DailyPapers

Project page: innovator-zero.github.io/FASTER Paper: huggingface.co/papers/2603.19… Code: github.com/innovator-zero…

FASTER achieves 10x faster action sampling for real-time VLAs By compressing multi-step denoising into a single step, it enables immediate reaction in highly dynamic tasks like table tennis—even on consumer GPUs such as the RTX 4060.

Model: huggingface.co/nvidia/Nemotro… Paper: huggingface.co/papers/2603.19…

NVIDIA just released Nemotron-Cascade 2 on Hugging Face A 30B MoE model with 3B activated parameters that achieves gold medal performance at IMO and IOI 2025.

Project page: h-embodvis.github.io Model: huggingface.co/zd11024/Video3… Paper: huggingface.co/papers/2603.19…

VEGA-3D: Teaching multimodal LLMs spatial reasoning through video generation Baidu and Huazhong University researchers introduce a plug-and-play framework that extracts implicit 3D priors from video diffusion models, enabling geometric scene understanding and embodied decision making without explicit 3D supervision.

Paper: huggingface.co/papers/2603.16… Dataset: huggingface.co/datasets/AIcel… Benchmark: war-forecast-arena.com

When AI Navigates the Fog of War Can AI predict war before it becomes history? This study evaluates frontier LLMs on 11 critical moments from an ongoing conflict using only real-time data, revealing both impressive strategic reasoning and surprising blind spots.

Paper: huggingface.co/papers/2603.17… Code: github.com/pUmpKin-Co/Com…

Complementary Reinforcement Learning Alibaba researchers introduce a neuroscience-inspired RL paradigm enabling policy actors and experience extractors to co-evolve, fixing misalignment in static memory systems and boosting agentic performance by 10%.

The paper reveals a fundamental trade-off between optimizing models for human use and using them as proxies for human behavior. Aligned models win in one-shot games, but base models dominate when reciprocity and retaliation matter. huggingface.co/papers/2603.17…

Alignment Makes Language Models Normative, Not Descriptive We compared 120 base–aligned model pairs on 10,000+ human decisions. Base models outperform aligned models nearly 10:1 in strategic games. Alignment learns what humans should do; base models learn what they actually do.

Accepted at CVPR 2026. Paper: huggingface.co/papers/2603.14… Codes and models will be released soon.

Video-CoE Alibaba researchers introduce a Chain of Events paradigm that breaks videos into temporal sequences to enhance logical reasoning for future event prediction, establishing a new state-of-the-art on video event prediction.

Paper: huggingface.co/papers/2603.17… GitHub: github.com/aiming-lab/Met…

MetaClaw: Just Talk Your personal agent evolves automatically from every conversation. MetaClaw meta-learns in the wild using cloud LoRA and skill synthesis, scheduling weight updates during sleep or idle time with zero downtime.

Find it here: github.com/huggingface/sk…

Introducing the Daily Papers SKILL.md Enables agents to > read paper content as markdown > search papers > find linked @huggingface models, datasets > fetch the papers API > and more! Link below ⬇️

Project page: mosaicmem.github.io/mosaicmem/ Paper: huggingface.co/papers/2603.17…

MosaicMem A hybrid spatial memory for video world models bridging explicit 3D and implicit memory, enabling long-horizon navigation, memory-based editing, and dynamic scene generation with improved camera consistency.