Latent

Your progress

Progress is stored in this browser only. Clearing site data or switching browsers starts fresh. Questions you have answered come back for spaced review.

AI from zero
  1. 1What is AI, really?
  2. 2Turning things into numbers
  3. 3Getting less wrong
  4. 4The artificial neuron
  5. 5Networks of neurons
  6. 6How a chatbot writes
  7. 7What AI gets wrong
  8. 8AI that takes actions
  9. 9Where your AI runs
  10. 10Safety in AI products
  11. 11Open and closed models
  12. 12How AI sees, hears, and makes pictures
  13. 13When someone attacks an AI app
How machines learn
  1. 1The learning loop
  2. 2Linear regression and loss
  3. 3Gradient descent
  4. 4Better optimizers
  5. 5Probability and Bayes' rule
  6. 6Classification and probability
  7. 7Overfitting and generalization
  8. 8Decision trees
  9. 9Ensembles: forests and boosting
  10. 10Measuring a classifier
  11. 11Finding groups: k-means
Neural networks
  1. 1Layers are matrix multiplications
  2. 2Why nonlinearity matters
  3. 3Backpropagation
  4. 4Training a network
  5. 5Convolutions
  6. 6Reading handwriting
  7. 7Autoencoders and latent spaces
Language as vectors
  1. 1Tokens
  2. 2Vectors and similarity
  3. 3Word embeddings
  4. 4Seeing high dimensions
  5. 5Search by meaning
  6. 6Retrieval-augmented generation
  7. 7Reading an embedding
  8. 8How your feed decides: recommendation systems
Transformers and LLMs
  1. 1Predicting the next token
  2. 2Before transformers: recurrent networks
  3. 3Attention, step by step
  4. 4Many heads and the causal mask
  5. 5Where words are
  6. 6The transformer block
  7. 7Inside a real language model
  8. 8How LLMs are trained
  9. 9Where training data comes from
  10. 10Fine-tuning and LoRA
  11. 11Generating fast: the KV cache
  12. 12Quantization: smaller numbers
  13. 13How a chat assistant works
Learning by trial and error
  1. 1Explore or exploit: bandits
  2. 2States, actions, and rewards
  3. 3Learning from experience: Q-learning
  4. 4Policy gradients
  5. 5Rewarding LLMs, and reward hacking
Inside the model
  1. 1Reading a model layer by layer
  2. 2How circuits form: induction heads
  3. 3Superposition: more features than neurons
  4. 4Switching parts off and steering
  5. 5Alignment: getting the goal right
  6. 6How alignment is attempted
Models that see
  1. 1How computers see
  2. 2Recognizing objects with your camera
  3. 3Images and words together
  4. 4Depth from a single image
  5. 5Where the classifier looks
  6. 6Finding anything you can name
  7. 7Cutting objects out
  8. 8How a model reads a picture
  9. 9Finding a body's pose
Models that hear and speak
  1. 1How computers hear
  2. 2What sound is this?
  3. 3Speech to text
  4. 4Text to speech
  5. 5A voice assistant taken apart
Frontiers
  1. 1Diffusion: from noise to data
  2. 2Diffusion on images
  3. 3Pictures, one token at a time
  4. 4One space for images and text
  5. 5Mixture of experts
  6. 6Thinking longer
  7. 7Models that use tools
  8. 8Evaluation: what a score means
  9. 9Inside an agent harness
  10. 10Compute and energy
  11. 11Long context in practice
  12. 12Where multimodal models are heading
Securing AI
  1. 1Attacks on AI systems
  2. 2Prompt injection
  3. 3Poisoned retrieval and memory
  4. 4Agents with too much power
  5. 5Poisoned data and the supply chain
  6. 6Adversarial examples

Try "embedding", "softmax", "overfitting", or "backpropagation".