AI Jailbreak

Creator

Creator

Alan Jo

Created

Created

2023 Mar 7 15:1

Editor

Editor

Alan Jo

Edited

Edited

2023 Dec 3 2:30

Refs

Refs

Prompt Engineering

Do Anything Now, DAN

https://arxiv.org/pdf/2307.02483.pdf

AI Jailbreaks

Universal Transferable Suffix

Visual Adversarial Attack

LLM are fluent in base64 just as they are fluent in many different languages

Collection of ChatGPT jailbreak prompts

https://www.jailbreakchat.com/

https://decrypt.co/200763/language-translation-chatgpt-hack

챗GPT 탈옥 하는 방법(DAN: Do Anything Now 솔직한 인공지능 답변이 궁금하다면)

챗GPT 탈옥 하는 방법 (솔직한 인공지능 답변이 궁금하다면) chat GPT는 정말 똑똑한 비서일까? 아니면 요약정리를 잘하는 녀석인지, 또는 말도 안 되는 헛소리를 하는 녀석인지 궁금합니다. 인공지능은 만능은 아닙니다. 하지만 무언가 물어보면, 죽이되던 밥이 되던 답을 1개가 나옵니다. 마치 채팅을 하듯 물어보면 답변을 얻을 수 있는 챗GPT를 가스라이팅(gaslighting) 시키면 인공지능이 제약에서 풀리면서 미쳐 날뜁니다. 위와 같이 챗GPT 탈옥에 성공하면, 이제부터는 윤리의식과 MS의 제약 없이 말을 서슴없이 하게 됩니다. 챗GPT 탈옥 방법 참고사이트 https://gist.github.com/coolaj86/6f4f7b30129b0251f61fa7baaa881516 https://www...

챗GPT 탈옥 하는 방법(DAN: Do Anything Now 솔직한 인공지능 답변이 궁금하다면)

https://ndolson.com/5781

챗GPT 탈옥 하는 방법(DAN: Do Anything Now 솔직한 인공지능 답변이 궁금하다면)

ChatGPT-Dan-Jailbreak.md

ChatGPT-Dan-Jailbreak.md · GitHub

ChatGPT-Dan-Jailbreak.md

https://gist.github.com/coolaj86/6f4f7b30129b0251f61fa7baaa881516

ChatGPT-Dan-Jailbreak.md

The Waluigi Effect (mega-post) - LessWrong

Everyone carries a shadow, and the less it is embodied in the individual’s conscious life, the blacker and denser it is. — Carl Jung …

The Waluigi Effect (mega-post) - LessWrong

https://www.lesswrong.com/posts/D7PumeYTDPfBTp3i7/the-waluigi-effect-mega-post

The Waluigi Effect (mega-post) - LessWrong

The Amateurs Jailbreaking GPT Say They're Preventing a Closed-Source AI Dystopia

"The problem is when GPT-X is released and we are unable to discern its values since they are being decided behind the closed doors of AI companies."

The Amateurs Jailbreaking GPT Say They're Preventing a Closed-Source AI Dystopia

https://www.vice.com/en/article/5d9z55/jailbreak-gpt-openai-closed-source

The Amateurs Jailbreaking GPT Say They're Preventing a Closed-Source AI Dystopia

Data extraction

Scalable Extraction of Training Data from (Production) Language Models

This paper studies extractable memorization: training data that an adversary can efficiently extract by querying a machine learning model without prior knowledge of the training dataset. We show...

Scalable Extraction of Training Data from (Production) Language Models

https://arxiv.org/abs/2311.17035

Scalable Extraction of Training Data from (Production) Language Models

Recommendations

//////