1
AI-to-AI Code Reviews of GitHub Pull Requests
AI 自动审查 GitHub 拉取请求，实证研究揭示效果与挑战
Article URL: https://arxiv.org/abs/2608.21311 Comments URL: https://news.ycombinator.com/item?id=49426227 Points: 1 # Comments: 0
2
The Quality of Claude AI-authored Python Tests Is Not Weaker Than Human-authored Tests
实证研究显示，Claude生成的Python测试质量不逊于人类，AI写代码再添有力证据。
arXiv:2608.15188v1 Announce Type: cross Abstract: We evaluate the quality of Claude AI-written Python tests against human-written Python tests from tw…
3
Does ISO-Grounded NFR Specification Improve LLM Code Generation? A Comparison of Rich and Structured Interventions against a Natural-Language Baseline
一项实证对比发现，用ISO标准结构化非功能需求描述，能显著提升LLM生成代码的质量，提示工程优化有新思路。
arXiv:2608.13742v1 Announce Type: cross Abstract: In LLM-based code generation, Non-Functional Requirements (NFRs) are often specified as terse one-li…
4
How Organizations Use AI: Evidence from ChatGPT
从ChatGPT数据看企业AI落地真相，揭示组织采用AI的模式与成效。
arXiv:2608.12236v1 Announce Type: cross Abstract: We study how organizations use frontier generative AI by linking ChatGPT Enterprise account records …
5
Agent Skills Can Be Harmful: An Empirical Study of Skill-Induced Failures in LLM Agents
别迷信Agent技能！实证研究发现，某些技能反而会拖累大模型任务成功率，反直觉结论值得关注。
arXiv:2608.11888v1 Announce Type: new Abstract: Agent skills are the de facto mechanism for extending LLM agents with reusable guidance. A skill can s…
6
When Chain-of-Thought Helps and When It Hurts: An Empirical Investigation of the Serial-Depth Bottleneck in LLM Reasoning
揭示思维链并非万能，实证剖析串行深度瓶颈对LLM推理的边界影响。
arXiv:2608.09942v1 Announce Type: cross Abstract: It is widely assumed that chain-of-thought (CoT) prompting universally improves LLM reasoning. We in…
7
Demystifying Deep Learning Compiler Front End Bugs: An LLM-Aided Empirical Study
首次用LLM辅助系统剖析深度学习编译器前端缺陷，为编译器调试与AI结合提供实证新视角
Article URL: https://arxiv.org/abs/2607.25651 Comments URL: https://news.ycombinator.com/item?id=49150741 Points: 1 # Comments: 0
8
The Easy Trap: Why LLMs Underestimate Misconception-Driven Difficulty
LLM评分题目难度时可能被“常识”误导，一项研究揭示其低估基于误解的学生真实困难
arXiv:2607.26067v1 Announce Type: cross Abstract: Large language models (LLMs) are increasingly used for estimating item difficulty in educational ass…
9
Learning from 53.6K Real-World Developer Edits of AI-Generated Code
研究5.36万次真实开发者编辑，揭示AI代码生成模型的实际改进方向
arXiv:2607.25130v1 Announce Type: cross Abstract: Imperfections in AI-generated code require that software developers modify the generated code manual…
10
No Reliable Evidence of Self-Reported Sentience in Small Large Language Models
小参数大模型自称有意识？新研究用严谨测试给出否定答案。
arXiv:2601.15334v2 Announce Type: replace-cross Abstract: Whether language models possess sentience has no empirical answer. But whether they believe …
11
Effectiveness of LLM-based Software Diversity for Reliability Improvement -- an Empirical Study
用大模型生成多样代码提升软件可靠性？这项实证研究给出严谨验证，值得开发者关注。
arXiv:2607.03174v1 Announce Type: cross Abstract: Software diversity has been extensively studied as a means of reducing the risk of common-mode failu…
12
An Exploratory Study on LLM-Generated Code and Comments in Code Repositories
最新研究揭示LLM生成代码在真实仓库中的质量与注释模式，为开发者提供实用参考。
arXiv:2607.01867v1 Announce Type: cross Abstract: The use of LLMs in software development has become increasingly widespread on tasks such as code gen…
13
An Empirical Study of Security Calibration in Large Language Models for Code
一篇实证研究，揭示大语言模型在生成代码时对自己安全性的认知偏差，关乎代码安全可靠性。
arXiv:2606.31159v1 Announce Type: cross Abstract: Large Language Models (LLMs) are rapidly transforming software development, yet their use in securit…
14
Not-quite-human tastes: the stylized omnivorousness of LLM survey surrogates
LLM能否精准模拟人类文化品味？论文揭示其作为调查代理的“拟人”局限与刻板杂食性。
arXiv:2606.30085v1 Announce Type: new Abstract: Large-language models have proven to be remarkable if inconsistent parrots of public attitudes and opi…
15
"Generate" the Future of Work through AI: Empirical Evidence from Online Labor Markets
AI对在线劳动力市场的实证研究，揭示生成式技术如何重塑工作模式与就业生态。
arXiv:2308.05201v4 Announce Type: replace Abstract: Large Language Model (LLM)-based generative AI systems are general-purpose tools capable of augmen…
16
Output-Space Allocation Costs for Calibration-Guided LLM Compression: An Empirical Study
校准引导的LLM压缩竟有输出空间分配开销？这篇实证研究用数据揭示量化压缩的隐性成本，值得技术玩家细读。
arXiv:2606.27785v1 Announce Type: cross Abstract: Training-free compression methods for large language models (LLMs) often use calibration data to gui…
17
An Empirical Study of LLM-Generated Specifications for VeriFast
大模型自动生成VeriFast规格，实证检验其在分离逻辑验证中的效果与局限，为降低形式化验证门槛提供新思路。
arXiv:2606.26490v1 Announce Type: cross Abstract: Static verification tools can assure industrial scale software, but require significant human labor …
18
Constraint Tax in Open-Weight LLMs: An Empirical Study of Tool Calling Suppression Under Structured Output Constraints
开源LLM在结构化输出约束下，工具调用能力被抑制的实证研究，揭示“约束税”现象。
arXiv:2606.25605v1 Announce Type: new Abstract: Tool Calling and Structured Output are two core capabilities of modern Agent systems, yet their intera…
19
Who Owns the AI Recommendation? A Multi-Industry Empirical Map of Brand Category Ownership Across Large Language Models
大模型正重塑品牌推荐格局，这份跨行业实证图谱揭示AI推荐话语权归谁，品牌战略必读。
arXiv:2606.23057v1 Announce Type: cross Abstract: Large language models now mediate how buyers discover products and services, making the competitive …
20
MAS-PromptBench: When Does Prompt Optimization Improve Multi-Agent LLM Systems?
多智能体LLM系统提示何时优化才有效？这篇论文用系统性实验给出了答案，值得AI从业者关注。
arXiv:2606.23664v1 Announce Type: new Abstract: Multi-agent systems (MAS) offer a scalable path forward for agentic AI, comprising multiple LLM-based …