How Large Language Models Work - Quiz

Question 1

What is a language model fundamentally designed to do?

To learn language patterns and make predictions about sequences of tokens To store every website on the internet To replace the computer operating system To translate every sentence into numbers and never produce text

Question 2

What is a token?

A unit of text processed by a language model A physical part inside the computer A complete database record A type of neural network layer

Question 3

What is next-token prediction?

Predicting a likely next token based on the available input and context Predicting the next computer that will connect to the internet Automatically retraining the model after every question Searching a database for an exact sentence

Question 4

What happens to model parameters during training?

They are adjusted during training so the model becomes better at its task They are permanently deleted during training They are replaced by the users question They never change during training

Question 5

What is the difference between training and inference?

Training adjusts parameters; inference uses the trained model on new input Training and inference are exactly the same process Inference always retrains the entire model Training only happens after deployment

Question 6

Why is context important to a language model?

Context provides information that helps determine meaning and an appropriate continuation Context permanently changes every model parameter Context is only used to display the final answer Context has no effect on language generation

Question 7

Is a language model simply a database containing every sentence from its training data?

No. The model represents learned patterns and relationships through its parameters Yes. Every sentence is stored exactly as a separate answer Yes. The model only copies complete sentences Yes. The model cannot generate new combinations of text

Question 8

What is one important limitation of language models?

It can generate convincing text that contains incorrect information It can never make a mistake It only works with numbers It always knows when an answer is incorrect