Classwork 17

ML and LLM Concepts

Author

Byeong-Hak Choe

Published

November 18, 2026

NoteInstructions

Choose one best answer for each question. Submit the answer letter and a short reason in your own words on the corresponding Brightspace page.

You can answer these questions from AI and Data Analytics. You do not need an AI tool, R code, or model training. Follow the instructor’s assignment-specific AI directions and the course AI policy; permission for another classwork does not automatically apply here.

1. Fixed rules or learning?

Which process follows a fixed rule, rather than learning a pattern from examples?

  • A. Learn from emails labeled spam or not spam to classify a new email.
  • B. Add recorded prices and divide by their count to calculate an average.
  • C. Find groups of customers with similar spending and visit patterns.
  • D. Learn from earlier sales records to estimate sales for a new week.

2. A known numerical outcome

A retailer learns from past weeks to estimate sales in the following week. Which value is the known outcome in each historical training example?

  • A. The calendar week used to identify the record.
  • B. The number of columns in the training table.
  • C. The sales estimate produced by the model.
  • D. The actual sales recorded in the following week.

3. Finding customer groups

A retailer has each customer’s visits per month and average spending per visit. It wants to find similar customers, with no preassigned group labels. Which approach fits?

  • A. Unsupervised learning that finds groups from the recorded patterns.
  • B. Supervised learning that uses each customer’s known group as the answer.
  • C. Applying an existing VIP/non-VIP label already stored in the file.
  • D. Waiting for someone to supply the correct group for every customer.

4. Deep learning

Which description best fits deep learning?

  • A. Any calculation that uses a very large spreadsheet.
  • B. A program that applies only hand-written rules.
  • C. Machine learning that uses neural networks with many layers.
  • D. Sorting observations into categories already recorded in the data.

5. Where an LLM fits

Which statement correctly connects a large language model (LLM) with machine learning?

  • A. An LLM is outside machine learning because its output contains words.
  • B. An LLM is a machine-learning model that can generate language.
  • C. Every machine-learning model generates language in the same way as an LLM.
  • D. Machine learning predicts numbers, so it cannot produce categories or text.

6. Tokens

In a text-based language model, what is a token?

  • A. A piece of text, such as a word, part of a word, or a symbol.
  • B. Only a complete word, never part of a word or punctuation.
  • C. One entire row from a dataset supplied to the model.
  • D. A score showing that the model’s answer is correct.

7. Generating a response

Which description best matches the next-token process introduced in the lecture?

  • A. The model always copies a complete paragraph from one published source.
  • B. The model chooses the entire response in one step without using earlier text.
  • C. A person must approve each word before the model can continue.
  • D. The model scores possible next tokens, selects one, and repeats the process.

8. Using context through attention

What does an attention mechanism help an LLM do?

  • A. Give every earlier token exactly the same importance in every situation.
  • B. Guarantee that every statement in the response is factually correct.
  • C. Give different importance to parts of the context when generating the next token.
  • D. Replace all the training examples with a list of hand-written answers.

9. Further training with examples

An existing model receives further training using prompts paired with examples of desired responses. Which term best fits this process?

  • A. Fine-tuning.
  • B. Sorting records by date.
  • C. Generating a response without further training.
  • D. Clustering customers without answer labels.

10. Learning from human feedback

Which description best fits reinforcement learning from human feedback (RLHF)?

  • A. A model permanently changes its training whenever someone types a new prompt.
  • B. People’s comparisons or preferences between responses guide further training.
  • C. A program calculates a mean by following a fixed arithmetic formula.
  • D. A system groups records without receiving any human feedback.
Back to top