An open source approach to benchmark social biases in both proprietary and open source LLMs
2024-10-10, 15:00–15:30 (Europe/Luxembourg), C1.05.02

LIST presents a public leaderboard based on open source software that comprehensively assesses and benchmarks Large Language Models (LLMs) according to a set of seven social biases such as Ageism, LGBTIQ+phobia, Political bias, Racism, Religious bias, Sexism and Xenophobia. The initiative aims to raise awareness about the implicit social bias embedded in LLMs, and foster advances in trustworthy AI and the alignment to recent regulations in order to guardrail the societal impacts of both proprietary and open source AI.

R&T Engineer at Luxembourg Institute of Science and Technology (LIST)