HOBBYIST
Make something useful tonight
- Prompt basics
- Image + text workflows
- Personal projects
00 The illustrated field manual
Understand LLMs.
Choose the right one.
Put it to work.
A clear, independent field guide to large language models, without the hype.
No jargon requiredBuilt for real decisionsUpdated as the field changes
01 / INPUT03 / CONTEXT05 / OUTPUTROUTING CONSOLE / SELECT A PATH
What are you trying to do?
SAMPLE INPUT
Turn these rough notes into a clear, friendly project update.
EXPECTED OUTPUT
Start with prompting →A concise update with decisions, progress, and next steps.
CHAPTER 01THE 5-MINUTE GUIDE
01 Start here
Five things that make the rest make sense.
You do not need to understand neural-network mathematics to use an LLM well. You do need a useful mental model for what it can and cannot do.
- 01CORE IDEA
It predicts language
An LLM builds a response piece by piece from patterns learned during training. It does not retrieve a single stored answer from a database.
- 02INPUT
Context changes the answer
Your instructions, examples, source material, and conversation history shape what the model can produce right now.
- 03CAUTION
Fluency is not truth
A confident sentence may still be wrong. Verification matters whenever facts, money, safety, reputation, or people are involved.
- 04CHOICE
Different models fit different jobs
Quality, speed, price, privacy, context size, tool use, and deployment options all matter. There is no permanent universal winner.
- 05PRACTICE
Evaluation beats intuition
Test models with examples from your actual work. A small repeatable test set is more useful than a leaderboard alone.
02 Under the cover
One prompt.
Five useful stages.
This is a simplified map, not a literal diagram of every model. It is enough to reason about prompting, retrieval, and output quality.
01INPUT
Instructions + material
→02TOKENS
Text becomes pieces
→03CONTEXT
Relevant information
→04PATTERN
Relationships weighed
→05OUTPUT
Pieces become an answer
CHAPTER 02CHOOSE BY FIT
03 Model fit finder
Skip the model horse race.
The “best” model depends on what you are doing and what you value. Start with the shape of the job, then test named models against it.
QUALITYSPEEDCOSTPRIVACYCONTROL
CHAPTER 03PUT IT TO WORK
04 Prompt workbench
Give the model something to work with.
Good prompts are not magic spells. They reduce ambiguity by making the goal, audience, material, constraints, and quality bar visible.
PROMPT ASSEMBLY / FIVE PARTS
01
Role
What perspective or expertise is useful?
02
Goal
What should the result accomplish?
03
Context
What information should it use?
04
Constraints
What format, length, or boundaries apply?
05
Quality check
How should it inspect its own answer?
Token count is a rough estimate and varies by model.
Before sending:Is the goal clear?Did you include the source material?Do you know what must be verified?
05 Choose your route
You do not need to learn everything.
Follow the route that matches what you want to make.
BUILDER
Turn a model into a product
- APIs + structured output
- Retrieval
- Evaluation
TECHNICAL
Understand the machinery
- Tokens + context
- Open weights
- Fine-tuning
TEAM
Adopt it without chaos
- Use-case selection
- Risk controls
- Shared playbooks
CHAPTER 04TRANSLATE THE JARGON
06 Plain-language glossary
Know the term.
Keep moving.
Agent
A system that lets a model choose and perform steps toward a goal, often using tools.
Context window
The amount of information a model can consider in one interaction.
Embedding
A numerical representation used to compare meaning and retrieve related information.
Hallucination
A plausible-sounding response that is unsupported, incorrect, or invented.
Inference
The process of running a trained model to produce an output.
Open-weight
A model whose learned parameters are available under a specified license.
Prompt
The instructions and information provided to a model.
RAG
Retrieval-augmented generation: finding relevant material and supplying it as context.
Temperature
A setting that influences how predictable or varied generated text may be.
Token
A piece of text processed by a model. A word may be one token or several.
Transformer
The neural-network architecture behind most modern language models.
Tool use
A model requesting an external action such as searching, calculating, or calling an API.
07 Your next move
Pick one real task.
Run one small experiment.
Start with work you can inspect. Keep what helps. Measure what matters. Add autonomy only after the simple version is trustworthy.