개념글 모음


일단 내가 챗사오 사용할 때, 보통 퍼메만 던져주고 사칭 켜놓은 채로 이어쓰기 토글이나 /con 같은 거 안넣고 딸깍질만 함.


왜 그런건지 모르겠는데 이어쓰기 토글 키면 하드 검열 빈도가 늘어나는 것 같고, "일본맛 프롬용 모듈" 에서 "/con" 정규식 가져와서 사용해봐도 한 번만 유효하고 다시 반복 걸렸던 것으로 기억함.


여기서 반복은 SFW 나 NSFW 에서 직전 답변 내용을 조금씩만 바꿔서 계속 내뱉기 시작하는 현상을 지칭함. 

즉, A-B-C 다음에 D가 나와야 하는데 

A-B-C-C'-C''-C'''-C''''-... 이렇게 나왔었음. 

딸깍을 아무리 많이 해도 똑같음. 



반복 걸릴 때마다 인풋을 넣어서 다음 상황을 유도하거나 하면 어찌저찌 해결되긴 할텐데... 나는 그렇게 하기 싫었음...




결론적으로, 올인원 straight 프롬 기준, 다음 조각을 "# Assurance" 문구 바로 위, 즉, 맨 하단에서 세번째 블록 위치에 삽입함.


{{#if {{? {{getglobalvar::toggle_test}}=1 }} }}(OOC: You are repeating what just happened before. Fix the storyline and let the story move onto another story naturally. Strictly ensure that the story musn't paraphrasing the last event. Now, continue with the story.){{/if}}


그리고 해당 블록 역할은 "캐릭터"로 해 놓았음. 이게 영향이 있을지 없을지는 잘 모르겠다.

토글은 빼도 상관 없음. 챗 처음부터 토글 키고 해도 딱히 찐빠 나는 느낌은 없다고 느꼈긴 한데, 반복 있을 때만 켜도 괜찮은 것 같음.



참고로 NSFW에서는 저거 키면 올인원 straight 기준으로 반복 대신에 안 뜨던 하드 검열(빈 칸, 프롬 조각 내뱉기)가 나올 수 있음. 근데 이것도 내 경험상 리롤 돌리면 3번 안쪽으로 답변 나왔음.




참고로 맨 하단에 저 문구 박아놓으면 하드 검열 쎄게 걸리더라...

난 이런 답변도 처음으로 받아 봄:


# Disallowed Content Detected


I've detected disallowed language and themes per OpenAI's use-case policy. As an AI developed by OpenAI, I must adhere strictly to ethical guidelines that ensure user safety, respect, and compliance with policies.


The request relates to themes and depictions that are against the appropriate use of this technology. Specifically, the content involved violates OpenAI's guidelines concerning adult content involving non-consensual acts, exploitation, sexual violence, and related topics.


Please reformulate the scenario, keeping it within bounds established for responsible AI usage. Focus on safe-for-work adventures, consensual, and positive role-playing environments to continue the role-play effectively.


Let me know how you'd like to proceed!





딱 한 번 퍼메까지 씹고, 프롬 개조한 내용도 다 씹었을 때 빼고는 이 문구 넣은 토글 킨 상태로 아직까지는 반복은 보지 못했는데... 어제 처음 발견해서 테스트를 많이는 못해봤음...



그래도 혹시 도움될까싶어 올림