Apple improves temporal awareness in egocentric video understanding

New training method incentivizes MLLMs to respect event order in egocentric video reasoning. Outperforms baselines on Ego4D across multiple temporal tasks.