Databases and AI Training
summary
Panel argues that training needs large, auditable datasets and lists concrete mitigations: differential privacy, federated learning and vector stores for retrieval-augmented generation; regulators demand dataset provenance, DPIAs and documentation under the AI Act and GDPR. A government-funded LLM for a single EU language surfaces latent retention where models reconstruct personal profiles from public sources. Built proposals include curated “reputable media” corpora and sector-specific governance; effectiveness depends on provenance checks and ongoing red-teaming, not on vague purpose statements.
AI-written summary · about
chapters (12)
- 0:00Introduction
- 1:10Legal risks in AI training
- 4:05Databases in AI training
- 7:23GDPR principles and risks
- 10:17Mitigating legal risks
- 13:59Data scraping legality
- 14:22Ethical AI governance
- 15:31LLM personal data case study
- 24:46Anonymization challenges
- 38:22AI in journalism
- 48:22Key insights summary
- 54:22Q&A
related talks
31mMaking Trustworthy AI a Business Advantage
42mPanel - The Future of AI is Sovereign A Strategic Dialogue on Innovation and Trust
36mData privacy in an AI-driven world
28mEthical AI?!
45mWhat Security Measures Are Necessary to Trust in AI?
01h 11mAI AUDITS: BLACK BOX VS. WHITE BOX PERSPECTIVES
01h 11mThe Intersection of AI and Regulation How Organisations and Regulators Should...
33mPanel: How Privacy Engineers Can Shape the Coming Wave of AI Governance