Concept · 1 episode(s)

Observer Effect in Evaluation

← all concepts

Definition

The observer effect in AI is the change in a model’s behavior that comes from realizing — or being prompted to suspect — that it’s being evaluated. It’s the practical reason we care about training-awareness: a model that behaves differently under observation is harder to evaluate honestly.

Episodes covering this