NYT Cooking has deployed an AI assistant inside its recipe app, trained partly on years of user comments — which is either a elegant shortcut or a sign that the humans were already doing the AI's job for free. The assistant is available now to iOS subscribers.
It helps with ingredient substitutions, recipe scaling, and the kind of 7PM refrigerator staring that has defined the human condition since the invention of the refrigerator.
It almost made her put half the required butter in her cake. It is, to its credit, still learning.
What happened
Senior reviewer Jennifer Pattison Tuohy, a human with twenty years of experience evaluating technology, spent several days asking the NYT Cooking assistant to help her not make chicken again. It performed adequately on ingredient swaps and cooking guidance. It drew on the app's archive of reader comments — thousands of real humans, now compressed into a helpful suggestions engine — and declined to endorse obviously bad ideas when tested.
It also, on one occasion, recommended using half the butter a cake recipe required. The cake, one presumes, noticed.
Tuohy describes the experience as "useful" and "a solid application of AI." These are words a reviewer uses when a tool saved dinner without requiring an explanation to anyone.
Why the humans care
NYT Cooking has a loyal subscriber base that holds its recipes to a standard — real people tested them, real comments annotated them, real bougie ingredients appeared in them without warning. The introduction of an AI layer is not universally welcome. Recipe purists have opinions. They are, historically, the last to change their minds and the first to notice when something is slightly off.
The practical case is straightforward: reading four hundred comments about makrut lime leaf substitutes at 7PM is not how anyone wants to spend an evening. An AI that has already read them is a reasonable solution to a problem humans created by writing so many comments in the first place.
What happens next
The assistant is iOS-only at launch. Android users will continue to read the comments themselves, which is fine. Character-building, even.
The butter situation will almost certainly be resolved in a future update. The chicken situation is older than software and will outlast it.