From a single token,
to a working agent.
A complete, cinematic course on how generative AI actually works. No heavy mathematics — every idea is built from clear analogies, animated visuals and a narration that takes it one gentle step at a time. We start at the smallest thing a model ever sees, and finish with a working, reliable product.
Foundations — What Generative AI Actually Is
Prediction, not magic: what these models are, how they differ from every AI that came before, and why 2022 changed.
- The map — token to application
- Generation vs retrieval
- Discriminative vs generative
- The three pillars
- Why 2022, and not 1995
Inside the Model — Tokens, Meaning, and Attention
Text becomes tokens, tokens become geometry, attention decides what matters — and the Transformer falls out.
- Tokens — the atoms of language
- Embeddings — meaning as geometry
- Attention — the secret sauce
- The Transformer, assembled
How Models Learn — and Why They Lie
Pretraining, fine-tuning and RLHF; the sampling dials you actually turn; and the real mechanism behind hallucination.
- Pretraining — reading the internet
- Fine-tuning and RLHF
- Sampling — temperature, top-p, top-k
- Hallucination, honestly
Beyond Text — Images, Sound, and Everything Else
Diffusion from pure noise, multimodal models that see and hear, and generation across code, speech, music and video.
- Diffusion — images from noise
- Multimodal — eyes, ears, voice
- Code, speech, music, video
Building Real Systems — Prompting, RAG, Agents, Production
Prompting that survives contact with reality, retrieval as memory, tool-calling agents, and the cost/latency/eval/safety bill that comes due in production.
- Prompting that actually works
- RAG — giving models a memory
- Tools, function calling, agents
- Cost, latency, evaluation, safety