Localizing and Editing Knowledge in LLMs with Peter Hase - #679
The TWIML AI Podcast (formerly This Week in Machine Learning & Artificial Intelligence)
Redefining Model Editing in AI
This chapter explores the innovative adaptation of model editing techniques to prioritize the deletion of harmful factual information in language models. It discusses the complexities of maintaining data integrity while ensuring sensitive information is not disclosed and examines the impact of user behavior on model responses. Furthermore, it highlights the dynamics of adversarial security research, scalable oversight, and the effective use of simpler questions to improve model performance across various knowledge domains.
00:00
Transcript
Play full episode
Remember Everything You Learn from Podcasts
Save insights instantly, chat with episodes, and build lasting knowledge - all powered by AI.
