What if AI could understand 3D space just by watching videos? Researchers from Huazhong University of Science and Technology and Baidu present VEGA-3D—a plug-and-play framework that repurposes a pretrained video diffusion model as a "Latent World Simulator." Instead of relying
@jiqizhixin
-
AI Generates ‘Maxed Out Sexy’ 80s Anime Girl
By
–
Why treat all classes equally when picking prompt weights? Researchers from Univ. of Melbourne, Southeast Univ., & RIKEN present CARPRT. It scores each prompt’s relevance per class by averaging over predicted images—no training needed. Outperforms class-independent methods
-
Grok Feature Now Available for Premium and Premium+ Subscribers
By
–
Can continuous diffusion finally close the gap with discrete models for language generation? UIUC researchers present LangFlow, the first continuous diffusion language model to rival top discrete approaches. They bridge embedding-space diffusion and Flow Matching using a
-
Google DeepMind TIPSv2 Boosts Vision-Language Dense Patch-Text Alignment
By
–
Can vision-language models truly see the fine-grained details in images? Google DeepMind presents TIPSv2. They boost dense patch-text alignment using three novel tricks: a distillation method where the student outperforms the teacher, an upgraded masked image objective
-
SHAPE Method Rewards Reasoning Progress Over Verbosity in LLMs
By
–
What if LLMs could reason smarter, not just longer? Researchers from Huawei Taylor Lab, Peking University, and Shanghai University of Finance and Economics introduce SHAPE. The method rewards actual progress in reasoning — not verbosity — by using a two-level system: a
-
Tencent Hunyuan Script-a-Video Introduces Smart Caption System MTSS
By
–
What if your video captions could be edited without rewriting the entire scene? Tencent Hunyuan Team introduces Script-a-Video with MTSS — a smart caption system that splits video into separate streams (shots, events, references) and links them with explicit timestamps and
-
Scaling RL Training Boosts Larger LLMs in Math Reasoning
By
–
Why do larger LLMs get even better with reinforcement learning post-training? Researchers from USTC, Oxford, and Shanghai AI Lab reveal how scaling RL training works for math reasoning. They tested the Qwen2.5 series (0.5B to 72B) and found: – Larger models are more compute-
-
ASCII Emoticons Can Trick LLMs Into Generating Harmful Code
By
–
Can smiley faces break AI? Researchers from Xi’an Jiaotong University, NTU, and UMass Amherst reveal a new LLM vulnerability: emoticon semantic confusion. ASCII emoticons like 🙂 can trick models into misinterpreting intent, leading to harmful code generation. Their study,
-
Stanford and UC Berkeley Present LLM-as-a-Verifier Framework
By
–
What if your LLM could verify its own reasoning with near-human precision? Stanford & UC Berkeley researchers present LLM-as-a-Verifier: a general-purpose framework that gives fine-grained feedback by breaking tasks into smaller criteria, scoring with higher granularity, and
-
MathForge Method Boosts AI Reasoning With Harder Math Problems
By
–
What if making math problems harder actually boosts AI reasoning? Researchers from Renmin University, Alibaba, and other institutions introduce MathForge—a new method that flips the script. Instead of avoiding tough questions, it actively seeks them out. The approach combines