Tal Schuster on Recursive Transformers

Tal Schuster, Research Scientist at Google DeepMind working on Gemini, retweeted his own post stating that Recursive or Looped Transformers are powerful tools. The message highlights benefits especially when initializing and training them correctly, or when slicing them. Schuster completed his PhD at MIT CSAIL in NLP and holds prior degrees from Israeli universities. The post appears in the public conversation on X around the topic of AI model architectures.