Volume 8 — Large Language Models¶
What they are and what they are not
Tokenisation, pretraining, fine-tuning, alignment, inference, context windows, prompting and evaluation — plus an honest treatment of failure modes.
Outlined, not yet written
This volume is fully outlined but the lessons are not yet authored. The chapter list below is the plan of record, not a promise that content exists.
What you will be able to do¶
- Explain what a token is and why it drives cost and behaviour
- Design a prompt that is testable rather than lucky
- Build an evaluation harness for a non-deterministic system
- Reason about when an LLM is the wrong tool
Chapters¶
- Tokenisation
- Pretraining and the objective
- Fine-tuning and instruction tuning
- RLHF and alignment
- Inference, sampling and temperature
- Context windows and attention cost
- Prompt engineering as an engineering discipline
- Structured output and tool use
- Evaluating LLM systems
- Hallucination, and what actually reduces it
- Cost and latency engineering