Meta Platforms is moving more quickly to substitute large language models for human content moderators as part of a twin strategy: reduce operating costs and concentrate investment on AI initiatives, according to people familiar with the situation.
The $1.4 trillion company has accelerated plans to apply large language models to the review of posts and advertisements across its services. Sources said the shift has the potential to lower labor costs by billions of dollars annually.
Within the current year Meta has already diverted about 50% of human review requests to large language models, the people said. The company intends to further curtail human involvement before the end of the year, and for certain categories of content aims to reduce human review to better than 90% automation.
Historically, Meta has combined automated tools with human reviewers, including engagement of third-party contractors, to judge whether content or ads violate company policies. When users appeal moderation decisions, those appeals have typically been processed by human reviewers.
The initiative to expand AI-driven moderation arrives alongside chief executive Mark Zuckerberg's decision to channel billions of dollars into talent and infrastructure for what he describes as "personal superintelligence." That effort is intended to support a suite of highly personalized AI products and agents for users.
The company’s move reflects a reallocation of resources toward AI capabilities while seeking operating efficiencies in areas that previously relied on substantial human review. Details on the exact deployment across all content categories and the operational implications for third-party reviewers were not disclosed by the sources.
Contextual note: The information in this article is based on accounts from people familiar with Meta's internal planning. The company’s expressed goals include both cost savings and a shift in spending toward the development of advanced AI services.