RoyceWQ: Hi, been using various AI writers to see what one is detected the most by the GPT-2 Output Detector.
19/20 generations were accurately detected.
Any idea why that is?
Is there a way to prevent this?
Can you share links of some such tools
I've been using huggingface.co/openai-detector and tested about 10 AI Writers including Jarvis.
I have tested a tool, while writing live post on it, but it showed it ,90 percent negative, it seems that there result is not accurate
i have tested the same while writing the live on huggingface.co/openai-detector its showed 60% negative, it means that this tool is not perfect at all.
Wow! Thanks for bringing huggingface.co/openai-detector to my attention.
Are you both saying that Creaitor.ai or the huggingface.co/openai-detector is not accurate/perfect?
Because I’ve tested several AIs and they all get detected. I don’t know how huggingface.co knows. Scary!!!
I’ve also noticed that if you paraphrase the text afterwards then it’s not detected as AI generated.
OpenAi Detector basically knows some of the methods GPT-2/3 uses to create the text. This tool calls them tokens. The more text you give it, the more accurate it becomes. Far from perfect for sure. At least these are what I noticed while testing like 10 AiWriters with 20 outputs each.
I've also noticed that Content Rewriters and Paraphrases do help with the detection rate.
Royce, what AI Writers came out on top, with less detections?