Hacker Newsnew | past | comments | ask | show | jobs | submitlogin

Recipes have been one application that OpenAI/ChatGPT is actually really good at. No ads, interactive substitutions/scaling/conversions.


"hey bot, I have {foodstuffs} available, I need to make dinner for n people. Give me some options in the x style"

Probably one of my most-used prompts, and it's batting close to 1.000.

Every now and then it will make a mistake, like forgetting the salt, putting a step in the wrong order, or the like, but far less often than you'd think. If you already have even a middling amount of kitchen experience, it's a fantastic use case.


Seems like an oddly high risk process. Not only are you risking bad taste from major missing ingredients like salt, but also ruining things with the wrong temperature/ cooking times.

A reasonably comprehensive cookbook like The Joy Of Cooking or something specific to your preferences seems like less effort overall vs Google or a LLM.


> Not only are you risking bad taste from major missing ingredients like salt, but also ruining things with the wrong temperature/ cooking times.

they are operating under the assumption you are a decent cook. any decent cook will know how to taste for salt and what is safe temperature wise. if you blindly follow instructions you are not a decent cook.


I don’t read “have even a middling amount of kitchen experience” as being a decent cook. IMO, a decent cook doesn’t really need full recipes they can just wing it based on a basic idea, but it takes a long time to get to that level.


It... really doesn't, though? Cooking is hardly by any means trivial, and achieving recipe-quality results based on a quick skim and winging it from there is certainly more difficult, but I'd argue that anybody who can't make something decent without religiously following a recipe isn't even of "middling kitchen experience".

I may be biased though, "skim a recipe and wing it" is my default style anymore after all.


I’m not talking about skimming a recipe and wringing it.

Think easting something in a restaurant then replicating something similar at home the next day just based on the flavor. Doing that for a few things is easy, but it takes significant time to get to that level across a wide range regional cuisines.


It really, honestly isn't a high-risk process at all. You just have to think a little bit. All I can say is try it yourself and see. For me, half the fun is seeing what it comes up with based on what I feed it, especially if I encourage fusion cuisine. I've hit some real winners that way.

Also, portion adjustments are quite useful, as I mostly only cook for one or two people. Scaling isn't always linear, so that's been a helpful step.


Cooking times don't seem very reliable to me anyway, every recipe I try takes longer than it says in the recipe.


I am more skeptical of the ingredient proportions than the cooking times...


That likely comes down to your preferences or equipment.


That, and recipe writers lie like a rug when it comes to time. :D

https://news.ycombinator.com/item?id=27632835


I would treat it analogous to asking ChatGPT to code for you (but for food) - some times it works great, other times you gotta nudge it a bit.

The tool is currently best used by people who already know how to code/cook and don't want to spend too much cognitive bandwidth, but have the skills to mofidy as needed.


That's pretty much exactly how I see it, too. Offload a good chunk of the scut work, but I'm sanity checking the results. Still a strong net positive, at least in my experience.


It is more likely going to have a pretty good grasp on ingredients that pair well than a random unrated post in allrecipes.com.

Likely won't pair hot sesame oil with milk, for example. (someone please tell me that's not a thing).


The Joy of Cooking is my favorite cookbook but only certain editions. Buying the wrong one is pretty high risk.




Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact

Search: