Featured image of post AI Paper Daily | 2026-03-27

AI Paper Daily | 2026-03-27

今日概览

共收录 14 篇论文 | Audio LLM: 2 篇 | LLM Training: 0 篇 | AI Agents: 5 篇 | 通用热门: 3 篇 来源:arXiv(0) | HuggingFace(100) | Semantic Scholar(0)

本期日报聚焦最新研究,涵盖 Audio LLM、LLM Training 和 AI Agents 方向。


重点推荐 ⭐

CUA-Suite: Massive Human-annotated Video Demonstrations for Computer-Use Agents

(暂无摘要)

  • 作者: Xiangru Jian, Shravan Nayak, Kevin Qinghong Lin et al.
  • 来源: huggingface (69 upvotes)
  • 链接: arXiv | PDF
  • 关键贡献: (需人工补充)
  • 代码/权重: 待确认
📄 Abstract 中文翻译

Why Does Self-Distillation (Sometimes) Degrade the Reasoning Capability of LLMs?

(暂无摘要)

  • 作者: Jeonghye Kim, Xufang Luo, Minbeom Kim et al.
  • 来源: huggingface (27 upvotes)
  • 链接: arXiv | PDF
  • 关键贡献: (需人工补充)
  • 代码/权重: 待确认
📄 Abstract 中文翻译

UI-Voyager: A Self-Evolving GUI Agent Learning via Failed Experience

(暂无摘要)

  • 作者: Zichuan Lin, Feiyu Liu, Yijun Yang et al.
  • 来源: huggingface (29 upvotes)
  • 链接: arXiv | PDF
  • 关键贡献: (需人工补充)
  • 代码/权重: 待确认
📄 Abstract 中文翻译

GameplayQA: A Benchmarking Framework for Decision-Dense POV-Synced Multi-Video Understanding of 3D Virtual Agents

(暂无摘要)

  • 作者: Yunzhe Wang, Runhui Xu, Kexin Zheng et al.
  • 来源: huggingface (16 upvotes)
  • 链接: arXiv | PDF
  • 关键贡献: (需人工补充)
  • 代码/权重: 待确认
📄 Abstract 中文翻译

🔊 Audio LLM

BioVITA: Biological Dataset, Model, and Benchmark for Visual-Textual-Acoustic Alignment

(暂无摘要)

  • 链接: arXiv | PDF
  • 摘要: (暂无摘要)
📄 Abstract 中文翻译

YingMusic-Singer: Controllable Singing Voice Synthesis with Flexible Lyric Manipulation and Annotation-free Melody Guidance

(暂无摘要)

  • 链接: arXiv | PDF
  • 摘要: (暂无摘要)
📄 Abstract 中文翻译

🤖 AI Agents

CarePilot: A Multi-Agent Framework for Long-Horizon Computer Task Automation in Healthcare

(暂无摘要)

  • 链接: arXiv | PDF
  • 摘要: (暂无摘要)
📄 Abstract 中文翻译

Retrieval Improvements Do Not Guarantee Better Answers: A Study of RAG for AI Policy QA

(暂无摘要)

  • 链接: arXiv | PDF
  • 摘要: (暂无摘要)
📄 Abstract 中文翻译

🔥 通用热门

OmniWeaving: Towards Unified Video Generation with Free-form Composition and Reasoning

(暂无摘要)

  • 链接: arXiv | PDF
  • 摘要: (暂无摘要)
📄 Abstract 中文翻译

Marchuk: Efficient Global Weather Forecasting from Mid-Range to Sub-Seasonal Scales via Flow Matching

(暂无摘要)

  • 链接: arXiv | PDF
  • 摘要: (暂无摘要)
📄 Abstract 中文翻译

Toward Physically Consistent Driving Video World Models under Challenging Trajectories

(暂无摘要)

  • 链接: arXiv | PDF
  • 摘要: (暂无摘要)
📄 Abstract 中文翻译

📌 其他值得关注

Understanding the Challenges in Iterative Generative Optimization with LLMs

(暂无摘要)

  • 链接: arXiv | PDF
  • 摘要: (暂无摘要)
📄 Abstract 中文翻译

SpectralSplats: Robust Differentiable Tracking via Spectral Moment Supervision

(暂无摘要)

  • 链接: arXiv | PDF
  • 摘要: (暂无摘要)
📄 Abstract 中文翻译

GenMask: Adapting DiT for Segmentation via Direct Mask

(暂无摘要)

  • 链接: arXiv | PDF
  • 摘要: (暂无摘要)
📄 Abstract 中文翻译

来自 HuggingFace 热门或 Semantic Scholar 的较早论文,虽超出严格时间窗口但仍值得关注。

Can LLM Agents Be CFOs? A Benchmark for Resource Allocation in Dynamic Enterprise Environments

(暂无摘要)

  • 链接: arXiv | PDF
  • 摘要: (暂无摘要)
📄 Abstract 中文翻译

On the Direction of RLVR Updates for LLM Reasoning: Identification and Exploitation

(暂无摘要)

  • 链接: arXiv | PDF
  • 摘要: (暂无摘要)
📄 Abstract 中文翻译

EVA: Efficient Reinforcement Learning for End-to-End Video Agent

(暂无摘要)

  • 链接: arXiv | PDF
  • 摘要: (暂无摘要)
📄 Abstract 中文翻译

Omni-WorldBench: Towards a Comprehensive Interaction-Centric Evaluation for World Models

(暂无摘要)

  • 链接: arXiv | PDF
  • 摘要: (暂无摘要)
📄 Abstract 中文翻译

WildWorld: A Large-Scale Dataset for Dynamic World Modeling with Actions and Explicit State toward Generative ARPG

(暂无摘要)

  • 链接: arXiv | PDF
  • 摘要: (暂无摘要)
📄 Abstract 中文翻译


生成时间:2026-03-27 00:05:24 UTC | 数据来源:arXiv、HuggingFace、Semantic Scholar


Cover image source: Pixiv

Licensed under CC BY-NC-SA 4.0