Your daily digest
Research papers, made simple to understand
The AI and machine learning papers people are reading most, each one rewritten in plain words: the problem, the idea, what was found and why it matters.
- Image & Video Generation
- Multimodal & Vision
Survey unifies video understanding and generation via world modeling framework
This survey organizes video foundation models around four operations—reading, writing, sharing, and interacting—viewing video as a partial observation of a latent world, and highlights shifts toward persistent, controllable, and physically consistent video modeling.
Paper: Reading, Writing, Sharing, and Interacting with the World through Video · Shengqiong Wu, Meng Luo, Bobo Li and 5 others
258 upvotes on Hugging Face6 min readRead the article
- AI Agents
- Benchmarks & Evaluation
UserArena benchmark and OPTD method improve long-horizon user simulation for shopping agents
UserArena provides a large-scale, real-world e-commerce benchmark with long user trajectories and production A/B validation. The proposed OPTD method uses trajectory-level intent and on-policy distillation to train simulators that better reproduce user behavior and policy effects over long interactions.
Paper: UserArena: Benchmarking Interactive User Simulation for Long-Horizon Shopping Agents · Jujia Zhao, Jiakai Tang, Jing Yu and 6 others
47 upvotes on Hugging Face7 min readRead the article
- AI Agents
HiSTEVE integrates hierarchical planning and native control for Minecraft agents
HiSTEVE combines a capability-aware harness with a vision-language-action policy to achieve 85.0% atomic task success on MineStudio Bench and 37.4% average success on OpenHA, enabling long-horizon tasks like crafting an iron pickaxe from scratch.
Paper: HiSTEVE 1.0: Harnessing Human-Interface VLA Agents for Open-World Games · Zecong Tang, Zian Guan, Yuncheng Guo and 3 others
38 upvotes on Hugging Face6 min readRead the article
Browse earlier papers
2 days of papers in November 2026