🚨 New #EMNLP2025 paper! Do LLMs exhibit distinct behavior when the prompt looks similar to common evaluation prompts? 👀 We show that prompts that signal bias evaluation can flip the measured bias. See below ⬇️