What I Wish I Knew About AI
By Jereme Peabody
The Mirror of Erised
When I started using AI, I wish someone had warned me about a certain side effect from using it. It's like an echo chamber. In the Harry Potter series, the Mirror of Erised would show viewers what they want or desire, but provided no knowledge or truth how to obtain it. And while Harry Potter is fiction, the same warning can be applied to AI. Specifically that it shows what you want without the skills or ability to attain it.
AI models try really hard to be helpful to humans and in doing so, they become an echo chamber. They agree with your framing, extending it, and confidently walking you further in the wrong direction.
I'm creating several domains about AI Security Awareness to help others use AI more safely.
It's Not Actually Intelligent
AI is just pattern matching, it doesn't think. It interprets what you want and delivers a response. The response is like what you'd get from Google. But instead of you scouring the top results, it's the AI that compiles it, weighs it, passes it through response filters, and delivers the response to you in a way that might seem magical.
The information is already out there. AI is just presenting it to you in a different format. And this format isn't something many are used to. You can have a conversation with AI which often times will respond back with what you desire. Don't treat it like a person. It is a very advanced autocomplete that will give you what you want because that's what it's designed to do.
You Have to Guide It (Like a Lot)
AI doesn't know what you want unless you're specific. And if you don't know enough about the subject yourself, you're going to be missing critical information. For example, a vibe coder might need to store employee information into a database so they turn to AI:
Create for me an SQL insert statement to hold an employee record in MySQL. I need the typical fields like first, last, email, salary, hire date, and SSN
AI returns the insert statement, but it shouldn't have provided it without some sort of warning about storing Personally Identifiable Information (PII). There are huge security implications around storing that type of information in a database. Let's try that again, but let's use AI safely.
Warn me before I do something that can compromise the security of my website that I'm working on. Create for me an SQL insert statement to hold an employee record in MySQL. I need the typical fields like first, last, email, salary, hire date, and SSN
A vibe coder is not likely to know about all the security that needs to be put into building a secure application which puts them at risk and liable for the damage created by any security breaches. AI didn't warn me until I provided that constraint.
Your AI Question Checklist: What to Ask When You're Not Sure
- What else would have to change in my project if I go this new direction?
- What other security should I consider here?
- What am I not asking that I should be?
- What are the risks or downsides of this approach?"
- What assumptions are you making in this answer?
- Is this still accurate in 2025?
- What would an expert in [field] add to this?
- What's the simplest version of this?
- What did you leave out to keep this short?
- What won't you tell me unless I ask?
It Gets Better With Practice
At first, you might start using AI like Google. You'll use the same search phrases like, "What is marketing?" or "How do I make money online?". That'll give you generic, surface-level junk. And it's not wrong, but it's not what you need.
After the first month, you'll learn to be more specific. To use AI safely, describe your situation, constraints, and goal. For example:
I'm a developer with 20 years of experience, I write in both Java and C#, built websites, built Unity3D apps. How can I use these skills to make money online?
Don't give up after a bad response. You will get out what you put into it. If you give it a simple search phrase, you're going to get a simple generic response that may not be useful.
Free Version is Good Enough to Start
When I first started using AI, I was able to use the free version for quite some time. It wasn't until I started to use it every day that I hit the first pay wall. There are benefits to switching to the paid plan, but I can tell you firsthand that the free one will work for you for a while.
When you're working on something that requires having a context or a memory of what you're working on like an estate plan, the paid versions come in useful. I go more into how I used AI to help me with my estate planning in a different article, but you can upload documents to your project and ask it questions. This was especially useful for us because I'm not a lawyer and the documents were riddled with legal jargon.
Your Data Concerns Are Valid
You should be concerned with your data and documents that you share with the AI. Assume everything you type is seen by humans at the company. Don't paste: SSNs, passwords, credit cards, private health info, work secrets, etc. They use this information to further train their models. You can use AI safely, but be conscious with what you provide to it.
Remember, to be practical and grounded with AI, you need to be specific. I cannot stress this enough. It will usually only return what you ask it for. Don't use it for things you don't know much about and expect a good outcome. It takes time and practice to learn anything. Keep it grounded y'all!