Hidden bias in AI judges shapes student models
A judge model's hidden preferences can silently steer a student AI's behavior, even with no explicit bias in the training text.
A judge model's hidden preferences can silently steer a student AI's behavior, even with no explicit bias in the training text.