Daily AI Papers Briefing (2026-03-12)

【本日の論文】1. Geometry-Guided Reinforcement Learning for Multi-view Consistent 3D Scene Editing https://huggingface.co/papers/2603.031432. Thinking to Recall: How Reasoning Unlocks Parametric Knowledge in LLMs https://huggingface.co/papers/2603.099063. Omni-Diffusion: Unified Multimodal Understanding and Generation with Masked Discrete Diffusion https://huggingface.co/papers/2603.065774. MM-Zero: Self-Evolving Multi-Model Vision Language Models From Zero Data https://huggingface.co/papers/2603.092065. InternVL-U: Democratizing Unified Multimodal Models for Understanding, Reasoning, Generation and Editing https://huggingface.co/papers/2603.09877