A Recipe for Disaster? ChatGPT Has A Mission To Create Unique, Delicious Dishes And Fail Miserably
So you think your new ability to inform ChatGPT of AI-generated recipes can result in culinary masterpieces?
Hold that thought, he advises World of Vegan, a popular health website dedicated to the vegan lifestyle. The site recently launched an interesting experiment powered by AI for production, where it enabled ChatGPT to compile more than 100 recipes. The team inspired the AI bot to create new and innovative recipes for various occasions from late night dishes to brunch and dessert ideas. From there, the site’s team of chefs tested each recipe to see how they tasted.
The result? It’s not good.
All of this lobbying led to what the group described as “incredibly sad results.” With many recipes, the team of chefs at World of Vegan saw the composition of the ingredient “which would immediately clash and where a mistake would happen.” The team also noticed that the recipes could be “deceptive,” appearing familiar at first but often described as “rich” and “bad” when they were very different.
“I had a feeling ChatGPT would be difficult with recipe development, as recipe making is a delicate mix of fine art and science,” World of Vegan founder and chef Michelle Cehn told Spoon. “But I was shocked at how hard it was to find a single spring recipe written by ChatGPT that works with passing. This is an important warning to both food bloggers looking for shortcuts. again home cooks looking for quick recipes. You’ll save yourself a lot of trouble (and wasted time, energy, and money) by bypassing ChatGPT and opting for a high-quality blogger recipe instead.”
Another big fail cooked up by the World of Vegan team was the vegan scalloped potato dish (pictured above), a recipe the chef said had a list of ingredients and cooking instructions that didn’t work. The resulting dish had a bad color, a spicy sauce, and a bad taste.
According to World of Vegan, of the 100 or so recipes the team cooked up, only one – the cauliflower taco dish – produced an interesting result.
Cehn believes that the 1% success rate may be due to ChatGPT’s reliance on what data is actually flawed, i.e. millions of cheap recipes taken from the internet. With this as its basis, things are destined to go wrong when a bot is tasked with making a unique recipe.
“The human brain cannot access all that information, so people may independently (and unintentionally) create duplicate recipes online. As ChatGPT has to create truly unique recipe, it has to be a little complicated to create one that is not copied.”
While one might expect a site that focuses on creating recipes to be skeptical about AI filling its shoes, I have no doubt that the negative results are far from what others might find if they did the same experiment. Good recipes often come from a lot of experimentation and experience, something you don’t get when a bot freewheels up a new dish idea out of thin air.
And while a highly specialized AI trained on the compatibility of various cooking ingredients — something like a chatbot based on Chef Watson — might produce better results, we don’t have that, at least not yet.
Bottom line: Human sponsored recipe creators are still needed…for now.
