Don't Worry About the Vase Podcast AI #138 Part 2: Watch Out For Documents
Oct 17, 2025
The discussion navigates the intriguing implications of the GAIN Act and the geopolitical stakes of chip exports. California's child-safety AI bills spark debate on technology lobbying against stricter regulations. Public sentiment strongly favors holding AI companies accountable for damages. The potential for document poisoning in AI reveals alarming vulnerabilities, while broader issues of alignment and misalignment in AI systems are also explored. The talk emphasizes the psychological strain on AI professionals, urging a balance between innovation and mental health.
AI Snips
Chapters
Transcript
Episode notes
Don't Grant Blanket Liability To AI
- Do not give the public exactly what it wants on liability for AI; blanket strict liability would cripple useful systems.
- Instead, enable suits for negligence and clear failures to meet reasonable standards to preserve helpful AI while allowing redress.
Favor Lightweight Federal Transparency
- Support targeted, federal transparency requirements rather than broad state-by-state bans or moratoria.
- Use modest, well-scoped rules (e.g., labels, thresholds) to generate evidence and avoid reactive overreach.
AI Water Use Is Not The Big Issue
- AI's water usage is trivial compared with other industries and won't be a meaningful constraint for now.
- Even large AI growth only modestly increases water demand relative to agriculture and power sectors.