AI-powered
podcast player
Listen to all your favourite podcasts with AI-powered features
Exploring Residual Streams, Expert Models, and Potential Outliers in LM Space
Exploring the concept of residual streams in models, leveraging Duarkesh's analogy of a river and boat to illustrate how information flow can enhance output without heavy computational burden. Delving into expert models like Arctic to examine the impact of smaller expert models in comparison to larger ones.