News

Mar 23, 2023

Artificial Intelligence, Real Flavor

I t started Erev Pesach with a bottle of olive oil. “What are we having for

It started Erev Pesach with a bottle of olive oil.

"What are we having for dessert?" I muttered to myself as I looked at my laptop. My last-days-of-Pesach menu stared back at me, every meal planned with dips, soup, main, and sides. My husband and I had agreed to host last-minute guests for meals, and I wanted to serve something more elevated than the cut fruit I had originally planned.

Something. But what? I left my laptop and ducked into the kitchen, looking for ingredient inspiration. I found cocoa powder, almond flour, and sugar — not a bad start.

Maybe brownies, I thought, and turned to hunt for some vegetable oil so I could start baking. Only there was no vegetable oil. I had run out over Chol Hamoed. All I had left was a nearly empty bottle of extra-virgin olive oil. You know, the kind for salad dressings and savory sautés and definitely not brownies?

I sighed and went back to my desk, resigned to the fact that we would just have fruit for dessert. I didn't have time to either shop for new ingredients or trawl through the Internet for a Pesach dessert recipe involving olive oil. I decided to take my mind off the disappointment by knocking a few things off my to-do list for work and opened my project-planning app.

My first task was to draft an email informing a group of students about an upcoming exam. I opened a new tab and navigated to ChatGPT, the artificial intelligence chatbot that I had been using to write messages for the past several weeks in order to boost my productivity. I prompted the bot with my request, and immediately, a professionally worded email draft started to appear. As I watched the text grow on the screen, an idea began to grow in my mind.

I started a new chat with the prompt: "Generate a dessert recipe based on almond flour that includes olive oil. The recipe must be gluten-free, dairy-free, and suitable for Passover." Within seconds, I had a recipe. Within minutes, I had dessert in the oven. And within a few hours, I was serving a delicious Pesachdig orange-olive oil cake to my guests, to rave reviews.

Out of curiosity, I asked ChatGPT how it generated the cake recipe. I received this response:

As a language model, I was able to generate that recipe by utilizing my natural language processing capabilities and knowledge of cooking.

First, I searched for a recipe that included almond flour and olive oil, and then I narrowed my search to recipes that are dairy-free, gluten-free, and suitable for Passover. I then used my knowledge of baking and cooking techniques to create a recipe that included the appropriate measurements, ingredients, and baking instructions.

While my recipe may not have been tested in a kitchen, it was generated using ingredients and techniques that are commonly used in almond flour baking recipes. I hope you find the recipe useful and enjoy making it!

After this initial success, I was curious to learn whether ChatGPT could consistently produce reliable and delicious recipes — or whether I had experienced beginner's luck. My goal was to use ChatGPT to generate a full Shabbos dinner menu, including challah, two different dips, a soup, a salad, one main dish, two sides, and a dessert.

To start, I hopped onto ChatGPT and prompted it to create a "Shabbos dinner menu that adhered to the laws of kashrus" and contained the courses I had decided upon. The bot spat out a stereotypical mix of dishes — gefilte fish and chopped liver, matzah ball soup, brisket, and babka. Ever the considerate chat bot, it even added a note at the end of the menu that I should make sure to consult my local "rabbi or other knowledgeable authority regarding the specific dietary requirements for [my] community." It seems that while ChatGPT has a narrow understanding of what can be served at a Shabbos seudah, it does know about asking an LOR.

I then requested a menu that was more unique and flavorful, and was immediately rewarded with a menu filled with warm Middle Eastern spices that included baba ghanoush, Moroccan lentil soup, and orange semolina cake. There was just one issue — the main was chicken shawarma sandwiches, which I’d happily serve for a weekday meal, but not as a main for Shabbos.

I decided to keep experimenting with different prompts, which taught me a few things about how ChatGPT currently functions when generating recipes. First, I found that if I made multiple menu requests within the same chat, the bot would "forget" about the kosher requirements and give me recipes involving dairy and bacon. I had to "remind" the bot multiple times to generate recipes that were dairy-free and suitable for a kosher meat meal.

I also found that the way I worded my prompts greatly impacted the results. Writing "Shabbos" or "Friday night dinner" in the prompt would invariably return menus with recognizably Jewish food, such as kugels for side dishes. Including "dairy-free" or "pareve" in the prompt would give me results from the vegan world, including vegan apple crisp, chocolate-date truffles, and cookie recipes without dairy and eggs. Asking for full menus usually resulted in simpler recipes, like simple roasted chicken, while asking for individual recipes with specific descriptors (like unique, avant-garde, easy) generated more exciting results.

After prompting ChatGPT to produce several menus and not finding one that hit all the notes I was hoping for, I decided to prompt the bot for each course individually.

Challah: I’ve always wanted to try a challah recipe that uses honey instead of sugar, but never had time to find a recipe that looks promising. I typed in the prompt "Generate a recipe for one loaf of challah that uses honey as the only sweetener." I got a recipe quickly. When I tried it, I found that the dough came together easily but was too sticky to knead. I ended up adding about a quarter cup of additional flour to the dough — but after that, it was smooth sailing, and the loaf I baked was fluffy and sweet, with a golden crumb.

Dips: I prompted ChatGPT with "Generate recipes for two flavorful dairy-free dips" and received recipes for guacamole and chummus. Not wrong, but not exactly dishes I need recipes for. I followed up with "Please give me two different recipes with more complex flavors," and was rewarded with a roasted red pepper eggplant dip and a white bean dip with roasted garlic. Both were delicious and came together easily, though the bot did forget to include parchment paper or oil on the roasting pan for the eggplant and pepper. It's clear that the human cook needs to use their common sense when using artificial intelligence for recipe inspiration!

Soup: When I asked for a full menu, most of the soups suggested by ChatGPT were fairly basic — chicken noodle, lentil, butternut squash. I decided to get more specific and prompted the bot to "Give me a recipe for a unique, dairy-free soup that can be made in advance and freezes well." In return, I got a recipe for Coconut Curry Butternut Squash Soup — and while I was initially nervous about the combination of spices (cinnamon and curry?), I was pleasantly surprised at the balance of flavors, especially when paired with the richness of the coconut milk. My toddler ate this up, so it's a win in my book!

Salad: Like the soup, ChatGPT initially gave me very basic salads, like mixed greens with vinaigrette. It also had a hard time sticking to pareve recipes and kept sneaking in goat cheese and feta when I asked for something more exciting. I decided to change my strategy and ask for a salad with a "unique and flavorful dressing," and the bot created a recipe for Thai Mango Salad. Now this was a winner! Everyone at our table inhaled it.

Main: I had a London broil in my freezer, and the cardinal rule of London broil is to keep preparation simple. I asked ChatGPT for an "easy marinade" for the meat and it generated a recipe that included all the usual suspects from recipes I’ve used in the past (soy sauce, balsamic, Worcestershire, etc.). I tried it out as written, and it was great! The sweetness from the balsamic and brown sugar mingled with the umami of the soy sauce to produce a crust that reminded me of good beef jerky.

Sides: This one was a bit of a struggle, since ChatGPT tended to provide ultra-simple recipes (think roasted potatoes and garlic green beans) unless given very specific prompts. Below are the prompts and the results.

"Generate a recipe for classic potato kugel" — I got a fairly standard-looking recipe that baked up fine. I’m not sure why the flour and baking powder were needed, but no complaints here.

"Give me a recipe for a unique kugel" — ChatGPT produced a recipe for a sweet potato- apple kugel and it was…interesting. It included a full onion and quite a bit of spice. I decided to make the recipe as written and hope for the best. I wouldn't make it again as is, but might try again with a quarter of an onion (or none at all) and less spice.

"I have a bag of frozen cauliflower. Generate an avant-garde recipe using it." This one was cool! ChatGPT wrote a recipe for cauliflower- quinoa pilaf with roasted balsamic grapes. After making the recipe as written, though, my husband and I agreed that while the quinoa was great and the grapes were delicious, the spices in the pilaf didn't quite match the balsamic grape vibe. I would absolutely make this again, but either leave out the grapes or change the flavor profile of the pilaf to a more Italian theme so that the roasted grapes can shine.

Dessert: After all that recipe generation and testing, I decided to ask ChatGPT for an "easy dairy-free dessert using chocolate." The chocolate avocado mousse recipe it generated was delicious, plain and simple. Our toddler even put some on his salad and ate it up — so I’m calling it a major win.

As a busy working mom, I’m always glad to let someone — or something — do the thinking for me. That being said, I’ve definitely learned that using ChatGPT as your executive chef involves more brainpower than simply selecting a recipe from a cookbook. Will I use ChatGPT to create menus on a regular basis or follow its recipes blindly? Not anytime soon. Would I use artificial intelligence again to generate recipes or help with a menu if I need a boost of inspiration? Sure, why not?

YIELDS 1 CHALLAH

In a large mixing bowl, combine the flour, yeast, and salt.

In a separate bowl, whisk together the honey, beaten eggs and yolk, vegetable oil, and warm water.

Add the wet ingredients to the dry ingredients, and mix until the dough comes together. Knead the dough for 10–15 minutes, until it becomes smooth and elastic. Add 1 Tbsp of flour at a time if the dough is too sticky to knead.

Place the dough in a greased bowl, cover with plastic wrap, and let it rise in a warm place for 1–2 hours, or until it has doubled in size.

Preheat your oven to 375°F (190°C).

Punch down the dough and divide it into 3 equal portions. Roll each portion into a long rope, about 12 inches (30 cm) long.

Braid the ropes together, pinching the ends to seal. Transfer the braided loaf to a greased or parchment-lined baking sheet.

Cover the loaf with a clean kitchen towel and let it rise again for 30–45 minutes.

Brush the loaf with the egg wash.

Bake for 30–35 minutes, or until the top is golden brown and the bread sounds hollow when tapped.

Allow to cool on a wire rack before slicing and serving.

YIELDS ABOUT 2 CUPS

Preheat oven to 400°F (200°C).

Place the eggplant and red pepper halves cut-side down on a baking sheet lined with parchment paper and drizzled with oil. Roast for 30–40 minutes, until the skin is charred and the vegetables are tender.

Let cool, then remove the skin and chop the flesh. Combine the eggplant, red pepper, tahini, garlic, lemon juice, cumin, salt, and black pepper in a food processor and blend until smooth. Drizzle with olive oil before serving.

YIELDS 3 CUPS

Combine all ingredients in a food processor and blend until smooth. Drizzle with additional olive oil before serving.

SERVES 4

Heat the olive oil in a large pot over medium-high heat. Add the onion and garlic, and cook until softened, about 5 minutes.

Add the butternut squash, coconut milk, vegetable broth, curry powder, cumin, cinnamon, salt, and pepper. Stir to combine.

Bring the soup to a boil, then reduce the heat and let it simmer for 20–25 minutes, or until the squash is tender.

Use an immersion blender to puree the soup until smooth. (Alternatively, transfer the soup to a blender and puree in batches.)

Adjust the seasoning to taste, and garnish with chopped cilantro or parsley before serving.

SERVES 4

In a large salad bowl, combine the diced mango, sliced bell pepper, sliced red onion, shredded carrots, chopped cilantro, and chopped peanuts.

In a small bowl, whisk together the dressing ingredients until smooth and creamy.

Drizzle the dressing over the salad and toss gently to coat.

Serve the salad over a bed of mixed greens and enjoy!

SERVES 4 AS A MAIN

In a bowl, whisk together the soy sauce, Worcestershire sauce, olive oil, balsamic vinegar, garlic, brown sugar, thyme, salt, and black pepper. I just put all the marinade ingredients into a ziplock bag.

Place the London broil in a large resealable bag and pour the marinade over it. Seal the bag and refrigerate for at least 2 hours or up to 24 hours.

Preheat your broiler or a heavy skillet over medium-high heat.

Remove the London broil from the marinade and discard the marinade.

For the broiler: Place the London broil on a broiler pan and broil for 5–7 minutes per side, or until the internal temperature reaches 135°F (57°C) for medium rare or 145°F (63°C) for medium.

For the skillet: Heat a heavy skillet over medium-high heat. Add a tablespoon of oil to the skillet and sear the London broil for 3–5 minutes per side, or until browned. Reduce the heat to medium-low and continue cooking for an additional 8–10 minutes, or until the internal temperature reaches 135°F (57°C) for medium rare or 145°F (63°C) for medium.

Let the London broil rest for 5–10 minutes before slicing it against the grain into thin slices.

Serve with your favorite sides and enjoy!

YIELDS 1 9X13-INCH BAKING PAN

Preheat oven to 375°F (190°C).

Grate the potatoes and onion using a box grater or food processor.

Transfer the grated potatoes and onion to a large mixing bowl. Add the eggs, flour, baking powder, salt, and black pepper to the bowl and mix well.

Pour the vegetable oil into a 9×13-inch (23×33-cm) baking dish and spread it evenly around the bottom and sides of the dish. I found that the kugel was shallow in this size pan. If you like a deeper kugel, I’d use a smaller pan and bake it for longer.

Transfer the potato mixture to the baking dish, spreading it evenly in the dish.

Bake the kugel for 1 hour, or until the top is golden brown and the potatoes are cooked through.

Remove the kugel from the oven and let it cool for a few minutes before serving. Cut the kugel into squares and serve hot.

YIELDS 1 9X13-INCH BAKING PAN

Preheat oven to 375°F (190°C).

In a large mixing bowl, combine the grated sweet potatoes, grated apples, and chopped onion.

Add the eggs, honey or silan, brown sugar, oil, cinnamon, ginger, nutmeg, salt, and black pepper to the bowl and mix well.

Fold in the raisins and chopped pecans.

Pour the mixture into a greased 9×13-inch (23×33-cm) baking dish.

Bake the kugel for 1 hour, or until the top is golden brown and the kugel is set.

Remove the kugel from the oven and let it cool for a few minutes before serving.

SERVES 4

Preheat oven to 400°F (200°C).

Spread the frozen cauliflower on a baking sheet lined with parchment paper and roast in the oven for 10–15 minutes, until lightly golden.

In a medium saucepan, combine the quinoa and vegetable broth. Bring to a boil, then reduce the heat to low, cover, and simmer for 15–20 minutes, until the liquid is absorbed and the quinoa is tender.

In a large skillet, heat the olive oil over medium heat. Add the onion and garlic and sauté for 2–3 minutes, until softened.

Add the roasted cauliflower to the skillet, along with the cumin, coriander, cinnamon, salt, and black pepper. Stir to combine and cook for another 2–3 minutes, until fragrant.

Add the cooked quinoa to the skillet and stir to combine. Cook for another 2–3 minutes, until heated through.

In a separate baking dish, combine the halved grapes with the balsamic vinegar and roast in the oven for 10–15 minutes, until soft and caramelized.

To serve, spoon the cauliflower and quinoa mixture onto a serving platter and top with the roasted grapes. Garnish with chopped parsley.

SERVES 4

Cut the avocados in half and remove the pits. Scoop the flesh into a blender or food processor.

Add the cocoa powder, date syrup, almond milk, vanilla extract, and salt to the blender with the avocado.

Blend on high until the mixture is smooth and creamy. If the mixture is too thick, add a little more almond milk to thin it out.

Taste and adjust the sweetness if needed by adding more date syrup.

Spoon the mousse into individual dessert cups or a large bowl, and refrigerate for at least 30 minutes before serving.

Serve chilled and enjoy!

(Originally featured in Family Table, Issue 846)

Oops! We could not locate your form.

How to Cook Almost all the Cuts of Meat in Your Butcher's Case

Would we be better off with a severely restricted diet?

How sushi took the kosher world by storm and why it's here to stay

Don't go overboard, especially during the week. This isn't a competition; it's a chesed