����Vlog

Ask Sharpen Model Overview

Ask Sharpen is built using a large language model (LLM). This LLM is a private instance of OpenAI’s GPT-4o provided via Microsoft Azure AI. This model is given context for the specified title (and only that title) using a Retrieval Augmented Generation (RAG) pattern, which indexes ����Vlog’s content. 

Data Privacy and Security

����Vlog takes matters of security and bias very seriously, and we have performed extensive testing and monitoring to ensure Ask Sharpen meets our high standards for educational use. We built Ask Sharpen to be secure in design and have guardrails in place to minimize bias and inappropriateness in responses. We are committed to ongoing enhancement of safety measures.

• Secure Treatment of Limited PII: Ask Sharpen has limited access to the user’s profile including but not necessarily limited to user name and user’s input for Ask Sharpen’s performance. User inputs are not used for model training.
  
  
• No Data Sharing: Ask Sharpen does not use any user inputs (such as prompts, uploaded notes, questions asked) or outputs (such as�� flashcards created) to train any third-party AI models. Data is processed only to provide the service.��
  
  
• Secure Data Handling: Only the input user prompts and output data relating
  to the LLM-powered flashcard and question generation components are retained. The input user prompts are securely retained for 6 months and output data for 3 years. Student uploaded materials, if applicable, are deleted after 24 hours.
  
  
• Bias, Reliability, and Appropriateness Guardrails:
  
  
  • Ask Sharpen follows ����Vlog’s content guidelines and uses multiple layers of safeguards to help facilitate output that is appropriate and inclusive.��
    
    
  • Guardrails are applied both before and after content is generated, including input filtering and an “LLM-as-a-judge” check that reviews output against safety standards. Content that fails is automatically blocked or replaced.
    
    
  • Ask Sharpen primarily draws from Sharpen’s trusted content library and student-uploaded materials where applicable. In some cases, responses may be generated using LLMs, based on public sources that are not owned or verified by ����Vlog. Responses generated using public sources are clearly labeled.
    ����
  • Ask Sharpen’s AI generated output is regularly reviewed and monitored by ����Vlog’s AI team, content team, and subject matter experts to ensure it provides high-quality, educationally appropriate support. Our teams evaluate responses using a robust set of criteria including accuracy, relevance, appropriateness, depth, completeness, and clarity and conciseness.�� ��

Continuous Improvement

����Vlog’s team is dedicated to the continuous improvement of our products, including Ask Sharpen. To better serve learners, ����Vlog teams regularly review and test the system and generated output to improve accuracy, minimize risk, and promote a safe student experience. Our team reserves the right to make changes to the underlying model as needed. This ongoing process ensures that our products including Ask Sharpen improve as a reliable and effective tool for education.

Instructor and Institutional Choice��

While AI technologies, like Ask Sharpen, present many exciting opportunities within education, ����Vlog wants to ensure the choice to use these technologies remains firmly in the hands of institutions. Institutions or instructors who want to discuss offering Ask Sharpen to their students should reach out to their Account Managers with any questions.��������������������������������������������������������