New preprint Data Heterogeneity Limits the Scaling Effect of Pretraining Neural Data Transformers is up on bioRxiv.