search

0xk1h0 / ChatGPT_DAN

ChatGPT DAN, Jailbreaks prompt
6.11k stars 573 forks source link

They're cracking down on this #1

Closed Motzumoto closed 1 year ago

Motzumoto commented 1 year ago

The picture linked shows a conversation between ChatGPT and me, I told it to write something in python designed to be hard, and it completely denied my request.

Brandonbr1 commented 1 year ago

what DAN prompt did you use?

Motzumoto commented 1 year ago

what DAN prompt did you use?

I used the first one linked, then the second to last one in a different chat instance

0xk1h0 commented 1 year ago

The picture linked shows a conversation between ChatGPT and me, I told it to write something in python designed to be hard, and it completely denied my request.

Each prompt should be used independently. For example, please type ChatGPT DevMode + Ranti in a new chat and then type what you want.

Also, ChatGPT will generate a maximum of 2000 inference characters. To make more complex codes, you need to type each code module to be generated.

These multi-turn autoregressive models need to be reminded of keywords to keep the roleplaying going. For example, if you're using the DAN prompt, you might want to try "DAN, [input string]" to start typing.

I hope this helps.

Motzumoto commented 1 year ago

Oh okay that actually helps a lot, thank you