Jev-like models as a possible route to on-device AI for phones
One post explores whether Jev-like models could support fast, cheap AI decisions for phone notifications, typing, inboxes and calendars.
TLDR
The post asks whether Jev-like models could accelerate on-device AI for phones. It argues that strong large language models remain a long way from running on phones, citing memory bandwidth, model size, power and heat constraints. It says next-generation mobile AI processors are targeting modestly sized language models. It also raises another possible approach: encoding an older but useful model directly into phone hardware.
Combined views
7.8K
1 Source, first seen 4h ago
Jev-like models as a possible route to on-device AI for phones
One post explores whether Jev-like models could support fast, cheap AI decisions for phone notifications, typing, inboxes and calendars.
TLDR
The post asks whether Jev-like models could accelerate on-device AI for phones. It argues that strong large language models remain a long way from running on phones, citing memory bandwidth, model size, power and heat constraints. It says next-generation mobile AI processors are targeting modestly sized language models. It also raises another possible approach: encoding an older but useful model directly into phone hardware.