M Saiful Bari (Maruf) is a researcher at Amazon AGI, currently studying ablations and scaling laws for Nova Series Models. Previously, he worked on enhancing multilingual capabilities. Earlier before joining Amazon, he was the Training Lead
and one of the Core Maintainers
of ALLaM, a sovereign foundational model for English and Arabic language technologies. His research focuses on understanding and advancing large language models, particularly in the areas of scaling
, training dynamics
, and systematic evaluation of frontier
(-> superintelligence) models. He investigates these aspects through three primary areas: (1) scaling behaviors and training dynamics of large language models in terms of data
and truthfulness
, (2) developing robust evaluation methodologies for frontier class (-> superintelligence) models with scalable-oversight
, and (3) exploring efficient learning paradigms through transfer learning at scale
. This research combines theoretical frameworks with large-scale empirical studies, leading to both methodological innovations and practical applications.