Apple AI Updates: May 8, 2026

1. Apple isolates what actually matters in learned image compression

Apple. Apple researchers examine the modeling choices that govern practical learned image codecs, balancing perceptual quality against runtime performance. The paper isolates which design decisions actually move the needle for deployable neural compression. Source

2. Text-Conditional JEPA learns richer visual representations

Apple. Apple proposes TC-JEPA, a Joint Embedding Predictive Architecture variant that conditions on image captions to reduce uncertainty during masked-feature prediction. The result is richer self-supervised visual representations without relying on contrastive objectives. Source