
Ari Morcos is co-founder and CEO of DatologyAI, building a self-service data curation platform for AI teams. Prior to founding Datology, Ari spent five years at FAIR (Meta AI), most recently as a Senior Staff Research Scientist, where his research on data curation and self-supervised learning received Outstanding Paper Awards at NeurIPS 2022 ("Beyond neural scaling laws: beating power law scaling via data pruning") and ICLR 2023. Before Meta, he was a Research Scientist at DeepMind, applying tools from neuroscience to understand generalization, representation learning, and the dynamics of training in deep networks. He holds a PhD in Neuroscience from Harvard and a BS in Neuroscience from UC San Diego.
Using publicly available information we constructed an analysis to help you get a feel for this speaker before deciding to attend their session.
Morcos makes the empirically-grounded case that 'models are what they eat' - that smarter data curation (filtering, rebalancing, curriculum sequencing, synthetic data) beats brute-force scaling of models or compute, citing results like up to 7.7x faster training and a 3B model matching an 8B baseline. Attend for a clear, data-first mental model of where pretraining is heading, drawn from his FAIR/DeepMind research background and DatologyAI's trillion-token production pipeline.