This essay was written with Barath Raghavan, and originally appeared in The Guardian . Major benchmarks measure what AI can do. None measure whether it does what you mean: the distance between what you ask an AI to do and the unspoken assumptions about how you want the AI to do it.

Lire l’article original →