Hello,

How’s your first AI app coming along from the last lesson?

I hope you’re having fun and feeling excited playing with your new AI app. :)

I wasn’t planning to send you an email today because I really want you to enjoy working on and improving your app.

But I realized that you might be upset with me later if I didn’t share these important lessons.

So, in this email, I want to tell you about some mistakes I made on my way to building a million-dollar business.

Let’s learn from them!

#1: AI Isn’t Built for Facts

This was a big mistake I made when I didn’t fully understand how generative AI works.

The AI we’re using, like ChatGPT, is what we call “Generative AI.”

This type of AI is designed to create something new based on the information it has learned.

Sometimes, you can ask AI to find a fact, like:

The AI might give you the correct answer.

But here’s the thing — "generative AI" wasn’t made to find facts!

Most generative AIs learn from a lot of information found on the internet, which includes both correct and incorrect facts.

So, you can’t always trust the AI to give you the right information.

I’m not saying it can’t find facts, but it just wasn’t designed for that purpose.

In the future, AI might get better at this, especially with new technologies like RAG (Retrieval-Augmented Generation).

RAG helps AI find facts based on information we provide, instead of relying on what the AI already knows.

So, if your idea is focused on finding facts, you have two options:

1. If you’re serious about it, consider using RAG technology in your AI app. Dify.AI actually supports some great RAG tools.
2. Or, skip the fact-finding idea and create something that focuses on creativity, opinions, or anything else that doesn’t rely on facts.

I didn’t know about RAG at first and tried hard to make my AI app perfect for providing factual answers.

But 70% of the time, it gave the wrong answers lol

So, my advice is to avoid using AI for facts, at least in the beginning.

#2: AI Doesn’t Know Real-Time Information

I’ll keep this one short because it’s similar to mistake #1.

Generative AI usually works with old information from the internet, so it doesn’t have up-to-date facts.

You need to be careful when creating an AI app that relies on the latest information because most of the data the AI has is outdated!

For example:

The AI might give you a result that looks great, but it’s probably wrong because it doesn’t know the latest trends.

You might say that ChatGPT gives you up-to-date answers. That’s true, but let me explain.

When you use ChatGPT.com, it “cheats” by opening a web browser in the background to search for the latest information before giving you an answer.

But when our AI app works, it directly prompts OpenAI’s GPT-4o-mini, which doesn’t have this “cheating” feature. So, it only uses old information to generate an answer.

We can make OpenAI’s GPT “cheat” like ChatGPT.com, but that’s too advanced to explain here.

Just remember: for now, don’t rely on AI to give you answers that need up-to-date information.

#3: Don’t Try to Get a Long Result with One Prompt

I’ll keep this short.

If you want your AI app to create something long, like writing a book, don’t expect it to do that with just one prompt.

It’s not possible, so don’t waste your time trying!

For example, if you ask your AI app to:

Even if you give more details, like:

It still won’t work.

The AI might try to follow the prompt, but it won’t get it right.

Instead, if you want your AI app to create long content, break it down into smaller tasks.

For example, if you want to write a book, build your AI app to do it step by step:

1. Table of Contents Generator - First, create an app that generates the Table of Contents (ToC) for the book.
2. Content Generator - Next, create an app that generates the content for each chapter based on the ToC. Run this app one chapter at a time. If your book has 12 chapters, you’ll run the app 12 times.

So, to recap: Don’t waste time trying to get your AI app to produce a long result with one prompt. Design it to run and generate content chunk by chunk.

#4: You Don’t Need to Stick with One AI Engine

The last mistake I want to share is this:

Sometimes, when you can’t get the result you want from one AI (like ChatGPT), even after trying hard, it’s okay to switch to another AI.

For example, if you’re using OpenAI’s GPT-4o-mini and it’s not giving you the result you need, try switching to Google’s Gemini and see if you can tweak the prompt to get what you want.

If that doesn’t work, try another AI like Anthropic’s Claude, Meta’s LLaMA, or Mixtral.

Different AIs can give you different results, even with the same prompt.

So, it’s more about mixing and matching AIs to get the result you want, not just modifying the prompt.

Press the Pedal

I have plenty more mistakes to share with you, related to AI apps and other things.

If you’re serious about building your AI app as a SaaS business, let me know, and I’ll find a way to share more.

Tomorrow, I’ll send you another email with the “game plan” I used to earn from AI apps.

It’s time to start thinking about how to make some money, right? ;)

See you tomorrow!

—Bank K.