08:47
2026-07-08
tej.as
large-language-models
Flagship models are not always better
A developer building a realtime note-taking app found that Apple's on-device Foundation Models failed to process a 27-minute conversation, while a smaller Qwen model succeeded using a hierarchical map…