This week, we launched a new project called The Slokie Chef, which combines old-school optimization algorithms to create ingredient combinations that are optimized for their net nutritional value, and uses the OpenAI's API to have ChatGPT combine these into sensible dishes. There are also high-fiber and high-protein options, and an ingredient selector that you can use to make use of what you already have on-hand.
You might wonder why a company that makes personalized children's books is putting something like this out. Well, the answer is that I'm also a parent and know the struggle it is to come up meal plans every day... especially ones that are healthy. This has a ways to go before it entirely solves that problem, but that's where I hope to take it. I'm also hoping to extend this project to a point where we can offer personalized cookbooks that augment your own family recipes with additional suggestions and meals to help fill the pages with meals that compliment those old favorites, either nutritionally or because they adhere to certain flavor profiles.
This project is still in its early alpha stage. You can peruse about 50 thousand pre-generated optimized ingredient mixes using the app as it stands now, and new mixes are being added regularly. Many already have ChatGPT recipes associated with them. Others don't. If you click the "Generate Meal Plan" button for a given list, and it appers to hang, just wait a minute or so... our system is just waiting for ChatGPT to finish its piece. The next time you click that combination, it should respond immediately.
ChatGPT is obviously a huge step forward for machine learning. However, there are a lot of limitations. As its own disclaimers state, "ChatGPT may produce inaccurate information about people, places, or facts." If you just ask it to create a healthy meal plan for you including whatever core ingredients you feed it, it's unlikely, in my experience at least, that what it gives you will actually meet all the various FDA requirements. That's because meeting all of those is a notoriously complex problem. Most of the diet trends/fads out there focus on maximizing or minimizing one or two things (like protein, carbs, fiber or fat), but disregard everything else, or give them passing attention.
As a machine learning engineer, I have attempted to solve twice before (once back in 2010, and then again in 2020) using traditional optimization algorithms. The problem I've had with my previous attempts to solve this in a purely mathematical way was that the "optimal" ingredient lists that I found were odd, to say the least. I embraced that back in 2020 when I used optimization algorithms to come up with a recipe for what I was calling "Matrix Porridge" in honor of the sludge from the 1999 blockbuster. The idea was that if people who caught COVID lost their sense of taste, then maybe they'd be open to something that provides "everything your body needs," without worrying too much about the strange mix of flavors.
But once ChatGPT launched, I started thinking that maybe this might be the thing I'd been waiting for all along... a way to take a list of odd-sounding ingredients and combine them into main dishes, sides and snacks in a way that made sense. That turned out to sort of work, as you'll see with some of the solutions you get with the app. There are definitely still some kinks to work out.
But when the app does work, it works quite well. I think with some more time and effort, I'll be able to filter out the odder solutions and train this system to make some meal plans that even my picky eater kids might like.
Check it out here.