覆盖北京时间日期：2026-07-10、2026-07-11。聚焦 Agent、LLM reasoning/planning/tool use/memory、RAG、coding agents、evaluation/benchmark 与训练/推理基础设施。
Top Picks
#1 · Total 21/25
QiHong Chen, Aaron Imani, Iftekhar Ahmed · 2026-07-10 · arXiv export API
一句话结论：Repository-level code generation requires implementing target functions while accounting for complex cross-file dependencies and project-specific conventions. Existing retrieval methods pred…
为什么重要：贴近 Agent / LLM reasoning / coding / evaluation / personalization 研究线，适合快速转成复现实验或产品验证。
方法要点：Repository-level code generation requires implementing target functions while accounting for complex cross-file dependencies and project-specific conventions. Existing retrieval methods predominantly rely on lexical, structural, or semantic similarity, often o…
实验/证据：Evidence 4/5；需全文核验 benchmark、baseline、ablation 与代码可得性。
局限/风险：快筛基于官方元数据/摘要，结论强度以论文全文为准。
Lucian 下一步：抽取任务定义、指标与 baseline，加入 Auto Research 阅读/复现实验队列。
Relevance 5Novelty 3Substance 4Evidence 4Actionability 5
arXiv · PDF
#2 · Total 24/25
Xiaoshuai Song, Liancheng Zhang, Kangzhi Zhao, Yutao Zhu, Zhongyuan Wang, Guanting Dong · 2026-07-10 · arXiv export API
一句话结论：Large language model (LLM)-based web search agents are transforming information seeking from simple factoid question answering into complex, deep-and-wide search and research-oriented tasks.…
为什么重要：贴近 Agent / LLM reasoning / coding / evaluation / personalization 研究线，适合快速转成复现实验或产品验证。
方法要点：Large language model (LLM)-based web search agents are transforming information seeking from simple factoid question answering into complex, deep-and-wide search and research-oriented tasks. A single ReAct-style agent is constrained by one long trajectory and…
实验/证据：Evidence 5/5；需全文核验 benchmark、baseline、ablation 与代码可得性。
局限/风险：快筛基于官方元数据/摘要，结论强度以论文全文为准。
Lucian 下一步：抽取任务定义、指标与 baseline，加入 Auto Research 阅读/复现实验队列。
Relevance 5Novelty 4Substance 5Evidence 5Actionability 5
arXiv · PDF
#3 · Total 20/25
Chuning Zhu, Eva Xu, Jose Barreiros, Krishnan Srinivasan, Paarth Shah, Abhishek Gupta · 2026-07-10 · arXiv export API
一句话结论：Human decision-making is highly flexible -- some actions are taken immediately; others require longer deliberation. Language models have exhibited a similar capacity for adaptive "reasoning.…
为什么重要：贴近 Agent / LLM reasoning / coding / evaluation / personalization 研究线，适合快速转成复现实验或产品验证。
方法要点：Human decision-making is highly flexible -- some actions are taken immediately; others require longer deliberation. Language models have exhibited a similar capacity for adaptive "reasoning." However, transferring this capability to continuous control policies…
实验/证据：Evidence 3/5；需全文核验 benchmark、baseline、ablation 与代码可得性。
局限/风险：快筛基于官方元数据/摘要，结论强度以论文全文为准。
Lucian 下一步：抽取任务定义、指标与 baseline，加入 Auto Research 阅读/复现实验队列。
Relevance 5Novelty 4Substance 3Evidence 3Actionability 5
arXiv · PDF
#4 · Total 23/25
Wenbo Xu, Zhimin Chen, Xiaojie Liang, Hengrui Liu, Wei Lu · 2026-07-10 · arXiv export API
一句话结论：Rapid advancements in video diffusion models and temporal editing tools have enabled the generation of highly realistic human-centric videos, posing unprecedented challenges to digital conte…
为什么重要：贴近 Agent / LLM reasoning / coding / evaluation / personalization 研究线，适合快速转成复现实验或产品验证。
方法要点：Rapid advancements in video diffusion models and temporal editing tools have enabled the generation of highly realistic human-centric videos, posing unprecedented challenges to digital content forensics. Existing benchmarks primarily focus on either face-swapp…
实验/证据：Evidence 4/5；需全文核验 benchmark、baseline、ablation 与代码可得性。
局限/风险：快筛基于官方元数据/摘要，结论强度以论文全文为准。
Lucian 下一步：抽取任务定义、指标与 baseline，加入 Auto Research 阅读/复现实验队列。
Relevance 5Novelty 4Substance 5Evidence 4Actionability 5
arXiv · PDF
#5 · Total 17/25
Emanuele Quinto, Carlo Andrea Rozzi, Francesco Zanitti · 2026-07-10 · arXiv export API
一句话结论：Large language model (LLM) applications increasingly use explicit workflows for tool use, retrieval, branching, checkpointing, and human approval. Existing workflow systems already address m…
为什么重要：贴近 Agent / LLM reasoning / coding / evaluation / personalization 研究线，适合快速转成复现实验或产品验证。
方法要点：Large language model (LLM) applications increasingly use explicit workflows for tool use, retrieval, branching, checkpointing, and human approval. Existing workflow systems already address many execution concerns. This paper proposes a Lisp-inspired but langua…
实验/证据：Evidence 2/5；需全文核验 benchmark、baseline、ablation 与代码可得性。
局限/风险：快筛基于官方元数据/摘要，结论强度以论文全文为准。
Lucian 下一步：抽取任务定义、指标与 baseline，加入 Auto Research 阅读/复现实验队列。
Relevance 5Novelty 3Substance 3Evidence 2Actionability 4
arXiv · PDF
#6 · Total 20/25
Guoxuan Xia, Luka Ribar, Paul Balanca · 2026-07-10 · arXiv export API
一句话结论：Speculative decoding accelerates sampling from an autoregressive LLM by using a faster auxiliary model to draft tokens which are then verified in parallel by the LLM. Standard speculative de…
为什么重要：贴近 Agent / LLM reasoning / coding / evaluation / personalization 研究线，适合快速转成复现实验或产品验证。
方法要点：Speculative decoding accelerates sampling from an autoregressive LLM by using a faster auxiliary model to draft tokens which are then verified in parallel by the LLM. Standard speculative decoding is lossless: its rejection and resampling steps exactly preserv…
实验/证据：Evidence 3/5；需全文核验 benchmark、baseline、ablation 与代码可得性。
局限/风险：快筛基于官方元数据/摘要，结论强度以论文全文为准。
Lucian 下一步：抽取任务定义、指标与 baseline，加入 Auto Research 阅读/复现实验队列。
Relevance 4Novelty 3Substance 5Evidence 3Actionability 5
arXiv · PDF
#7 · Total 24/25
Kaifeng Zhao, Mathis Petrovich, Haotian Zhang, Tingwu Wang, Siyu Tang, Davis Rempe · 2026-07-10 · arXiv export API
一句话结论：Generating realistic 3D human motions in real-time within interactive applications is key for animation, simulation, and humanoid robotics. While recent offline motion generation approaches…
为什么重要：贴近 Agent / LLM reasoning / coding / evaluation / personalization 研究线，适合快速转成复现实验或产品验证。
方法要点：Generating realistic 3D human motions in real-time within interactive applications is key for animation, simulation, and humanoid robotics. While recent offline motion generation approaches offer precise control via text and kinematic constraints, they lack th…
实验/证据：Evidence 5/5；需全文核验 benchmark、baseline、ablation 与代码可得性。
局限/风险：快筛基于官方元数据/摘要，结论强度以论文全文为准。
Lucian 下一步：抽取任务定义、指标与 baseline，加入 Auto Research 阅读/复现实验队列。
Relevance 5Novelty 4Substance 5Evidence 5Actionability 5
arXiv · PDF
#8 · Total 23/25
Cheng-De Fan, Chun-Wei Tuan Mu, Chen-Wei Chang, Chin-Yang Lin, Kun-Ru Wu, Yu-Chee Tseng · 2026-07-10 · arXiv export API
一句话结论：Recovering high-quality video from sparse event streams is a challenging task. Regression methods often blur textures, while existing generative models struggle with long-term stability. We…
为什么重要：贴近 Agent / LLM reasoning / coding / evaluation / personalization 研究线，适合快速转成复现实验或产品验证。
方法要点：Recovering high-quality video from sparse event streams is a challenging task. Regression methods often blur textures, while existing generative models struggle with long-term stability. We propose LongE2V, a novel approach that leverages pre-trained video dif…
实验/证据：Evidence 5/5；需全文核验 benchmark、baseline、ablation 与代码可得性。
局限/风险：快筛基于官方元数据/摘要，结论强度以论文全文为准。
Lucian 下一步：抽取任务定义、指标与 baseline，加入 Auto Research 阅读/复现实验队列。
Relevance 5Novelty 4Substance 4Evidence 5Actionability 5
arXiv · PDF
#9 · Total 18/25
Yuantian Shao, Peisong Wang, Zhilei Liu, Chuangyi Li, Yuanteng Chen, Pengcheng Xie · 2026-07-10 · arXiv export API
一句话结论：Large language models (LLMs) are increasingly constrained by memory capacity, weight bandwidth, and checkpoint storage during deployment. Existing low-bit compression methods mainly follow t…
为什么重要：贴近 Agent / LLM reasoning / coding / evaluation / personalization 研究线，适合快速转成复现实验或产品验证。
方法要点：Large language models (LLMs) are increasingly constrained by memory capacity, weight bandwidth, and checkpoint storage during deployment. Existing low-bit compression methods mainly follow two directions. Scalar or group-wise quantization is simple and compati…
实验/证据：Evidence 1/5；需全文核验 benchmark、baseline、ablation 与代码可得性。
局限/风险：快筛基于官方元数据/摘要，结论强度以论文全文为准。
Lucian 下一步：抽取任务定义、指标与 baseline，加入 Auto Research 阅读/复现实验队列。
Relevance 5Novelty 4Substance 3Evidence 1Actionability 5
arXiv · PDF
#10 · Total 23/25
Yifan Zhou, Qihao Yang, Yan Li, Donggang Li, Xiru Hu, Hokin Deng · 2026-07-10 · arXiv export API
一句话结论：Scientific ideas rarely start from a blank page. They inherit mechanisms, repair known limitations, and recombine pieces of earlier work, much like biological genomes. Current benchmarks sti…
为什么重要：贴近 Agent / LLM reasoning / coding / evaluation / personalization 研究线，适合快速转成复现实验或产品验证。
方法要点：Scientific ideas rarely start from a blank page. They inherit mechanisms, repair known limitations, and recombine pieces of earlier work, much like biological genomes. Current benchmarks still say little about whether AI systems can follow this inheritance str…
实验/证据：Evidence 5/5；需全文核验 benchmark、baseline、ablation 与代码可得性。
局限/风险：快筛基于官方元数据/摘要，结论强度以论文全文为准。
Lucian 下一步：抽取任务定义、指标与 baseline，加入 Auto Research 阅读/复现实验队列。
Relevance 4Novelty 4Substance 5Evidence 5Actionability 5
arXiv · PDF
#11 · Total 22/25
Xinlong Zhao, Dongsheng Liu, Hengyu Zhao, Zixuan Fu, Zheng Wang, Jie Cai · 2026-07-10 · arXiv export API
一句话结论：As available training data approaches its physical limit, gains from Scaling Laws have begun to diminish. Consequently, improving Large Language Models (LLMs) now depends less on data expans…
为什么重要：贴近 Agent / LLM reasoning / coding / evaluation / personalization 研究线，适合快速转成复现实验或产品验证。
方法要点：As available training data approaches its physical limit, gains from Scaling Laws have begun to diminish. Consequently, improving Large Language Models (LLMs) now depends less on data expansion and more on higher-quality data utilization. However, in the conte…
实验/证据：Evidence 4/5；需全文核验 benchmark、baseline、ablation 与代码可得性。
局限/风险：快筛基于官方元数据/摘要，结论强度以论文全文为准。
Lucian 下一步：抽取任务定义、指标与 baseline，加入 Auto Research 阅读/复现实验队列。
Relevance 4Novelty 4Substance 5Evidence 4Actionability 5
arXiv · PDF
#12 · Total 21/25
Siddharth Damodharan, Radhika Gupta, Ali Alshami, Ryan Rabinowitz, Jugal Kalita · 2026-07-10 · arXiv export API
一句话结论：Recent advances in Vision-Language Models, Large Language Models, and Multimodal Large Language Models have improved autonomous driving tasks such as scene understanding, decision making, tr…
为什么重要：贴近 Agent / LLM reasoning / coding / evaluation / personalization 研究线，适合快速转成复现实验或产品验证。
方法要点：Recent advances in Vision-Language Models, Large Language Models, and Multimodal Large Language Models have improved autonomous driving tasks such as scene understanding, decision making, trajectory prediction, and visual question answering. However, evaluatin…
实验/证据：Evidence 4/5；需全文核验 benchmark、baseline、ablation 与代码可得性。
局限/风险：快筛基于官方元数据/摘要，结论强度以论文全文为准。
Lucian 下一步：抽取任务定义、指标与 baseline，加入 Auto Research 阅读/复现实验队列。
Relevance 4Novelty 3Substance 5Evidence 4Actionability 5
arXiv · PDF
#13 · Total 21/25
Ali Larian, Qian Lin, Chang Zong Wu, Daniel S. Brown · 2026-07-10 · arXiv export API
一句话结论：As autonomous agents are increasingly deployed across diverse operational contexts, aligning their behavior with human intent demands reward functions that remain robust to such changes rath…
为什么重要：贴近 Agent / LLM reasoning / coding / evaluation / personalization 研究线，适合快速转成复现实验或产品验证。
方法要点：As autonomous agents are increasingly deployed across diverse operational contexts, aligning their behavior with human intent demands reward functions that remain robust to such changes rather than overfitting to any single environment. Inverse reinforcement l…
实验/证据：Evidence 3/5；需全文核验 benchmark、baseline、ablation 与代码可得性。
局限/风险：快筛基于官方元数据/摘要，结论强度以论文全文为准。
Lucian 下一步：抽取任务定义、指标与 baseline，加入 Auto Research 阅读/复现实验队列。
Relevance 4Novelty 4Substance 5Evidence 3Actionability 5
arXiv · PDF
#14 · Total 20/25
Ethan Leung, Elias Lumer, Corey Feld, Austin Huber, Vamse Kumar Subbiah, Kevin Paul · 2026-07-10 · arXiv export API
一句话结论：Reinforcement learning increasingly relies on an LLM judge to score each rubric criterion, and that judge acts as the reward model during training. Before such a signal can be trusted, we ne…
为什么重要：贴近 Agent / LLM reasoning / coding / evaluation / personalization 研究线，适合快速转成复现实验或产品验证。
方法要点：Reinforcement learning increasingly relies on an LLM judge to score each rubric criterion, and that judge acts as the reward model during training. Before such a signal can be trusted, we need to know how capable the judge must be and how biased it is. We stud…
实验/证据：Evidence 4/5；需全文核验 benchmark、baseline、ablation 与代码可得性。
局限/风险：快筛基于官方元数据/摘要，结论强度以论文全文为准。
Lucian 下一步：抽取任务定义、指标与 baseline，加入 Auto Research 阅读/复现实验队列。
Relevance 4Novelty 3Substance 5Evidence 4Actionability 4
arXiv · PDF
#15 · Total 16/25
Hongyu Liu, Chun Wang, Feng Gao, Xuanhua He, Yue Ma, Ziyu Wan · 2026-07-10 · arXiv export API
一句话结论：We propose OPSD-V, an on-policy self-distillation paradigm for post-training few-step autoregressive (AR) video diffusion models. Existing few-step AR video generators can produce long video…
为什么重要：贴近 Agent / LLM reasoning / coding / evaluation / personalization 研究线，适合快速转成复现实验或产品验证。
方法要点：We propose OPSD-V, an on-policy self-distillation paradigm for post-training few-step autoregressive (AR) video diffusion models. Existing few-step AR video generators can produce long videos with low latency, but still suffer from error accumulation and weake…
实验/证据：Evidence 2/5；需全文核验 benchmark、baseline、ablation 与代码可得性。
局限/风险：快筛基于官方元数据/摘要，结论强度以论文全文为准。
Lucian 下一步：抽取任务定义、指标与 baseline，加入 Auto Research 阅读/复现实验队列。
Relevance 4Novelty 4Substance 4Evidence 2Actionability 2
arXiv · PDF