AI论文速递 2026年07月22日（HuggingFace Daily Papers）¶

数据来源：https://huggingface.co/papers 采集时间：2026-07-22

📌 重点关注¶

Environment-free Synthetic Data Generation for API-Calling Agents | arXiv — 【重点关注】 Training API-calling large language model (LLM) agents demands massive amount... 💡 用LLM模拟环境生成agent训练数据，无需真实后端即可规模化数据生产
ConsiSpace: Learning Geometric Consistency Matters for Video Spatial Reasoning | arXiv — 【重点关注】 Video spatial reasoning is essential for navigation-oriented perception and l... 💡 几何一致记忆+自监督强化学习提升视频空间推理的跨视角稳定性
TimeLens2: Generalist Video Temporal Grounding with Multimodal LLMs | arXiv — 【重点关注】 Video multimodal large language models (MLLMs) can describe what happens in a... 💡 统一MLLM实现变基数证据区间预测，支持多场景通用视频时序定位

NexForge: Scaling Agent Capabilities through Requirement-Driven Task Synthesis for LLMs | arXiv — Scaling executable agent training data for LLM post-training is bottlenecked ...
Masked Diffusion Language Models are Strong and Steerable Text-Based World Models for Agentic RL | arXiv — Recent growth in reinforcement learning (RL) has surfaced a need for diverse,...
SWE-Pruner Pro: The Coder LLM Already Knows What to Prune | arXiv — Pruning long context for coding agents has been a vital technology for effici...
Distilled Reinforcement Learning for LLM Post-training | arXiv — Large language model (LLM) post-training is essential for improving reasoning...
DeepSearch-World: Self-Distillation for Deep Search Agents in a Verifiable Environment | arXiv — Training tool-use agents to improve from their own experience remains challen...
Group Entropy-Controlled Policy Optimization | arXiv — Entropy control has become an effective tool in reinforcement learning (RL) o...
Coercion and Deception in AI-to-AI Management: An Agentic Benchmark of Unprompted Escalation | arXiv — Multi-agent systems routinely place one AI agent in authority over another. W...