Developers can now incorporate Gemini Nano, a locally-running smartphone AI, into their own apps
Google has announced that developers can now incorporate the Gemini Nano, a smartphone model of its AI Gemini family, into their apps. This allows each app to use AI locally without sending user data to an external source.
Android Developers Blog: Gemini Nano is now available on Android via experimental access
Android Developers Blog: An introduction to privacy and safety for Gemini Nano
https://android-developers.googleblog.com/2024/10/introduction-to-privacy-and-safety-gemini-nano.html
Experimental access to Gemini Nano will be open from October 1, 2024, and all Android developers with the AI Edge SDK will be able to use Gemini Nano. Initially, it will be possible to experiment with text input prompts on the Pixel 9 series.
This processing happens directly on the device without calling a server, which offers many benefits, including sensitive user data remaining on-device, full functionality without an internet connection, and no additional cost for each inference.
According to Google, on-device generative AI models like the Gemini Nano run on less computing power than a cloud server, so they're better at handling specific tasks rather than 'jack-of-all-trades' like chatbots, such as rephrasing specific text to make it more casual, generating responses to messages, and structuring and summarizing sentences.
Gemini Nano has already been incorporated into Google's Pixel screenshot and screen reading functions, as well as recorder apps, and is said to be useful for image and voice understanding using multimodal functions.
'Integrating generative AI models directly into mobile apps is difficult because it requires significant computational resources and storage capacity. To address this challenge, we developed AICore, a new system service for Android. AICore lets you benefit from AI running directly on your device, without having to distribute your own runtime, models, or other components,' Google said.
According to Google, AICore complies with the Private Compute Core (PCC) environment, which keeps data private, and can only interact with limited system packages that comply with PCC, and cannot directly access the Internet. This not only allows AI to be used within the app without communicating with the outside world, but also reduces the risk of data being exposed to other apps, and makes it clear that no records of input data or output are stored.
Related Posts:
