# AI with Leo

> An interactive textbook on deep learning and large language models by Prof. Léonard Boussioux, with hands-on playgrounds at every step.

AI with Leo covers the full LLM stack — from tokenization and attention to training, alignment, multimodality, agents, evaluation, and societal impact — plus companion tracks on deep learning fundamentals and classical data analytics.

## Pages

- [Home](/): Overview of the textbook, all chapters, and progress tracking.
- [Glossary](/glossary): Searchable index of every key term across all chapters.

## Chapters

- [Overview: What Is a Language Model?](/chapters/llm-overview): Language models, autoregressive generation, GPT, and a roadmap of what you'll learn
- [The Rise of LLMs: A Historical Landscape](/chapters/historical-landscape): From GPT-1 to GPT-5, the ChatGPT moment, major players, and key milestones
- [Text to Tokens](/chapters/tokenization): How text becomes numbers: BPE, WordPiece, and subword tokenization
- [Embeddings & Positional Encoding](/chapters/embeddings): Representing meaning in high-dimensional space
- [The Transformer Architecture](/chapters/transformer): The building blocks: LayerNorm, FFN, residual connections
- [The Attention Mechanism](/chapters/attention): Queries, keys, values, and scaled dot-product attention
- [The Autoregressive Process](/chapters/autoregressive): Generating text one token at a time
- [Decoding Strategies](/chapters/decoding): Greedy, top-k, top-p, temperature, and beam search
- [Training: Pre-training & Fine-tuning](/chapters/training): From next-token prediction to RLHF
- [Advanced LLM Training Techniques](/chapters/training-advanced): Quantization, distillation, QLoRA, synthetic data loops, and reasoning-focused RL
- [Reasoning Models](/chapters/reasoning): Chain-of-thought, process reward models, and test-time compute
- [Mixture of Experts](/chapters/moe): Conditional computation: big models at small-model cost
- [Prompt Engineering Fundamentals](/chapters/prompt-engineering): Zero-shot, few-shot, chain-of-thought, system prompts, and core prompting best practices
- [Advanced Prompting & Production Patterns](/chapters/advanced-prompting): Self-consistency, ReAct, structured outputs, prompt security, workflows vs agents, and production patterns
- [Retrieval-Augmented Generation](/chapters/rag): Grounding LLMs in external knowledge with vector search and retrieval
- [Embedding Spaces, UMAP & Clustering](/chapters/embedding-spaces): Visualizing and clustering high-dimensional representations with UMAP and HDBSCAN
- [Agentic AI](/chapters/agentic): Tool use, memory, orchestration, sub-agents, and MCP
- [Multimodal Models](/chapters/multimodality): Vision, audio, video — unifying modalities in a single transformer
- [Image Models](/chapters/image-models): Diffusion, flow matching, world models, and modern image generation
- [LLM Evaluation](/chapters/evaluations): Benchmarks, metrics, LLM-as-judge, arena evaluation, and contamination
- [Alignment, Safety & Attacks](/chapters/alignment): RLHF, DPO, jailbreaking, prompt injection, and the open problems of AI safety
- [Hallucinations & the Limits of Understanding](/chapters/hallucinations): Why LLMs fabricate, contextual understanding gaps, detection, and mitigation
- [Search, Discovery & Agentic Optimization](/chapters/seo-agentic): Generative engine optimization, AI-mediated search, and content strategy for the AI era
- [Human-AI Collaboration & Cognitive Augmentation](/chapters/human-ai-collaboration): Cognitive offloading, delegation, the centaur model, and designing for human agency
- [AI & Environmental Sustainability](/chapters/sustainability): Energy, water, carbon footprint — the environmental cost of training and running LLMs
- [The Politics & Economics of AI](/chapters/politics-economics): Power concentration, open vs closed source, regulation, the OpenAI saga, and the AI arms race
- [Ethics, Bias, Fairness & Law](/chapters/ethics): Bias sources and mitigation, fairness frameworks, copyright, sustainability, and responsible deployment
- [A Brief History of AI](/chapters/deep-learning-history): Interactive timeline of landmark AI moments, from Dartmouth to Claude Code and ClawdBot.
- [The Perceptron](/chapters/deep-learning-perceptron): Interactive perceptron controls with weights, bias, and decision boundary intuition.
- [Activation Functions](/chapters/deep-learning-activation-functions): Visual comparisons of sigmoid, tanh, ReLU-family, and derivative behavior.
- [Multilayer Perceptron](/chapters/deep-learning-mlp): XOR-focused MLP playground with training and live network state.
- [Loss Functions](/chapters/deep-learning-loss-functions): Hands-on MSE and cross-entropy visualizations with editable predictions.
- [The Loss Landscape](/chapters/deep-learning-loss-landscape): Interactive convex/non-convex landscape simulation with trajectory playback.
- [Gradient Descent](/chapters/deep-learning-gradient-descent): Faithful GD, momentum, and Adam trajectories on analytic surfaces with exact gradients.
- [Backpropagation](/chapters/deep-learning-backpropagation): Single-neuron forward/backward pass walkthrough with interactive derivatives.
- [Convolutional Neural Networks](/chapters/deep-learning-cnn): Interactive convolution chapter covering kernels, stride, and feature maps.
- [Recurrent Neural Networks](/chapters/deep-learning-rnn): Hands-on RNN state updates with memory-retention intuition and gating context.
- [Deep Learning & Beyond](/chapters/deep-learning-beyond): Dropout and overfitting intuition with modern deep learning context.
- [Neural Network Builder Lab](/chapters/deep-learning-lab): Left-panel builder with live code generation, architecture breakdown, and training simulation.
- [Linear Regression](/chapters/data-analytics-linear-regression): Interactive linear modeling chapter with regularization and coefficient paths.
- [Logistic Regression](/chapters/data-analytics-logistic-regression): Interactive binary classification with cross-entropy, probabilities, and thresholds.
- [Classification Metrics and Imbalance](/chapters/data-analytics-classification): Interactive classification diagnostics with confusion matrix, ROC/PR curves, and imbalance tradeoffs.
- [CART: Decision Trees](/chapters/data-analytics-cart): Interactive CART visualization with depth controls and split interpretation.
- [Random Forest](/chapters/data-analytics-random-forest): Bootstrap sampling, ensemble voting, and overfitting resistance in practice.
- [XGBoost](/chapters/data-analytics-xgboost): Residual-driven boosting dynamics and confidence evolution over steps.
- [LightGBM](/chapters/data-analytics-lightgbm): LightGBM internals with histogram bins, leaf-wise growth, and speed tradeoffs.
- [Model Comparison](/chapters/data-analytics-compare): Switch between classification and regression mode to compare model families.
- [Clustering](/chapters/data-analytics-clustering): Interactive clustering playground to contrast centroid and density-based methods.
- [Time Series Forecasting](/chapters/data-analytics-time-series-forecasting): Interactive forecast simulator with trend, seasonality, and holdout-error comparisons.