
How I Learned to Stop Worrying and Love the Transformer
Ashish Vaswani, co-author of the original Transformer paper "Attention is All You Need" and co-founder of Adept AI Labs, returns to Stanford's CS25 Transformers United seminar to trace the model's origins and evolution. He walks through the motivations that led to the Transformer, replacing recurrence with attention, and surveys how the architecture has changed since its 2017 debut, from scaling practices to adaptations across language and other domains. He closes with a discussion of open research directions he sees as promising for the field. The talk is pitched at students already familiar with deep learning basics, giving the inventor's own perspective on why the design choices were made and where the field might go next. Runtime is about 81 minutes, recorded as part of Stanford's CS25 course.