The Model Era (2022) Quiz

Exam-prep questions for the The Model Era (2022) track of the Terminal Academy. Work language models from a terminal: local model runners, tokenization, sampling, system prompts, tool calls, RAG, injection defence.

This category currently has 60 questions in the SERVBG quiz bank. Below are a few sample questions: the full interactive quiz shuffles through the whole set with instant scoring.

Sample questions

What does a large language model actually read and bill for, instead of whole words?

Individual ASCII bytes only
Whole paragraphs
Unicode code points, never merged
Tokens (sub-word units)
Whole sentences

In a terminal tokenizer demo, why might the word 'internationalization' become two or more tokens instead of one?

Capital letters always force a split
Tokenizers only understand whole dictionary words
The tokenizer only accepts words under 5 letters
Every noun is split by definition
Long or uncommon words are often split into smaller sub-word pieces

What is a 'context window' in the context of an LLM?

The number of documents a RAG system has ingested
The maximum number of tokens (prompt plus reply) the model can attend to at once
The number of turns a user is allowed to type per session
The maximum number of tool calls allowed per request
The physical RAM size of the GPU running the model

A conversation grows past the model's context window. What typically happens to make room for new tokens?

The model automatically doubles its own context window
Tokens are compressed losslessly with no information lost
Older messages are truncated or dropped from what gets sent to the model
The temperature setting is lowered to compensate
The model refuses to respond until the user upgrades hardware

Why does the token count of a prompt usually differ from its plain word count?

Only vowels count as tokens
Whitespace is counted as its own separate word every time
Tokenizers always produce exactly one token per word
Tokens are always exactly 4 characters long
Punctuation, numbers, and rare/long words often split into multiple tokens, while short common words may be a single token

Start the full quiz ›