100 Things to Know About AI
The vocabulary of modern AI, explained one entry at a time — each a single idea, stated plainly enough to quote and precisely enough to trust.
This is a reference, not a course. Read it in order or arrive at a single entry from a search; each stands on its own. Every entry opens with a one-sentence answer, then earns that sentence in the paragraphs below.
- 1
Artificial intelligence is software that learns from data instead of following rules a person wrote.
Artificial intelligence is software that learns to perform a task by finding patterns in data, rather than following step-by-step instructions written by a programmer. Today's most capable AI systems are large language models, which learn from vast amounts of text.
- 2
Machine learning is how most AI is built today.Aug 4
Machine learning is the practice of getting software to learn a task from examples instead of explicit rules, and it underpins nearly every modern AI system.
- 3
A large language model predicts the next token from everything before it.Aug 11
A large language model is a neural network trained to predict the next unit of text; fluent writing, reasoning, and answers all emerge from that one objective at scale.
- 4
A token is the unit of text a model actually reads.Aug 18
Models read tokens — sub-word fragments — not words or letters, which is why cost, context limits, and speed are all counted in tokens.
- 5
The context window is everything a model can hold in mind at once.Aug 25
The context window is the fixed span of tokens a model can attend to in one request; anything outside it is invisible unless you retrieve and re-supply it.
- 6
Retrieval-augmented generation feeds a model facts it was never trained on.Sep 1
RAG retrieves relevant documents at query time and places them in the context window, letting a model answer from current or proprietary sources.
- 7
A hallucination is a confident answer with no basis in fact.Sep 8
A hallucination is fluent, well-formed output that is simply untrue; it happens because a model optimises for plausible next tokens, not verified truth.
- 8
An embedding turns meaning into a list of numbers.Sep 15
An embedding maps text to a vector so that similar meanings sit close together, which is what makes semantic search and retrieval possible.
- 9
Fine-tuning adjusts a trained model's weights on your own examples.Sep 22
Fine-tuning continues training a base model on a smaller, task-specific dataset so it adopts a style, format, or domain — changing the model itself.
- 10
An agent is a model given tools and a goal to pursue on its own.Sep 29
An agent wraps a model in a loop where it can call tools, observe results, and decide the next step, turning one-shot answers into multi-step work.
- …
90 more, publishing weekly.
Common questions
- Do I need a technical background to read this?
- No. Every entry is written to be understood by a decision-maker, not a machine-learning engineer, while staying accurate enough that an engineer would not object.
- How often does a new entry appear?
- Roughly one a week. The full list of a hundred is fixed from the start; each entry becomes a link on the morning it publishes.