NYT Cooking Tests AI Recipe Modifications: Convenience and Controversy Coexist

The New York Times is testing a new AI feature for its cooking app NYT Cooking, designed to help users modify recipes. According to an announcement last Friday by Genevieve Ko, senior editor at NYT Cooking, the feature can recommend substitutes for ingredients and cooking equipment, and scale recipes to any quantity. However, the news has sparked some controversy among users.

Feature Highlights: Smart Substitutions and Flexible Scaling

The core of this AI feature lies in addressing common pain points in cooking: when you're missing an ingredient or piece of equipment, or need to adjust recipe portions, it can provide intelligent suggestions. Specifically, the AI can recommend ingredient substitutes—for example, using other seasonings in place of a missing spice—and can also suggest alternatives for cooking equipment, such as offering viable options when you don't have a particular pot. Additionally, users can scale recipes to any quantity, and the AI will automatically adjust ingredient proportions and cooking steps.

For those who cook regularly, these features undoubtedly enhance the experience. As a cooking platform with a large subscriber base, NYT Cooking has long been known for its high-quality recipe content. Introducing AI assistance means it is attempting to increase interactivity and personalization while maintaining content professionalism.

User Controversy: Will AI Ruin the Joy of Cooking?

Although the features seem practical, the announcement "sparked some drama" among users. Some users expressed concerns about AI intervening in recipe modifications. Cooking itself is a creative activity, and many users enjoy following recipes as they are or adjusting them based on their own experience. AI recommendations may cause recipes to lose their original character or even lead to flavor deviations.

On the other hand, some users are skeptical about the accuracy of AI. Ingredient substitution is not always a simple equivalent exchange; different substitutes may affect the texture and flavor of a dish. For example, replacing white sugar with a certain sugar substitute may require adjusting the amount and cooking time. Whether AI can understand these nuances remains to be verified.

Additionally, privacy and data usage are potential concerns. Although the report did not mention specific details, users may worry that their cooking habits and preferences could be used to train AI models.

The Science and Art of Ingredient Substitution

Ingredient substitution is far from simple name replacement; it involves functional matching in food science. For example, baking soda (sodium bicarbonate) commonly used in baking needs an acidic component (such as yogurt or lemon juice) to activate and produce carbon dioxide to make dough rise. If randomly replaced with baking powder, differences in gas production speed and volume may cause the final product to collapse or become overly alkaline. Another example: when replacing butter with applesauce, because applesauce has high water content and lacks the shortening effect of fat, you need to reduce other liquids and accept a denser texture. For AI to give reliable suggestions, it must understand these chemical and physical interactions, rather than relying solely on semantic similarity.

Industry Trends: AI Is Entering Daily Life

NYT Cooking's attempt is not isolated. In recent years, AI has been gradually penetrating the food and home life sectors. From smart recipe recommendations to meal planning, various apps are exploring AI application scenarios. As a traditional media giant, the New York Times' test also reflects its digital transformation strategy—using AI to enhance user engagement and potentially provide more value for future paid subscriptions.

However, the implementation of AI in cooking still requires caution. Recipes are not just collections of ingredients and steps; they also contain cultural context and cooking techniques. When AI modifies recipes, how to balance innovation with respect for the integrity of the original recipe is a question developers need to consider.

Challenges of Applying Large Language Models in Vertical Domains

NYT Cooking's AI feature is likely built on large language models (LLMs). LLMs are good at generating fluent text, but their knowledge comes from statistical patterns in training corpora rather than explicit modeling of cooking chemistry. When a user asks to "replace white sugar with honey," the model may ignore honey's high water content, stronger sweetness, and tendency to caramelize, giving incorrect advice for an equal-quantity substitution. In addition, recipe scaling involves nonlinear changes—for example, spice amounts are not strictly linear, and cooking time needs to be adjusted based on heat transfer—which remain difficult for current models. Therefore, such features usually need to combine domain knowledge bases and human review rather than pure end-to-end generation.

Future Outlook: Testing Phase and User Feedback

Currently, the feature is still in testing, and the New York Times has not announced an official launch date. User feedback will be crucial, especially evaluations of the accuracy and practicality of AI suggestions. If user concerns can be properly addressed and the reliability of AI recommendations ensured, this feature could become a highlight of NYT Cooking.

For cooking enthusiasts, AI assistance may bring new inspiration, but ultimately, the joy of cooking still lies in hands-on practice and continuous experimentation. AI should serve as a tool, not a replacement for human creativity.