video1.0<iframe src="https://www.loom.com/embed/bf914ef4961d43c9800014d43fe03b85" frameborder="0" width="1152" height="864" webkitallowfullscreen mozallowfullscreen allowfullscreen></iframe>8641152Loomhttps://www.loom.com8641152https://cdn.loom.com/sessions/thumbnails/bf914ef4961d43c9800014d43fe03b85-ff9da7490cc49376.gif300.458In this video, I present my project for the AI Safety Ethics & Society Spring 2025 cohort (https://www.aisafetybook.com/virtual-course), focusing on reducing medical hallucinations using Sparse Autoencoders & Chain of Thought Prompting. I highlight the critical issue that one in five medical AI answers can be wrong, which can have serious consequences for patients. My proposed solution is simple and accessible, requiring no technical skills, and aims to improve the accuracy of AI in healthcare. I also share my progress, including a significant improvement in recall scores from 63% to 93%. Please take a look and let me know your thoughts!