Breaking the Mirror: Activation-Based Mitigation of Self-Preference in LLM Evaluators

Open in new window