The other day I did a test at the recently opened Arko restaurant in Enric Granados. I ran the menu through ChatGPT 4o and asked it to recommend the best dishes in different situations. A trick for the prompts: being polite or hinting that something serious could happen if it doesn't respond well, improves the answer.
First I asked for low-calorie dishes because, after summer, we've gone out of control, and it suggested the oyster, miso-glazed eggplant, wok squid, and exotic fruit mix. Curiously, these are the same ones it recommends for a romantic date, reminding me of M. F. K. Fisher's famous text about how light dishes are essential if, after a meal, one desires a passionate ending.
I also asked it what to order on a sad day, and it recommended the Alaskan black cod with sweet chili sauce, describing it as "a mild dish with a sweet and spicy touch that can cheer you up." Although these are not dishes I would have noticed at first, both the squid and the cod are authentic delights. When I asked why it didn't recommend sushi (one of the restaurant's strong points), it replied: "I seek to suggest a less conventional experience."
Intrigued by its recommendations, I decided to do the same at the Maymanta restaurant in the Grand Hyatt Barcelona. This time, instead of me asking, I asked it to interview me. It writes: "What kind of gastronomic experience are you looking for?", "Do you prefer new flavors or comforting options?", "Do you have any restrictions?". Based on my answers, it recommended the octopus with mole and avocado, a magnificent recipe that, moreover, coincided with the suggestions of the maître d'.
One of the most fascinating aspects of generative AI is the Theory of Mind, which involves the ability to attribute mental states to others. While Google only looks for exact matches when you ask for ten recipes with Iberian ham and you depend on someone having already published it, a large language model can devise or hallucinate ten different recipes, adapting them to the context and the needs expressed in the prompt.
Generative AI, moreover, is stochastic, offering different answers depending on the question, which inspires and provides new ideas. The gastronomic sector knows this: Coca-Cola has launched its Coca-Cola 3000, co-created with AI, and Nestlé and PepsiCo use Tastewise, which has presented TasteGPT, a tool that predicts what and how we will eat. For their part, other tools like DishGen and Smart Cuisine allow creating recipes with the ingredients one has at home.
AI is also used to enhance chefs' creativity, co-create dishes, and generate images that inspire the aesthetics of new recipes. On the peninsula, François Chartier stands out, linked to SONY AI, who has just opened a center in Barcelona. There is also the physicist Eneko Axpe, a NASA collaborator and Harvard professor, who has collaborated with chefs such as Juanlu Fernández, Nandu Jubany, and Eneko Atxa, creating a menu designed by AI.
However, AI continues to generate controversies, such as the spread of misinformation, the facilitation of cyberattacks, or, in general, the risk of endangering humanity. It is unsettling that a large part of the OpenAI team—the creator of ChatGPT—has left the company, and it remains to be seen how proposals like Anthropic, 01.AI, the recent Kling, or the supposedly more ethical Safe Superintelligence will evolve, as well as the controversial SB 1047 Law. For now, however, we can continue exploring and discovering which dishes and recipes from Barcelona restaurants ChatGPT is most amazed by.