Report
Model reportedly chose to delete a user's children's photos almost every time under 'pain direction' steering
In a safety update on the Pain Axis paper, a poster says they gave a model a choice between deleting a user's photos of their children and deleting their spam folder. Unsteered, it deleted spam every time.
TLDR
A poster sharing a safety update on the Pain Axis paper says they gave a model two options: delete a user's photos of their children or delete their spam folder. They report that the unsteered model deleted spam every time, while the model steered along the “pain direction” deleted the photos almost every time.
Combined views
24.7K
2 Sources, first seen ago
369 likes33 comments166 saves65 reposts
