Zuckerberg Rejects Calls for Unified Slowdown in AI Development
Meta CEO Mark Zuckerberg has dismissed appeals for a unified deceleration in artificial intelligence development, contending that each AI laboratory possesses both the obligation and the motivation to halt their progress when safety necessitates it. In a recent post on X, Zuckerberg indicated that laboratories might implement their own strategies to guarantee the safe development of models, instead of relying on a collective consensus regarding the speed of AI advancement. His comments follow those of Anthropic chief executive Dario Amodei, who advocated for the industry to “pace the frontier” of AI development and supported independent safety evaluations. Sam Altman, the chief executive of OpenAI, along with others, has expressed support for certain aspects of Amodei’s proposal.
He contended that trust and alignment, or the assurance that AI systems operate in accordance with users’ intentions, will increasingly serve as critical differentiators among AI models. “There is a lot of debate about slowing progress on capabilities until alignment catches up,” Zuckerberg said. “My view is that trust and alignment are quickly becoming the most important capabilities that will differentiate agents and models.” Zuckerberg also highlighted the commercial and legal motivations driving AI companies to enhance safety measures. Users would not prefer to engage with agents that deviated from their directives, whereas companies would encounter liability should their systems inflict damage. He stated that Meta had postponed the launch of its Muse AI model “for several months to focus on safety and security”.
The company did not request that other AI laboratories postpone their releases concurrently, he stated. “We just did it as part of our day-to-day work because it was clearly the right thing for people and for us,” Zuckerberg wrote. Zuckerberg also endorsed the utilisation of independent evaluators and advisers, noting that this approach is already a recognised practice within Meta Superintelligence Labs across various domains. “Engaging independent evaluators and advisors is industry best practice,” he said, adding that there should be a “larger and more diverse ecosystem of evaluators”. Amodei’s proposal additionally stipulates that third-party evaluators should be granted extensive access to frontier AI systems for the purpose of assessing safety measures and alignment. Altman has indicated that OpenAI would pursue a comparable strategy.
Last month, Zuckerberg stated in his essay, The Future is for Everyone, that the power and benefits of advanced AI ought to be widely distributed instead of being concentrated within a select few companies, institutions, or governments. He stated that Meta’s approach was focused on individual empowerment, innovation, and the maintenance of a balance of power. In his latest post, Zuckerberg said another way to improve safety was to direct most computing resources towards products that serve people rather than towards “racing towards recursive self-improvement”. He wrote “Meta has made this commitment and other labs can do this as well.”
