New York City, NY – As a dedicated home cook, coming up with new and innovative dishes for my family every night can be a challenge. In a quest for inspiration, I decided to test the capabilities of AI models to see if they could assist me in creating unique recipes. To truly put these AI models to the test, I enlisted the help of Tom’s Guide U.S. Editor-in-Chief Mike Prospero to capture a random selection of ingredients in a photograph, mystery box style.
The ingredients included pizza dough, canned peaches, anchovies, hot sauce, and fresh mozzarella. While my immediate instinct was to create a pizza with a hot sauce base and mozzarella topping, I deferred to AI to choose our menu for the evening. Utilizing Google Gemini, OpenAI ChatGPT, and Anthropic Claude for the experiment, each equipped with vision models, we set out to see how well they could identify ingredients.
The first task for the AI chatbots was to recognize the ingredients in the photograph. While both Claude and Gemini identified the word “ingredients,” ChatGPT described them as “food items” and excelled in recognizing the ingredients accurately. However, Gemini missed one ingredient and mislabeled another, while Claude invented a fictitious sixth ingredient not present in the picture.
Moving on to the recipe creation phase, I asked each AI to come up with three recipes using the given ingredients to test their prompt following ability and creativity. Each AI was tasked with creating dishes that incorporated all the ingredients in unique ways.
ChatGPT developed a gourmet peach and anchovy stuffed bread, utilizing additional ingredients like honey, balsamic vinegar, ricotta cheese, and garlic. On the other hand, Gemini created a Galette with cornmeal, eggs, butter, chives, lemon, vanilla extract, cinnamon, and sugar. Lastly, Claude prepared a stuffed mozzarella bread with sun-dried tomatoes, olives, rosemary, sesame seeds, and an egg.
While the dishes created by the AI chatbots may not be something I would personally eat, visually, they appeared more appealing when visualized by MidJourney. Overall, Claude’s recipes were deemed more favorable, although ChatGPT had a better understanding of the image and ingredients, showcasing the strengths of vision AI technology.
In exploring the intersection of cooking and artificial intelligence, this experiment shed light on the potential of AI to assist home cooks in creating unique and innovative dishes. Through collaborations with AI models, the culinary world is evolving to embrace technology in the kitchen, offering new possibilities for creativity and inspiration.