Will ai take over the world? Or are we just overthinking it?

If you think AI will take over the world, you're crazy. Just kidding! (but not if you don't follow).

Unless You live under a rock, you probably know what ai is, it's an artificial intelligent brain with digital neurons in short. AI isn't dangerous itself; it actually depends on How people use it. AI won't do anything; It's just a bunch of code. You need to decide how to use it responsibly. Companies even give restrictions to the AI, but Not everything is perfect...

Let's see how it's trained so you know what's happening behind the scenes, or should I say, neurons.

1. Training

There are many types of AI Training methods, but let's cover the Main 4 -

I. Supervised Learning

Supervised learning is basically when the AI is given data tagged with its meaning. It will try to configure the patterns, colours, punctuations, and the relevant type of data between the images. For example, a model trained with dog and cat images will be able to distinguish between them in no time.

II. Semi-supervised learning

As the name suggests, this is only partly supervised, not everything is labelled… and that’s where it gets interesting.

It’s basically supervised learning—but with some data missing labels. The model learns from the labelled part, then tries to figure out the rest.

This means more flexibility…but also more chances to mess up.

III. Unsupervised Learning

Your neurons will likely guess that this type of learning has no tagged data and uses untagged data patterns to tag it itself. Because it is totally independent, it trains longer and identifies way more patterns in data than the other supervised ones.

IV. Reinforcement learning

Jokes aside, this is my favourite type It learns through trial and error—and honestly, it’s fun to watch.

Think of a soccer robot Every move gets a reward or a penalty. Over time, it learns what works… and what doesn’t.

Fun fact: Supervised learning is still the most commonly used type.

2. Ruling

Now, let’s take control… not WW3, just your AI model.

There are 3 types of ruling systems:

Supervised Fine-Tuning (SFT) -

Think of it as a second round of training. A pre-trained model is trained again using labelled data to guide its responses.

This helps the AI give more accurate and useful answers.

It’s still part of overall training, but we separate it into stages so we can easily fix and improve specific parts of the model.

Reinforcement Learning from Human Feedback (RLHF) -

This type of Controlling system is like Reinforcement learning but combined with human feedback, for example an ai is being trained at RLHF for negative and positive responses -

Humans decide what is negative and what is positive, the ai learns like "This is what humans don't like".

Constitutional AI -

This is like giving a checklist to ai, the ai checks all of it to give a near-perfect answer. It checks all factors in the list before completing the answer. This type of ai is a newer beta type of system used in advanced systems.

AI systems today mostly use SFT and RLHF.

So now that you know how ai is trained and how it's restricted, you'll be able to confidently say "AI is dangerous if the user is dangerous."

To summarize, AI is just a trained model, without intentions or feelings. AI might be manipulated with specific illegal prompt engineering to provide some inappropriate answers and the company or ai model is not responsible.

Share for more knowledge!