A few insights from Apple’s ML team.

A few insights (https://machinelearning.apple.com/papers/apple_intelligence_foundation_language_models.pdf) from Apple’s ML team.
The input data includes various sensors and breathing and audio, apparently. There’s a local little model (still unclear how good it is, the benches are ok). I think this will make it to production next year and only on the last 2 generations of devices (because memory, as I understand it).