Model Editing Insights
Peter discusses the traditional approach to machine learning as a black box, emphasizing the use of fine-tuning to update factual information in models without retraining from scratch. He highlights the challenges of maintaining accurate knowledge, such as updating sports player affiliations, and addresses the need for methods to delete outdated information, particularly in visual models. The conversation sheds light on innovative strategies for efficient model editing in AI.In this clip
From this podcast
The TWIML AI Podcast (formerly This Week in Machine Learning & Artificial Intelligence)
Localizing and Editing Knowledge in LLMs with Peter Hase - 679
Related Questions