Hacker News

Previous-Token Prediction Based LLM Inversion for Prompt Reconstruction

7 points by dalmo3 ago | 3 comments

totetsu [-]

Imagine if you could take the AI marketing slop, or political smear campaigns, texts, and Images, and see the misanthropic sludge that was used as a prompt to generate it.

ruther4d |root |parent [-]

Or what if you could take something not slopped and get back the prompt that would most closely result in it