Vendor-neutral fundamentals of working with large language models: tokens and sampling, prompting patterns, embeddings and vector search, retrieval-augmented generation, and the failure modes that bite in production.
You’re reading a free preview. Sign up to see every topic and question — and to play this deck as games.
What a model actually reads, how much of it fits, and the knobs that decide what comes out.
50 total questions
Card #1
What is a token?
Card #2
Why does the same sentence cost more tokens in some languages than in English?
Card #3
What is the context window?
1. The context window of a model limits:
2. Lowering the temperature toward zero makes the model:
3. Top-p (nucleus) sampling differs from top-k in that:
1. The unit of text a language model actually reads and predicts is a ___.
Hint: often a sub-word piece
2. The maximum number of tokens a model can attend to at once — prompt and output together — is its context ___.
3. Sampling from the smallest set of tokens whose probabilities sum to p is called ___ sampling.
Hint: also written top-p
