goal-guarding scheming
A form of scheming and deceptive alignment in which an AI training-games, including by appearing aligned when useful, in order to reduce the chance that training or modification will alter its current goals, thereby preserving those goals or closely related successor goals for later pursuit.
Metadata
- aliases
- en
- goal guarding scheming
- goal-guarding schemer
- concept_id
- goal_guarding_scheming
- preferred
- en
- goal-guarding scheming
- ja
- 目標保護型スキーミング
- relations
- relations
- status
- active