1
LLM Sparsity Prior for Robust Feature Selection
利用大模型蕴含的稀疏性先验，为高维数据特征选择提供鲁棒策略，理论贡献显著。
arXiv:2605.23102v1 Announce Type: cross Abstract: Large language models (LLMs) offer a scalable mechanism to elicit domain-informed prior information …
2
蔚来乐道 L80 首批车主超 90% 选择 BaaS 电池租用方式购买，15.68 万元起
IT之家 5 月 25 日消息，蔚来旗下乐道汽车今日公布了 L80 首批车主购买数据： 90%+ 的车主选择了 BaaS 电池租用方式购买 62% 的车主选择了 Max+ 车型 58% 的车主选择了静岳黑外观 70%+ 的车主选择了月榕棕内饰 55%+ 的车主选择了 21 英寸六辐曜黑轮圈 70%+…
3
How does an AI agent pick from 686 skills in a second?
实测AI代理如何在一秒内从686个技能中精准选中最优解，准确率68%的秘密武器是语义路由+向量检索。
I ran an empirical test on the "skills as semantic router" pattern for Claude Code agents. I indexed 686 randomly sampled skills from a 4,556-skill co…
4
Unified Data Selection for LLM Reasoning
提出统一数据选择框架，为LLM推理任务高效筛选高质量训练数据，显著提升推理能力。
arXiv:2605.22389v1 Announce Type: new Abstract: Effectively training Large Language Models (LLMs) for complex, long-CoT reasoning is often bottlenecke…
5
Balancing Uncertainty and Diversity of Samples: Leveraging Diversity of Least, High Confidence Samples for Effective Active Learning
提出Leveraging Diversity of Least方法，通过平衡不确定性与多样性优化主动学习样本选择，提升模型训练效率
arXiv:2605.22169v1 Announce Type: new Abstract: Deep learning models, including Convolutional Neural Networks (CNNs) and Vision Transformers (ViTs), h…
6
What and When to Distill: Selective Hindsight Distillation for Multi-Turn Agents
多轮对话代理只能“一刀切”蒸馏？这篇论文给出何时蒸馏、蒸馏什么的智能选择策略
arXiv:2605.19447v1 Announce Type: new Abstract: Reinforcement learning can train LLM agents from sparse task rewards, but long-horizon credit assignme…
7
Reconnecting. – – 5/5 why don't they fix codex
用户直言Codex问题长期未修复，暗示将倒向Claude，OpenAI团队却沉默以对
it's been a month tagging sama openai tibo on X for this issue and no one seem to reply and eveyone is falttering codex, im sure im not the only one f…
8
Sustainability via LLM Right-sizing
评估11款专有模型，揭示何时小模型更优，兼顾可持续性与成本效益
arXiv:2504.13217v3 Announce Type: replace Abstract: Large language models (LLMs) have become increasingly embedded in organizational workflows. This h…
9
Dr.LLM: Dynamic Layer Routing in LLMs
提出动态层路由机制，让LLM推理时跳过无关层，显著提升效率与精度。
arXiv:2510.12773v2 Announce Type: replace-cross Abstract: Large Language Models (LLMs) process every token through all layers of a transformer stack, …
10
Difficulty-Based Preference Data Selection by DPO Implicit Reward Gap
新方法用DPO隐式奖励差距衡量样本难度，自动筛选高质量偏好数据，提升模型训练效率。
arXiv:2508.04149v2 Announce Type: replace-cross Abstract: Aligning large language models (LLMs) with human preferences is a critical challenge in AI r…
11
SMART Fine-tuning Factor Augmented Neural Lasso
提出SMART框架，将预训练模型融入高维非参数变量选择，为微调提供理论基础。
arXiv:2604.12288v2 Announce Type: replace-cross Abstract: Fine-tuning is a widely used strategy for adapting pre-trained models to new tasks, yet its …
12
LeanSearch v2: Global Premise Retrieval for Lean 4 Theorem Proving
LeanSearch v2提出全局前提检索，一次性找出Lean 4定理所需全部引理，突破现有单步或语义匹配局限。
arXiv:2605.13137v2 Announce Type: replace-cross Abstract: Proving theorems in Lean 4 often requires identifying a scattered set of library lemmas whos…
13
Stage-adaptive Token Selection for Efficient Omni-modal LLMs
阶段自适应token选择策略，显著提升全模态大语言模型推理效率，突破多任务性能瓶颈。
arXiv:2605.20035v1 Announce Type: new Abstract: Omni-modal large language models (om-LLMs) achieve unified audio-visual understanding by encoding vide…
14
Robust Checkpoint Selection for Multimodal LLMs via Agentic Evaluation and Stability-Aware Ranking
提出基于代理评估与稳定性感知排名的多模态大模型检查点选择新方法，提升选点鲁棒性与模型性能。
arXiv:2605.18852v1 Announce Type: new Abstract: Checkpoint selection for multimodal large language models (MLLMs) presents significant challenges when…
15
Causal Intervention-Based Memory Selection for Long-Horizon LLM Agents
用因果干预方法优化LLM代理长期记忆选择，告别语义相似度检索的局限性
arXiv:2605.17641v1 Announce Type: cross Abstract: Long-horizon LLM agents rely on persistent memory to support interactions across sessions, yet exist…
16
Scales++: Compute Efficient Evaluation Subset Selection with Cognitive Scales Embeddings
用认知嵌入高效筛选评估子集，大幅降低大模型评测成本，保持预测准确性。
arXiv:2510.26384v2 Announce Type: replace-cross Abstract: The prohibitive cost of evaluating large language models (LLMs) on comprehensive benchmarks …
17
An Approximation Algorithm for Graph Label Selection
被ICML 2026收录的图标签选择近似算法，9页7图含理论分析与证明。
arXiv:2605.18623v1 Announce Type: cross Abstract: In the graph label selection problem, one is given an $n$-vertex graph and a budget $k$, and seeks t…
18
Forecasting Downstream Performance of LLMs With Proxy Metrics
用代理指标提前预判LLM下游表现，为模型选型提供可靠决策依据
arXiv:2605.18607v1 Announce Type: cross Abstract: Progress in language model development is often driven by comparative decisions: which architecture …
19
Learning-Zone Energy: Online Data Selection for Efficient RL Post-Training
提出Learning-Zone Energy方法，在线选择数据以提升RL后训练效率，避免均匀分配浪费计算。
arXiv:2605.17003v1 Announce Type: new Abstract: Reinforcement Learning (RL) post-training has emerged as the dominant paradigm for eliciting mathemati…
20
Convex Dataset Valuation for Post-Training
提出凸数据集估值方法，解决LLM后训练中数据集选择的成本与性能权衡问题
arXiv:2605.16704v1 Announce Type: new Abstract: Improving LLM performance on downstream tasks sometimes requires leveraging auxiliary datasets during …