DeepMind’s Gemma Scope peers under the hood of large language models

Google DeepMind has released Gemma Scope, a series of sparse autoencoders (SAEs) that help interpret Gemma 2 models.Read More

​VentureBeat