What are you basing this on?
Last Thanksgiving I asked an AI for a chocolate chip cookie recipe using steel cut oatmeal. It gave me one. At first I judged the result a failure because they spread too much, giving me greasy paper thin cookies. However, they tasted ok and people liked them. Was that "creative?” I would say no, because the AI had no way to anticipate or judge the result. I think @simplicitus mentioned this kind of thing above.
He doesn’t eat a lot strawberries…
There has be a ton of recipes on internet, so coming up with examples wouldn’t be tough. It knowing what is going to taste like is not possible though.
I didn’t check but I don’t believe there are many. If there are, they probably do something special to reduce spreading. Like maybe mixing in some regular oats would be best. I tried the usual fixes for excess spreading in a later batch and they didn’t help at all. I think because the steel cut oats just can’t absorb the moisture during cooking so you get loosely packed grains oozing butter. But I’m not a baker. The AI is also not a baker, so it gave me a more or less standard recipe with a steel cut substitution. That’s not very creative by most standards.
There’s already AI cookbooks. I’ve also used chatGPT to figure out what to make for dinner based on what was in my fridge and it worked fine. Admittedly the AI can’t taste a recipe and give itself feedback but at that point a human who has little experience cooking could do it. Basically vibe cooking
Isn’t the AI just looking up existing recipes?
1 Like
I’m obviously not very creative, but maybe you should try asking for an oatmeal cookie recipe that uses chocolate chips instead of raisins… 
It is nice that you can get that from an AI, but it is only remarkable if you don’t know that there are recipe apps that will do the same thing (give you recipes based on a list of ingredients) predating LLMs.
I would be VERY surprised if the current state of ChatGPT could pull off one of the best in the city level. For starters, there are too many bad/stunt/fad recipes on the internet. ChatGPT is going to try to shoehorn cottage cheese or jammy eggs into way too many places just because those things pop off on TikTok. Also, because it can’t taste or actually make the recipes, I don’t think it will ever balance flavors and know when novel combinations actually work quite as well as an elite or even good chef.
This video shows that ChatGPT can make a recipe that kinda works, but can be greatly improved by humans who can bring additional wisdom and can adjust on the fly as the dish is cooking.
Where I do think ChatGPT could be useful (and I suspect is already being used) is for menu development for places like Wonder. I think it started in New York and recently opened in DC and the basic idea is that there is one central kitchen churning out dishes from multiple different cuisines. So, you have a “Greek” menu with a few things like grape leaves, and mousakka, a “Thai” menu serving papaya salad, pad Thai, and some curries, an “American” menu with burgers and fries, etc. You can order anything off of any of those menus, so ordering mozzarella sticks, kungpao chicken and baklava is definitely a thing that can happen.. It’s not promising the best version of any of those dishes, but it is promising an ok version that allows people to satisfy multiple cravings at the same time and that allows family members to not have to fight over what type of food to get. For that application, I’m sure ChatGPT could come up with at least a decent first pass if you told it to whip up basic versions of, say, a Greek menu that contained 3 appetizers, 4 entrees (including one vegetarian) and a dessert.
2 Likes
Yeah you’re right ChatGPT would probably not be able to do it, but I bet an AI specifically designed to be good at recipes with current technology and techniques would be able to do it. Esepcially if they had a foodie, not a trained chef but a trained palate essentially vibe cooking and giving it feedback on the recipes. Maybe you wouldn’t get NOMA but you’d get something better than your typical culinary institute of America grad turned head chef and definitely better than the corp chain restaurants.
I just saw a post on Reddit about a restaurant that used chat gpt to edit its menu and streamline. Most of the criticism was a lot of humans could do that as well but it is interesting that businesses are coming up with unique ways to apply ai to problems.
ChatGPT: Tastes fine to me! Maybe add a pinch of salt if you want.
How long before ChatGPT will be able to beat me at chess?
I think you are massively underestimating the level of skill that a trained chef brings to the job. I generally think we are all terrible at estimating the expertise of experts, especially when they do something we kind of know how to do.
Lots of the reasons that chain restaurants are bad is not because of a lack of skill in menu design, it’s a combination of
a) economics demanding cheap food, badly paid people and low quality ingredients
b) the average American having the taste of an average American
In fact, within these constraints, chain restaurants menu design is probably close to maximised and elite
1 Like
How about
AI is a useful tool now, it comes with substantial immediate risks to society now, today based on current tech, that can be managed but probably won’t be because we are in worst timeline. Plus AGI is probably coming but on the outer end of estimates. And no one knows what AGI will do, but could be bad.
That’s a more measured take and I can’t say I disagree, but it might not poll well on account of tldr. I’m just curious where the forum stands. You’re welcome to put up a poll of course. I feel kind of out of it on these pills so I’d rather not atm. Maybe I should ask an AI to create a poll based on my options. What does AI think we might think of it? That might be fun.
To continue the food discussion, I had hospital food today. Although it’s probably close to optimum for its purpose, I’m quite sure AI could do better given some freedom from those constraints.
If there are two sides, Im starting to shift.
I was definitely team hype early on. Because these tools are OBVIOUSLY incredible, and team sceptic was saying stupid shit like it’s glorified auto complete only good for writing bad LinkedIn posts.
But over time. Team hype has got crazier and the tech isn’t particularly improving in useful ways. At the same time team sceptic is using it a lot more and conceding it’s usefulness, while pointing out lots of very significant risks. The teen suicide note up thread for example.
Those seem like accurate observations. Definitely, as a skeptic, I am using it more and more and finding it very useful for some things. But I’m still firmly in the skeptical camp.
I’m not close to an expert, and maybe I just have a soft spot for him because of his guitar book, but Gary Marcus is an expert and seems perfectly worth listening to. Much more so than the Altmans and Musks. My opinions are provisional though. Maybe the big breakthrough will happen next year. I mean I can’t say for sure it won’t.
We’re getting a giant data center here in Utah. One thing that is for sure is that thing will drain power and water we don’t have to spare. I don’t think the trade-offs favor projects like this. Unless you’re an oligarch I guess.
1 Like