Reward Function Tuning

Definition

The idea that a predictive model must have its objective deliberately tuned toward the desired outcome, or it will optimize a trivial proxy (e.g., predicting that rare events never happen, or finding only replacement-level players).

Cast within

No topic-region cast yet — this appears once Reward Function Tuning's compiled claims are aligned into a topic region's argument tree.

Claims mentioning Reward Function Tuning (1)

Your notes

Loading notes...