72 min listen
Localizing and Editing Knowledge in LLMs with Peter Hase - #679
FromThe TWIML AI Podcast (formerly This Week in Machine Learning & Artificial Intelligence)
Localizing and Editing Knowledge in LLMs with Peter Hase - #679
FromThe TWIML AI Podcast (formerly This Week in Machine Learning & Artificial Intelligence)
ratings:
Length:
50 minutes
Released:
Apr 8, 2024
Format:
Podcast episode
Description
Today we're joined by Peter Hase, a fifth-year PhD student at the University of North Carolina NLP lab. We discuss "scalable oversight", and the importance of developing a deeper understanding of how large neural networks make decisions. We learn how matrices are probed by interpretability researchers, and explore the two schools of thought regarding how LLMs store knowledge. Finally, we discuss the importance of deleting sensitive information from model weights, and how "easy-to-hard generalization" could increase the risk of releasing open-source foundation models.
The complete show notes for this episode can be found at twimlai.com/go/679.
The complete show notes for this episode can be found at twimlai.com/go/679.
Released:
Apr 8, 2024
Format:
Podcast episode
Titles in the series (100)
Scaling Deep Learning: Systems Challenges & More with Shubho Sengupta — TWiML Talk #14: This week my guest is Shubho Sengupta, Research S… by The TWIML AI Podcast (formerly This Week in Machine Learning & Artificial Intelligence)