From grey goo to grandma's napalm the ethics of AI and what you can't do with it
summary
Explains concrete risks and workarounds: generative models need many tries (often 7th or 8th prompt) to get usable images, attackers circumvent safeguards with "story" prompt to elicit recipes like Grandma’s napalm, and model hallucinations are non‑commutative so queries and assertions give different answers. Cites real harms: ArtStation artists withdraw work after mass scraping, automated mortgage/medical decisions have been deployed and denied people, and fly‑by‑wire human‑in‑the‑loop design is proposed as a practical mitigation.
AI-written summary · about
chapters (6)
- 0:00Introduction
- 0:58About Ben
- 6:53Ethics
- 13:30AI vs Robotics
- 18:00Extra dangerous
- 20:17Whats most dangerous
related talks
55mAI Showdown: The Good, The Glitchy, and The Risky
43mAI Engineer Ethics: Shaping Responsible Generative AI
35mAI ethics for tech writers
28mRound Table | Generative AI: Threat or Ally?
42mCode Without Caution: AI's Path to the Next Digital Disaster
45mOverview of Risks, Vulnerabilities and Paths to Safer Use of Generative AI
37mThe Shadows That Follow AI Generative Models
52mAI Is Just Software, What Could Possibly Go Wrong?