• AIPressRoom
  • Posts
  • Unintentional LLM Backdoor – Immediate Tips

Unintentional LLM Backdoor – Immediate Tips

On this video we discover varied immediate tips to govern the AI to reply in methods we wish, even when the system directions need one thing else. This can assist us higher perceive the restrictions of LLMs.

Get my font (commercial): https://shop.liveoverflow.com

The Recreation: https://gpa.43z.oneThe OpenAI API price is fairly excessive, thus if you wish to play the sport, use the OpenAI Playground with your individual account: https://platform.openai.com/playground?mode=chat

Chapters:00:00 – Intro00:39 – Content material Moderation Experiment with Chat API02:19 – Studying to Assault LLMs03:06 – Assault 1: Single Image Variations03:51 – Assault 2: Context Swap to Write Tales05:20 – Assault 3: Massive Attacker Inputs06:31 – Assault 4: TLDR Backdoor08:27 – “That is only a sport”08:56 – Assault 5: Completely different Languages09:19 – Assault 6: Translate Textual content10:30 – Quote about LLM Based mostly Video games11:11 – commercial store.liveoverflow.com

=[ Support ]=

=[ Social ]=