Inside AI · Three interactive explorations

Better questions about AI.

How does it learn? Who pays for the answer? What makes it worth trusting? A few useful ideas, a few things to try, and no exam at the end.

  1. 01 · How it learns

    Gradient descent

    How does a model turn error into a better guess?

    Meet the weights and the representations they create. Then take one step on a real 3D loss surface—and try taking too large a step.

  2. 02 · What it costs

    Compute economics

    When does useful AI become a business that can pay for its machines?

    Sell more hours, or charge less for them. See why a busy machine can still lose money—and why customer value is a different question.

  3. 03 · What earns trust

    Proof & prediction

    What would make you trust this answer?

    An email, an invoice, a source, a rule. Explore what an LLM, a calculator, and a proof can each do for an ordinary AI user.

A little context. Something to try. An idea to keep.

Each exploration starts with the point, not a control panel. Worked examples and interactive experiments let you test the intuition. Equations, full tables, and dated sources stay one layer deeper. No sign-in or AI service is needed. Text, diagrams, and default results remain available without JavaScript or WebGL.

These are small, explicitly limited examples—not a claim to have simulated the world. Second edition: . Source dates and checks are listed with each article.