Apple is in talks with PrismML about tech that can shrink massive AI models enough to run fully on iPhone. They’ve already compressed a 27B-parameter model (Qwen 3.6) from ~54 GB down to under 4 GB (roughly 15× less memory) while keeping all parameters active. Why it matters: This kind of compression could enable significantly smarter on-device Siri and Apple Intelligence features that work offline, keep data private, and run more efficiently. Do you think this potential upgrade can finally justify iPhone’s price?(Lark Davis)