.
ViTaL Framework Combines Vision and Touch to Boost Robot Manipulation Success by 51%
ViTaL, a visuo-tactile inference-time steering framework, uses a bi-level optimization combining visual sampling and tactile diffusion to guide robot policies. On three real-world contact-rich manipulation tasks, it improved success by 51% over the base policy, outperformed unimodal steering by at least 33%, and exceeded naive multimodal fusion by at least 20%.