FT商學院

Meta plans big bet on AI’s secret ingredient: human brains

Decent data for training AI models is becoming scarce and people are needed to make more of it

Dystopian thriller “Soylent Green” ends with the hero discovering that the mystery food that sustains New York’s citizens, and whose name the movie bears, is actually made of human beings. Reports that Meta Platforms may acquire 49 per cent of artificial intelligence company Scale AI for $15bn illustrate a similar plot twist. AI, too, is people.

So far, building AI has mostly been an exercise in managing scarce resources of the inanimate kind. Top-end chips, for example, are in short supply; power constraints weigh. Most of Meta’s capital expenditure, estimated at up to $72bn this year, has been earmarked for servers and data centres, the tech equivalent of bricks and mortar.

But other ingredients are becoming scarce too. One is decent data, essential for training AI models effectively. OpenAI co-founder Ilya Sutskever has warned of reaching “peak data” in the same way the planet is heading towards “peak oil”.

您已閱讀31%(921字),剩餘69%(2057字)包含更多重要資訊,訂閱以繼續探索完整內容,並享受更多專屬服務。
版權聲明:本文版權歸FT中文網所有,未經允許任何單位或個人不得轉載,複製或以任何其他方式使用本文全部或部分,侵權必究。
設置字型大小×
最小
較小
默認
較大
最大
分享×