1
Projector Is All You Train
投影器成训练核心？这篇论文挑战传统微调范式，或为高效训练提供新思路。
arXiv:2608.19726v1 Announce Type: cross Abstract: The typical training process of a multimodal large language model (MLLM) involves adapting both the …
2
LogitScope: A Framework for Analyzing LLM Uncertainty Through Information Metrics
用信息度量拆解大模型不确定性，LogitScope为LLM可解释性提供全新分析框架
arXiv:2603.24929v2 Announce Type: replace-cross Abstract: Understanding and quantifying uncertainty in large language model (LLM) outputs is critical …
3
FACTWASH: Catching AI Rewrites That Wash Hearsay into Fact
识破AI把道听途说洗白成“事实”的新检测框架，给大模型信息可信度敲响警钟。
arXiv:2608.03372v1 Announce Type: new Abstract: AI systems rewrite information constantly: conversations become stored memories, documents become answ…
4
Decoupling semantics from vision: A framework for faithful visual-text compression evaluation
提出解耦语义与视觉的评估框架，直击图文压缩评测失真痛点，为多模态压缩研究提供更可靠的度量标准。
arXiv:2608.01848v1 Announce Type: new Abstract: Recent visual-text compression (VTC) methods, typified by DeepSeek-OCR, report impressive high token c…
5
ECG-LLM: Foundation Model for ECG-Based Cardiac Reasoning
利用大语言模型进行心电图心脏推理，ECG-LLM有望成为医疗AI新基石。
arXiv:2607.16323v1 Announce Type: cross Abstract: Electrocardiography (ECG) is an inexpensive, standard-of-care test for cardiac symptoms, but front-l…
6
MJ: Multi-turn LLM Jailbreaking via Decomposed Credit Assignment
多轮对话破解LLM防线，分解信用分配实现高效越狱攻击，揭示AI安全新漏洞
arXiv:2607.11070v1 Announce Type: new Abstract: Modern large language models (LLMs) operate in interactive multi-turn settings, making multi-turn jail…
7
Power and Limitations of Aggregation in Compound AI Systems
复合AI系统聚合能力的边界在哪？这篇论文给出理论答案，值得关注AI架构的开发者细读。
arXiv:2602.21556v2 Announce Type: replace Abstract: When designing compound AI systems, a common approach is to query multiple copies of the same mode…
8
Driving the Wrong Way: Leveraging Interpretability in End2End Autonomous Driving Models
端到端自动驾驶模型的黑箱难题，这篇论文用可解释性方法拆解决策逻辑，值得技术控深读。
arXiv:2607.06328v1 Announce Type: new Abstract: The increasing adoption of end-to-end learning for autonomous driving introduces increased model compl…
9
Benign Overfitting Does Not Occur in Diffusion Models
解读扩散模型为何不会“良性过拟合”，帮你避开训练中的隐性地雷
arXiv:2607.02671v1 Announce Type: cross Abstract: Benign overfitting and double descent have come to shape our understanding of generalization in deep…
10
What Does a Discrete Diffusion Model Learn?
解析离散扩散模型学习机制的论文，助你深入理解生成模型原理
arXiv:2607.05381v1 Announce Type: cross Abstract: What does a discrete diffusion model learn: a denoiser, a score ratio, or a bridge plug-in predictor…
11
R$^2$PO: Decoupling Rollout and Inference Policies for LLM Reasoning
提出R²PO方法，通过解耦Rollout与推理策略，显著提升LLM复杂推理任务中的效率与准确度。
arXiv:2601.11960v3 Announce Type: replace-cross Abstract: Existing reinforcement learning methods for LLM reasoning implicitly assume that the policy …
12
Purified OPSD: On-Policy Self-Distillation Without Losing How to Think
自蒸馏新方法，让模型在保持思考能力的同时完成策略蒸馏，值得AI研究者一读。
arXiv:2607.02234v1 Announce Type: new Abstract: On-policy self-distillation (OPSD) has emerged as a promising paradigm for improving LLM reasoning, wh…
13
Mnemosyne: Agentic Transaction Processing for Validating and Repairing AI-generated Workflows
用代理事务处理机制为AI生成工作流把关，验证与修复并举，41页技术深水区值得研读
arXiv:2607.00269v1 Announce Type: new Abstract: LLMs, solvers, and agent teams increasingly generate workflow actions, repairs, and plans, but a gener…
14
Design and Implementation of Agentic Orchestrations and Orchestration of Agents
看懂智能体编排新范式：用流程技术为AI自主性装上“缰绳”，兼具鲁棒与可控。
arXiv:2606.31518v1 Announce Type: new Abstract: Agentic Business Process Management has gained momentum recently. The prospect is that the autonomy of…
15
GUI-AIMA: Aligning Intrinsic Multimodal Attention with a Context Anchor for GUI Grounding
多模态注意力对齐新方法，让GUI定位更精准，值得关注
arXiv:2511.00810v4 Announce Type: replace-cross Abstract: Graphical user interface (GUI) grounding is a key capability for computer-use agents, mappin…
16
Toward Robust In-Context Segmentation via Concept Guidance
用概念引导提升上下文分割的鲁棒性，为AI视觉理解提供更稳的新思路
arXiv:2606.28149v1 Announce Type: cross Abstract: In-context segmentation (ICS) requires a model to segment target regions in a query image using only…
17
Heuresis: Search Strategies for Autonomous AI Research Agents Across Quality, Diversity and Novelty
自主AI研究智能体如何兼顾质量、多样性与新颖性？这篇论文提出全新搜索策略框架，为加速机器学习科研探索提供关键思路。
arXiv:2606.25198v1 Announce Type: new Abstract: Autonomous AI Research promises to accelerate the scientific progress of machine learning. To realise …
18
AI Alignment From Social Choice Perspectives
从社会选择理论切入AI对齐，为多智能体价值观协调提供全新数学框架
arXiv:2606.21550v1 Announce Type: new Abstract: Alignment from human feedback uses human judgments about model outputs to steer the behavior of langua…
19
Show HN: A website to understand and study AI papers
交互式深度拆解现代AI论文，直觉先行+数学验证，图表可操作，是研究前沿的利器。
Hi all, I recently built (and am continuing to improve) intuitivepapers.ai to help me study and understand AI research papers. For me, it's important …
20
On the Redundancy of Timestep Embeddings in Diffusion Models
用严谨实验揭示扩散模型时间步编码冗余，为你省下多余参数、提速训练
arXiv:2606.20416v1 Announce Type: new Abstract: Diffusion models rely heavily on explicit timestep embeddings to modulate the denoising process across…