Ever wonder why low-precision AI training with Flash Attention suddenly crashes? Tsinghua University researchers crack the code! They reveal it's a double whammy: internal attention data patterns become too similar, and subtle, biased rounding errors in low-precision math
@jiqizhixin
-
arXiv Becomes Independent From Cornell University Partnership
By
–
Every time you opened arXiv and saw that little “Cornell University” badge, you probably never thought it might disappear. But it might soon become history. After decades of partnership with Cornell and support from the Simons Foundation, arXiv is becoming an independent
-
FlowRVS: Language-Guided Video Object Segmentation
By
–
What if AI could precisely segment any object in a video, guided only by your natural language description? A joint effort by SGIT AI Lab, UC San Diego, HKUST, UTokyo, Cambridge, Zhejiang University of Technology, and Baidu presents FlowRVS. Instead of finding then
-
DDP-WM: Robots Plan Complex Actions Without Computational Overhead
By
–
What if robots could plan complex actions instantly, without massive computational overhead? Researchers from Sun Yat-sen University and X-Era AI Lab unveil DDP-WM. This new world model smartly separates a robot's world into critical physical actions and less important
-
FeatureBench: Evaluating LLM Coding Agents on Complex Software Features
By
–
How well do LLM coding agents truly perform on complex, end-to-end software feature development? Researchers from the Institute of Automation, Chinese Academy of Sciences and Huawei Technologies Co., Ltd. introduce FeatureBench, a new benchmark using a scalable, test-driven
-
LLMs Transform Reinforcement Learning in Recommendation Systems
By
–
Can recommendation systems truly understand and adapt to your evolving tastes? A team from University of Science and Technology of China, Kuaishou , and others is mapping out how Large Language Models (LLMs) revolutionize Reinforcement Learning (RL) in recommendation systems.
-
JTok: Scaling LLMs with Token-Indexed Parameters
By
–
Can LLMs achieve massive capacity gains without massive increases in computational cost? YES, say researchers from Shanghai Jiao Tong University and Xiaohongshu! They introduce JTok, a novel scaling method that uses lightweight "token-indexed parameters" to intelligently
-
IterResearch: AI Agents for Long-Duration Marathon Research Tasks
By
–
Can AI agents truly handle marathon research tasks without losing their way? Renmin University, Alibaba Group's Tongyi Lab, and OpenRLHF present IterResearch. This iterative method rethinks how agents process long tasks. Instead of a single, ever-growing memory context, it
-
GENIUS Suite Evaluates Multimodal AI Reasoning Capabilities
By
–
Can your favorite AI truly reason on the fly, or just remember what it's learned? A team from Peking University, CUHK, StepFun, PolyU, and MSRA introduces GENIUS, a groundbreaking evaluation suite. GENIUS challenges Unified Multimodal Models (UMMs) on Generative Fluid
-
GPTheology: How AI Reshapes Religious Thought and Practice
By
–
Prompts and Prayers: the Rise of GPTheology Paper: