Bias and fairness in language models
Measuring how language models reproduce, amplify, or mitigate social, political, and demographic biases across diverse representation prompts.
M.Sc. student in Computer Science · Institute of Computing, Unicamp
Anderson Luis Bento Soares is a Master's student at the Institute of Computing, University of Campinas (Unicamp), affiliated with the Laboratório de Informática Visual (LIV). My research concerns fairness, ideological sycophancy, and the identification and mitigation of bias in large language models, with an emphasis on Brazilian educational and societal benchmarks.
Anderson Luis Bento Soares is a Master's student in Computer Science at the Institute of Computing, University of Campinas (Unicamp), and holds a Bachelor's degree in Computer Science from the Federal University of Ceará (UFC).
He conducts research at the Laboratório de Informática Visual (LIV) at Unicamp. His work focuses on identifying and mitigating biases in language models, particularly in scenarios involving political behavior, socioeconomic attribution, and stereotypes. He also works with knowledge graphs and retrieval-augmented generation as tools for grounding and evaluating model outputs.
Measuring how language models reproduce, amplify, or mitigate social, political, and demographic biases across diverse representation prompts.
Building and analyzing benchmarks grounded in Brazilian social, educational (ENEM, FUVEST, Unicamp), and cultural settings.
Studying political personalization, echo chamber dynamics, and sycophantic responses in language models, along with mitigation through retrieval-augmented generation.
Evaluating small vision-language models on exam-based multimodal reasoning tasks and structured document understanding.
Loading publications…
Institute of Computing, Universidade Estadual de Campinas (Unicamp)
Research on knowledge graphs, retrieval-augmented generation, large language models, and the evaluation of ideological bias in artificial intelligence systems, together with the development of research projects and academic publications.
Universidade Federal do Ceará (UFC), Quixadá
Fellow in a program combining research, teaching, and outreach with an emphasis on cooperative learning. Co-organized outreach initiatives such as InfoGirl and GameNight, and served as UFC's state representative in the INTERPET movement, which promotes integration among PET groups in Ceará.