TwiddleLab / Courses / Artificial Intelligence
Generative AI: Making New Things
From noise to a picture: how machines make new things.
Best after: How Language Models Work, How Machines See and Hear
Lessons
- Generating vs Classifying (5 min): A classifier only needs the border between kinds: it says what a thing is.
- Autoencoders: Squeeze and Rebuild (7 min): An autoencoder has an encoder that squeezes the input into a short code and a decoder that rebuilds it.
- Latent Space Walks (7 min): Every code is a point in the latent space, and the decoder turns every point into a critter.
- GANs: Forger vs Detective (7 min): A GAN trains two networks against each other: a generator (forger) and a discriminator (detective).
- Add Noise, Learn to Undo (7 min): Forward noise buries a picture step by step: noisy = a · clean + b · noise, with a shrinking and b growing.
- From Pure Noise to Picture (7 min): Diffusion sampling starts from pure noise. At every step the denoiser guesses the noise and part of it is removed.
- Text to Image (7 min): Captions and pictures are trained to land near each other on one shared map, a text-image embedding.
- Music and Voice Generation (7 min): Sound is cut into short slices, audio tokens, often 50 per second.
- Copyright and Consent (7 min): "Is it allowed?" splits into separate questions: may it be trained on, may the output be shared, who is the author. Different people decide each one.
- Can We Detect Generated Content? (7 min): An AI detector gives a score, and honest and AI work overlap. Catching more fakes always means accusing more honest people, as with precision and recall.
- Project: Design a Critter Generator (8 min): A generator learns the cloud of its training data; directions in its code map let you steer one feature at a time.
Lessons are interactive and need JavaScript. Turn it on to play them.
