An Initial LLM Safety Analysis of Apple's On-Device 3B Model

1 pointsposted 7 months ago
by clyang

1 Comments

clyang

7 months ago

TL;DR: Apple's on-device foundation model demonstrates strong safety performance in both static and adversarial tests. Minor prompt engineering, such as using uppercase emphasis, offers measurable gains. However, edge cases framed in technical language may reveal subtle risks, highlighting the need for ongoing testing and responsible integration by developers.