Building AI responsibly with a GRC strategy
summary
A concrete GRC path: use model evaluation in Amazon Bedrock plus SageMaker Clarify, Model Monitor, Augmented AI, CloudWatch/CloudTrail and S3 logging to measure accuracy, safety, drift and human review; pick Claude 3 variants (Opus/Sonnet/Haiku) depending on cost and latency; Anthropic reports Opus as top-performing and a model 10x more resistant to jailbreaks, Haiku cutting DoorDash response latency ~50%, and Pfizer shrinking oncology analysis from months to days while running models inside AWS for simpler subprocessor governance.
AI-written summary · about
chapters (10)
- 0:00Introduction to Session
- 0:15Responsible AI Overview
- 1:49AI Risks and Business Impact
- 7:56AWS Responsible AI Strategy
- 12:27Responsible AI in Practice
- 20:52AWS Responsible AI Services
- 23:38Anthropic Safety Approach
- 31:52Constitutional AI Method
- 36:02AWS and Anthropic Partnership
- 36:42Conclusion and Q&A
related talks
33mSolving business challenges in Gen AI with Anthropic
38mBuilding Responsible AI in the Age of GPT
24mBuilding GenAI applications with Amazon Bedrock
47mEmbedding AI: Don’t Burn Your Fingers in the Heat of the AI revolution
29mElevating generative AI and LLM observability
48mHow to Make Smart Architecture Decisions when Building Gen AI Apps
39mBuilding Safe AI Agents
32mBuilding a GenAI-Driven Developer Organization