This is the first article in the “Bare minimum” series — a concise look at how AI works. Each piece will cover one idea or concept; I’ll try to keep them short and in order.

We’ll start with common terms; elsewhere in the series — prompt engineering, RAG systems, and LLM agents for complex multi-step tasks.

Video: Watch on YouTube

What a language model is

A language model is a system that, given what has already been said, predicts the next word. And so on in a loop:

The → The cat → The cat sat → The cat sat on → The cat sat on the → …

Filling its own context with words it has just produced.

Where we run into them

Any chatbot — whether it’s a support “assistant” that drives us crazy until we ask for a human, or “smart” chats like DeepSeek or ChatGPT — relies on a language model.

The only difference between a support bot and a “smart” one from DeepSeek is scale: the scale of the dataset and the server capacity needed to process it.

Parameter scale

That leads to parameter scale:

• 1–7 billion — models you can run on a local laptop
• Trillions — models that need server clusters

Models in the trillion-parameter range are called foundation models: they have absorbed not just statistics of word sequences, but knowledge encoded in language. Language is not only a medium for communication but also a record of collective experience.

AI, ML, and neural networks — what’s the difference

AI (artificial intelligence) is the broadest term: systems and programs that behave “intelligently” — they solve tasks that usually need human intelligence (speech understanding, chess, image recognition, decision-making, and so on). AI can be built without machine learning (e.g. rules, expert systems).

ML (machine learning) is a subset of AI: a way to build AI so the system learns from data instead of hand-written rules. The goal is to find patterns in examples (data) and use them for predictions or decisions. ML includes more than neural networks: decision trees, linear models, clustering, and others.

NNs (neural networks) are a subset of ML: models inspired by neurons in the brain (layers, weights, activations). One of the most powerful ML tools, especially for images, text, and speech. Deep learning is ML with deep (many-layer) neural networks.

The relationship: AI ⊃ ML ⊃ NN — neural networks are a kind of machine learning, and machine learning is one way to implement artificial intelligence.

Practice what you learned

Short games help you check how well you’ve absorbed the material.

Quiz:

Sort the shelf — drag statements onto the right shelves (AI, ML, NN, parameter scale):