You could have invented transformers from memory

The article argues that transformers can be understood as a natural extension of associative memory models, showing how attention mechanisms emerge from the goal of storing and retrieving key-value pairs. By building up

Read full story at hackernews →