—
I’ve experimented with Generative AI in the kitchen last Saturday. Since I didn’t want to order pizza, I used ChatGPT to help me how much food and drinks I needed when I hosted for a large crowd.
Anyone who has ever tried to cook for 8+ people probably knows the struggle. Recipes are usually written for four, and manually multiplying quantities can be annoying and difficult when cooking for an uneven amount of people. So, I decided to see whether ChatGPT could help me. I gave it a pasta recipe for four people and asked it to scale it up for 17 people. It gave me a detailed shopping list with adjusted quantities of pasta, sauce and vegetables.
I was impressed since it saved me a lot of thinking time. But I noticed a big limitation: AI assumes you can just multiply everything linearly. For example, it told me to use 15 tablespoons of olive oil in one pan. That doesn’t actually work, because when cooking for large groups you don’t cook everything at once in a single pan. You need to split the recipe into batches, because doing it all it one pan won’t fit. Or at least, my pan wasn’t big enough.
I also tried the same experiment with cocktails. This worked surprisingly well: scaling up espresso martinis for 17 people gave me an accurate shopping list for coffee, vodka bottles, coffee liquor and edible coffee beans. Still, ChatGPT forgot that the drinks should be made in pitchers or batches, not in one giant shaker. Given that I had to make the food and cocktails for 17 people, I still had to figure out how big every batch needed to be and what fits in my pans and cocktail shakers.
This experiment on Saturday showed me that AI is a great starting point for meal and party planning, but human judgment is still needed when looking at ChatGPT’s generated answers.
While I believe that you article does underline a very important point in regards to our relationship to ChatGPT, namely that we cannot trust the AI blindly. However I believe you also underline the importance of people learning how to formualte high quality prompts so that ChatGPT can produce high quality output. If we do not know which parameters to determin in a prompt, Chat GPT will assume information, ultimatly leading to low quality outputs.
This is a super interesting post! I sometimes ask ChatGPT to come up with recipes for me according to what I have in my pantry, and this works surprisingly well. I agree that you can not just blindly add whatever ingredients or spices as it says, but it’s a very useful tool to brainstorm or come up with new ideas for sure! I also wonder if you upload the recipes you like already as a reference, maybe this is gonna work even better.