| id | generative-ai-and-llms | ||||||||||||||||||||||||||||||||||||||||||||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| title | Generative AI and language models | ||||||||||||||||||||||||||||||||||||||||||||||||||||
| order | 1 | ||||||||||||||||||||||||||||||||||||||||||||||||||||
| discoveryCycle | true | ||||||||||||||||||||||||||||||||||||||||||||||||||||
| checkpoints |
|
||||||||||||||||||||||||||||||||||||||||||||||||||||
| objectives |
|
Before looking inside a language model, put the common terms in the right relationship. They overlap, but none is a replacement name for all the others.
Artificial intelligence (AI) is the broad goal of making computer systems perform tasks that seem to require perception, language, prediction, reasoning, or decision-making. An AI system can use explicit rules, learned models, search, or several techniques together.
Machine learning (ML) is one way to build AI behaviour. Instead of programming every decision, we give a model adjustable values called parameters and use data to find useful settings.
Generative AI describes learned systems that produce new content such as text, images, audio, or code. A photo classifier can use machine learning without being generative: it chooses a label for an existing image.
A language model assigns probabilities to sequences of language. A modern large language model (LLM) has many learned parameters and is trained on a large amount of text or multimodal data. “Large” describes scale, not a new kind of objective.
A chatbot product is not only an LLM. It normally combines a model with a conversation format, system instructions, safety policy, tools, retrieval, memory supplied in the prompt, and a user interface. The model generates tokens; the surrounding application decides what context and capabilities the model receives.
Predict: Is a learned photo classifier generative AI? Is a chat product identical to the LLM inside it? State which boundary makes each answer true or false.
:::explorable{src="../explorables/generative-ai-map/index.ts" title="AI, machine learning, generative AI, and LLM term map" height="430" id="generative-ai-map"} Choose a rule-based system, learned classifier, image generator, or LLM chat product. The panel shows whether each is learned, generative, language-focused, and a complete product system. :::
The examples form overlapping sets:
- a rule-based spell checker can be an AI system without machine learning;
- a learned photo classifier is ML but does not generate new content;
- an image generator is generative ML but is not a language model; and
- a chat application may use an LLM while still having important non-model components.
The last boundary matters when debugging. A wrong answer might come from model behaviour, missing context, a retrieval error, a tool result, decoding settings, or application instructions. Calling the whole product “the model” hides those different responsibilities.
Take a coding-agent chat product as an example. With your tutor, name one responsibility belonging to the model and three belonging to the surrounding product. Then choose which component you would inspect if the system read the wrong file but produced fluent text about it.
The phrase “generative AI understands everything on the internet” collapses several claims. A model learns statistical structure from its training process; it does not carry a verified database of every source, and a chat product may add current information through tools or retrieval. Fluency is not proof of factual access or correctness.
You are ready to continue when you can draw the nested relationship and explain why a model, its training process, and the product that calls it are three different things.