Resources

The full video playlist for this week — recordings and demos, watch in order or jump to a session.

Step-by-step: install Ollama, pull a local model, and run a tiny fully-commented Python chat client that sets system prompts and streams replies. Runs offline, no API key — a clean demo of what a chatbot actually is.

The annotated chat client plus three example system-prompt personas, zipped. Standard-library Python only — nothing to pip install. (Setup guide linked above.)

Paste text and watch it split, across several model families. The fastest way to show a class that "cybersecurity" is not one token.

Same idea, side-by-side across tokenizers — good for showing that token counts differ per model for identical text.

The BPE tokenizer library itself. This is what the Day-2 guardrail labs pre-cache so TokenLimit works offline.

How BPE actually builds a vocabulary. The explainer to reach for when someone asks why rare strings tokenize badly.

Cost per million tokens, speed and quality compared across frontier models in one table.

Per-million-token input and output pricing, with the input/output asymmetry visible.

Gemini per-token rates, including the free tier limits.

Per-token rates for the GPT family. (Canonical source; blocks automated checks, so verify in a browser.)

Count tokens before you send a request — how to estimate cost and stay inside a context window.

The ~10-minute canonical explainer of LLM01 from the person who coined the term. Discuss: which of today's controls would blunt each of his examples?

The source document for the 40% domain — the exact numbering used all day. Discuss: which two risks are NEW in 2025 (LLM07, LLM08) and why did RAG make them necessary?

The ATT&CK-style matrix for AI. Pull up one case study and trace it across the tactics. Discuss: what tactic did the attacker open with, and where would you have broken the chain?

All-in-one desktop app: point it at documents and chat over them locally. The quickest way to show RAG to a class without building a stack.

The local model runtime used in every lab this week — serves qwen2.5 and nomic-embed-text on CPU.

The vector store behind the Day 1 RAG pipeline. Embeddings in, nearest-neighbour search out.

Desktop alternative to Ollama with a built-in model browser and chat UI.

Self-hosted chat front-end that speaks to Ollama, with document upload for RAG.

Python framework for the ingest -> chunk -> embed -> retrieve pipeline when you outgrow a hand-rolled script.

The other common framework for chaining retrieval and generation steps.

On-demand cloud GPUs for running larger models than your laptop can host — rent by the minute, no long-term commitment. Referral link. Plain sign-up: runpod.io.

The CLI coding agent used to build this course site. Referral link. Plain sign-up: claude.ai/code.