Targeted Model Removal
Richard discusses a novel approach to selectively remove concepts from a pre-trained model without compromising its overall integrity. By using a supervised training process, he explains how this method allows for quick adjustments while maintaining the model's prior knowledge. The conversation highlights the balance between efficiency and the potential need for retraining as removal requests scale up, addressing a crucial aspect of contemporary generative AI.In this clip
From this podcast
The TWIML AI Podcast (formerly This Week in Machine Learning & Artificial Intelligence)
Visual Generative AI Ecosystem Challenges with Richard Zhang - 656
Related Questions