AI垃圾内容正在摧毁在线社区
推荐指数 36.0 NO. 019 · 2026.05.08
发布2026/05/07Score68Comments20
为什么值得看
文章指出LLM生成的低质内容（AI slop）正泛滥于Reddit、Stack Overflow等社区，导致真实用户流失、平台信任崩塌。对AI从业者警示：你训练模型所依赖的社区数据，可能正在被你自己的工具污染。
媒体预览
编辑判断
这篇文章的价值不在于它描述的现象——圈内人早已感知到——而在于它把'社区死亡'和'模型训练数据来源'直接挂钩了。大多数团队做RLHF、做知识蒸馏时，默认Reddit、Stack Overflow、Quora是'干净的人类标注数据'，但这个假设正在快速失效。
更隐蔽的风险是反馈循环：模型A用污染后的社区数据训练，生成更多slop涌入社区，模型B再抓取时数据质量进一步劣化。已经有团队在清理预训练数据时发现2023年后的Stack Overflow回答中AI生成比例超过30%。
如果你还在用未经时间筛选的公开社区数据做训练，建议立刻加入'发布时间过滤'和'人机检测'两个预处理步骤，或者优先选用2022年前的归档数据作为可信基线。
社区反馈
负面 31 条评论
核心争论：AI slop是否已大规模渗透社区，以及人类能否有效识别LLM生成内容
The writing here is good. Quote of the day "Any fool can feed coins into a fruit machine and pull the arm."
> AI slop is driving up the noise, and making the signal more and more difficult to discern in communities. Thank you, puts into words why I no longer look at Show HNs.
I have largely written Reddit off and no longer visit it after an experiment I did where I had an agent karma farm for me and do some covert advertising. As I went through the posts it wrote I realized that as a reader I would have NO idea that these were just written by a computer. Many many peopl