https://aiscience.uk/tags/evaluation/ Recent content in Evaluation on AI Science Report Hugo en-us Wed, 15 Jul 2026 00:00:00 +0800 https://aiscience.uk/posts/llm-judge-bias-mechanistic-interpretability-activation-steering/ Wed, 15 Jul 2026 00:00:00 +0800 https://aiscience.uk/posts/llm-judge-bias-mechanistic-interpretability-activation-steering/ <p>If you ask ChatGPT to rate a response that begins with &ldquo;GPT-4:&rdquo; versus the same response labeled &ldquo;GPT-2:&rdquo;, you know what happens. The score drops. Not because the content changed (it didn&rsquo;t) but because the label whispered something the model couldn&rsquo;t ignore. This is LLM-as-judge bias, and until now, it&rsquo;s been studied almost entirely from the outside: tweak the input, measure the score shift, repeat. A new paper from researchers at Alibaba, MBZUAI, USC, and Michigan asks a different question. When an LLM judge gives an unfair score, what&rsquo;s happening inside the model?</p>