hackcv · 实时 AI 资讯聚合

自然语言处理 · 大模型、推理与对齐

全部生成式 AI Agent / 智能体多模态计算机视觉自然语言处理强化学习 AI 安全编程助手公司与产品政策与监管研究进展 AI 视频

论文精选 82

WearableQA: A Benchmark for Health Reasoning over Real-World Wearable Data· 可穿戴健康推理基准

Recent advances in wearable sensing enable continuous monitoring of physiological and beha…

领域：cs.CL作者：Ji Soo Lee、Xilun Chen、Pierce Chuang

#cs.CL

📎 arXiv🕒 09-05 01:52🔗 arxiv.org

论文精选 82

LexFlip: A Dissociation Diagnostic for Legal Meaning Preservation Metrics

Does a simplified legal clause still say what the original said? The checks in current use…

领域：cs.CL作者：Gaurab Baral

#cs.CL

📎 arXiv🕒 09-04 23:49🔗 arxiv.org

论文精选 75

Self-Supervised Lexical Representation Learning for Fast, Large-Scale Phylogenetic Inference· 自监督词表示学习

Computational phylogenetics has become an essential tool in historical linguistics, yet it…

领域：cs.CL作者：Tim Wientzek

#cs.CL

📎 arXiv🕒 09-04 23:23🔗 arxiv.org

论文精选 82

A Verifier-Guided Explainable Reasoning Framework with Gold-Anchored QLoRA, Task-Aware Mixture-of-Experts, and Group-Relative RLVR· 基于验证器的可解释推理框架

Large language models (LLMs) show strong reasoning ability, but their explanations can rem…

领域：cs.CL作者：Thi Kim Trang Vo、Nam Tien Le、Thi Kim Nguyet Vo

#cs.CL

📎 arXiv🕒 09-04 22:52🔗 arxiv.org

论文精选 82

Can Large Language Models Anticipate Behavioral Responses to Social Policies? A Case of Pension Enrollment Prediction among China's Flexible Workers· 用大模型预测养老金参保行为

Assessing the impacts of social policy changes is a widely acknowledged challenge for poli…

领域：cs.CL作者：Yumiao Li、Peixin Liu、Donglin Di

#cs.CL

📎 arXiv🕒 09-04 22:24🔗 arxiv.org

论文精选 82

Compile by Training: Turning Natural-Language Specifications into Local Neural Functions· 编译时训练：将自然语言规范转换为本地神经函数

Many recurring text functions are easy to describe but difficult to implement with rules, …

领域：cs.CL作者：Yuntian Deng、Pengyu Nie、Stuart Shieber

#cs.CL

📎 arXiv🕒 09-04 01:59🔗 arxiv.org

论文精选 82

ESPO: Error-Structured Prompt Optimization via Diagnose, Diversify, and Stabilize· ESPO: 基于错误结构的提示优化

Evolutionary prompt optimizers such as GEPA suffer from prompt bloat: each iteration appen…

领域：cs.CL作者：Lihao Liu、Peng Tang、Kunwar Yashraj Singh

#cs.CL

📎 arXiv🕒 09-04 01:59🔗 arxiv.org

论文精选 82

Legibility is Not Interpretability: Comparing Judged and Actual Importance in Chain-Of-Thought Reasoning· 可读性与可解释性：链式思考中的重要性对比

Reasoning traces from chain-of-thought models appear to offer a legible window into how a …

领域：cs.CL作者：Kevin Du、Alexander Hoyle、Laura Ruis

#cs.CL

📎 arXiv🕒 09-04 01:59🔗 arxiv.org

论文精选 80

Knowledge Acquisition During Pre-training? Large Language Models Learn Better With Auxiliary Views· 大模型预训练时如何更好学习？辅助视角很重要

Gaps remain in our understanding of how large language models (LLMs) acquire knowledge dur…

领域：cs.CL作者：Joseph Lee、Yidi Huang、Dokyoon Kim

#cs.CL

📎 arXiv🕒 09-04 01:57🔗 arxiv.org

论文精选 83

Last Translation Benchmark· 最后的翻译基准测试

For scientific progress, we need benchmarks that test the limits of state-of-the-art model…

领域：cs.CL作者：Vilém Zouhar、Niyati Bafna、Mukund Choudhary

#cs.CL

📎 arXiv🕒 09-04 01:54🔗 arxiv.org

论文精选 82

User Feedback Provides a Unique Signal that LLMs Can not Detect· 用户反馈的重要性

Harnessing naturally occurring feedback from user interactions offers a promising learning…

领域：cs.CL作者：Shachar Don-Yehiya、Leshem Choshen、Omri Abend

#cs.CL

📎 arXiv🕒 09-03 01:42🔗 arxiv.org

论文精选 85

DiscoSign: Discourse-Aware Text to Sign Language Gloss Translation· DiscoSign: 面向话语的文本到手语词汇翻译

Sign language processing systems have traditionally operated at the sentence level, ignori…

领域：cs.CL作者：Vasileios Baltatzis、Mert Inan、Connor Gillis

#cs.CL

📎 arXiv🕒 09-03 00:31🔗 arxiv.org

论文精选 82

EarlyEval: Cheaper Agent Evaluation via Early Outcome Prediction· 早期评估：通过早期结果预测降低代理评估成本

Evaluating LLM agents is essential for guiding their development, yet it has grown prohibi…

领域：cs.CL作者：Yuling Shi、Zhensu Sun、Junsen Dong

#cs.CL

📎 arXiv🕒 09-03 00:15🔗 arxiv.org

论文精选 85

HyperStyler: Low-resource Authorship Style Transfer via Context-aware Style Navigation and Hypernetworks· HyperStyler：基于上下文感知风格导航和超网络的低资源作者风格迁移

Low-resource authorship style transfer (LAST) aims to rewrite text into the style of an ar…

领域：cs.CL作者：Jongkyung Shin、Minguk Jeon、Chanwoo Park

#cs.CL

📎 arXiv🕒 09-03 00:07🔗 arxiv.org

论文精选 83

From Reweighting to Rewriting: Unlocking the Intervention Effects of Influential Samples in Training Data Attribution· 从重权到重写：解锁训练数据归因中的干预效应

Training data attribution (TDA) aims to identify training examples that shape model behavi…

领域：cs.CL作者：Yuzhang Luo、Chenpeng Wang、Jianhui Chen

#cs.CL

📎 arXiv🕒 09-03 00:07🔗 arxiv.org

论文精选 82

Beyond Scores: Understanding LLM-as-a-Judge Mechanisms in Summarization Evaluation· 超越评分：理解大语言模型裁判机制

LLM-based evaluators of natural language generation (NLG) quality are widely deployed as s…

领域：cs.CL作者：Himil Vasava、Ming Jiang

#cs.CL

📎 arXiv🕒 09-02 01:59🔗 arxiv.org

论文精选 81

CordisBench: Can Language Models Reason About Component Lifecycles in Dynamic Agent Harnesses?· CordisBench：语言模型能否推理动态代理生命周期？

Dynamic agent harnesses let language models change the software that shapes their own exec…

领域：cs.CL作者：Damien Sileo、Dimitri Kachler

#cs.CL

📎 arXiv🕒 09-02 01:59🔗 arxiv.org

论文精选 82

The Rise of Verbal Reinforcement Learning· 口头强化学习的兴起

Natural language is emerging as a primary feedback channel for improving language agents, …

领域：cs.CL作者：Kshitij Tayal、Arun Sharma、Genta Indra Winata

#cs.CL

📎 arXiv🕒 09-02 01:58🔗 arxiv.org

论文精选 84

StudentSim: Training LLM-based Student Simulators· StudentSim: 训练基于 LLM 的学生模拟器

AI tutors are most useful when they adapt to each student's strengths, weaknesses, and pre…

领域：cs.CL作者：Ke Yang、Chenglong Wang、Michel Galley

#cs.CL

📎 arXiv🕒 09-02 01:55🔗 arxiv.org

论文精选 85

Closing Cost-Quality Gap in Document VLMs: Difficulty-Aware Data Curation and Quality-Adjusted Deployment Economics· 关闭文档 VLM 成本-质量差距

Extracting structured fields from hundreds of millions of documents annually remains costl…

领域：cs.CL作者：Maksim Evdokimov、Matvey Ivanov、Dmitrii Tsiupin

#cs.CL

📎 arXiv🕒 09-02 01:40🔗 arxiv.org

论文精选 82

Context-Aware Interleaved Batching for WhisperX· WhisperX的上下文感知交错批处理

While WhisperX accelerates speech transcription via intra-audio batching, it isolates audi…

领域：cs.CL作者：Carlos Bain、Max Bain

#cs.CL

📎 arXiv🕒 09-01 01:59🔗 arxiv.org

论文精选 82

Configurable Semantic Chunking for Biomedical Information Extraction in Retrieval-Augmented Generation· 生物医学信息抽取中的可配置语义分块

BioMedRAG introduced retrieval-augmented generation with a learned chunk scorer for biomed…

领域：cs.CL作者：Riya Ahuja、Tim Kacprowski、Roya Shiasi Sardoabi

#cs.CL

📎 arXiv🕒 09-01 01:44🔗 arxiv.org

论文精选 85

DIASENTINEL: An Auditable Multi-Agent System for Guideline-Grounded Diabetes Risk Screening· DIASENTINEL：用于糖尿病风险筛查的多代理系统

Large language models (LLMs) offer promising clinical decision support but remain vulnerab…

领域：cs.CL作者：Yung Wei Shueh、Zhi-Jie Chen、Chia-Hsuan Hsu

#cs.CL

📎 arXiv🕒 09-01 01:40🔗 arxiv.org

论文精选 82

PaperGym: Rubric-Centered Evolution for Research-Plan Generation· PaperGym: 以评分标准为中心的研究计划生成进化

Research planning is the decisive capability of AI scientists. Yet a research plan admits …

领域：cs.CL作者：Yuhan Wang、Zhengxi Lu、Yuchen Yan

#cs.CL

📎 arXiv🕒 09-01 01:31🔗 arxiv.org

论文精选 85

Aspire: Can Models Self-Evolve from Vague Goals?· AI模型能否从模糊目标自我进化？

Many important forms of human learning begin with a vague goal, such as "become a better p…

领域：cs.CL作者：Yuhao Wu、Jingyuan Zhang、Jiajun Shi

#cs.CL

📎 arXiv🕒 09-01 01:14🔗 arxiv.org

论文精选 82

A Formal Limitation on Learning Human Language From Textual Corpora· 从文本语料库学习人类语言的正式限制

Can a listener recover what a speaker means from the form of an utterance alone? We answer…

领域：cs.CL作者：Emily Cheng、Ryan Cotterell

#cs.CL

📎 arXiv🕒 08-29 01:38🔗 arxiv.org

论文精选 82

Phoneme- and Word-Level Metrics Using Self-Supervised Speech Representations for Forced Alignment Evaluation· 使用自监督语音表示的音素和单词级评估指标

Forced alignment evaluation typically requires manually annotated timestamps, limiting lar…

领域：cs.CL作者：V. S. D. S. Mahesh Akavarapu、Michael Daniel、Gerhard Jäger

#cs.CL

📎 arXiv🕒 08-29 00:39🔗 arxiv.org

论文精选 82

Ladders in Chaos: When, How, (and Perhaps Why) Does Test-Time Scaling Improve LLM Machine Translation· 混沌中的阶梯：测试时缩放如何改善大语言模型机器翻译

Two forms of test-time scaling for Large Language Models (LLMs) have emerged as effective …

领域：cs.CL作者：Di Wu、Sergey Troshin、Christof Monz

#cs.CL

📎 arXiv🕒 08-29 00:22🔗 arxiv.org

论文精选 83

NL2AGBench: Benchmarking LLM Auto-Formalization for AlphaGeometry· 自然语言到 AlphaGeometry 基准

Recent advances in large language models (LLMs) have demonstrated strong capabilities in n…

领域：cs.CL作者：Samuel Xiao、Judy Song、Rory Hu

#cs.CL

📎 arXiv🕒 08-29 00:07🔗 arxiv.org

论文精选 83

Blind Men and the Elephant: Probing the Epistemic Myopia of LLMs under Long-Tail Divergent Knowledge· 盲人摸象：探索LLM在长尾分歧知识下的认知局限

Factual question answering (QA) typically assumes a single canonical answer, obscuring whe…

领域：cs.CL作者：Zhuoshi Pan、Junru Lu、Yan Qian

#cs.CL

📎 arXiv🕒 08-29 00:06🔗 arxiv.org

论文精选 82

CritICL: Inference-Time Weak-to-Strong Generalization from Small Language Model Failure Modes· CritICL：小模型失败模式改进大模型推理

Recent advances in inference-time scaling have significantly improved the reasoning perfor…

领域：cs.CL作者：Yufan Wu、Yinghui He、Zhengyi Hu

#cs.CL

📎 arXiv🕒 08-28 01:59🔗 arxiv.org

论文精选 83

TTPO: Test-Time Policy Optimization· 测试时策略优化

Recent prominent post-training methods, such as Reinforcement Learning (RL) and On-Policy …

领域：cs.CL作者：Aozhe Wang、Zhengxi Lu、Jianze Wang

#cs.CL

📎 arXiv🕒 08-28 01:58🔗 arxiv.org

论文精选 82

Stochastic Estimation of Transduced Language Models

Transduced language models (TLMs) compose a pretrained \emph{source} language model with a…

领域：cs.CL作者：Vésteinn Snæbjarnarson、Samuel Kiegeland、Manuel de Prada Corral

#cs.CL

📎 arXiv🕒 08-28 01:50🔗 arxiv.org

论文精选 85

Boosting LLM Exploration via Weak-Model Guidance in RLVR· 通过弱模型引导增强大模型探索

Reinforcement Learning with Verifiable Rewards (RLVR) significantly improves LLM reasoning…

领域：cs.CL作者：Xingyu Shen、Huishuai Zhang、Peng Li

#cs.CL

📎 arXiv🕒 08-28 01:45🔗 arxiv.org

论文精选 83

Consolidating RLVR Capabilities Across Domains: A Deep Dive into Fusion Paradigms· 跨域整合RLVR能力：融合范式深度解析

Reinforcement learning with verifiable rewards (RLVR) improves specific capabilities of la…

领域：cs.CL作者：Siye Wu、Kai Yang、Yuchen Cai

#cs.CL

📎 arXiv🕒 08-28 01:38🔗 arxiv.org

论文精选 75

Prefix Sliding for efficient test-time scaling· Prefix Sliding：高效的测试时扩展方法

Test-time scaling uses extra test-time compute to improve performance, such as letting lan…

领域：cs.CL作者：Niklas Muennighoff、Zhengyang Wang、Zeyi Chen

#cs.CL

📎 arXiv🕒 08-27 01:37🔗 arxiv.org

论文精选 81

Fine-Tuning Whisper for Automatic Speech Recognition in Baniwa: A Preliminary Study· 微调 Whisper 用于班瓦语自动语音识别

Automatic Speech Recognition (ASR) technologies have achieved remarkable performance in re…

领域：cs.CL作者：Leonardo Duart、Tiago Fonseca、Thiago Chacón

#cs.CL

📎 arXiv🕒 08-27 01:29🔗 arxiv.org

论文精选 83

Beyond Local Surprise: Grounded Dialogue as Selective Belief Revision under Referential Uncertainty· 超越局部惊奇：对话中的选择性信念修正在参照不确定性下

When a speaker refers to a scene that the listener cannot directly see, the listener must …

领域：cs.CL作者：Ziming Liu、Bhanu Chaitanya Jasti、Ziyang Xu

#cs.CL

📎 arXiv🕒 08-27 01:13🔗 arxiv.org

论文精选 82

VISA: Agentic Self-Evolving Data Synthesis for Multimodal Instruction Following· VISA：代理自我进化数据合成

Multimodal instruction-following models require training data that is accurate, diverse, v…

领域：cs.CL作者：Min Zeng、Guanxin Tan、Libin Cen

#cs.CL

📎 arXiv🕒 08-27 00:55🔗 arxiv.org

论文精选 82

Distinct dynamics of conceptual and referential disruptions in human reading and large language model processing· 人类阅读与大模型处理中的不同中断动态

Linguistic meaning is grounded in conceptual content, from which reference to particular e…

领域：cs.CL作者：Rui He、Nihal Altay、Wolfram Hinzen

#cs.CL

📎 arXiv🕒 08-27 00:47🔗 arxiv.org

论文精选 82

BrowserForge: Scaling Web Episode via Parallel Browser Sandboxes· BrowserForge：通过并行浏览器沙箱扩展网页片段

Web agents that act from rendered pixels avoid the fragility and heavy token cost of readi…

领域：cs.CL作者：Fei Tang、Huawen Shen、Zhiqiong Lu

#cs.CL

📎 arXiv🕒 08-26 01:35🔗 arxiv.org

论文精选 85

Reading Is Not Using: Retrieval, Judgment, and the Design of AI Financial Research Workflows· 读取不等于使用：AI金融研究工作流程的设计

Large language models (LLMs) are increasingly deployed as AI analysts to process financial…

领域：cs.CL作者：Miao Liu、Zhizhe Liu

#cs.CL

📎 arXiv🕒 08-26 01:31🔗 arxiv.org

论文精选 83

Structurally-bounded Agentic Graph Exploration for Evidence-Grounded Scholarly DeepSearch· 结构受限的代理图探索

We present Crase, a bounded and inspectable alternative to deep research agents for schola…

领域：cs.CL作者：Rima Hazra、Sayan Layek、Somnath Banerjee

#cs.CL

📎 arXiv🕒 08-26 00:51🔗 arxiv.org

论文精选 83

Linear Probing Provides Robust and Efficient Detection of Machine-Generated Text· 线性探测提供稳健高效的机器生成文本检测

Distinguishing machine-generated text (MGT) from human-written text (HWT) becomes increasi…

领域：cs.CL作者：Gerrit Quaremba、Hanqi Yan、Elizabeth Black

#cs.CL

📎 arXiv🕒 08-26 00:21🔗 arxiv.org

论文精选 83

ExpConCAD: Experience-Guided Text-to-CAD Generation from Shape Descriptions with Implicit Spatial Constraints· 基于经验的文本到CAD生成

Text-to-CAD aims to generate executable CAD programs from natural-language descriptions. H…

领域：cs.CL作者：Jingyao Liu、Jinkang Tang、Chen Huang

#cs.CL

📎 arXiv🕒 08-25 23:58🔗 arxiv.org

论文精选 82

SWE Refactor Bench: Can Coding Agents Complete a Long-Horizon, Whole-Repository Stack Migration?· SWE Refactor Bench：代码代理能否完成全库迁移？

Modern software systems accumulate technical debt over decades of development, which makes…

领域：cs.CL作者：Deyao Hong、Yizhe Chi、Wenyi Li

#cs.CL

📎 arXiv🕒 08-25 01:59🔗 arxiv.org

论文精选 83

ConvergeFlow: Language Flow with Provable Convergence to Token Embeddings· ConvergeFlow: 词嵌入收敛的语言流

Recent advances in continuous diffusion and flow-based language models (LMs) have achieved…

领域：cs.CL作者：Na Li、Yuchen Jiao、Changxiao Cai

#cs.CL

📎 arXiv🕒 08-25 01:54🔗 arxiv.org

论文精选 80

When Names Cross Scripts: A Source-Grounded Benchmark for Historical Entity Reconciliation in the Mongol World· 多语言历史人物身份匹配基准测试

Historical people may appear under different languages, scripts, and transcription traditi…

领域：cs.CL作者：Xiang Chen、Zeyu Zhang

#cs.CL

📎 arXiv🕒 08-25 01:10🔗 arxiv.org

论文精选 82

On the Threat Model of Weird Generalization and Emergent Misalignment· 微调数据对奇怪泛化和突发偏差的影响

Narrow fine-tuning on small, domain-specific datasets can produce broad and surprising cha…

领域：cs.CL作者：Miriam Wanner、Mark Dredze、William Walden

#cs.CL

📎 arXiv🕒 08-25 00:41🔗 arxiv.org

论文精选 82

What's the Catch? Evaluating Temporal Consistency in Vision-Language Models· 时间一致性评估：视觉-语言模型的挑战

Vision-language models (VLMs) achieve strong performance on video and image-sequence bench…

领域：cs.CL作者：Marek Hradil、Danae Sánchez Villegas

#cs.CL

📎 arXiv🕒 08-25 00:40🔗 arxiv.org

论文精选 85

Move by Move: Measuring and Steering How LLMs Conduct Psychotherapy· 逐举动分析：大模型如何开展心理治疗

Users increasingly turn to large language models for emotional support, yet little is know…

领域：cs.CL作者：Afonso Baldo、Hugo Pitorro、Areti Vassilopoulos

#cs.CL

📎 arXiv🕒 08-22 01:32🔗 arxiv.org

论文精选 81

Prompt-Model Interaction Reaches the Fixed Points: A deterministic, task-free structural readout -- and the factorizations of it that failed· 提示词-模型交互达到固定点：确定性的结构读取

That a prompt's effect is not a property of the prompt is established: prompts optimised f…

领域：cs.CL作者：Nicolás Vera Zúñiga

#cs.CL

📎 arXiv🕒 08-22 01:25🔗 arxiv.org

论文精选 83

Memory Augmentation Unlocks Efficient Chain-of-Thought Reasoning· 记忆增强解锁高效链式思考

Large language models often rely on Chain-of-Thought (CoT) reasoning to solve complex task…

领域：cs.CL作者：Simeng Zhang、Yilong Chen、Wenyuan Zhang

#cs.CL

📎 arXiv🕒 08-22 00:22🔗 arxiv.org

论文精选 85

EnSI-RAG: Entity-Structure-Indexed Retrieval-Augmented Generation for Long-Document Question Answering· 长文档问答的实体结构检索增强生成模型

Question answering (QA) over long, connected documents remains challenging because relevan…

领域：cs.CL作者：Xuanyu Meng、Jiashuo Sun、Jash Rajesh Parekh

#cs.CL

📎 arXiv🕒 08-22 00:05🔗 arxiv.org

论文精选 82

Benchmarking Patent Drafting from Inventor-Style Disclosures· 从发明人风格披露基准测试专利撰写

While recent large language models (LLMs) have achieved promising results on individual pa…

领域：cs.CL作者：Lekang Jiang、Wenjun Sun、Stephan Goetz

#cs.CL

📎 arXiv🕒 08-22 00:00🔗 arxiv.org

论文精选 83

ConceptGuard: Benchmarking Context-Sensitive Unlearning in Large Language Models· ConceptGuard：评估大模型上下文敏感卸载

Large Language Models (LLMs) increasingly require selective removal of harmful or sensitiv…

领域：cs.CL作者：Sahil Kale、Ian Harris

#cs.CL

📎 arXiv🕒 08-21 01:59🔗 arxiv.org

论文精选 83

G-CARL: Grounded Checklist-Aligned Reward Learning for Patient-Oriented Medical Report Interpretation· G-CARL：面向患者的医疗报告解读

Personalized interpretation of medical reports has emerged as an increasingly important ne…

领域：cs.CL作者：Shiao Xie、Siyu Chen、Jianwei Lv

#cs.CL

📎 arXiv🕒 08-21 01:59🔗 arxiv.org

论文精选 82

Inducing Task Models from Computer-Use Traces

Naturalistic computer-use traces, passively recorded screenshots and mouse or keyboard act…

领域：cs.CL作者：Yucheng Jiang、Zora Zhiruo Wang、Ruishi Chen

#cs.CL

📎 arXiv🕒 08-21 01:57🔗 arxiv.org

论文精选 85

Inject, Align, Recover: Staged Post-Training for Retrieval-Free Document Knowledge Internalization· 无检索文档知识内化的大模型后训练

Large language models often fail to answer questions about a bounded document collection w…

领域：cs.CL作者：Qian Kou、Xiaofeng Shi、Xiaosong Qiu

#cs.CL

📎 arXiv🕒 08-21 01:14🔗 arxiv.org

论文精选 82

Task-CoEvolve: Efficient Harness Optimization via Adaptive Validation Task Selection· Task-CoEvolve: 通过自适应任务选择优化代理框架

We present a novel approach to efficient LLM agent harness optimization through adaptive v…

领域：cs.CL作者：Atsuyuki Miyai、Kiyoharu Aizawa、Toshihiko Yamasaki

#cs.CL

📎 arXiv🕒 08-20 23:24🔗 arxiv.org