… Processing AI requests on-device is free, whereas using a data center model can be incredibly costly. Apple is reportedly using a version of Google& 039;s large Gemini model to train a smaller version that could be capable of running on-device. …
… Overall, Apple& 039;s researchers found that the SafetyParis approach is "effective at highlighting weaknesses in state-of-the-art vision-language models, and can serve as a useful data augmentation strategy for training sample-efficient guard models." The testing approach could be used to enhance … …
… An on-device AI that stays out of the way Most users will likely interact with Apple& 039;s AI systems and Siri using these base models and nothing else. It will be the default, and after the Gemini training, will likely be more than enough for the features Apple will reveal during WWDC. …
… Apple can ask for high-quality results and the Gemini model& 039;s "chain of thought," which can be used to train its own small model. The end result is the creation of a smaller model that performs a specific function just like the larger model. …
… It is not that Apple has replaced Siri with Gemini, nor that Google will now get to train its service on users' data. Instead, Apple is licensing a Google Gemini model to help improve its own Apple Foundation Models. …
… Some rumors pointed to a spring 2026 release of a more contextual Siri powered by the Gemini-trained Apple Foundation Models. Of course, that could still occur at any time since Apple Intelligence updates occur independently from OS updates. …
… While some rumors pointed to the spring for a release of the new Apple Foundation Models trained by Gemini, there still aren& 039;t any signs that it is coming. Apple could release the new AI at any time via a backend update, but that seems unlikely due to the proximity of WWDC. …
… The new models enable users to ask questions about objects they are looking at through their visionOS passthrough video feeds. Other AI features announced for iOS 27 like the improved Spotlight search and indexing are also coming to Apple Vision Pro. …
… Really, the training should be the other way around. In comparison to the rest of the AI industry, Siri has to catch up to a massive degree. …
… As for what SFI-Bench tests specifically, the study explains this in a fairly straightforward manner: "Beyond spatial cognition, SFI-Bench incorporates functional and knowledge-grounded reasoning, probing whether models understand what objects in the scene are for, how they are operated, and how fa… …