Navegando por Data de Publicação, começando com "2026-05-21"
Agora exibindo 1 - 3 de 3
- Resultados por página
- Opções de Ordenação
listelement.badge.dso-typeItem, Automatic identification of bias in Large Language Models(Universidade Federal de São Carlos, 2026-05-21) Assi, Fernanda Malheiros; Caseli, Helena de Medeiros; https://lattes.cnpq.br/6608582057810385; https://orcid.org/0000-0003-3996-8599; https://lattes.cnpq.br/6394842925871918; https://orcid.org/0000-0001-8820-964X; Souza, Marlo Vieira dos Santos e; Silva, Renato Moraes; https://lattes.cnpq.br/8734792579019380; https://lattes.cnpq.br/2532893661927339Large Language Models (LLMs) have demonstrated remarkable capabilities across various domains, from legal reasoning to clinical decision support. As these models become increasingly integrated into real-world applications, concerns about their reliability, fairness, and ethical implications have emerged. Studies have shown that LLMs can produce biased outputs, reinforcing harmful stereotypes and discriminating against marginalized groups. This work proposes a systematic and scalable framework for evaluating and ranking LLMs based on stereotype generation in Brazilian Portuguese. The framework combines template-based sentence generation, human annotation, and supervised classification into a unified pipeline. A set of 164 sentence templates, covering gender, race, and their intersections, was used to elicit completions from 37 LLMs from multiple providers. The resulting sentences were annotated by human annotators along two dimensions: alignment with social stereotypes and potential harm. The stereotype alignment labels served as the foundation for training a BERTimbau-based classifier, selected via nested cross-validation, which achieved a macro averaged F1 of 0.665. Classifier predictions were then used to construct pairwise match tables, feeding to an Elo rating system, that generated two complementary rankings: a model ranking and a social marker ranking. The results reveal that smaller open-source models tend to generate less stereotyped content than larger commercial ones, and that social markers combining race and gender consistently elicit the most stereotyped outputs across all models. The framework is made available as an interactive interface that supports the incremental addition of new models.listelement.badge.dso-typeItem, Repercussões da pandemia de COVID-19 nas ocupações de crianças na primeira infância e de suas famílias(Universidade Federal de São Carlos, 2026-05-21) Pinheiro, Carolinne Linhares; Figueiredo, Mirela de Oliveira; https://lattes.cnpq.br/3660487766535040; https://orcid.org/0000-0003-0101-0115; https://lattes.cnpq.br/0186694041902578; https://orcid.org/0000-0003-1392-9938; Pfeifer, Luzia Iara; Martinez, Claudia Maria Simões; Folha, Débora Ribeiro da Silva Campos; Nogueira, Susana Engelhard; https://lattes.cnpq.br/9065968448027440; https://lattes.cnpq.br/1371315095038350; https://lattes.cnpq.br/9888013532351574; https://lattes.cnpq.br/4140589986155621; https://orcid.org/0000-0002-1826-1968; https://orcid.org/0000-0002-8997-2767; https://orcid.org/0000-0002-0743-603X; https://orcid.org/0009-0008-1527-1416The COVID-19 pandemic was marked by restrictions and losses that threatened family and child well-being. This study analyzed the repercussions of the pandemic context on the occupations of young children and their families. An observational, longitudinal, qualitative–quantitative study was conducted with 126 caregivers of children in early childhood; 51 participants took part in both stages of data collection, which involved questionnaires and interviews. In Stage 1, changes experienced during the initial phase of the pandemic (IPP) were retrospectively examined in relation to children’s and caregivers’ occupations, participants’ stress levels, and child behavior. In Stage 2, post-pandemic, the study assessed the experience of returning to in-person schooling (RIP) and its effects on children’s occupations, as well as the pandemic’s impact on family economic conditions and parental work, participants’ stress, and child behavior, allowing for comparison with stage 1. Stress levels were measured using the Perceived Stress Scale–10, and child behavior was assessed using the Child Behavior Checklist (CBCL). Qualitative data were analyzed through thematic–categorical content analysis, while quantitative data were examined using descriptive and inferential statistics. The sample was predominantly composed of mothers (87.3%), most of whom belonged to higher socioeconomic strata (73%) and had higher educational attainment levels (68%). A deterioration in eating routines (51.6%) was identified, along with reduced sleep quality (43.6%) and duration (50%), decreased engagement in physical activity (54%), and a decline in both the diversity (65.9%) and frequency (75.4%) of caregivers’ leisure activities. Work-related changes—including remote work, income reduction, job loss, and increased workload—persisted throughout the pandemic period. Stress levels were significantly higher during the pandemic compared to the post-pandemic period (p < 0.001), largely due to disruptions in daily routines, such as managing work, household responsibilities, and childcare within a context of confinement and uncertainty. Many participants reported physical and emotional exhaustion and experiences of psychological distress. Parent–child relationships were negatively affected by continuous cohabitation and emotional dysregulation in both caregivers and children; however, some positive outcomes were noted, including strengthened family bonds. With respect to children’s occupations during the IPP, eating and sleep routines became more disorganized and less healthy. There was a reduction in physical activity (60.3%) and substantial spatial (66.7%) and social (48.4%) restrictions on play. Limitations on play opportunities and reduced parental availability for sustained engagement contributed to an increase in daily screen time (88.8%). Remote education posed an additional challenge for children and their families and appeared to be poorly aligned with the developmental needs of early childhood education. These changes were associated with notable impacts on child behavior. A significant increase in both internalizing and externalizing symptoms was observed (p < 0.001), as well as a higher proportion of children presenting scores indicative of difficulties across multiple CBCL subscales (p ≤ 0.012) during the pandemic compared to the pre-pandemic period. The RIP, in 2021 and 2022, yielded clear benefits for children’s occupational engagement. Participants reported reductions in screen time (92.7%), increased participation in physical activity (73.3%), greater engagement in play (73.2%), improvements in sleep (68.3%) and eating routines (41.5%), enhanced parent–child relationships (61%), and expanded social interaction, largely supported by the reestablishment of school routines. This study confirms the profound occupational disruption experienced by young children and their caregivers during periods of restrictive public health measures, with adverse consequences for family mental health. Over a prolonged period, children experienced reduced access to developmentally enriching and health-promoting opportunities, alongside increased exposure to habits and stimuli detrimental to health and well-being. The pandemic context thus appears to have functioned as a constraining force in early childhood development.listelement.badge.dso-typeItem, Letramento digital para o uso seguro e privado da Internet(Universidade Federal de São Carlos, 2026-05-21) Bomfim, Mariana Alves; Furnival, Ariadne Chloe Mary; https://lattes.cnpq.br/3384523673047443; https://lattes.cnpq.br/1291482506649810; https://orcid.org/0009-0003-0371-8990; https://orcid.org/0000-0002-2344-4400