Google AI Overview
My Understanding
Google’s AI ecosystem is primarily centered around the Gemini family of models, which replaced previous models like PaLM and Bard. They range from massive data-center models to small, on-device models. Google also provides specialized tools for developers and researchers to build on top of these models.
AI Draft
The Gemini Family
- Gemini Ultra: The largest and most capable model for highly complex tasks.
- Gemini Pro: An optimized model for scaling across a wide range of tasks (Powering Gemini Advanced).
- Gemini Flash: A lightweight, low-latency model designed for speed and efficiency.
- Gemini Nano: The most efficient model built for on-device tasks (running locally on phones like Pixel or S24).
Other Notable Products
- Gemma: A family of lightweight, state-of-the-art open models built from the same research and technology used to create the Gemini models.
- Vertex AI: Google Cloud’s enterprise platform for training and deploying ML models.
- Google AI Studio: A fast, web-based tool for developers to prototype with Gemini models using API keys.
- NotebookLM: An AI-powered research and writing assistant that uses your own documents as context.
Key Points
- Multimodality: Gemini was built from the ground up to be multimodal (understanding text, images, video, and audio).
- On-Device AI: Gemini Nano is a “banana-sized” (ultra-small) model that allows AI features to run without an internet connection, preserving privacy.
- Integration: Deeply integrated into Google Workspace (Docs, Gmail, Sheets).