Sherry Marcus Ph.D., AWS | theCUBE + NYSE Wired: Media Week - Cyber & AI Innovators Summit
summary
AWS describes concrete Bedrock work: distillation to turn large models into smaller, cheaper runtimes; new Model Evaluation and Knowledge Basis Evaluation tools to benchmark many models on customer examples; Nova family that claims ~75% lower inference cost with similar accuracy; Trainium/Inferentia-based training and "hundreds of thousands" GPU-scale training capacity; connectors, knowledge graphs and RAG for harmonizing data; intelligent prompt routing that routes ~40% of simple queries to small models for cost and latency wins.
AI-written summary · about
chapters (15)
- 0:00Introduction and welcome
- 0:27AWS Bedrock and Applied Sciences
- 1:11Science behind Bedrock capabilities
- 2:34Model optionality and trade-offs
- 4:42Model interactions and distillation
- 6:10Scaling laws and future of AI
- 8:41Why Amazon builds Nova models
- 10:11Cost efficiency and Graviton analogy
- 11:39Accuracy and guardrails in AI
- 13:08Amazon’s training infrastructure
- 14:38Agents and data harmonization
- 16:02Agentic AI and prompt optimization
- 18:31Promise of agentic AI
- 20:21Amazon’s internal AI applications
- 21:22Closing remarks and thanks
