Abstract:
According to Bloomberg, Meta CEO Mark Zuckerberg said on Tuesday that AI labs should rely on independent evaluators and consultants to ensure the safety of models. Zuckerberg on Tuesday joined a growing number of tech executives weighing in on a global debate on how best to prevent AI systems from harming humanity.
Zuckerberg
"Introducing independent evaluators and consultants is an industry best practice." Zuckerberg wrote in a post on the X platform. He noted that, in general, having a larger, more diverse ecosystem of evaluators helps.
Zuckerberg said that there is currently a lot of debate about whether the development of AI capabilities should be slowed down and waiting for alignment technology to catch up. His point is that trust and alignment are quickly becoming the most important capabilities that differentiate agents and models. Any laboratory that does not take alignment seriously will fall behind. Alignment refers to aligning the way AI does things with human interests.
Zuckerberg's post
At present, some company executives have expressed support for slowing down the development of AI and advocate cooperation to reach consensus. Zuckerberg wrote that Meta decided to slow down the development of the Muse AI tool on its own without first asking its peers to review their progress.
"Meta delayed the release of Muse for several months to focus on safety and security," he wrote. "We didn't ask everyone else to do it first and then follow up ourselves. We just did it as part of our daily routine because it was obviously the right thing for our users and for us."
Zuckerberg believes that devoting the vast majority of computing power to serving people, rather than racing to promote recursive self-improvement of AI, is one of the best ways to ensure the safe development of this technology. Meta has made this commitment, and other labs can do the same.
Last weekend, a 3,800-word article by Anthropic CEO Dario Amodei further heated up the heated debate surrounding the economic and security risks of AI. In it, he called for slowing down the development of state-of-the-art systems so researchers can better understand potential threats. Amodei’s call received rare support from OpenAI CEO Sam Altman and SpaceX CEO Elon Musk.
Comments