Module 2: How These Tools Work
LLM mechanics, context windows, hallucinations, sycophancy, and the landscape of AI coding tools.
Slides
Outline
- How LLMs generate code: next-token prediction and why it produces both useful code and confident nonsense
- Context windows and attention: what happens as conversations get long
- Hallucination patterns specific to code: fabricated APIs, invented packages, plausible-but-wrong algorithms
- Sycophancy and user-framing bias: why the model tends to agree with you
- Landscape of AI coding tools: autocomplete, chat interfaces, agentic CLI harnesses, autonomous agents
Learning Objectives
- Explain, at a practical level, how LLMs generate text and why this produces both useful code and confident nonsense
- Describe what a context window is, why conversations degrade over length, and what “lost in the middle” means for long coding sessions
- Identify common code-specific hallucination patterns: fabricated APIs, invented library functions, plausible but incorrect algorithms, and fake package names
- Map the current landscape of AI coding tools (autocomplete, chat, agentic) and understand the tradeoffs between them