Peking University's EvoPhys-World Tops Stanford Benchmark
Peking University's EvoPhys-World ranks first on Stanford's WorldScore, trained natively on Moore Threads S5000 GPUs.
Latest articles in Research
Peking University's EvoPhys-World ranks first on Stanford's WorldScore, trained natively on Moore Threads S5000 GPUs.
Lehigh University team introduces OpenSkill, enabling self-evolving AI agents without supervision. It achieves state-of-…
New research reveals AI digital twins for heart care may fail women due to historical data biases in medical training se…
Machine learning boosts weather forecasting but cannot replace physics-based models. Hybrid approaches remain essential …
AI digital twins revolutionize cardiology but risk excluding women due to historical data gaps and physiological differe…
Google DeepMind launches a faster AlphaFold model, revolutionizing drug discovery and biological research with unprecede…
Spirit AI and Tsinghua University's FP3 model targets 3D robotic perception gaps, earning an ICRA 2026 nomination.
CVPR 2026 closes with DeepMind's D4RT taking top honors, Oxford VGG's second win, and a Chinese undergrad's GPU hack goi…
Google Research introduces Agentic RAG to Gemini Enterprise, boosting factuality by 34% with a Sufficient Context Agent.
Vision embeddings transform images into data vectors, enabling advanced AI analysis and cross-modal search capabilities …
Yunzhu Li proposes structured world models blending 3D physics and 2D data to solve robot learning bottlenecks at ICRA 2…
Aniket Bera argues that current autonomous systems rely on simplified environments. He proposes a new framework for true…