Keynote: Tracking the Pulse of Generative AI
summary
More than 30 model variants are easily accessible and providers follow scaling laws so cutting‑edge models get bigger and costlier to train, making post-training — fine-tuning, quantization, distillation — the real competitive edge. A healthcare survey finds over a third plan to double GenAI budgets and nearly half increase them by 50% or more; close to half already use multiple models. Concrete systems include Google’s Data Commons knowledge graph (≈240 billion points) and Metaphor’s three‑layer production graph RAG stack, while tooling to watch includes Ray, VM, MergeKit, Sycamore and llm‑blender. Risk management and AI incident plans remain surprisingly rare.
AI-written summary · about
chapters (19)
- 0:00Introduction and welcome
- 0:27Foundation models landscape
- 3:19Adoption trends in generative AI
- 5:28Challenges in AI projects
- 6:52Early usage patterns for models
- 8:20Model ensembles and cascades
- 11:17Data strategies for generative AI
- 15:34Retrieval augmented generation
- 18:00Graph RAG examples
- 22:42Generative AI agents
- 25:33Analysts and mathematicians
- 28:22AI infrastructure platforms
- 31:16Multimodality and video understanding
- 33:03Open ecosystem for models
- 34:51Hardware and software stacks
- 36:19Evaluating AI applications
- 37:24Risk management in AI
- 39:56Building generative AI apps
- 41:24Key takeaways and conclusion
related talks
17mGenerative AI: Keeping it real
11mKeynote: Building an Interoperable Ecosystem for Generative AI
01h 20mGenerative AI for Businesses
21mOpportunities of Generative AI
23mScaling GenAI Applications and Agentic AI for Business Competitive Differentiation
26mGenerative AI: Get a Grip … and a Look Under the Hood
42mFrom theory to practice: The rise of Generative AI
13mKeynote: State of Open GenAI