@misc{oswald2023uncovering, title={Uncovering mesa-optimization algorithms in Transformers}, author={Johannes von Oswald and Eyvind Niklasson and Maximilian Schlegel and Seijin Kobayashi and Nicolas Zucchet and Nino Scherrer and Nolan Miller and Mark Sandler and Blaise Agüera y Arcas and Max Vladymyrov and Razvan Pascanu and João Sacramento}, year={2023}, eprint={2309.05858}, archivePrefix={arXiv}, primaryClass={cs.LG}, url={https://arxiv.org/abs/2309.05858} }