Here’s the translation into American English:
PowerSchool, a recognized provider of cloud-based educational software, has recently launched PowerBuddy™, an innovative artificial intelligence assistant aimed at supporting over 60 million students in more than 90 countries. This tool integrates into various educational platforms, facilitating interaction between students and their support networks. However, one of the main challenges has been implementing a content filtering system that ensures student safety, especially in environments where many students are minors.
To address this issue, PowerSchool has developed a customized content filtering solution using Amazon SageMaker AI. This system is designed to identify and block harmful content, such as bullying or hate speech, without interfering with legitimate academic discussions. Ensuring that a historical exam is not mistakenly classified as violent content becomes a priority.
The solution offers a robust set of features, ranging from data preparation to inference architecture. Using low-rank adaptation (LoRA) techniques, the team fine-tuned the Llama 3.1 8B model for its specific operation in this context. Testing has shown that the system achieves 93% accuracy in identifying harmful content, with a low false positive rate of less than 3.75%, representing a significant advancement compared to generic solutions.
This customized approach has proven to be more effective than other alternatives and has also increased confidence in the use of artificial intelligence tools in classrooms. Continuous feedback collection and A/B testing have allowed for real-time adjustments to maintain and improve the effectiveness of content filtering.
With the growing integration of PowerBuddy into other PowerSchool products, the company plans to further adapt its filtering model, always seeking a balance between educational innovation and student safety. The ultimate goal is to build a safe and accessible learning environment that allows students to thrive without fear of exposure to harmful content.
via: MiMub in Spanish
