I thought they had confirmed at least some of the image generation stuff happening locally. I am in the intelligence beta now and went offline and played around with Siri and lots of stuff worked. Not really doing much new right now, but the speed and quality of understanding and dealing with when you stumble over words are way better.