LLM bias and fairness
Measuring how language models reproduce, amplify, or mitigate social and political biases.
Computer Science Researcher · Unicamp · Kunumi Institute
Anderson Luis Bento Soares
Researcher working on the identification and mitigation of biases in language models, with emphasis on fairness, ideological behavior, socioeconomic attribution, and evaluation of large language models in the Brazilian context.
About
Academic profile
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 currently conducts research at Unicamp within the Graphs4AI project at the Kunumi Institute. His work focuses on identifying and mitigating biases in language models, especially in scenarios involving political behavior, socioeconomic attribution, stereotypes, and retrieval-augmented generation.
Research
Interests
Brazilian AI evaluation
Building and analyzing benchmarks focused on Brazilian social, educational, and political contexts.
Ideological behavior
Studying ideological sycophancy, personalization effects, and echo chamber dynamics in LLMs.
Multimodal evaluation
Evaluating small vision-language models on exam-based multimodal reasoning tasks.
Publications
Publications
LLMs are ideological chameleons: personalized echo chambers in the Brazilian political context
Scientific Reports Nature Portfolio
DOIEnem-VL: A Multimodal Benchmark of Brazilian Exams for Evaluating Small Vision-Language Models
Accepted at BRACIS 2026 Springer LNAI
How Large Language Models Profile Brazilians: Socioeconomic Attribution and Stereotypes
Accepted at BRACIS 2026 Springer LNAI
Can Retrieval-Augmented Generation Reduce Ideological Sycophancy in Large Language Models?
Accepted at LACCI 2026 IEEE Xplore
Who Gets Recommended What? Gender and Race/Color Biases in LLM Recommendations for Brazilian Undergraduate Fields
Accepted at LACCI 2026 IEEE Xplore
Online