AI Risks and Sensible Regulation

Combining AI and Human Work πŸ€– β€” with Marco Trombetti

Topic 7 of 9

Published Feb 20, 2025 Β· Clip 45:20–55:32 (10:12)

Marco raises concerns about 'personal alignment': companies optimizing models to tell users what they want to hear, enabling super persuasion β€” a capability possible with today's technology. He points to closing data sources like X and Reddit deals with OpenAI as fuel for persuasion-optimized models. He argues for regulating AI that initiates conversations with humans, bans on autonomous weapons (like biochemical weapon bans), and clear limits on government surveillance β€” noting current regulation like the AI Act governs citizens' tools while governments exempt themselves. He criticizes the current legislation as designed for models of five years ago, but insists regulation should guarantee fair competition and protect from misuse: "we don't want to say you cannot use a knife, we want to say you cannot kill people" β€” regulate the outcome, not the tool.