What's Scarier Than Agents Taking over Internet? CEO Cartel Trying Take over AI
4 hours ago
- Dario Amodei proposes slowing AI capability growth due to serious risks, including embedded evaluators, democratic coordination, and global coordination.
- The author, a 'geezer,' objects to slowing AI development, especially if it delays curing major diseases, and sees proposals as anti-competitive.
- Embedded evaluators in AI labs are criticized as a costly burden for startups, benefiting incumbents like Anthropic and OpenAI.
- Waiving antitrust laws for AI coordination is seen as enabling a cartel, harming competition, and potentially slowing innovation for new entrants.
- Global coordination is deemed unverifiable and unrealistic, with risks of defection (e.g., by China, Russia, North Korea), leaving the US behind.
- The author suspects tech titans push these ideas to address financial pressures (e.g., high costs, low profitability) and to fend off regulation and competition from open-source and Chinese AI.
- The article argues AI is a neutral tool that can be used for good or evil, and slowing development is like defunding the police due to bad actors.
- Instead of pacing, the author suggests creating a White Hat Institute (WHI) to use AI for defense, and outlawing ransom payments to hackers to reduce cyber threats.
- The author concludes that slowing AI through cartelization serves only the big tech companies, but hurts society and the country.