Baking With AI: Is It Possible?

When baking, if you don’t have a trusted go-to recipe it’s hard to know where to turn without trying out multiple recipes and judging the results. An enjoyable task, sure, but with the emergence of artificial intelligence for the masses (in this case ChatGPT) I wondered if AI could generate the perfect recipe and save us some time.

So in the interest of science, I baked an AI-generated chocolate yule log over the Christmas holidays to answer the question: How close are we to baking perfection? Read on, or jump to the recipe.

Getting the question right

Generating anything useful from AI is all about the prompts, i.e. the questions you ask the all-knowing machine. I started off asking it:

Please tell me how to make a delicious traditional yule log cake.

The recipe it produced used cups, but as any good baker will tell you, measurements should be by weight, not by volume, and I prefer grams so I tried again:

Using metric measurements, please tell me how to make a delicious traditional yule log cake.

The resulting recipe was the same but with conversions into grams, however some things were still in cups and some of the metric conversions were wrong. For example, 1¼ cups of granulated sugar is about 250g, not 150g. Time to try a different approach:

Using European measurements, please tell me how to make a delicious traditional yule log cake.

Aha, now the ingredients were all metric, although interestingly it generated a different recipe to before. Anyway, it was something I could bake with confidence so it’s time to preheat the oven and put on the apron!

Putting theory into practice

On the face of it, the recipe seemed OK. No weird ingredients and everything that would normally go into a cake, but a couple of things stood out:

• That’s a lot of flour! I’ll do as I’m instructed, but I’m already having doubts.
• The “log” is covered in cream icing with a bark-like texture. I don’t think AI understands that trees are generally brown.

Mixing and baking the sponge went fine but sadly the result was not great. The excessive amount of flour meant that although it did rise a bit, the sponge was generally rubbery and impossible to bend without cracking.

As you can see in the photo, the final roll was more like a series of folds than an elegant curve. And it was tough to eat too. The cream-based filling and frosting was lovely but it couldn’t save the rest of the cake. Time to try again but this time we’ll make an adjustment. Based on human-generated recipes I looked at, using around 30% of the suggested flour amount should be better, and everything else could stay the same. On to round 2!

The revenge bake

Success! Just one simple change and the resulting sponge was soft and fluffy, making it easy to roll (while still hot) and it held its shape nicely until it was cool enough to add the filling. Even the white “tree” and brown “snow”, although unnatural, looked pretty and tasted good. Not too rich and a sweetness I was happy with.

The verdict

It’s kind of a win, to my surprise. Although the first bake was a failure, fixing the single mistake of too much flour resulted in a recipe that I’ll probably come back to in a year’s time – or sooner. There’s nothing I’d change other than perhaps getting more fancy with the decoration. Maybe AI can help with some suggestions…

---

Here’s the full recipe, including my correction for the flour. For my experiment I baked the cake using half the amounts listed.

Ingredients

• 200g 60g all-purpose flour
• 40g unsweetened cocoa powder
• 1 tsp baking powder
• 1/4 tsp salt

• 4 eggs
• 100g granulated sugar
• 1 tsp vanilla extract
• 1 tbsp hot water

• Confectioners’ sugar, for dusting

Filling and frosting

• 250ml heavy cream
• 50g confectioners’ sugar
• 1 tsp vanilla extract
• Chocolate shavings or cocoa powder, for garnish (optional)

Instructions

1. Preheat your oven to 180°C (350°F) and line a 33cm x 23cm (13in x 9in) baking sheet with parchment paper.
2. In a large mixing bowl, whisk together the flour, cocoa powder, baking powder, and salt.
3. In a separate bowl, beat the eggs and sugar together until light and fluffy. Add the vanilla extract and hot water, and mix until well combined.
4. Gradually add the dry ingredients to the egg mixture and mix until just combined.
5. Pour the batter onto the prepared baking sheet and spread it evenly with a spatula.
6. Bake the cake for 10-12 minutes, or until a toothpick inserted into the center comes out clean.
7. While the cake is still warm, dust it lightly with confectioners’ sugar. Then, carefully roll the cake up into a tight cylinder, using the parchment paper to help you roll it. Set aside to cool completely.
8. In a medium mixing bowl, beat the heavy cream, confectioners’ sugar, and vanilla extract until stiff peaks form.
9. Unroll the cooled cake and spread the whipped cream evenly over the surface. Roll the cake back up into a tight cylinder.
10. Transfer the yule log cake to a serving platter and frost the outside with the remaining whipped cream. Use a fork or a potato masher to create a bark-like texture on the surface of the cake.
11. Garnish the yule log cake with chocolate shavings or cocoa powder, if desired. Serve and enjoy!

---

Addendum: I’ve noticed that since I ran this experiment, ChatGPT has been updated and is now showing a very different yule log recipe for the prompt I used. Is it better? There’s only one way to find out!