Series · 5 parts · 15 min

How AI is made

Most explanations of AI describe what a model does when it answers you. This series goes back a step, to how the model was made. It starts with pre-training, where a model reads a vast amount of text and learns by guessing the next word, then moves to the people who shape it into an assistant, and the labelled data and numbers that come out of it all.

Watch them in order. Each stands on its own, but the steps build.

Start the series →
  1. The video's title card: the label How AI is made above the headline Nobody taught it to write, and in blue, It learnt by guessing. 3:30
    Video Part 1

    How AI learns from text

    Before an AI model can answer you, it reads a vast amount of text and sits one test, guessing the next word, trillions of times. What goes in, what gets thrown out, who marks the guesses, what the model's numbers are, what it costs, and why there's no text inside it at all.

  2. The video's title card: the label How AI is made above the headline Reading isn't helping, and in blue, People teach it to help. 2:55
    Video Part 2

    How AI learns to help

    A model that has only read the internet can continue text but can't yet help you. How people turn it into an assistant, by writing good answers, ranking its own, and training a second model to score them the way they would. Reinforcement learning from human feedback, in plain words, and whose taste it learns.

  3. The video's title card: the label How AI is made above the headline Someone has to label it, and in blue, Now an AI goes first. 2:28
    Video Part 3

    How AI helps label data

    Training data often needs a person to say what each example is. Now an AI labels first, a check scores how far to trust each label, and people settle the hard cases. Where labels fit in making a model, what it saves, and why the AI's own confidence isn't enough.

  4. The video's title card: the label How AI is made above the headline 8B. 70B. 405B. and, in blue, What the number tells you. 2:56
    Video Part 4

    What the B in 8B means

    AI models come in sizes written like 8B, 70B or 405B. The B is billions of parameters, the numbers training sets. What that number tells you about the memory and hardware a model needs, how quantisation shrinks it, why bigger isn't always better, and why some huge models only use a slice at a time.

  5. The video's title card: the label How AI is made above the headline Open weights and, in blue, What you get, and why it matters. 3:05
    Video Part 5

    What open weights means

    Some AI makers let anyone download their models. What's in the download is the weights, the numbers set during training, usually without the data or code behind them and always under a licence. Who releases them and why, what open weights is and isn't, why it matters to you, and the risk on the other side.

Want a video on something?

Tell me what you're stuck on. I take requests, and the good ones end up in here.

Ask for a video Subscribe on YouTube Follow on LinkedIn