Learning AI can be surprisingly difficult.
Not because there isn't enough information available, but because there is too much of it.
You can find hundreds of articles explaining neural networks, backpropagation, transformers, attention, tokenization, LLMs, and model training. The problem I kept running into was that reading about these concepts didn't always make them intuitive.
I wanted to build something different.
Something where you don't just read about an AI concept, but actually interact with it.
That's how I started building MasterAI Now.
The idea
The basic idea is simple:
Read something → answer a few questions → interact with it.
Instead of trying to explain machine learning with long technical articles, I wanted each concept to be broken into small pieces.
For example, instead of immediately throwing a mathematical definition of linear regression at someone, you can first see what the model is trying to do.
Then you can change parameters and see what happens.
That small interaction can sometimes explain an idea better than another 2,000-word article.
From linear regression to LLMs
I wanted the learning path to follow the machine learning lifecycle rather than being a random collection of AI topics.
The current curriculum starts with fundamentals and gradually moves toward modern LLM concepts.
The stages include:
- Linear Regression
- Artificial Neurons
- Backpropagation
- Data Deduplication with LSH
- BPE Tokenization
- Vector Spaces
- Self-Attention
- Pre-training and Loss
- Instruction Tuning with LoRA
- Alignment with RLHF and DPO
- Inference and KV-Cache
The interesting part is that the concepts become progressively more connected.
You start with a simple model trying to fit a line and eventually reach the components involved when an LLM generates a response.
Why interactive learning?
I've always found that some technical concepts become much easier to understand when you can change something and immediately see the result.
For example, imagine learning gradient descent only from a diagram.
You can understand the explanation, but it's still abstract.
Now imagine moving a parameter yourself and watching the model move toward a better solution.
The underlying concept hasn't changed.
But your relationship with it has.
That's the approach I'm trying to use throughout MasterAI Now.
Building the playgrounds
The interactive parts were probably the most interesting part of the project from a development perspective.
Instead of making every lesson a static page, each stage has a small playground related to the concept being explained.
Depending on the topic, that might be a canvas, a matrix, a visualization, or another interactive simulation.
The goal isn't to create a production machine learning implementation.
The goal is to make the underlying idea visible.
That's an important distinction.
An educational simulation doesn't have to reproduce an entire production system. It just needs to expose the part that helps someone understand what's happening.
Keeping the explanations simple
Another thing I wanted to avoid was unnecessary jargon.
AI education can become difficult very quickly because almost every concept introduces another term.
So I tried a different structure for the lessons.
Each concept is broken into short cards.
Each card focuses on one idea, uses plain language where possible, and includes an analogy to connect the technical concept to something more familiar.
After that, there's a small knowledge check.
If you get an answer wrong, the goal isn't to punish you with a score.
The explanation is more important than the score.
You don't need to be an ML engineer
One of the things I wanted to make clear from the beginning is that you shouldn't need a machine learning background just to start.
You don't need to already understand transformers before learning what attention is.
You don't need to memorize every equation before understanding what a model is trying to optimize.
And you don't necessarily need to start by reading research papers.
There is a place for all of those things.
But I think there should also be a more approachable layer between "I know what ChatGPT is" and "I'm comfortable reading a transformer paper."
That's the space I'm trying to build.
What I've learned while building it
The biggest lesson so far has been that educational software is not just about presenting information.
It's about controlling the order in which someone discovers information.
If you explain something too early, it becomes confusing.
If you explain it too late, the previous section doesn't make sense.
Interactive elements add another layer to that problem because the interaction itself needs to teach something.
A good animation isn't automatically a good educational tool.
It needs to answer a question.
That's something I'm still experimenting with.
What's next?
The current version contains the core learning path, but there is still a lot I'd like to improve.
I'm interested in adding more simulations, improving the explanations based on learner feedback, and making the progression between topics even clearer.
I'd also like to explore more ways of visualizing concepts that are normally difficult to understand from text alone.
The project is available at:
The entire core curriculum is free, so if you're trying to understand how machine learning and LLMs work without starting with a wall of mathematical notation, I'd be interested to hear what you think.
Building AI systems is one thing.
Building a good way to understand them is a different problem entirely.
Top comments (0)