Jacob Pfau

Let’s Think Dot by Dot: Hidden computation in transformer language models