TechSpot means tech analysis and advice you can trust. Read our ethics statement.
The big picture: The most complex and capable generative AI models require the computational power of remote servers and cloud platforms to process and respond to users' prompts. However, with the advent of new silicon technology incorporating embedded AI accelerators, less sophisticated versions of these models can still offer some intriguing features even in the absence of an internet connection.
Google has unveiled Gemini, its multimodal AI system designed to usher in a new era for the company's business. Gemini is described as Google's most capable AI model to date, with the flexibility to operate on remote servers for the most complex tasks (Gemini Ultra), scale across a broader range of tasks (Gemini Pro), and even function on devices without a cloud connection requirement (Gemini Nano).
Gemini Nano, Google's most efficient next-gen AI model, is now integrated into the recently launched Pixel 8 Pro line of premium smartphones. The model leverages the phone's Tensor G3 system-on-chip unit, which is equipped with a new-generation accelerator unit (NPU) for enhanced AI-related computing tasks.
The first two AI features coming to Pixel 8 Pro are Summarize in the Recorder app, and Smart Reply in Gboard, The Recorder app, with the assistance of Gemini Nano, has the capability to extract summaries from recorded conversations, interviews, presentations, or other audio feeds. With Smart Reply, users will receive suggestions for the next message to send in a conversation.
Summarize will be available on all Pixel 8 Pro phones starting with the Android December update. The second Gemini Nano feature, meanwhile, will remain in developer preview for the time being.
According to Google, Gemini Nano will provide "high-quality" response suggestions, thanks to the enhanced "conversational awareness" delivered by the on-device AI model. Initially, this feature will be compatible with WhatsApp only, but more apps are expected to adopt the AI-powered suggestion system in 2024.
Designed to run on low-power systems such as Android smartphones and tablets, Nano aims to assist users in keeping their sensitive data secure. For more complex AI-related tasks, such as uncanny image generation or internet-wide searches, Android enthusiasts will also have the option to use the larger, cloud-focused Gemini model via Bard. Additionally, in 2024, Bard will be integrated into the Assistant app on Pixel devices.