AI论文速递 2026年05月29日（HuggingFace Daily Papers）

数据来源：https://huggingface.co/papers 采集时间：2026-05-29

📌 重点关注

1. Rethinking How to Remember: Beyond Atomic Facts in Lifelong LLM Agent Memory | arXiv — 【重点关注】 To enable reliable long-term interaction, LLM agents require a memory system ...

💡 长期记忆系统让Agent稳定交互，提升实用性 2. LACUNA: Safe Agents as Recursive Program Holes | arXiv — 【重点关注】 LLM agents increasingly act by writing code, yet a split persists between the...

💡 代码安全新思路，解决LLM执行代码的风险 3. Don't Guess, Just Ask: Resolving Ambiguity in Referring Segmentation via Multi-turn Clarification | arXiv — 【重点关注】 Referring segmentation aims to segment the target objects in images or videos...

💡 多轮澄清提升分割精度，避免误识别

📋 其他值得关注

1. Beyond 3D VQAs: Injecting 3D Spatial Priors into Vision-Language Models for Enhanced Geometric Reasoning | arXiv — Vision-Language Models (VLMs) often struggle with robust 3D spatial reasoning...
2. UI-KOBE: Knowledge-Oriented Behavior Exploration for Lightweight Graph-Guided GUI Agents | arXiv — Recent advances in mobile GUI agents have shown strong potential for automati...
3. Fast-dDrive: Efficient Block-Diffusion VLM for Autonomous Driving | arXiv — End-to-end autonomous driving via Vision-Language-Action (VLA) models demands...
4. GUI-CIDER: Mid-training GUI Agents via Causal Internalization and Density-aware Exemplar Reselection | arXiv — Despite the rapid progress of multimodal large language models in building Gr...
5. AgentHijack: Benchmarking Computer Use Agent Robustness to Common Environment Corruptions | arXiv — Autonomous computer use agents that powered by multimodal large language mode...
6. Efficient and Scalable Provenance Tracking for LLM-Generated Code Snippets | arXiv — Large language models (LLMs) for code completion and generation are increasin...
7. AgentFugue: Agent Scaling for Long-Horizon Tasks through Collective Reasoning | arXiv — Recent progress on long-horizon agentic tasks has been driven largely by scal...