Previous-Token Prediction Based LLM Inversion for Prompt Reconstruction (arxiv.org)
9 pointsby dalmo3an hour ago3 comments
totetsu 36 minutes ago
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 34 minutes ago
Or what if you could take something not slopped and get back the prompt that would most closely result in it
totetsu 29 minutes ago