覆盖北京时间日期：2026-07-05、2026-07-06。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 20/25
Juliette Decugis, Fabian Gloeckle, Francis Bach, Taco Cohen, Gabriel Synnaeve · 2026-07-03 · arXiv export API
一句话结论：How can Large Language Models (LLMs) solve problems they currently cannot? Repeated sampling scales test-time compute but GPU cost grows linearly with attempts, while reinforcement learning…
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：How can Large Language Models (LLMs) solve problems they currently cannot? Repeated sampling scales test-time compute but GPU cost grows linearly with attempts, while reinforcement learning (RL) with verifiable rewards improves single-attempt accuracy at the e…
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 3Substance 4Evidence 5Actionability 3
arXiv · PDF
#2 · Total 20/25
Juanwu Lu, Junyu Zhu, Ziran Wang · 2026-07-03 · arXiv export API
一句话结论：Realistic traffic simulation requires agents that imitate logged behavior and can also be steered along interpretable axes. Such controllability enables engineers to isolate variables, repro…
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：Realistic traffic simulation requires agents that imitate logged behavior and can also be steered along interpretable axes. Such controllability enables engineers to isolate variables, reproduce specific edge cases, and test autonomous systems without real-wor…
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
#3 · Total 20/25
Qiaowei Miao, Kehan Li, Yawei Luo, Yi Yang · 2026-07-03 · arXiv export API
一句话结论：Generative diffusion models excel at synthesizing high-quality images, videos, and 3D content under multimodal control. However, arbitrary user-defined modality-to-4D (X-to-4D) generation re…
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：Generative diffusion models excel at synthesizing high-quality images, videos, and 3D content under multimodal control. However, arbitrary user-defined modality-to-4D (X-to-4D) generation remains challenging due to the high cost of constructing diverse dataset…
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
#4 · Total 20/25
Xuehui Wang, Xuankun Yang, Wei Shen · 2026-07-03 · arXiv export API
一句话结论：Visual token pruning is a crucial strategy for accelerating VLMs by compressing redundant image patches, yet existing methods often fail to preserve critical cues under dense instructions an…
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：Visual token pruning is a crucial strategy for accelerating VLMs by compressing redundant image patches, yet existing methods often fail to preserve critical cues under dense instructions and fine-grained queries. In this paper, we investigate this failure and…
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 3Novelty 4Substance 5Evidence 4Actionability 4
arXiv · PDF
#5 · Total 20/25
Nina Hosseini-Kivanani, Sandipana Dowerah · 2026-07-01 · arXiv export API
一句话结论：State-of-the-art speech datasets predominantly focus on widely spoken languages, often overlooking low-resource languages such as Luxembourgish, which remain underrepresented in speech techn…
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：State-of-the-art speech datasets predominantly focus on widely spoken languages, often overlooking low-resource languages such as Luxembourgish, which remain underrepresented in speech technology research. In this work, we introduce LuxEmo, a 21-hour conversat…
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 3Novelty 3Substance 5Evidence 5Actionability 4
arXiv · PDF
#6 · Total 19/25
Josh Hills, Ida Caspary, Asa Cooper Stickland · 2026-07-03 · arXiv export API
一句话结论：As AI coding agents become more autonomous, they increasingly ship code iteratively, with the codebase persisting across sessions. This persistence creates a new attack surface: a misaligned…
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 AI coding agents become more autonomous, they increasingly ship code iteratively, with the codebase persisting across sessions. This persistence creates a new attack surface: a misaligned or prompt-injected agent can distribute attacks across pull requests…
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 4Substance 3Evidence 2Actionability 5
arXiv · PDF
#7 · Total 19/25
Achint Mehta · 2026-07-03 · arXiv export API
一句话结论：Agentic coding assistants are increasingly given extra capabilities, such as browser based testing tools and design oriented system prompts, on the assumption that more capability yields bet…
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：Agentic coding assistants are increasingly given extra capabilities, such as browser based testing tools and design oriented system prompts, on the assumption that more capability yields better software. This study tested that assumption directly. Ninety indep…
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 4Evidence 1Actionability 5
arXiv · PDF
#8 · Total 19/25
Cristian-Gabriel Florea, Stelian Spînu · 2026-07-03 · arXiv export API
一句话结论：Over 285 million people worldwide live with a visual impairment, for whom everyday tasks such as avoiding obstacles, locating personal belongings, recognizing familiar faces, or handling cas…
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：Over 285 million people worldwide live with a visual impairment, for whom everyday tasks such as avoiding obstacles, locating personal belongings, recognizing familiar faces, or handling cash remain persistent obstacles to personal autonomy. Existing assistive…
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 4Substance 3Evidence 2Actionability 5
arXiv · PDF
#9 · Total 19/25
Kent K. Chang · 2026-07-03 · arXiv export API
一句话结论：Language models are increasingly used to quantify cultural phenomena, but what makes such measurement distinctively cultural? This paper argues that NLP work on culture is a material-discurs…
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：Language models are increasingly used to quantify cultural phenomena, but what makes such measurement distinctively cultural? This paper argues that NLP work on culture is a material-discursive practice: the apparatus -- model, data, annotation, evaluation --…
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 5
arXiv · PDF
#10 · Total 19/25
Yi Ren · 2026-06-28 · arXiv export API
一句话结论：Medical question-answering benchmarks rarely cover the maternal, neonatal, child, and reproductive-health questions a nurse-midwife asks, and, to our knowledge, no public chunk-level relevan…
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：Medical question-answering benchmarks rarely cover the maternal, neonatal, child, and reproductive-health questions a nurse-midwife asks, and, to our knowledge, no public chunk-level relevance benchmark exists for maternal-health guideline retrieval. We releas…
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 4Evidence 3Actionability 5
arXiv · PDF
#11 · Total 19/25
Gil Harari, Yoel Zimmermann, Ola Tangen Kulseng, Laura Zichi, Chuin Wei Tan, Marc L. Descoteaux · 2026-07-03 · arXiv export API
一句话结论：Machine learning interatomic potentials (MLIPs) have become a hallmark of AI for scientific simulation. While efforts on new architectures and datasets have led to increasingly accurate and…
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：Machine learning interatomic potentials (MLIPs) have become a hallmark of AI for scientific simulation. While efforts on new architectures and datasets have led to increasingly accurate and general models, the choice of optimizer for training has largely remai…
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 3Novelty 3Substance 5Evidence 5Actionability 3
arXiv · PDF
#12 · Total 19/25
A. S. Anudeep, Vaanathi Sundaresan · 2026-07-03 · arXiv export API
一句话结论：For clinical deployment, it is essential that automated diagnostic systems remain reliable when confronted with previously unseen cases, yet deep models routinely misclassify out-of-distribu…
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：For clinical deployment, it is essential that automated diagnostic systems remain reliable when confronted with previously unseen cases, yet deep models routinely misclassify out-of-distribution (OOD) inputs with high confidence, underscoring the need for more…
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 3Novelty 4Substance 5Evidence 4Actionability 3
arXiv · PDF
#13 · Total 19/25
Manuel Alonso-Carracedo, Ruben Fernandez-Boullon, Pedro Celard, Francisco J. Rodriguez-Martinez, Lorena Otero-Cerdeira · 2026-07-03 · arXiv export API
一句话结论：Scalable and reliable grading of command-line examinations remains a challenge in computing education, where rising enrolments make manual marking difficult and rule-based autograders cannot…
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：Scalable and reliable grading of command-line examinations remains a challenge in computing education, where rising enrolments make manual marking difficult and rule-based autograders cannot handle partial credit, equivalent solutions, or syntactic variation.…
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 3Novelty 4Substance 5Evidence 4Actionability 3
arXiv · PDF
#14 · Total 18/25
Thomas Winninger · 2026-07-03 · arXiv export API
一句话结论：Steering and monitoring activations in Large Language Models (LLMs) are increasingly used for both safety and interpretability. Early work assumed behaviours are encoded along single linear…
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：Steering and monitoring activations in Large Language Models (LLMs) are increasingly used for both safety and interpretability. Early work assumed behaviours are encoded along single linear directions, but recent findings suggest complex behaviours, such as th…
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 3Evidence 2Actionability 5
arXiv · PDF
#15 · Total 18/25
Wentao Zhang, Liliana Hotsko, Woojeong Kim, Pengyu Nie, Stuart Shieber, Yuntian Deng · 2026-07-03 · arXiv export API
一句话结论：Many everyday programming tasks resist clean rule-based implementation, such as alerting on important log lines, repairing malformed JSON, or ranking search results by intent, and are increa…
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：Many everyday programming tasks resist clean rule-based implementation, such as alerting on important log lines, repairing malformed JSON, or ranking search results by intent, and are increasingly outsourced to large language model APIs at the cost of locality…
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 3Novelty 4Substance 3Evidence 4Actionability 4
arXiv · PDF