Post #2449496
2026-05-05 17:05 UTC
Replies (8)
-
@bens@mastodon.xyz 2026-05-05 17:07
@paul@tapbots.social this would be great and plays to their privacy-first strength.
-
@hoboshrimps@mastodon.social 2026-05-05 17:09
@paul@tapbots.social I agree with the local model approach, but I don’t understand why Apple would want to develop something much more computationally expensive that could only be used by the top percentile of their user base for no additional cost. The only way I could see them doing what you’re describing is if they rolled it into iCloud plus or something.
-
@TambourineMan@mastodon.social 2026-05-05 17:14
@paul@tapbots.social if only Apple sold a Mac where you could fit lots of RAM. #MacPro #NeverForget
-
@_holger@mastodon.social 2026-05-05 17:17
@paul@tapbots.social if they do it, they would probably train them on Swift & SwiftUI. No ObjC.
-
@ottost@mastodon.social 2026-05-05 17:37
@paul@tapbots.social I hope the new CEO come with a better roadmap.
-
@km321@mastodon.online 2026-05-05 17:59
@paul@tapbots.social Don’t give up on small models for edge devices. There have been break throughs in the last couple of months that allow much larger models to run in the same amount of RAM. I’ve been surprised at what my 16 GB iPad can do now. When will Apple update their built in model to take advantage? Who knows? I’m hoping their deal for Gemini might also include on device Gemma models.
-
@Patryk@mas.to 2026-05-05 18:49
@paul@tapbots.social Alibaba's Qwen models prove we are getting close (the 27b variant is exceptional at mundane computer tasks) and Apple is probably the best company to make it work well, with proper safety guardrails, sandboxing and privacy. I really hope they are looking closely at what Claude Code/OpenClaw/Hermes Agent are doing, this is their best opportunity at making Siri useful.
-
@rooster@mastodon.social 2026-05-05 21:43
@paul@tapbots.social and make it so your iPhone can use the model on your Mac from anywhere