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

arXiv: https://arxiv.org/abs/2509.03647 · HTML: https://mlsanity.com/paper/2509.03647/

## Key facts

- Cited by 3 papers in the map (data through 14 Oct 2026).
- Placed in Human values and AI feedback alignment, one of 10 directions in LLM alignment / RLHF / preference optimisation.
- Builds on 22 of its 25 references that are in the map.
- First posted on arXiv 3 Sep 2025.

## Placed in

- LLM alignment / RLHF / preference optimisation / [Human values and AI feedback alignment](https://mlsanity.com/alignment/human-values-and-ai-feedback-alignment/)
- Mechanistic interpretability / [Probing deception, bias and beliefs](https://mlsanity.com/mechinterp/probing-deception-bias-and-beliefs/)

## Builds on (most cited first, 22 of 22)

- [The Llama 3 Herd of Models](https://mlsanity.com/paper/2407.21783/) (2024)
- [Judging LLM-as-a-Judge with MT-Bench and Chatbot Arena](https://mlsanity.com/paper/2306.05685/) (2023)
- [Direct Preference Optimization: Your Language Model is Secretly a Reward Model](https://mlsanity.com/paper/2305.18290/) (2023)
- [A Survey on LLM-as-a-Judge](https://mlsanity.com/paper/2411.15594/) (2024)
- [Don't Give Me the Details, Just the Summary! Topic-Aware Convolutional Neural Networks for Extreme Summarization](https://mlsanity.com/paper/1808.08745/) (2018)
- [Steering Llama 2 via Contrastive Activation Addition](https://mlsanity.com/paper/2312.06681/) (2023)
- [Phi-4 Technical Report](https://mlsanity.com/paper/2412.08905/) (2024)
- [LLM Evaluators Recognize and Favor Their Own Generations](https://mlsanity.com/paper/2404.13076/) (2024)
- [Justice or Prejudice? Quantifying Biases in LLM-as-a-Judge](https://mlsanity.com/paper/2410.02736/) (2024)
- [Self-Preference Bias in LLM-as-a-Judge](https://mlsanity.com/paper/2410.21819/) (2024)
- [Personalized Steering of Large Language Models: Versatile Steering Vectors Through Bi-directional Preference Optimization](https://mlsanity.com/paper/2406.00045/) (2024)
- [VisIT-Bench: A Benchmark for Vision-Language Instruction Following Inspired by Real-World Use](https://mlsanity.com/paper/2308.06595/) (2023)
- [LLMs as Narcissistic Evaluators: When Ego Inflates Evaluation Scores](https://mlsanity.com/paper/2311.09766/) (2023)
- [CodeUltraFeedback: An LLM-as-a-Judge Dataset for Aligning Large Language Models to Coding Preferences](https://mlsanity.com/paper/2403.09032/) (2024)
- [A Unified Understanding and Evaluation of Steering Methods](https://mlsanity.com/paper/2502.02716/) (2025)
- [Do LLM Evaluators Prefer Themselves for a Reason?](https://mlsanity.com/paper/2504.03846/) (2025)
- [Beyond the Surface: Measuring Self-Preference in LLM Judgments](https://mlsanity.com/paper/2506.02592/) (2025)
- [One-shot Optimized Steering Vectors Mediate Safety-relevant Behaviors in LLMs](https://mlsanity.com/paper/2502.18862/) (2025)
- [Leveraging Uncertainty Estimation for Efficient LLM Routing](https://mlsanity.com/paper/2502.11021/) (2025)
- [Inspection and Control of Self-Generated-Text Recognition Ability in Llama3-8b-Instruct](https://mlsanity.com/paper/2410.02064/) (2024)
- [Rerouting LLM Routers](https://mlsanity.com/paper/2501.01818/) (2025)
- [Configurable Preference Tuning with Rubric-Guided Synthetic Data](https://mlsanity.com/paper/2506.11702/) (2025)

## Built on by (most cited first, 3 of 3)

- [Self-Preference Bias in Rubric-Based Evaluation of Large Language Models](https://mlsanity.com/paper/2604.06996/) (2026)
- [Are LLM Evaluators Really Narcissists? Sanity Checking Self-Preference Evaluations](https://mlsanity.com/paper/2601.22548/) (2026)
- [One Axis, No Brake: Self-Knowledge Limits the Filtering of Harmful Peer Conformity in LLMs](https://mlsanity.com/paper/2609.18998/) (2026)

## Abstract (arXiv)

Large language models (LLMs) increasingly serve as automated evaluators, yet they suffer from "self-preference bias": a tendency to favor their own outputs over those of other models. This bias undermines fairness and reliability in evaluation pipelines, particularly for tasks like preference tuning and model routing. We investigate whether lightweight steering vectors can mitigate this problem at inference time without retraining. We introduce a curated dataset that distinguishes self-preference bias into justified examples of self-preference and unjustified examples of self-preference, and we construct steering vectors using two methods: Contrastive Activation Addition (CAA) and an optimization-based approach. Our results show that steering vectors can reduce unjustified self-preference bias by up to 97%, substantially outperforming prompting and direct preference optimization baselines. Yet steering vectors are unstable on legitimate self-preference and unbiased agreement, implying self-preference spans multiple or nonlinear directions. This underscores both their promise and limits as safeguards for LLM-as-judges and motivates more robust interventions.