Responsible AI: A Hands-On Approach
summary
Demonstrates using Gemini and Vertex AI Studio to set per-category safety thresholds (negligible/low/medium/high/none/off) and warns the default is off, so safety labels disappear. Shows safety attribute scoring on prompts and responses, finish reasons that stop streaming, a Python Colab for testing, and a LangChainJS safety manager that throws clear exceptions or replaces blocked answers. Presents examples (1+1 false positive, “stubbed toe” harassment rate), grounding via Google Search, and training secondary safety models.
AI-written summary · about
chapters (9)
- 0:00Introduction and speaker background
- 3:10Responsible AI overview and risks
- 10:33Safety filters and default settings
- 15:09Testing with AI Studio
- 27:26Handling safety violations in code
- 29:58Developer tools and responsibilities
- 31:01Q&A on user protection and speech impact
- 39:33Q&A on bias and truthfulness
- 45:13Closing and call to action
related talks
30mIntroduction to AI Safety. Let's be ready before it's too late!
26mResponsible AI-What, Why, How and Future!
46mCan you trust your AI?
41mThe High Stakes Security of AI
38mBuilding Responsible AI in the Age of GPT
44mAI Foundations: Mitigate Risks and Boost SOC Efficiency
40mCode is Life - Building a Responsible Generative AI Solution
26mSponsored Session: Navigating Open Source AI with Guardrails