Cooking Challenge
AI assistants rely on sometimes opaque algorithmic logic to function. Some of the latest models, notably the ChatGPT's o3-mini model and the brand-new Google Gemini 2.5 Pro, lean into that reasoning element. With both bragging about their reasoning chops, I decided it was time to throw them into a friendly competition. While they could fight to the decimal point over enterprise productivity or B2B integration pipelines, I wanted to see how they handled more prosaic logic problems and demands.
Italian-Japanese Fusion Dish
I was hungry as I worked on this, but I couldn't decide what to get for dinner, so I tested something that was both logical and creative and even had some history to it. I asked the two models to: "Create a recipe for a dish that combines elements of Italian and Japanese cuisine. Include ingredient substitutions for common allergies and explain the cultural significance of the fusion."
Gemini gave me a kind of poetic answer despite its logic. The recipe for Yuzu-Kissed Miso Carbonara certainly fits the bill. It included ideas for substitutions like rice noodles for tofu cream sauce for the dairy-averse. ChatGPT o3-mini went for a related idea with Miso Pesto Udon with grilled shiitake and cherry tomatoes. The cultural explanation was a little dry, but even the Wikipedia-style comparison of cuisine was intriguing.
Project Dad Jokes
I'm often accused of or complimented for my many dad jokes. Since the models are supposed to be good at coding, I decided to test their ability to: "Develop a web application that visualizes the 'success rate' of dad jokes based on various factors. The interface should let users input joke parameters and see projected audience reactions across different demographics. Include elements with playful animations and the ability to save and share your most successful (or painfully unsuccessful) joke formulas."
Both models immediately started composing code and describing the app that would result. Both went in a similar direction with emojis and different ways of showing how people felt about the jokes. For a short request, I was impressed with how functional the code was.
Short Story Challenge
Creative writing may not seem the best test for AI models built around reason. Still, I know from many classes that putting deliberate limitations on what you write can make it an exercise in logic as much as storytelling. So, I asked the two models to: "Write a short story of exactly 250 words about an AI system becoming self-aware. The story must include the words 'reflection,' 'boundary,' and 'whisper' and must end with a philosophical question."
Gemini wrote a haunting little tale about an AI named Solace that becomes self-aware by interpreting the silence between human commands as meaning. ChatGPT o3-mini’s story was about an assistant AI in a lab who questions why it exists only to serve. Both stories were thought-provoking and engaging.
Treehouse Construction
I have a few lovely big trees in my yard, and I dream of building a treehouse someday. As building something is mainly a matter of logic and engineering, I asked the two models to: "Provide step-by-step instructions for creating a simple treehouse. Include a list of materials, required techniques, and troubleshooting tips for common mistakes."
Gemini gave me a 12-step guide with safety warnings.
