logoalt Hacker News

Previous-Token Prediction Based LLM Inversion for Prompt Reconstruction

8 points • by dalmo3 • today at 1:27 AM • 3 comments • view on HN

Comments

totetsu • today at 2:00 AM

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.

➕ show 1 reply