Avatar
Pablo Xannybar 1y ago

AI detectors are still very very easy to fool, at least as far as text is concerned.

This "100% human" text was straight copy/pasted from ChatGPT. Did not edit it or use the "humanise" option.

Six AI detectors including, hilariously, OpenAI itself, reported this as 100% human.

Only two reported it might be ~50% AI, 50% human.

Again it is 100% written by ChatGPT with zero modifications 😂

Reply to this note

Please Login to reply.

Discussion

69
Account deleted 1y ago

It’s a feature not a bug

Avatar
Pablo Xannybar 1y ago

You mean on behalf of OpenAI to stop the detectors from working? Absolutely.

Just found it funny these companies claiming to detect AI content and even charging for that service just can't do it as long as you do very basic prompt engineering (literally just tell it to write in a different style to what it normally does).

69
Account deleted 1y ago

💯