Pangram doesn't work; it only damages people's reputations.

时间与来源

发布: 09/05 16:57
收录: 09/05 21:00
来源类型: 开发者社区
档位: 社区
信源状态: 正常

AI 摘要

一位用户反映，AI检测工具Pangram错误地将其15年前撰写的七篇论文标记为AI生成，尽管该工具声称其误报率仅为千分之一。与此同时，用户指出，Pangram反而将实际由AI生成的文本识别为人类撰写。这表明Pangram可能存在不准确性，并可能因错误识别作品而损害个人声誉。

https://preview.redd.it/j316ve0jfqnh1.png?width=1585&format=png&auto=webp&s=afa5df0f49118ae72a8d0beaad8af6d5373ff5be

I uploaded seven essays, and it flagged all seven as AI-generated in at least some parts. The funny thing is that I wrote those essays almost 15 years ago. How can they claim to have a false-positive rate of 1 in 1,000?

This text, on the other hand, was actually generated by AI, yet it identifies it as human-written.