zlacker

What's the limiting factor for model replication by others? Amount of compute? Model architecture? Quality / Quantity of training data? Would really appreciate insights on the subjects