Keynote: What's New in AI
summary
Overview of concrete advances: Transformers (2017) and attention enable long-context models with tens of billions of parameters trained on terabyte-scale corpora using GPUs and Google TPUs (launched 2015); reinforcement learning with human feedback improves judgment; DeepMind’s AlphaFold delivers 200 million predicted protein structures; productized models include T5, LaMDA, PaLM and Gemini plus Gemini Nano optimized to run on-device for features like encrypted reply suggestions; tooling and platforms listed are Vertex AI, Kaggle, Duet AI, Bard and Caris 3.0 enabling multi-framework large-scale training and deployment.
AI-written summary · about
chapters (12)
- 0:00AI innovations and human needs
- 1:03Large language models and impact
- 2:08Traditional programming to neural networks
- 3:09How LLMs learn and scale
- 4:54Computational power and data advances
- 6:16Techniques for model performance
- 7:44No-code AI access
- 8:48Device-based AI optimization
- 9:32Responsible AI applications
- 9:52Google’s AI products
- 10:54Introducing Gemini
- 11:35AI’s future and human impact
related talks
12mKeynote: What's New in AI
24mKeynote: Next-generation frameworks for Large-scale AI
20mWhat's new in Machine Learning Keynote
43mBeyond the AI Hype: What's Real, What's Next
14mAI for Science
27mKeynote - Google Cloud, Machine Learning
01h 01mKeynote: Transformers: The Rise of ChatGPT
12mKeynote: AI Infrastructure