Building Responsible AI in the Age of GPT
summary
Claims that responsible AI requires integrating six risk tenants into every development stage and that early checks save cost compared with late retraining (cites the ~85% McKinsey stat about projects not reaching production). Recommends model selection from providers like OpenAI Anthropic or Cohere, red teaming, measuring truthfulness, and documenting choices. Shows a long system prompt used to guide Bing Chat (GPT‑3.5/GPT‑4), builds guardrails like input/output filters and human-in-the-loop review (as in Adobe) to reduce hallucinations.
AI-written summary · about
chapters (6)
- 0:00Introduction
- 1:05Responsible AI in LLM era
- 8:34Responsible AI in development lifecycle
- 18:42Designing responsible AI systems
- 29:33Q&A on responsible AI challenges
- 37:58Closing remarks
related talks
34mResponsible AI: The Key to Ethical, Safe and Inclusive Software Development
26mResponsible AI-What, Why, How and Future!
48mPractical Strategies for Successful AI Adoption in your Organisation
40mSocially responsible AI: Practical tools for robust design
35mResponsible AI: Ethics in Software Development
35mAI ethics for tech writers
39mBuilding a Responsible AI Governance Program
06m7 practical tips for safer AI and language models