Learn how AI actually works — by touching it
MasterAI Now teaches the full machine learning and LLM lifecycle through short, jargon-free cards and browser-based playgrounds. Drag a slider, watch a gradient flow, understand the idea for good.
- interactive stages
- 11
- reader cards
- 55
- live playgrounds
- 11
- free forever
- 3h
How every stage works
Read
Five cards written for humans: plain language, bold key terms, one honest analogy each.
Check
A one-click micro-quiz per card. Wrong answers are explained, never punished.
Play
Unlock a live simulation — a canvas, a 3D space or a matrix — and change the parameters yourself.
The 11-stage track
Three modules, from a single straight line to the KV-cache inside a running LLM. Every stage ends with something you can actually manipulate.
Build intuition for learning itself: fitting a line, bending it into a decision, and assigning blame across layers.
- STAGE 010%Unlocked
Linear Regression
Teach a machine to draw the best straight line through messy data.
11 min5 cards5 checks0%0/5 cards - STAGE 020%Locked
The Artificial Neuron
One weighted sum plus one bend equals a decision machine.
13 min5 cards5 checks0%0/5 cards - STAGE 030%Locked
Backpropagation
The credit-assignment algorithm that makes deep learning possible.
15 min5 cards5 checks0%0/5 cards
Everything a model knows comes from its data. Deduplicate the corpus, tokenize it, and place meaning in vector space.
- STAGE 040%Locked
Data Deduplication with LSH
Duplicate text poisons a model. Find near-duplicates at web scale.
14 min5 cards5 checks0%0/5 cards - STAGE 050%Locked
BPE Tokenization
How raw text becomes the integers a transformer actually reads.
14 min5 cards5 checks0%0/5 cards - STAGE 060%Locked
3D Vector Space
Meaning becomes geometry: every word is a direction in space.
15 min5 cards5 checks0%0/5 cards
From attention to alignment to inference: the five steps that turn raw text into a served assistant.
- STAGE 070%Locked
Self-Attention
Every token looks at every other token and decides what matters.
16 min5 cards5 checks0%0/5 cards - STAGE 080%Locked
Pre-training & Loss
Next-token prediction, scaled until general intelligence falls out.
15 min5 cards5 checks0%0/5 cards - STAGE 090%Locked
Instruction Tuning (LoRA)
Turn a text completer into an assistant — without retraining the whole model.
16 min5 cards5 checks0%0/5 cards - STAGE 100%Locked
Alignment (RLHF / DPO)
Teach the model not just what is correct, but what people actually prefer.
16 min5 cards5 checks0%0/5 cards - STAGE 110%Locked
Inference & KV-Cache
What happens at request time: prefill, decode, sampling and VRAM.
17 min5 cards5 checks0%0/5 cards
Your progress
0 / 11stages complete
Nothing completed yet. Stage 1 is unlocked and free - the stages build on each other, but you can start anywhere.
Progress is saved to your Google account: sign in on any device and pick up where you left off.
Finish everything. Prove it with a certificate.
All 11 stages, readers and playgrounds are free. When every stage is complete you can generate a verifiable certificate with your name and a shareable card for LinkedIn.
Common questions
- Is the whole curriculum really free?
- Yes. All 11 stage readers, every micro-quiz and every interactive playground are free with no account required. Only the optional verifiable certificate is paid.
- Do I need maths or programming experience?
- No. The reader cards are written in plain language with one analogy per concept, and every formula is explained before it appears. The playgrounds let you feel each parameter instead of deriving it.
- How long does the track take?
- Roughly 3 hours of reading, plus time spent experimenting in the playgrounds. Each stage is short enough to complete in one sitting.
- What does the certificate prove?
- It records that you completed all 11 stages - readers, knowledge checks and simulations - and it is cryptographically signed and publicly verifiable through a shareable link.