TE
科技回声
首页24小时热榜最新最佳问答展示工作
GitHubTwitter
首页

科技回声

基于 Next.js 构建的科技新闻平台,提供全球科技新闻和讨论内容。

GitHubTwitter

首页

首页最新最佳问答展示工作

资源链接

HackerNews API原版 HackerNewsNext.js

© 2025 科技回声. 版权所有。

Tell HN: ChatGPT subtly censored my code in a way that hurt

6 点作者 ji_zai大约 1 年前
I asked it to add functionality to a piece of code I was writing.<p>It replaced: &quot;# this is the important line (THIS HAS TO MATCH THE ACTUAL FORMAT OF THE FUCKING FILE COMING IN!)&quot;<p>with: &quot;# Ensure this matches the actual file format&quot;<p>It updated this comment above the area that I had asked it to edit. And I just copied and pasted the entire snippet without checking.<p>Now, I&#x27;ve been using &quot;FUCK&quot; as my search term to quickly jump to this line. It served a damn purpose. It was intentional.<p>This makes me more cautious about copying and pasting code, not because it might not work, but because the code might be altered in subtle ways that aren&#x27;t easily perceivable, that could negatively impact the codebase in the long-run.<p>Of course, a simple base prompt should fix this, but I might not catch this happening in many other cases, or other mediums for that matter.<p>I believe that AI will significantly help us with all forms of creation: code, videos, films, music, etc. Soon, we won&#x27;t be able to imagine creating without it.<p>We&#x27;ll need to prevent it from censoring us in ways that we don&#x27;t want to be censored.

2 条评论

taylorfinley大约 1 年前
It&#x27;s nice to ask for diffs, then you can see exactly what&#x27;s changing.
bunbun69大约 1 年前
if you don&#x27;t enjoy the censorship in ChatGPT then consider rolling your own LLM