Module 6 · GenAI · transformers · LLMs · agents · ~33 min · Engine AI-v6

The machine
that talks.

Modern AI, opened all the way up — and grounded in real telecom work. How text becomes numbers (tokenization), how words gain meaning (embeddings), the attention mechanism with its query·key·value arithmetic worked by hand, the transformer block, how a model learns the whole internet by playing one game (next-token prediction), how it is aligned into an assistant, then the practical layer: sampling, prompting, RAG, and AI agents that use tools — with every idea earning its keep in the network operations center.

▶  Play the module ✦  Open the playground
16chapters · ~33 min
Q·K·Vattention, worked
24animated scenes
2live playgrounds
Watch · Module 6 · ~33 min

From a single weighted sum to a system that reads your specs

The stage animates every idea as it is spoken: text splitting into tokens, tokens becoming vectors of meaning, the word “its” reaching across a sentence to “cell” through attention, multi-head attention weaving many relationships, the transformer block with its residual shortcuts, next-token prediction at internet scale, the RAG retrieval loop, and an AI agent calling tools to investigate a degradation. English karaoke subtitles.

output · live render — Module 6 · Generative AI, LLMs & Agentscue-synced engine
CT-AITEL-1000 · Module 6 · Generative AI, LLMs & Agents
Generative AI, LLMs & Agents — the machine that talks
0:00 / 0:00
Playground · touch the mechanism

Run attention on a sentence — then tune the sampler

Pick a word and watch attention compute its query·key scores against every other word and blend their values — the exact library-search the video works by hand. Then move the temperature dial and watch the next-token distribution go from focused to creative.

⭐ attention on a sentence — Q · K · Vthe library search
click a word below — watch its attention
Each word publishes a query (what it seeks), a key (what it offers) and a value. The query meets every key, softmax turns the scores into weights, and the word rebuilds itself from the weighted values.
⭐ the sampler — temperaturefocused ↔ creative
0.7
Reference · the modern-AI stack

Fifteen cards, one concept each

Tokenization, embeddings, attention, the transformer block, pretraining, alignment, sampling, context, prompting, RAG, agents — what each is, and the trap that bites. Filter as you revise.

Reference · the vocabulary

Every term of generative AI, searchable

From the token to the AI agent — the working vocabulary of transformers and LLMs, each with its meaning and, where it helps, its formula.

termfamilyformula / originwhat it means
Mastery check

The Module 6 mastery check

Ten questions from a pool of nineteen, reshuffled every load. Pass at 70%+ — and you hold the whole arc of modern AI, from a neuron to an agent.

Q 1 / 10
0 ✓