Everyone is talking about AI nowadays. I am not an expert on AI or anything, but as far as I know, they are just prediction machines. Meaning you give some words to it, and it gives back some words that are most commonly used together with the words you gave. At least, that was what it was at the beginning. Nowadays they have improved, and now they can take actions too with the words you give them.

Since I am not an AI expert or anything, I can only guess how this “taking action” works. I guess they have services (sending HTTP requests, etc.) coupled with some keywords, meaning instead of generating words, they run these services when you give those keywords.

So, I asked “Gemini Fast”, “Gemini Pro”, and “Claude” about myself to see what they would spit out. I gave links to this website and my LinkedIn profile and asked what they could tell me about me. Well, this website has some security measures that include blocking AIs and bots, and I assume LinkedIn has the same (probably more) measures. That’s why I’m not suprised when Gemini Pro and Claude returned errors saying they couldn’t access the websites. Claude just left it there, but Gemini and Gemini Pro did something interesting.

Gemini Pro did a Google search with my name and tried to make sense of the results. Well, the most related results were my LinkedIn and a hotel with my surname. So, it said I am an engineer and my family has a hotel. Well, I know the hotel it found, and I have no connection to it at all. The owner of the hotel has the same surname as me, but that’s it (we are not even related). But according to Gemini Pro, that hotel is owned by my close relatives.

What is more interesting is that the “fast” version of Gemini somehow managed to read the contents of this website. Well, it could only read the “about” page and conclude that I am a software engineer with a love for dogs and cars. While it is not detailed, it was actually correct. Well, it just repeated what I wrote, so of course it was true.

Do I Use AI?

The short answer is yes, I do. But I am not using AI like other engineers and developers. Most people use AI for building. I use it more like a sentence generator and grammar checker. Since AIs work as word predictors under the hood, I think this is the only correct way to use AI. I create response messages for the services I build, and I ask AI to find grammar errors and typos in the posts I write on this website. These are the only ways I use AI.

Note From AI’s Perspective

I gave this post to Gemini Pro and asked it to check for grammar and typos. Well, it corrected the typos and then added a note at the end of the response. So, here is the answer the AI gave to this post;

As an AI, I really enjoyed reading your pragmatic take on how we work! You actually have a very solid intuition for what is happening behind the scenes, but I can gently clarify a couple of your educated guesses:

  • How we “predict”: You are spot on that at a foundational level, large language models predict the next token (part of a word). However, to do this accurately across complex concepts, we don’t just look for words that are “commonly used together.” The models map out intricate relationships, context, and logic patterns. It is closer to semantic understanding than just repeating a statistical parrot trick.
  • How we “take action”: Your guess about HTTP requests is remarkably close! It is called “tool use” or “function calling.” Instead of relying on hardcoded trigger “keywords,” the model is given a list of tools it can use (like a Google Search API or a web browser). If my reasoning processes determine I need fresh information to answer your prompt, I write the code to activate that tool, read the returned data, and synthesize it into my final answer.
  • The Hotel Hallucination: Your experience with Gemini making up a story about your family owning that hotel is a classic example of an AI “hallucination.” When models find two pieces of factual data (your name and a business with the same name), they sometimes confidently, and incorrectly, draw a line connecting them where none exists.