Clawdemy Lessons

Orientation for the lesson on shipping multimodal AI: what you'll learn, the reasoning-stack prerequisite, learning outcomes, and time and difficulty.

Overview of agent security: the threat model, three attack categories and defenses, the defense-in-depth toolkit, and the Agents Rule of Two.

A roadmap to the Taylor series: why factorials matter, the e^x, sine, and cosine series, and how gradient descent and Newton's method are Taylor at work.

Overview of higher derivatives: notation, the second derivative as acceleration and curvature, the second-derivative test, and ML curvature uses.

A map of reinforcement learning fundamentals: how the agent-environment-reward loop works, what makes RL hard, and the exploration-exploitation tension.

An overview of value iteration: what you will learn, the prerequisites, the math involved, and how the update reappears later in Q-learning and DQN.

Overview of value functions and the Bellman equations: what V and Q are, how the one-step recursion is derived, and why optimal policies are greedy.

A brief on temporal-difference learning: the TD(0) update, a worked four-episode chain, the MC vs TD bias-variance tradeoff, and where TD powers modern RL.

Overview of the Q-learning lesson: what you will learn, prerequisites, the off-policy vs SARSA distinction, the DQN preview, and time and difficulty.

What policy iteration covers: the evaluate-then-improve loop, the improvement theorem, a worked MDP, prerequisites, and where it sits in RL planning.

Overview of the policy gradient capstone: what REINFORCE, actor-critic, PPO, and RLHF cover, prerequisites, the math involved, and time and difficulty.

Model-free policy evaluation explained: first-visit vs every-visit Monte Carlo, a 3-state worked example, and the unbiased but high-variance trade with TD.

A study brief for the Markov Decision Process lesson: what you will learn, where it fits in the track, prerequisites, and a note on the notation ahead.

An orientation to function approximation and deep RL: why tables fail, the linear-Q gradient step, the deadly triad, and how DQN's fixes stabilize learning.

Preview of the multimodal AI opener: learning outcomes, prerequisites, where the lesson fits, and the read and practice time to expect.

Diffusion image generation, in brief: learning outcomes, prerequisites, where DiT and MM-DiT fit, and read and practice time.

Brief on transformer video generation: what spacetime patches change, the two compression strategies, the captioned-data constraint, and failure modes.

Brief for the multimodal reasoning lesson: learning outcomes, prerequisites, scope, and the four-layer stack of perception, reasoning, tools, and alignment.

Preview of the native multimodal lesson: learning outcomes, prerequisites, and time needed to grasp how one transformer trains on all modalities at once.

What this lesson covers: applying the multimodal world model framing to drug discovery, plus how to tell ML benchmark claims from clinical claims.

Orientation for the JEPA lesson: what you will learn, prerequisites, how JEPA's embedding-space objection to generative pretraining sets up world modeling.

Overview of the encode-then-fuse vision-language recipe: the three pieces, CogVLM's visual expert, two-stage training, prerequisites, and time to read.

A survey brief mapping six directions the LLM field is moving, with learning outcomes, prerequisites, and how three of them set up the deeper lessons ahead.

The five core LUI UX patterns (streaming, citations, regeneration, hedging, recoverable failure), where the lesson fits, prerequisites, and supporting details.

Overview of when and how to fine-tune your own LLM: the three-part decision test, the staged pipeline, the tools, the economics, and learning outcomes.