Fetching the paper…

Transformers Implement Functional Gradient Descent to Learn Non-Linear Functions In Context · Around