Skip to content
澎湃新闻 · 检索「AI风险」· 澎湃新闻马上测工作室 朱轩 吴琪 实习生 林霄 陈霜·Indexed by this site · Originally published· Evaluation →· Its place in the scan record →

Same-text tests on CNKI, Wanfang and VIP: AI-generation rate detection results differ widely, original content wrongly flagged

马上测丨知网、万方、维普三平台同文实测: AI 生成率检测结果悬殊,原创内容遭冤

AI summary

The Paper's Instant Testing Studio submitted the same batch of manuscripts to CNKI, Wanfang, and VIP for AIGC detection. For an approximately two-thousand-character paper that was 100% AI-generated, the suspected AI rates on the three platforms were 99.9%, 24.68%, and 49.97%, respectively. For another approximately 3500-character purely AI-written manuscript, CNKI detected 67.6% and VIP 69.8%, while Wanfang's total proportion of suspected AI-generated text was 7.15%, with a significantly suspected proportion of 0.00%. A nearly ten-thousand-character manuscript written by a student based on field research and interviews was judged by the three respectively as 30.1%, 25.22%, and 35.94%, and the transcribed content of interviewee recordings was also marked in red. All three reports noted that the detection results are for reference only; some universities have already set limits based on the proportion of AI-generated content, for example, Sichuan University requires no more than 20% for humanities and no more than 15% for science, engineering, and medicine.

Read originalthepaper.cn