Google Expands Gemini Nano AI Access to All Android Developers and More Devices
Mountain View, California - October 2024
Google is expanding the experimental access to its advanced Gemini Nano model, part of the Gemini AI family introduced in late 2023. Initially restricted to select partners, this new access provides all Android developers with the opportunity to explore the on-device AI technology using the AI Edge SDK through the AICore app. The experimental access kicks off with the Pixel 9 series devices, with plans to extend to broader functionalities and additional devices in the near future.
Designed to function seamlessly across diverse platforms from data centres to mobile devices, the Gemini Nano model is key for various AI-driven applications. Developers can now experiment with text-to-text prompts, although it should be noted that Google’s access guidelines explicitly state it is for development purposes and not for production use.
Google touts several benefits of on-device processing with Gemini Nano. Eliminating server interactions ensures user data privacy, as data is handled directly on the device. Furthermore, it provides consistent functionality even without internet access and avoids per-inference costs often associated with cloud-based models. Despite operating on devices with limited computational power compared to cloud servers, Gemini Nano's optimisation makes it effective for tasks such as rephrasing, proofreading, generating smart replies, and summarising texts.
According to Google, the current model available to developers, "Nano 2," is nearly twice the size of its predecessor "Nano 1." The new variant offers substantial improvements in quality, excelling in both academic benchmarks and real-world applications, providing capabilities that rival much larger models.
The AI Edge SDK allows developers extensive control over the model’s performance, where they can fine-tune parameters like randomness, token limits, and response generation. Notably, several Google apps, including the Pixel Recorder and Talkback, already utilise Gemini Nano, underscoring its potential in enhancing Android user experiences and improving accessibility.
Recently, Google announced the complete rollout of Gemini Live to all Android users. As of this week, the service is fully available, enabling users to engage in spoken conversations with the AI bot through the Android app. This conversational interaction is reportedly more engaging compared to the text-based approach previously used.
Meanwhile, Google’s Gemini Nano model is also making its way to more Android devices beyond Google’s own offerings. The Motorola Edge 50 Ultra has become the latest smartphone to support Gemini Nano, marking Motorola's first foray into on-device large language models (LLM). This rollout includes an update to the AICore app, enabling the download and utilisation of the approximate 1.25GB Gemini Nano model via a new API in private preview.
Notably, Magic Compose, a feature in the Google Messages app, leverages Gemini Nano for generating suggested replies by analysing the last 20 messages in a conversation. Through on-device processing, it ensures that user privacy is maintained by keeping data local.
Currently, the Motorola Edge 50 Ultra joins other premium phones like the Google Pixel 8 and 9 series, Samsung Galaxy S24 series, and Xiaomi 14T models in supporting Gemini Nano. Google has plans for extending more functionalities across these and other upcoming devices, ensuring the sizeable download is poised to add significant value through various applications.
Developers and users alike eagerly await further features and updates that fully harness the potential of Gemini Nano, promising a more integrated and AI-enhanced Android experience.
Source: Noah Wire Services