Bounded AI Self-Improvement: A Recursive Loop You Can Supervise
中文摘要
有界AI自我改进可被监督，规避了代理改变自身目标的危险。这是一个允许人类监督的递归循环。
English Summary
Bounded AI self-improvement is supervisable, mitigating dangers of agents changing their own goals. It's a recursive loop allowing human oversight.
Original Excerpt
What this article answers (plain language): Self-improving AI is dangerous when the agent can move the goalposts — author its own success… Continue reading on Medium »