AI Safety Has an Enforcement Problem
中文摘要
AI安全面临执行难题：相比对齐，缺乏可验证的权限限制是更紧迫的系统性问题。
English Summary
Beyond alignment, the most immediate AI safety concern is the lack of verifiable limits on AI authority.
Original Excerpt
Alignment matters. But intelligence without verifiable limits on authority is the more immediate systems problem. Continue reading on Medium »