Apple's PrismML: Or, How to Stop Worrying About Model Size and Start Worrying About Efficiency
Apple is reportedly developing PrismML, a technique intended to enable 27-billion-parameter AI models to run directly on iPhones. This represents a pivot in the industry from raw scale toward efficiency, portability, and on-device deployment. The source frames this as a recognition that bigger is not always better, particularly when constrained by thermal limits, battery capacity, and network latency.
This teaches you to evaluate models by task-appropriate metrics rather than parameter count alone, a habit the field has been embarrassingly slow to adopt. On-device inference eliminates network dependency and reduces latency, which matters for real-time applications; you should now ask whether your use case actually requires cloud connectivity or merely inherited it from technical limitations.
Apple is developing PrismML, as reported by Tom Kydd on 1950.ai. No public release timeline or Apple confirmation appears in the source.
Step 1: On your iPhone, open the Shortcuts app, create a new shortcut, and add the "Get Text From Input" action followed by "Show Result" to build a minimal local automation. Step 2: Download Apple's Core ML-optimized model from its developer examples or use the built-in Live Text and Visual Look Up features to observe existing on-device inference behavior. Step 3: Time the response of this local processing against the same query sent to Siri requiring server lookup, noting the latency difference to internalize why on-device execution matters.