Independent auditors and better incentives for AI oversight
The author says their estimate of AI catastrophe risk is low because they expect society to act, and calls for models that can reliably audit other AI.
TLDR
A post proposes building AI models that can reliably audit other AI, recruiting independent auditors with diverse perspectives, and giving them incentives to catch real dangers without blocking everything. It also calls for tackling cyber threats the author says are already hitting society and politics, and for universities to get involved and move faster. The author frames these as opportunities for researchers to work on AI risks and participate directly in oversight.
Combined views
21.4K
2 Sources, first seen 17d ago