Blog
Thoughts, tutorials, and notes on things I am learning.
-
Post 2: Deeper Dive: Untangling Tasks in a Toy Transformer
This is a sister article to 'Untangling Tasks in a Toy Transformer'. This article is more technically complete and provides some of the motivations for certain architectural decisions. This serves as a deeper dive to someone more familiar with the traditional aspects of machine learning.
-
Post 1: Untangling Tasks in a Toy Transformer
We explore model thinking by training a model to do two tasks, and then intervene in the model's brain, disabling the task we choose while leaving the other untouched.