In the poster session, Hadi Mohammadi from Utrecht University explained his poster titled “Enhancing Human-Likeness of Generative AI through Explainable Techniques”. His research focused on using Explainable AI (XAI) methods, such as SHAP, to identify and replace influential tokens in AI-generated text, enhancing the naturalness and coherence of the content.

Qixiang Fang shared his work titled “PATCH! Psychometrics-Assisted Benchmarking of Large Language Models: Evaluating Proficiency in 8th Grade Mathematics“, introducing a novel approach for evaluating large language models (LLMs). This method applies psychometric techniques to assess LLMs’ performance using benchmarks like MMLU (Massive Multitask Language Understanding).

For more details, contact Hadi Mohammadi at h.mohammadi@uu.nl or Qixiang Fang at q.fang@uu.nl.