Navigating Data Governance in the Age of Generative AI
summary
Argues that governance must track new artifacts (prompts, embeddings, model outputs) and shows concrete harms and fixes: demonstrations of Gemini 1.5 with 1M tokens, examples of prompt injection and data extraction exploits, and research that poisons training sets for about $60; cites a customer bot handling 2.3M conversations ≈ 700 FTE to show real-world scale. Covers vector DBs, lineage snapshots, MLOps/versioning and detection strategies so teams can assess whether hands-on engineering and security work follows the claims.
AI-written summary · about
chapters (12)
- 0:00Data governance and GenAI intro
- 2:38Generative AI history and types
- 8:58How LLMs work and tokenization
- 11:25GenAI impact on workforce
- 17:24GenAI in analytics transformation
- 24:00Threats and governance risks
- 27:29Hallucinations plagiarism lineage
- 34:06Data poisoning risks
- 36:13Prompt injection vulnerabilities
- 39:24Mitigating GenAI risks
- 42:16Governance adaptations for GenAI
- 47:11Key takeaways and next steps
related talks
14mBuilding Transparent and Ethical Generative AI Models in High-Stakes Industries
26mGenerative AI: Impact of Responsible AI
28mEthical AI?!
19mMachine Intelligence With Humans in Mind
29mTraditional Governance Dead: How AI and GenAI Are Revolutionizing Governance
40mPannel Discussion | From Hype To Value: Real World Applications Of Generative AI And LLMs
17mGenerative AI: Keeping it real
45mSecuring and Governing Generative AI: Learnings from Microsoft