📑DeepMind: "영상 만들라고 시켰더니 3D를 통째로 배웠다"

이 호에서 다룬 논문

• REASONING

  Accurate, Interdisciplinary and Transparent Structure-property Understanding with Deep Native Structural Reasoning

  Independent · Structural Reasoning, Multi-domain LLM, Scientific AI

• VISION

  Video Generation Models are General-Purpose Vision Learners

  Google DeepMind, MIT, University of Toronto · Video Generation, Vision Pretraining, World Prior

• AGENTS

  Long-Horizon-Terminal-Bench: Testing the Limits of Agents on Long-Horizon Terminal Tasks with Dense Reward-Based Grading

  Independent · Agent Benchmark, Long-Horizon, Dense Reward

• VISION

  Scalable Visual Pretraining for Language Intelligence

  Shanghai AI Lab · Visual Pretraining, Document Understanding, Multimodal

• REASONING

  Ideas Have Genomes: Benchmarking Scientific Lineage Reasoning and Lineage-Grounded Idea Generation

  Independent · Scientific Reasoning, Idea Lineage, Benchmark

• AGENTS

  UniClawBench: A Universal Benchmark for Proactive Agents on Real-World Tasks

  Independent · Proactive Agent, Real-World Benchmark, Tool Use

• VISION

  LongE2V: Long-Horizon Event-based Video Reconstruction, Prediction, and Frame Interpolation with Video Diffusion Models

  Independent · Event Camera, Video Diffusion, Reconstruction

• AGENTS

  Remember When It Matters: Proactive Memory Agent for Long-Horizon Agents

  Meta · Memory Agent, Long-Horizon, Proactive Recall

• WORLD MODEL

  ARDY: Autoregressive Diffusion with Hybrid Representation for Interactive Human Motion Generation

  NVIDIA, ETH Zurich · Motion Generation, Autoregressive Diffusion, Real-Time

• NLP

  Enhancing In-context Panoramic Generation via Geometric-aware Pretraining

  Independent · Panoramic Generation, Geometry-aware, In-context