AI Model's Inner Workings Broken into Understandable Features

Original: Towards Monosemanticity Decomposing Language Models With Dictionary Learning

Source: anthropic.com ↗