Why CurlSbot is not AI

AI is a hot topic right now. Some people love it, others hate it. Wherever you stand, know that CurlsBot ingredient analysis does not use AI. Here's why (and how it actually works)

Defining AI

When most people think of AI, they are thinking of an AI type called LLMs, which stands for large language models. LLMs are trained on a ton of data and can do some pretty impressive stuff. That said, there have been many controversies about whether the devs of LLMs got permission to use the data they are trained on.

CurlsBot on the other hand uses old-fashioned programming, the kind you would learn in an average programming class. It just takes some text, splits it into items, and compares each item to our database. It is not "smart" enough to be considered AI. If you are a coder, even a beginner, you can likely even read the code and understand how it works.

The only "AI" feature I have is the experimental photo feature, that uses an open source tool called Tesseract to try to find text from an image. It does not store any data, it runs completely on your computer/phone, and uses a minuscule amount of energy compared to LLMs.

1. It's unpredictable

LLM-based AIs are largely created and run by big companies like OpenAI (ChatGPT). There are many different models and they are always coming out with new models. These models may behave differently even given the same inputs.

In addition these models are so complicated that even their creators often can't tell why they give one answer for another.

CurlsBot is simple code and given the same inputs, it should always have the same results. Though keep in mind CurlsBot can also be wrong and have bugs! Lists might look the same to you but CurlsBot is just a couple of hundred lines of code and might not "understand" things if they are not comma separated, have weird characters, or many other text issues.

Text is one of the most complicated "problems" in computer programming. Think of all the languages, punctuations, and special characters that humans use. For example to a computer program like CurlsBot these two punctuation marks are totally different:

• "
• ”

You can help me by emailing me any issues you find, and please include the exact text so I can debug.

2. It would reduce user privacy

Since LLMs are so complicated, they generally have to be run by the companies that built them. So I'd have to send your data to OpenAI (ChatGPT) or another similar company, which is not only really complicated, but introduces legal and privacy issues.

3. It's expensive and uses a lot of energy

While CurlsBot does have some affiliate links and advertising, it doesn't make enough money for me to afford using LLMs.

LLMs also use a ton of energy which can have environmental costs.

Why spend a lot of money and energy on something that simple code like CurlsBot can do?

4. It's often wrong

Think about LLMs as just really fancy autocomplete. Does your autocomplete always get things right? Mine certainly does not. LLMs also are trained on data that might be wrong or from a long time ago. I've seen a lot of people post LLM generated hair advice on Reddit and sometimes it's OK, but sometimes it's also wrong. I've seen people get advice that's even unsafe!

At CurlsBot all our data is manually entered and reviewed by me. That doesn't mean it's always right, but it does mean that a real human being (me) checked it.

5. There's no reason to use AI in CurlsBot

In the end, CurlsBot just doesn't need AI. It's expensive, unpredictable, wastes energy, and all for something that simple code can do just as well if not better.

I also think turning CurlsBot into an AI tool would potentially reduce user trust. I personally lost my last job due to AI and I totally understand why many people are opposed to using it. I think it can be useful in some cases, but putting it in CurlsBot would be wasteful and unnecessary.