CPG members Xiaoxue, Bozhidar and Matthiu were awarded the best poster prize at the Pre-ICML 2025 meetup for their work titled “The Case for Informed Adversaries When Evaluating LLM Defenses.” The poster highlights the importance of considering adversaries with realistic and informed capabilities when assessing the robustness of Large Language Model (LLM) defenses. Euodia also presented a talk on “Free Record-Level Privacy Risk Evalution Through Artifact-Based Method”. Matthieu also presented his poster on memorisation in LLMs.

PreICML 2025

Know more