jsecretan / maven-public

6 stars 0 forks source link

Prompt injection vulnerability in interests #23

Open gdmagana opened 4 months ago

gdmagana commented 4 months ago

The AI descriptions of interests have several issues. One is in the image attached, where users can interface directly with the LLM using very simple jailbreak techniques. I have also seen the opening page of The Great Gatsby pasted in as an interest, where the AI description just tries to describe the first sentence. it was a UX nightmare with how large the title was, took up a full screen.

IMG_2403