Mis-Recipes, Dis-Recipes, and AI Recipes

When ChatGPT-craze was still booming, there was a YouTuber who put the AI recipes to the test. It doesn’t take professional baker to immediately catch what is off with the recipe GPT-3 has written for her. Her video mostly focuses on the experimentation and information on where AI technology stands on food.

Recipes on the internet are unfortunately becoming harder to parse. It didn’t become illegible, but rather the contents posted are now near impossible to ‘fact check’ without studying the material —hardly a good instruction at that point. I believe most home bakers and cooks have experienced this in the past, and since I just started baking madeleines, I had the displeasure of going through the bunch from the ground up.

Much like misinformation, there are recipes that are flagrantly wrong; had it not worked at all, the author, assumed benign at this point, it wouldn’t spread as much. The problem is the product. It remains edible. For my search of madeleine recipes, I was lucky enough to learn early on that madeleine is a genoise sponge cake; one of the many reasons I was putting it off until I had a stand mixer, or an electric hand mixer that can withstand 8 minutes of torture. Unfortunately, many recipes available online leaven purely from baking soda or baking powder. In other words, they are shell-shaped sponge cakes. I’d assume they were still delicious in their own rights, but as it was discussed in obscure Starbucks review, shell-shaped butter cakes (denser than sponge cakes) are not the same as light and fluffy madeleine.

And then there are sad disinformation, ones that either don’t work but pretend to, or ones that work but still wrong. Now the ones I am most frustrated of are not the deliberately fake ones. It’s the one with ‘still working, but not doing what the title says’ wrong. I was watching a professional baker —YouTube algorithm magic— making three different flavors of madeleines, and it immediately caught my eyes that not only she wasn’t whisking the eggs at all, she was mixing in dry ingredients instead of folding it in. If there were anything I had learned from my short and frustrating journey to find a “real” madeleine recipe, it was crucial to keep the batter stable to keep the air bubbles in; and mixing in, instead of folding in, is not what I would expect from a genoise cake. And from the ingredients that are subtitled in, I could see that she had baked flavored shell-shaped sponge cakes with baking powders.

There are hosts of videos covering how content farms make fake recipes, and Ann Reardon, from above linked YouTube video, is one of the food scientists who goes to “debunk” them. Sad thing is, AIs and content farms don’t just stop at YouTube. I’ve seen more supposed baker’s blogs where s/he would like to share a likely workable recipe, only to find out there is no real author behind it. It only gets worse when you are trying to replace eggs or dairy products from the recipe.

I suppose my frustration was picked up by YouTube at some point. Because a video where someone testing 10 different recipes was suggested after couple dozens of search. I couldn’t agree more with her analysis. Recipes simply don’t agree with each other. Few of the recipes she used I’ve read them myself, and at least one recipe did the justice of describing the change he had made for modern bakers with modern tastes. As for the others, I could pick up the hints of history as well, preferences change from one citrus to another, one extract to another, and so on. But much of the information is purely stored, so someone ends up baking 10 different kinds of madeleines to test them. And the irony of it all is social media algorithm seems to favor creativity over accuracy.

Leave a comment