Meta Bets Big on AI Moderators as Zuckerberg Hunts for Savings

Meta Platforms (NASDAQ: META) is moving fast to hand content moderation over to generative artificial intelligence, and the shift forms a central part of a wider cost-cutting campaign meant to offset chief executive Mark Zuckerberg’s enormous AI spending.

A Rapid Pivot to Large Language Models

The $1.4tn technology group has sped up its plans to deploy large language models that review content and advertising across its platforms. According to four people familiar with the matter, the move could save the company billions of dollars each year.

In fact, Meta has already swapped roughly 50 per cent of human review requests for LLMs this year, several of those people said. Moreover, the company wants to push that figure much higher before year-end — potentially by more than 90 per cent for certain categories of content.

Historically, Meta leaned on a blend of automated systems and human reviewers, including third-party contractors, to judge whether a post or advert broke its rules. Human reviewers also handled most user appeals. Now, however, the company is steadily dismantling that model.

Meta Frames the Shift as Performance, Not Penny-Pinching

Despite the obvious financial upside, Meta insists that better technology — not cost savings — drives the change. The company says it simply wants to use a fast-evolving tool more effectively.

To back that claim, Meta points to early results. Since March, it says, initial tests show that LLMs make 13 per cent fewer mistakes than humans when enforcing against violating content, while simultaneously catching 10 per cent more actual violations.

“The point of this work is to improve our enforcement efforts, and we’re deploying these more advanced AI systems once we’re sure they’re consistently performing better than our current methods of content enforcement,” the company added.

The Push Tracks Zuckerberg’s Superintelligence Ambitions

Notably, the moderation overhaul arrives as Zuckerberg funnels billions into the talent and infrastructure he needs to build what he calls “personal superintelligence” — a vision of hyper-personalised AI products and agents for users. For more on the buildout fueling this race, see PRISM’s AI Datacenter and Infrastructure index.

At the same time, Meta is squeezing operating costs elsewhere. The group is using AI to automate internal work such as coding, several people said. Beyond that, it has pushed through repeated rounds of layoffs and restructurings, which have hit staff morale hard.

Swapping Gemini for Meta’s Own Model

Until recently, Meta ran most of its moderation and customer support on Alphabet’s (NASDAQ: GOOGL) Gemini large language model. Lately, though, managers have instructed staff to switch over to Meta’s new foundational model, Muse Spark.

The change matters because conventional AI moderation — which typically relies on traditional machine-learning classifiers to flag rule-breaking material — often stumbles on nuance, whether that means satire or fast-shifting slang.

By contrast, Meta argues that today’s rapidly advancing LLMs, trained on vast troves of text and data, can sharpen efficiency, accuracy and response times. On top of that, the company expects them to extend reliable coverage across far more languages.

Related coverage on PRISM Marketview: Comms and Media index · PRISM Mid-Day Movers: NextGen Tech & AI