Scheduled system maintenance will be at 12/28/2022 19:50 PST. The system will be unavailable for a few minutes at this time.
With ChatGPT getting a lot of press, now's a fitting time to talk about how it can be used in yet another way to make a potentially positive impact on day-to-day life. In this article, I'm going to use real-world examples to show how ChatGPT can be used today to come up with recipes that use ingredients you already have on hand (such as those tracked through WizerFood).
This fits in well with the concept of shelf use offsetting, which I discuss in a prior article. With this, you have another option when you've scheduled a "Use Shelf Contents" day: instead of using left-overs, you can use ChatGPT to whip up a recipe that consumes things you need to get rid of. In doing so, you can both reduce food waste and save money.
WizerFood has a recently-added feature which makes it easy to import ChatGPT-generated recipes (and really many forms of plain-text recipes). I'll explain how this can be used, and also offer some observations about the pros and cons of using ChatGPT for this purpose.
It's worth noting when I say "ChatGPT" in this article, you can interchange that with a number of new Large Language Model AI chatbots that are available. Some may be better than others based on what I discuss here, but the general concepts apply to all of them.
If you're unfamiliar with ChatGPT... what is it? Fittingly, let's ask ChatGPT itself! Here's what it says when asked, "What is ChatGPT, described in one sentence":
ChatGPT is an AI chatbot developed by OpenAI that uses deep learning techniques to generate human-like text responses to natural language queries or prompts.
ChatGPT is able to synthesize a wide range of data, across many domains, and it does it both in its understanding of requests and in how it builds sensible output (... but not always!). The program is generally available to the public, and you can visit https://chat.openai.com to create a free account.
Once you have an account, you'll be able to "converse" with ChatGPT, as if it was a personal assistant. For example:
Is ChatGPT like having your own personal chef? It might sound like one on the surface, but that's because it's been trained to be knowledgeable. It's important to note there's a big difference between being knowledgeable and being truly creative. Let's look at some examples. Let's say we ask ChatGPT: "create a recipe that uses both honeydew melon and milk." The response I got was a recipe for a "Honeydew Melon Smoothie Recipe". This seems reasonable, if not smart: you wouldn't make something like a salad with this, or a soup. But it's also true that to know this, ChatGPT has seen far more real-world examples of smoothie recipes that use these ingredients than salads or soups that use them. So no magic (or general intelligence) here, just a lot of examples.
Let's try to stump ChatGPT with a list of highly incompatible ingredients: "create a recipe that uses all of these: butter, strawberries, yeast, fish sauce, oxtail, almonds and mint." To it's credit, it came back with:
I apologize, but it would be difficult to create a recipe using all of these ingredients together, as some of the ingredients have strong and distinct flavors that may not blend well with others. However, I can suggest a couple of recipes that use some of these ingredients...
It took a long time to give this response, so it was clearly trying hard to find a way to combine these - but it couldn't. (thankfully!!)
So is it better or worse than simply doing a Google (or Bing) search for recipes that use the ingredients you're interested in? If you're looking for human creativity - there's a lot of good recipes to find on the web. This is why WizerFood has always supported the ability to provide a URL address for a recipe, and it'll import the details from a wide range of pages. However, a case can also be made for a simple turn-key approach to getting a functional recipe that tries to maximize the use of the ingredients you have. The suggestions that ChatGPT generates usually "make sense," and if you're trying to combine a lot of ingredients, you might have better luck with ChatGPT than with a web search.
For example, let's do a web search for: "recipes that use garlic, ground turkey, potato, onion and chives". Google comes up with some suggestions, with thumbnail images, which is nice:
Looking at these recipes, there's what I would consider to be a "partial match" - most don't use all of the listed ingredients, together. However, they look complete, have well-balanced flavors, and honestly: pictures help a lot, and ChatGPT doesn't provide those.
What does ChatGPT come up with, with a very similar query? Here's an example response I got:
Here's a recipe for a delicious and hearty meal that uses garlic, ground turkey, potato, onion, and chives:
Garlic Turkey Potato Skillet with Chives
- 1 lb ground turkey
- 4 medium potatoes, peeled and diced
- 1 onion, diced
- 4 cloves garlic, minced
- 1/4 cup chopped chives
- 2 tablespoons olive oil
- Salt and pepper, to taste
1. In a large skillet, heat olive oil over medium-high heat. Add the diced potatoes and cook for 8-10 minutes, or until they start to soften and turn golden brown.
2. Add the diced onion and minced garlic to the skillet and cook for 2-3 minutes, or until the onion is translucent.
3. Add the ground turkey to the skillet and cook until browned, stirring occasionally to break up the meat.
4. Season the skillet with salt and pepper, to taste.
5. Once the turkey is fully cooked, sprinkle chopped chives on top and stir to combine.
6. Serve the garlic turkey potato skillet hot, garnished with additional chives if desired.
This recipe makes a great weeknight dinner or meal prep option. It's filling and flavorful, and the chives add a nice touch of freshness. Enjoy!
ChatGPT has done a better job in including all mentioned ingredients, and it offers sensible preparation steps. However it's also quite simplistic, not advancing many ingredients past those that were asked for. I'll give it points for effort, but where it falls short gives some clues about how we might refine our approach.
If we pursue the use case of having ChatGPT help us clear off our shelf of foodstuffs which might be getting close to expiry, WizerFood can participate in a couple of ways. First, we added the ability for importing recipes from plain text, like you can get by copying a ChatGPT response, directly. We can do this simply by clicking on the "copy to clipboard" icon, right within a ChatGPT response:
If you have a specific day in mind to schedule this recipe in WizerFood, you can click on the date on your Calendar to get a new item prompt. Here you can click the new "Create from text" option:
Clicking this launches a "new recipe" page and brings up a dialog box such as this:
Notice here WizerFood has looked at what are possible ingredient candidates, already on your shelf (and in your profile), and made a possible suggestion about a prompt you might use with ChatGPT. This prompt is built to do a few things that are based on the experience describe above: rather than pile on perhaps a dozen different ingredients on your shelf to create something from, instead:
It's easy at this point to complete the process of adding your recipe: a) click on the copy text icon next to the prompt, b) click the "ChatGPT" link to open a new window which, if you have a ChatGPT account (and you're logged in) is ready for accepting a prompt, and c) paste the prompt into the ChatGPT window and click "send".
The resulting recipe was, interestingly, "Grilled Eggplant with Pesto Sauce." It chose eggplant, not eggs, per my request. One of the cool aspects of ChatGPT is you can qualify your previous prompts. My second prompt: "The last recipe contained 'eggplant' instead of 'eggs' per my original request, revise it."
I apologize for the confusion. Here's a revised recipe that includes eggs instead of eggplant:
Title: BBQ Egg and Pesto Sandwich
The second try had better results: it included both eggs and pesto. Copying-and-pasting the recipe into WizerFood, we see this:
After clicking Submit, you'll get a populated new recipe:
You can find an "Enter from Text" link on the new recipe page, as well:
This works for more than just ChatGPT generated recipes. Notice by having the generated prompt ask for certain extra details like calorie count, WizerFood has a chance to pull in those types of details, too. A lot of possibilities exist here, so look for more refinements in the future.
If Artifical General Intelligence (AGI) becomes a reality... will that be the point at which this hypothetical tool can become creative, and suddenly it can come up with recipes that are on-par with a human chef? By definition, AGI might be able to do that - but can it taste its creations, like good chefs do and possibly adjust, mid-prep? Maybe we will build machines with taste receptors to support that some day! Beyond AGI, is there a something needed that gives not only intelligence, creativity and self-awareness - but a personality, a sense of ethics, emotions, inner monologue, etc. All of those elements, one might argue, are important for the kind of creativity that a true human chef would exhibit.
Do you really need a professional chef-level input for all your recipes? Probably not. The most sensible use case seems to be using ChatGPT to create quick, easy recipes that use specific ingredients. Expect nothing fancy - but definitely functional. In reality, anything ChatGPT suggests is a composite of its source data, so at some level, a human has participated in your suggested meal.
Would it be useful to take the integration that's available in WizerFood using textual recipe import, and take it a step further, by allowing you to import a recipe from ChatGPT, without having to visit chat.openai.com? Or having WizerFood make AI suggestions as part of its notifications? Let me know what you think through the survey questions, below!
Was this article interesting or useful? You can subscribe for updates at the bottom of the page, or better yet, get your free WizerFood account and start using it today! Here's a summary video that covers what was discussed in this article: