Free lesson · Module 1, lesson 6 of 6

The words everyone uses: model, parameters, training, inference, GPT

Decode the words in every AI headline (model, parameters, training, inference, prompt, GPT, open and closed models) with one everyday picture each, and read a model announcement without being lost.

The video for this lesson is being prepared

About 9 minutes, narrated in English, Hindi and Odia. The notes and the quiz below are open now.

Lesson notes

For the student

AI headlines use a handful of words again and again. Each one has an everyday picture, and once you know them, a model announcement stops being a wall of jargon.

Model, training, parameters, inference

  • A model is the trained program you talk to, like a trained cook. Inside, it is mostly a very large file of numbers
  • Training is the long learning stage, before anyone uses the model. One way to picture it: the model guesses the next piece of text, checks the real text, and gets a little better
  • Parameters are the numbers inside the model, adjusted again and again during training, like a cook changing the salt after every tasting. A large language model usually has billions of them, and no person sets them by hand: the training process nudges them
  • Inference is the trained model at work, answering you: the cook making your order, not learning
  • While a model answers, its numbers usually stay fixed. Some apps save your chats, and a maker may use them to train a later version, but the model answering you usually does not change its numbers as it answers

Prompt and GPT

  • A prompt is what you send the model: your question, your instructions, any text you paste in. Module 3 is all about writing them
  • GPT stands for generative pre-trained transformer, a type of model: it writes new text (generative), it was first trained on a huge amount of text (pre-trained), and its design is called a transformer (module 2, lesson 7)

Open and closed models

  • An open model usually means its weights (another word for the parameters, the learned numbers) are published, so anyone can download and run it, under the terms of its licence: a recipe printed in a book
  • Usually only the weights are published, not the text the model learned from
  • A closed model is used through a company's app or service; its weights stay with its maker: a restaurant's own dish
  • Neither kind is always better. Running an open model needs computers, which cost money; choose by your task, your budget and how private your data must stay

Reading a parameter count (example numbers)

  • A model with 70 billion parameters has 10 times as many learned numbers as one with 7 billion (70 ÷ 7 = 10)
  • It does not mean it read 70 billion documents, that it is faster (a bigger model usually needs more computing), or that it is better at every task

Three questions for any AI headline

  • Is the claim about learning (training) or answering (inference)?
  • Is there a parameter count? It tells you size, not proof of quality
  • Open or closed? It tells you who can get the weights, not which is better
  • Then test the model on a task of your own: only that tells you if it suits your work

The trap

  • Wrong: more parameters always means better answers; the model learns from my question as it answers; open models are always better, or always worse
  • Right: read a parameter count as size and test on your task; keep training and inference apart; read open or closed as who can get the weights; sort each claim as learning, answering or access

Try it

  • Find a news story about a new AI model, or ask a chat assistant to write a short made-up model announcement with five claims
  • In your notes file, write each claim on its own line and tag it: training, inference, size or access
  • For any parameter count, write one line on what it tells you and one on what it does not
  • Ask the assistant: what is the difference between training and inference? Check its answer against your notes

Lesson quiz

Checked in your browser. Nothing is sent or saved.

5 questions

Question 1 · easy

On the bus to college, Asha asks a chat assistant to explain photosynthesis, and it writes a reply. Which stage is the model in while it writes that reply?

Question 2 · easy

Meena, a teacher in Balasore, asks: "Does a chat assistant keep every page it learned from, and find the right page when I ask?" Which reply is right?

Question 3 · medium

Kiran tells a chat assistant a fact about his village in Ganjam. Do the model's parameters change right then, while it answers him, so it learns his fact?

Question 4 · medium

Sameer runs a small computer shop in Bhubaneswar. He wants to download a model and run it on his own office computer, so his customers' files never leave the shop. Which kind of model allows this?

Question 5 · hard

For a school project, a student in Koraput asks a chat assistant with no web search for three things:
- the year India became independent
- the name of the first headmaster of her small village school
- a polite closing line for her project report

Which answer needs the most checking against a record?

What LLMs do well and badlyThat was the free module. Tell me when the course opens