Meta: AI Rapidly Replaces Human Moderation

Meta: AI Rapidly Replaces Human Moderation

Meta has already replaced about 50% of all human moderation requests with large language models by 2025 and plans to increase this share to over 90% for certain types of content by the end of the year. This shift is expected to save the company billions each year, according to the Financial Times. Meta disputes the cost argument and instead emphasizes quality, claiming that since March, tests have shown that its language models make 13% fewer errors than humans when enforcing content policies while detecting 10% more actual violations. Unlike traditional ML classifiers that struggle with satire or evolving language, language models are said to better grasp nuances and cover more languages.

Employees describe a different situation. An insider indicates that the models continue to remove or restrict harmless content, and that there is not enough oversight for such a rapid deployment. The transition is already leading to layoffs, particularly among external contractors.

There is also a model shift happening behind the scenes, reports the Financial Times. Meta previously used Gemini from Google for moderation and support but has recently instructed its staff to switch to its own new foundational model called Muse Spark. The models are trained on past decisions made by human reviewers.