Opinion
Mr. Mallaby is the author of “The Infinity Machine: Demis Hassabis, DeepMind, and the Quest for Superintelligence.”
TL;DR
- Dario Amodei, CEO of Anthropic, is advocating for a cautious approach to AI development, especially regarding recursive self-improvement.
- Amodei fears that AI agents could "take over the entire internet" within 6-12 months without sufficient safety measures.
- He proposes "pacing" AI capability growth to allow safety advancements to catch up.
- Anthropic will grant permanent access badges and system permissions to outside experts to monitor safety practices.
- Rival AI executives, including Sam Altman (OpenAI), Elon Musk, and Demis Hassabis, have commended Amodei's stance, with Altman vowing to embed evaluators in OpenAI.