
Large Language Models Trivia Questions
Explore the technology behind tools that can write, summarize, translate, and chat. Test your knowledge of how language models are trained, what they can do, and the concepts that power them.
Large Language Models trivia questions and answers
During the pretraining of many language models, what task teaches the model to predict the next piece of text?
Show answerHide answer
Next-token prediction
A model learns by repeatedly predicting a missing next token and comparing its prediction with the training text. This simple objective can support many capabilities when applied to large, varied datasets.
What is the large collection of text and other material used to train a language model called?
Show answerHide answer
Training data
Training data can include books, web pages, code, and other sources. The mix and quality of that data influence what a model learns.
What are the units of text that a language model processes, which may be whole words, word fragments, or punctuation?
Show answerHide answer
Tokens
Token boundaries do not always match word boundaries. For example, an uncommon word might be split into several smaller pieces.
What tokenization method repeatedly merges frequent pairs of symbols or character sequences?
Show answerHide answer
Byte-pair encoding
Byte-pair encoding was originally developed for data compression. In language-model tokenization, it helps build a vocabulary of frequently occurring text pieces.
What transformer mechanism lets each token weigh information from other tokens in the input?
Show answerHide answer
Self-attention
Self-attention helps a model connect related words even when they are far apart in a sentence. Transformers use it as a central way to process context.
How to play
- • Each game has 10 questions. Before each one you see only a hint about its topic.
- • Choose what it’s worth, from $1 to $20. Each value can only be used once per game.
- • Type your answer. Small spelling mistakes are fine. Get it right and you win that amount.





