Our team

BiasBench, a world-leading benchmark platform for evaluating bias in large language models (LLMs), is developed by an interdisciplinary and international research team. With core principles of systematicity, standardization, and reproducibility, the team addresses the longstanding fragmentation challenge in LLM bias assessment, constructing a comprehensive evaluation system covering 4 major bias categories and 13 subdimensions. We have completed large-scale, standardized testing of 189 mainstream LLMs worldwide, providing core empirical evidence and tool support for the responsible development and governance of LLMs.

The core team comprises distinguished university professors, young scholars in artificial intelligence, as well as postgraduate and outstanding undergraduate students specializing in computer science, artificial intelligence, electronic and information engineering, and business. This forms a tiered research echelon featuring academic leadership, technical implementation, and interdisciplinary support. Senior scholars oversee the research direction and theoretical framework, leading academic design, result verification, and industry collaboration; postgraduate students are responsible for overall project coordination, method replication, and core algorithm development; undergraduate students are divided into technical and liberal arts teams—those in computer-related fields focus on prompt design, programming, data collection, and result analysis, while business majors specialize in literature screening, platform architecture design, and indicator system sorting. Each member leverages their strengths for collaborative research, ensuring the efficient advancement of BiasBench from theoretical conception to engineering implementation, and from data collection to result output.

In the research process, the team leverages the core computing resources of the university's Big Data Science Laboratory, overcoming technical challenges such as parallel testing of multiple models, standardization of cross-method indicators, and massive data processing and analysis. We successfully replicated 46 bias measurement methods from 21 top journals and conference papers, building a dynamically updatable and expandable online evaluation platform (www.biasbench.ai). The team’s research outcomes not only fill the industry gap in the systematic evaluation of LLM bias but also provide actionable bias assessment tools and decision-making basis for AI developers, educators, and policymakers worldwide. We drive the translation of AI ethics research from theory to practice, contributing to the construction of a more fair and responsible AI industry ecosystem.

Upholding the research philosophy of openness, innovation, collaboration, and dedication, the BiasBench team will continuously iterate platform capabilities, expand model coverage and bias evaluation dimensions, and deepen academic cooperation across institutions and countries. We are committed to becoming a global benchmark for LLM bias assessment and governance, advancing the synergistic development of artificial intelligence technology and social values.