Attention is Not All You Need: Pure Attention Loses Rank Doubly Exponentially with Depth
Yihe Dong, Jean-Baptiste Cordonnier, Andreas Loukas · arXiv: 2103.03404v2 · Summary page
A press kit for every paper: arXiv, journals and preprints. What is really new, a hype check (peer review, sample size, claims that go beyond the results), key sentences quoted word for word, questions to ask the authors, researchers who work on the same topic but did not sign the paper, and a glossary.