abxda academy

Generative AI: from neuron to chatbot

A complete, free course. Every lesson builds up step by step, with voice, animation, and pieces you can move so you really understand.

Start lesson 1Executive path3 h 55 min
  • 37 lessons
  • 6 h 57 min
  • Free, no sign-up
  • English and Spanish
  • Voice and captions
From the academy

Is this course useful? Support the academy It is free and will stay free.

How each lesson works

  1. One sentence, one change on screen

    Narration and image move together. Nothing is shown before it is explained.

  2. Your turn

    At key moments the lesson pauses so you can drag, tweak, or predict.

  3. Two depths

    The essentials are enough to understand. “Under the hood” adds the math and code for those who want it.

  4. Decision lens

    At the end, what the idea means for whoever decides: costs, risks, and what to ask.

The journey

Nine modules. Each builds on the last: nothing appears before it has been explained.

Part 1

How a machine learns

Available

From data to deep networks: error, neurons, backpropagation, vision, sequences, and meaning as geometry.

  1. Learning from data

    What it means for a machine to learn, how error is measured, and how it gets corrected.

    1. 01Programming or learningWhat changes when a machine learns instead of following rules?12 min
    2. 02The world as numbersHow does a machine see an image, a sound, or a text?10 min
    3. 03A model and its errorHow do you measure how wrong a model is?11 min
    4. 04Gradient descentHow does a machine find the right numbers without being told?11 min

  2. Neural networks

    From one neuron to a deep network: layers, backpropagation, and the craft of training well.

    1. 05The artificial neuronWhat exactly does an artificial neuron do?11 min
    2. 06What a neuron cannot doWhy is a single neuron not enough?6 min
    3. 07Layers and activationsHow do networks gain power by stacking layers?10 min
    4. 08BackpropagationHow is the blame for an error shared among thousands of weights?9 min
    5. 09Training well: batches, epochs, and optimizersWhich training decisions separate a network that learns from one that does not?12 min
    6. 10Generalizing and evaluatingHow do you know whether a model really works?9 min

  3. Deep learning

    Vision, sequences, GPUs, and meaning represented as geometry.

    1. 11Convolutions and visionHow does a network learn to see edges, shapes, and objects?6 min
    2. 12Sequences and memoryHow does a network process something that happens in order, like a sentence?10 min
    3. 13The deep learning big bangWhy did in 2012 what did not work in 1990?10 min
    4. 14Embeddings: meaning as geometryHow does a machine turn meaning into distances?11 min
Part 2

Generating: images and language

Available

Autoencoders, GANs, and diffusion; then tokens, attention, the transformer, scale, and how a model answers.

  1. Generating images

    Learning a distribution so you can sample from it: VAEs, GANs, and diffusion.

    1. 15Generating is learning a distributionWhat does it mean for a network to "generate" something new?10 min
    2. 16GANs: the forger and the detectiveHow does a generator learn by competing against a critic?11 min
    3. 17Diffusion: from noise to imageHow is an image created by removing noise step by step?12 min
    4. 18From text to imageHow does the model understand what you ask for in words?13 min

  2. Transformers and language

    Predicting the next word, splitting text into tokens, and looking with attention.

    1. 19Predicting the next wordHow can something that only predicts words write an essay?11 min
    2. 20Tokens: the model’s alphabetWhat does a language model actually read, and why does it matter?11 min
    3. 21AttentionHow does a word decide which other words to look at?12 min
    4. 22The complete transformerWhat is inside an LLM, block by block?12 min

  3. Scale

    Pretraining, scaling, paying the compute bill, and serving the model.

    1. 23PretrainingWhat does a model learn from reading half the internet?13 min
    2. 24Scaling laws and costsHow much does a model cost and what do you get by making it bigger?13 min
    3. 25GPUs, memory, and quantizationWhy does GPU memory rule over everything else?13 min
    4. 26Inference: how a model answersWhat happens, and what does it cost, between typing and the answer appearing?13 min
Part 3

From model to assistant to decision

Available

Fine-tuning, alignment, reasoning, memory, RAG, and agents; then hallucinations, evaluation, risks, and strategy.

  1. From model to assistant

    Fine-tuning, alignment, reasoning, memory, retrieval, and agents.

    1. 27Fine-tuning and LoRAWhen should you adapt a model, and how is it done cheaply?11 min
    2. 28From base model to assistantWhat turns a text predictor into a useful assistant?14 min
    3. 29Reasoning modelsWhy does thinking longer before answering improve the answer?12 min
    4. 30Context and memoryDoes the model remember what you said? Does it learn while you chat?12 min
    5. 31RAG: read before answeringHow do you give a model your own knowledge without retraining it?13 min
    6. 32Agents and toolsWhat changes when a model can act, not just talk?13 min

  2. Using AI with judgment

    Hallucinations, evaluation, risks, and how to decide where to apply it.

    1. 33Hallucinations and limitsWhy is it wrong with such confidence, and what can be done?12 min
    2. 34Evaluating AI systemsHow do you check that an AI system works for your case?12 min
    3. 35Risks and governanceWhat can go wrong, and who is accountable when it does?13 min
    4. 36Strategy for organizationsWhere to apply generative AI in your organization, and how to start?12 min

  3. Wrap-up

    The complete map, from neuron to chatbot.

    1. 37The complete mapHow do all the pieces fit together, from neuron to chatbot?11 min