Multimodal / world-model push beyond LLMs

Vision World Models survey + SANA-WM 2.6B; LaMI, Lance, Incantation, CogOmniControl, Aurora, LightSplat, Velox, COM4D, PhysX-Omni, CorText. New: Surya Ganguli call for unified physics/neuroscience/AI science; StepAudio 2.5 unified speech; SWIM vision-language alignment; Geo-Align geometry reward; SCOPE FPS world models; Seeing to Thinking decoupling perception/reasoning. New: Native Multimodal Modeling roadmap (comprehensive survey on NMM paradigm shift); Pantheon360 (3D digital twin from 360° video); TriSplat (sim-ready 3D reconstruction); AFD (on-policy adversarial flow distillation for video generation); WBench (multi-turn interactive video world model benchmark); MolmoAct2 (open-source robotics model beating π0.5). Gaps narrowing in sim-to-real/occlusion.

Sources (16)

Updated May 28, 2026

AI Research Digest

Multimodal / world-model push beyond LLMs

Gamma-World: Generative Multi-Agent World Modeling Beyond Two Players

GEM: Generative Supervision Helps Embodied Intelligence

From Pixels to Words -- Towards Native One-Vision Models at Scale

TriSplat: Simulation-Ready Feed-Forward 3D Scene Reconstruction

Toward Native Multimodal Modeling: A Roadmap

Pantheon360: Taming Digital Twin Generation via 3D-Aware 360° Video Diffusion

Open Source Robotics Model MolmoAct2: Ai2 Beats π0.5, Releases 720-Hour Bimanual Dataset

From Seeing to Thinking: Decoupling Perception and Reasoning Improves Post-Training of Vision-Language Models

SCOPE: Simulating Cross-game Operations in Playable Environments for FPS World Models

Geo-Align: Video Generation Alignment via Metric Geometry Reward

See What I Mean: Aligning Vision and Language Representations for Video Fine-grained Object Understanding

@ClementDelangue reposted: An interesting work on Physical AI: PhysX-Omni. First unified sim-ready generat...

StepAudio 2.5 Technical Report

@ylecun reposted: My new article "Toward a science of intelligence: unifying physics, neuroscience...

PhysX-Omni: Unified Simulation-Ready Physical 3D Generation for Rigid, Deformable, and Articulated Objects

Victoria Bosch: CorText - Chat with your brain data