Overview
This podcast episode features a deep dive into recent Gemini AI app updates, focusing on new features like Gemini Canvas, Deep Research enhancements, and advanced personalization capabilities. The hosts discuss user experiences, technical advances, and demo several key functionalities.
Gemini Canvas: Creation and Collaboration
- Gemini Canvas enables users to collaboratively create documents and web apps with AI assistance, supporting direct in-app editing.
- The UI allows users to modify specific document sections, similar to Google Docs, both on web and mobile.
- Canvas supports code generation and web app building, making it possible for non-coders to develop interactive experiences.
- Users can preview, publish, and share web apps developed through Canvas; shared apps do not require the recipient to have a Gemini account.
- The platform is expanding to support increasingly complex, sandboxed, and eventually API-driven web apps.
Deep Research: Enhanced Research Capabilities
- Deep Research leverages the Gemini 2.0 thinking model to automate long, agentic research tasks with improved reasoning and synthesis.
- The process includes a pre-planning step where users can customize the research outline before the AI begins.
- Research results provide inline citations at paragraph and sentence levels, improving transparency and credibility.
- Research output is editable in a Canvas-like interface, enabling further refinement.
- Audio overviews allow users to consume research content on-the-go across devices.
Personalization Features
- Users can opt-in to connect their Google Search history for a more personalized AI experience.
- Gemini remembers user preferences and prior conversations, reducing repetitive prompts and enhancing continuity.
- The AI uses selective context from user data to improve recommendations while ensuring user control over data access and utilization.
- Transparent citations are provided for personalized content, and users can review or disable data-driven personalization at any time.
Technical and User Experience Insights
- Transition from 1.5 Pro to 2.0 thinking model reduced the need for supervised fine-tuning and complex prompt engineering.
- Personalization features are launched experimentally, with a focus on user feedback, privacy, and control.
- Audio overviews are aligned with similar features in NotebookLM, providing a cohesive cross-app experience.
- The system aims to balance helpful personalization with minimal intrusiveness.
Decisions
- Launch Deep Research to all users for free
- Launch personalization features experimentally and gather user feedback before full rollout
Action Items
- TBD – Gemini App Team: Continue developing Canvas to support more complex and API-integrated apps.
- TBD – Gemini App Team: Gather and analyze user feedback on personalization features before broader deployment.
- TBD – Gemini App Team: Maintain transparency and user control around data use and personalization.
Questions / Follow-Ups
- Will a Deep Research API be released for developers?
- How will the personalization engine evolve with additional data sources across Google's ecosystem?
- What additional transparency features will be implemented as personalization expands?