Apple AI Updates: April 14, 2026

1. Apple Publishes “Cram Less to Fit More” Training Data Pruning Research

Apple. Apple researchers published a paper accepted at the ICLR 2026 Workshop on Navigating and Addressing Data Problems for Foundation Models, formalizing fact memorization from an information-theoretic perspective and developing data selection methods that prune training data based on training loss metrics to make frequency distributions more uniform. A GPT2-Small model (110M parameters) trained with the technique retained 1.3x more entity facts than standard training, matching models 10x its size. Source