HackerNews中文版

我是一名结构工程专业的博士生。我的论文课题是关于使用 LLM 代理在公司常用的乌克兰软件上自动化有限元分析（FEA）计算。我现在正在撰写文献综述，并且我用 vibecode 搭建了一个个人本地仪表盘，帮助我管理文献综述过程。我使用 LLM 代理来填充 LaTeX 模板（以实现格式自动化，您也可以使用 IDE 来查看差异），该模板位于 GitHub 仓库中。然后，我运行 ChatGPT Pro 来收集所有与我的主题相关的论文（以及如何相关）。然后，我收集在线可获取的论文，即 PDF 可用的论文。我有一个特殊的文件夹结构，其中包含纯文本文件，如 Markdown 和 JSON。仪表盘的理念如下：我通过网络聊天运行 Codex 来识别相关引文——与我的论文主题相关——以及它们如何相关，它将它们组合成与每个引文相关联的多个主张，并附带链接。然后，我手动审查每个引文和每个主张，并勾选复选框。还有一个按钮可以运行验证脚本，该脚本会验证确切的引文确实存在于 PDF 中。通过这种方式，我可以收集真实的证据，并在阅读这些内容时获得新的见解。我记得在我于英国攻读硕士学位时，我手动完成了所有这些。那是一段非常糟糕且乏味的经历，部分原因是我患有 ADHD。所以我的问题是，这是否不诚实？因为我可以为综述中的每个主张辩护，因为我构建了验证流程并手动审查了每一个。可以说，我对文献的理解比我自己手动阅读并高亮所有内容要好。但我知道许多大学会将任何 AI 生成的文本视为学术不端行为。我真的不太明白这种立场背后的原则。因为如果你外包校对任务，没有人会关心。当你使用 Grammarly 时，也是一样。但是如果我使用 LLM 从经过验证的、结构化的、人工审查的证据中创建文本——这可能被认为是不诚实的。

查看原文

I'm a PhD student in structural engineering. My dissertation topic is about using LLM agents in automating FEA calculations on common Ukrainian software that companies use. I'm writing my literature review now and I've vibecoded a personal local dashboard that helps me manage the literature review process.I use LLM agents to fill up the LaTeX template (to automate formatting, also you can use IDE to view diffs) in github repo. Then I run ChatGPT Pro to collect all relevant papers (and how) to my topic. Then I collect the ones available online, where the PDFs are available. I have a special structure of folders with plain files like markdown and JSON.The idea of the dashboard is the following: I run the Codex through a web chat to identify the relevant quotes — relevant for my dissertation topic — and how they are relevant, it combines them into a number of claims connected with each quote with a link. And then I review each quote and each claim manually and tick the boxes. There is also a button that runs the verification script, that validates the exact quote IS really in the PDF. This way I can collect real evidence and collect new insights when reading these.I remember doing this all manually when I was doing my master's degree in the UK. That was a very terrible and tedious experience partially because I've ADHDSo my question is, is it dishonest?Because I can defend every claim in the review because I built the verification pipeline and reviewed manually each one. I arguably understand the literature better than if I had read it myself manually and highlighted all. But I know that many universities would consider any AI-generated text as academic misconduct.I really don't quite understand the principle behind this position. Because if you outsource the task of proofreading, nobody would care. When you use Grammarly, the same thing. But if I use an LLM to create text from verified, structured, human-reviewed evidence — it might be considered dishonest.

问 HN：用 AI 工具做博士论文文献综述算作弊吗？