Skip to main content

Table 14 Percentage of undetected cases

From: Testing of detection tools for AI-generated text

Tool

03-AI

04-AI

05-ManEd

06-Para

Total

FNR

Recall

Check For AI

0

1

5

6

12

33.3%

66.7%

Compilatio

0

1

3

7

11

30.6%

69.4%

Content at Scale

9

9

9

9

36

100.0%

0.0%

Crossplag

0

2

4

7

13

36.1%

63.9%

DetectGPT

0

1

5

7

13

36.1%

63.9%

Go Winston

0

1

4

7

12

33.3%

66.7%

GPT Zero

1

0

1

1

3

8.3%

91.7%

GPT-2 Output Detector Demo

0

1

4

7

12

33.3%

66.7%

OpenAI Text Classifier

4

1

4

7

16

44.4%

55.6%

PlagiarismCheck

4

3

6

6

19

52.8%

47.2%

Turnitin

0

0

4

6

10

27.8%

72.2%

Writeful GPT Detector

1

3

6

8

18

50.0%

50.0%

Writer

4

3

5

7

19

52.8%

47.2%

Zero GPT

2

1

5

5

13

36.1%

63.9%

Average

19.8%

21.4%

51.6%

71.4%