Скачать или смотреть 51 - Humans Beat AI, But Cost 40 Times More

A recent study compared human moderators with the latest multimodal models like Gemini, GPT, and Llama to see who can truly ensure brand safety. The research comes from a team of media analysis and content safety specialists at Zefr Inc., a U.S. company that works with major advertisers to keep brands away from toxic content.

The verdict is clear: humans win. They are better at spotting borderline cases, understanding irony, reading cultural context, and telling satire from hate. Even the most advanced AI models fail where judgment matters most, letting harmful material slip through or blocking content that is not a problem.

But quality comes with a price — and a steep one: nearly 40 times higher than automated systems. For many companies, the temptation to cut costs and hand everything over to AI is huge. The problem is that savings can backfire fast. One mistake can trigger a reputational crisis that costs far more than what was saved.

The smart move is not to choose between humans and AI, but to combine them. Let the machine handle mass filtering, and leave the delicate calls to people. Because protecting a brand is not just about applying rules. It is a daily exercise in judgment. And for now, that judgment still belongs to humans.

#ArtificialDecisions #MCC #CamisaniCalzolari #MarcoCamisaniCalzolari

Marco Camisani Calzolari
marcocamisanicalzolari.com/biography