The Next Wave - AI and The Future of Technology I Made Games, Apps & Art Using Gemini 3 — Zero Code Needed
95 snips
Nov 25, 2025 Tulsee Doshi, Head of Product for Gemini at Google, dives into the transformative capabilities of Gemini 3 in creativity and productivity. She discusses how user feedback shaped its development and reveals unexpected strengths in multilingual and multimodal applications. Tulsee emphasizes safety in AI deployment and shares exciting use cases, such as creating interactive invitations and visualizing complex concepts. Tune in to discover how Gemini 3 is making AI more accessible for creators and enriching everyday tasks!
AI Snips
Chapters
Transcript
Episode notes
Multiple Levers Created The Leap
- Gemini 3's improvements come from combined architectural pre-training advances and targeted post-training fine-tuning.
- The team iterated on feedback from Gemini 2.5 to prioritize tool use and multimodal capabilities.
Prioritize Robust Safety Testing
- Invest heavily in external safety testing and reduce sycophancy to improve model behavior.
- Balance safety work with building better user experiences and instruction following.
Next Frontiers: Rich Outputs And Long Journeys
- Future work will push richer output modalities beyond text and longer multi-turn tool-driven journeys.
- The team aims to better fuse modalities and improve persistent, complex agentic behaviors.