AI generated recipes, a bad idea?
I found this page that genrate ai recipes (its in my native language). https://oppskrifter.app/
is it a bad idea to use ai recipes
1 Comment
702551April 5, 2025
Here's the deal with recipes: if the people sitting at your dinner table are enjoying what you have served the source of the recipe really doesn't matter. Grandma, some online recipe site, famous chef's cookbook, the back panel of a package of food, whatever.
However I will point out that all consumer-facing LLM-based AI chatbot assistants are dumb as rocks. They are essentially fancy probability calculators that have zero in the way of common sense, design, artistry, taste, or rationality. They just scan a bunch of documents and try to figure out what is the most likely/popular answer.
The problem is that these Large Language Models are heavily flawed.
Here's just one of many errors from last year:
https://www.livescience.com/technology/artificial-intelligence/googles-ai-tells-users-to-add-glue-to-their-pizza-eat-rocks-and-make-chlorine-gas
Now this is not some mom-and-pop AI shop. Google Gemini is run by Alphabet ($1.8 trillion market cap, ranked #5 in the Fortune 500). And yes, their AI recommends people to eat rocks and put glue on pizza.
There is *ZERO* common sense. Worse, its advice is hazardous to health.
Someday AI might be worthwhile but it really needs to be around 99.999% accurate.
Today's AI tools recommend people to do dumb stuff. And you don't want that because inevitably someone with zero common sense will follow the advice of an AI assistant with zero common sense.
Remember that *ALL* consumer-facing LLM-powered AI chatbots/assistants in 2025 are basically alpha quality software and are no where near to be ready for primetime.
Worse, there are documented regressions in how these LLMs are being trained and deployed. This may indicate a fundamental flaw in today's LLM design. Almost all of the programmers are using very similar methodology so there isn't one that stands out head-and-shoulders above the rest. They are all heinously unreliable, some more so than others.
AI chatbots can be entertaining though and are capable of a handful of relatively menial tasks for Joe Consumer at this time (like writing a business e-mail summarizing meeting notes or a simple expository article).
But there's absolutely no chance I'd put my health in the hands of some senseless AI tool in 2025. AI will get better eventually but there is no timeline when it might actually be reliable enough to recommend to the general public for tasks like this.
However I will point out that all consumer-facing LLM-based AI chatbot assistants are dumb as rocks. They are essentially fancy probability calculators that have zero in the way of common sense, design, artistry, taste, or rationality. They just scan a bunch of documents and try to figure out what is the most likely/popular answer.
The problem is that these Large Language Models are heavily flawed.
Here's just one of many errors from last year:
https://www.livescience.com/technology/artificial-intelligence/googles-ai-tells-users-to-add-glue-to-their-pizza-eat-rocks-and-make-chlorine-gas
Now this is not some mom-and-pop AI shop. Google Gemini is run by Alphabet ($1.8 trillion market cap, ranked #5 in the Fortune 500). And yes, their AI recommends people to eat rocks and put glue on pizza.
There is *ZERO* common sense. Worse, its advice is hazardous to health.
Someday AI might be worthwhile but it really needs to be around 99.999% accurate.
Today's AI tools recommend people to do dumb stuff. And you don't want that because inevitably someone with zero common sense will follow the advice of an AI assistant with zero common sense.
Remember that *ALL* consumer-facing LLM-powered AI chatbots/assistants in 2025 are basically alpha quality software and are no where near to be ready for primetime.
Worse, there are documented regressions in how these LLMs are being trained and deployed. This may indicate a fundamental flaw in today's LLM design. Almost all of the programmers are using very similar methodology so there isn't one that stands out head-and-shoulders above the rest. They are all heinously unreliable, some more so than others.
AI chatbots can be entertaining though and are capable of a handful of relatively menial tasks for Joe Consumer at this time (like writing a business e-mail summarizing meeting notes or a simple expository article).
But there's absolutely no chance I'd put my health in the hands of some senseless AI tool in 2025. AI will get better eventually but there is no timeline when it might actually be reliable enough to recommend to the general public for tasks like this.
Showing 1 out of 1 Comments
Recommended by Food52
Popular on Food52
Continue After Advertisement