AI LLM Basics MCQ Questions with Answers (Latest 2026)

Q1. Which option best describes a large language model?

A. A pretrained image classifier. B. A retrieval index of documents. C. A neural network trained on large text corpora to predict tokens. D. A random forest of features.

Select an answer to check.

Answer: A neural network trained on large text corpora to predict tokens.

Here, A neural network trained on large text corpora to predict tokens. is the right choice. LLMs predict next tokens. It aligns directly with what the question asks about which option best describes a large language model. A quick elimination of partially true options helps confirm it.

Q2. What is the primary purpose of a large language model?

A. A pretrained image classifier. B. A neural network trained on large text corpora to predict tokens. C. A retrieval index of documents. D. A random forest of features.

Select an answer to check.

Answer: A neural network trained on large text corpora to predict tokens.

In this case, A neural network trained on large text corpora to predict tokens. is correct. LLMs predict next tokens. It aligns directly with what the question asks about what is the primary purpose of a large. A quick elimination of partially true options helps confirm it.

Q3. Which statement about a large language model is most accurate?

A. A neural network trained on large text corpora to predict tokens. B. A retrieval index of documents. C. A pretrained image classifier. D. A random forest of features.

Select an answer to check.

Answer: A neural network trained on large text corpora to predict tokens.

The best option here is A neural network trained on large text corpora to predict tokens.. LLMs predict next tokens. It aligns directly with what the question asks about which statement about a large language model is. A quick elimination of partially true options helps confirm it.

Q4. How is a large language model best characterized?

A. A pretrained image classifier. B. A retrieval index of documents. C. A random forest of features. D. A neural network trained on large text corpora to predict tokens.

Select an answer to check.

Answer: A neural network trained on large text corpora to predict tokens.

For this question, A neural network trained on large text corpora to predict tokens. is correct. LLMs predict next tokens. It aligns directly with what the question asks about how is a large language model best characterized. A quick elimination of partially true options helps confirm it.

Q5. Which option best describes a token?

A. A vector embedding. B. A full sentence always. C. A single character always. D. A subword unit produced by the tokenizer.

Select an answer to check.

Answer: A subword unit produced by the tokenizer.

A subword unit produced by the tokenizer. is the correct answer here. Granularity varies by tokenizer. It aligns directly with what the question asks about which option best describes a token. A quick elimination of partially true options helps confirm it.

Q6. What is the primary purpose of a token?

A. A vector embedding. B. A subword unit produced by the tokenizer. C. A single character always. D. A full sentence always.

Select an answer to check.

Answer: A subword unit produced by the tokenizer.

Here, A subword unit produced by the tokenizer. is the right choice. Granularity varies by tokenizer. This matches the core idea being tested around what is the primary purpose of a token. A quick elimination of partially true options helps confirm it.

Q7. Which statement about a token is most accurate?

A. A subword unit produced by the tokenizer. B. A vector embedding. C. A full sentence always. D. A single character always.

Select an answer to check.

Answer: A subword unit produced by the tokenizer.

In this case, A subword unit produced by the tokenizer. is correct. Granularity varies by tokenizer. This matches the core idea being tested around which statement about a token is most accurate. A quick elimination of partially true options helps confirm it.

Q8. How is a token best characterized?

A. A subword unit produced by the tokenizer. B. A single character always. C. A full sentence always. D. A vector embedding.

Select an answer to check.

Answer: A subword unit produced by the tokenizer.

The best option here is A subword unit produced by the tokenizer.. Granularity varies by tokenizer. This matches the core idea being tested around how is a token best characterized. A quick elimination of partially true options helps confirm it.

Q9. Which option best describes BPE tokenization?

A. Character-only tokenization always. B. Sentence-level tokenization always. C. Byte-pair encoding that merges frequent pairs into subwords. D. Word-level tokenization only.

Select an answer to check.

Answer: Byte-pair encoding that merges frequent pairs into subwords.

For this question, Byte-pair encoding that merges frequent pairs into subwords. is correct. Common in modern LLMs. This matches the core idea being tested around which option best describes bpe tokenization. A quick elimination of partially true options helps confirm it.

Q10. What is the primary purpose of BPE tokenization?

A. Character-only tokenization always. B. Sentence-level tokenization always. C. Byte-pair encoding that merges frequent pairs into subwords. D. Word-level tokenization only.

Select an answer to check.

Answer: Byte-pair encoding that merges frequent pairs into subwords.

Byte-pair encoding that merges frequent pairs into subwords. is the correct answer here. Common in modern LLMs. This matches the core idea being tested around what is the primary purpose of bpe tokenization. A quick elimination of partially true options helps confirm it.

Q11. Which statement about BPE tokenization is most accurate?

A. Character-only tokenization always. B. Byte-pair encoding that merges frequent pairs into subwords. C. Word-level tokenization only. D. Sentence-level tokenization always.

Select an answer to check.

Answer: Byte-pair encoding that merges frequent pairs into subwords.

Here, Byte-pair encoding that merges frequent pairs into subwords. is the right choice. Common in modern LLMs. That is exactly the concept behind which statement about bpe tokenization is most accurate in this context. A quick elimination of partially true options helps confirm it.

Q12. How is BPE tokenization best characterized?

A. Character-only tokenization always. B. Sentence-level tokenization always. C. Word-level tokenization only. D. Byte-pair encoding that merges frequent pairs into subwords.

Select an answer to check.

Answer: Byte-pair encoding that merges frequent pairs into subwords.

In this case, Byte-pair encoding that merges frequent pairs into subwords. is correct. Common in modern LLMs. That is exactly the concept behind how is bpe tokenization best characterized in this context. A quick elimination of partially true options helps confirm it.

Q13. Which option best describes the context window?

A. Max tokens the model can attend to per call. B. The training corpus size. C. The vocabulary size. D. The model's parameter count.

Select an answer to check.

Answer: Max tokens the model can attend to per call.

The best option here is Max tokens the model can attend to per call.. Bounds prompt + completion size. That is exactly the concept behind which option best describes the context window in this context. A quick elimination of partially true options helps confirm it.

Q14. What is the primary purpose of the context window?

A. The vocabulary size. B. The model's parameter count. C. The training corpus size. D. Max tokens the model can attend to per call.

Select an answer to check.

Answer: Max tokens the model can attend to per call.

For this question, Max tokens the model can attend to per call. is correct. Bounds prompt + completion size. That is exactly the concept behind what is the primary purpose of the context in this context. A quick elimination of partially true options helps confirm it.

Q15. Which statement about the context window is most accurate?

A. Max tokens the model can attend to per call. B. The model's parameter count. C. The training corpus size. D. The vocabulary size.

Select an answer to check.

Answer: Max tokens the model can attend to per call.

Max tokens the model can attend to per call. is the correct answer here. Bounds prompt + completion size. That is exactly the concept behind which statement about the context window is most in this context. A quick elimination of partially true options helps confirm it.

Q16. How is the context window best characterized?

A. The training corpus size. B. Max tokens the model can attend to per call. C. The model's parameter count. D. The vocabulary size.

Select an answer to check.

Answer: Max tokens the model can attend to per call.

Here, Max tokens the model can attend to per call. is the right choice. Bounds prompt + completion size. It fits the requirement in the prompt about how is the context window best characterized. A quick elimination of partially true options helps confirm it.

Q17. Which option best describes a parameter?

A. A learned weight in the model. B. A hyperparameter set by users. C. An activation value. D. A token in the vocabulary.

Select an answer to check.

Answer: A learned weight in the model.

In this case, A learned weight in the model. is correct. Counts in billions for large LLMs. It fits the requirement in the prompt about which option best describes a parameter. A quick elimination of partially true options helps confirm it.

Q18. What is the primary purpose of a parameter?

A. An activation value. B. A learned weight in the model. C. A token in the vocabulary. D. A hyperparameter set by users.

Select an answer to check.

Answer: A learned weight in the model.

The best option here is A learned weight in the model.. Counts in billions for large LLMs. It fits the requirement in the prompt about what is the primary purpose of a parameter. A quick elimination of partially true options helps confirm it.

Q19. Which statement about a parameter is most accurate?

A. A learned weight in the model. B. A hyperparameter set by users. C. An activation value. D. A token in the vocabulary.

Select an answer to check.

Answer: A learned weight in the model.

For this question, A learned weight in the model. is correct. Counts in billions for large LLMs. It fits the requirement in the prompt about which statement about a parameter is most accurate. A quick elimination of partially true options helps confirm it.

Q20. How is a parameter best characterized?

A. An activation value. B. A hyperparameter set by users. C. A learned weight in the model. D. A token in the vocabulary.

Select an answer to check.

Answer: A learned weight in the model.

A learned weight in the model. is the correct answer here. Counts in billions for large LLMs. It fits the requirement in the prompt about how is a parameter best characterized. A quick elimination of partially true options helps confirm it.

Q21. Which option best describes a prompt?

A. A loss function. B. Input text guiding the LLM's response. C. A model checkpoint. D. A vector index.

Select an answer to check.

Answer: Input text guiding the LLM's response.

Here, Input text guiding the LLM's response. is the right choice. Quality of prompt drives quality of output. This is the most accurate statement for which option best describes a prompt. A quick elimination of partially true options helps confirm it.

Q22. What is the primary purpose of a prompt?

A. Input text guiding the LLM's response. B. A loss function. C. A vector index. D. A model checkpoint.

Select an answer to check.

Answer: Input text guiding the LLM's response.

In this case, Input text guiding the LLM's response. is correct. Quality of prompt drives quality of output. This is the most accurate statement for what is the primary purpose of a prompt. A quick elimination of partially true options helps confirm it.

Q23. Which statement about a prompt is most accurate?

A. Input text guiding the LLM's response. B. A model checkpoint. C. A vector index. D. A loss function.

Select an answer to check.

Answer: Input text guiding the LLM's response.

The best option here is Input text guiding the LLM's response.. Quality of prompt drives quality of output. This is the most accurate statement for which statement about a prompt is most accurate. A quick elimination of partially true options helps confirm it.

Q24. How is a prompt best characterized?

A. Input text guiding the LLM's response. B. A vector index. C. A loss function. D. A model checkpoint.

Select an answer to check.

Answer: Input text guiding the LLM's response.

For this question, Input text guiding the LLM's response. is correct. Quality of prompt drives quality of output. This is the most accurate statement for how is a prompt best characterized. A quick elimination of partially true options helps confirm it.

Q25. Which option best describes temperature?

A. The number of training tokens. B. The size of the context window. C. Sampling parameter controlling randomness. D. The number of layers.

Select an answer to check.

Answer: Sampling parameter controlling randomness.

Sampling parameter controlling randomness. is the correct answer here. Lower = more deterministic. This is the most accurate statement for which option best describes temperature. A quick elimination of partially true options helps confirm it.

Q26. What is the primary purpose of temperature?

A. Sampling parameter controlling randomness. B. The number of training tokens. C. The number of layers. D. The size of the context window.

Select an answer to check.

Answer: Sampling parameter controlling randomness.

Here, Sampling parameter controlling randomness. is the right choice. Lower = more deterministic. It aligns directly with what the question asks about what is the primary purpose of temperature. The other options are either incomplete or contextually incorrect.

Q27. Which statement about temperature is most accurate?

A. The size of the context window. B. The number of layers. C. The number of training tokens. D. Sampling parameter controlling randomness.

Select an answer to check.

Answer: Sampling parameter controlling randomness.

In this case, Sampling parameter controlling randomness. is correct. Lower = more deterministic. It aligns directly with what the question asks about which statement about temperature is most accurate. The other options are either incomplete or contextually incorrect.

Q28. How is temperature best characterized?

A. The number of training tokens. B. Sampling parameter controlling randomness. C. The size of the context window. D. The number of layers.

Select an answer to check.

Answer: Sampling parameter controlling randomness.

The best option here is Sampling parameter controlling randomness.. Lower = more deterministic. It aligns directly with what the question asks about how is temperature best characterized. The other options are either incomplete or contextually incorrect.

Q29. Which option best describes top-p sampling?

A. Sample from smallest set of tokens with cumulative prob ≥ p. B. A pure beam search. C. Always greedy decoding. D. Random sampling with no truncation.

Select an answer to check.

Answer: Sample from smallest set of tokens with cumulative prob ≥ p.

For this question, Sample from smallest set of tokens with cumulative prob ≥ p. is correct. Controls quality/diversity. It aligns directly with what the question asks about which option best describes top-p sampling. The other options are either incomplete or contextually incorrect.

Q30. What is the primary purpose of top-p sampling?

A. Random sampling with no truncation. B. Always greedy decoding. C. Sample from smallest set of tokens with cumulative prob ≥ p. D. A pure beam search.

Select an answer to check.

Answer: Sample from smallest set of tokens with cumulative prob ≥ p.

Sample from smallest set of tokens with cumulative prob ≥ p. is the correct answer here. Controls quality/diversity. It aligns directly with what the question asks about what is the primary purpose of top-p sampling. The other options are either incomplete or contextually incorrect.

Q31. Which statement about top-p sampling is most accurate?

A. Sample from smallest set of tokens with cumulative prob ≥ p. B. Always greedy decoding. C. A pure beam search. D. Random sampling with no truncation.

Select an answer to check.

Answer: Sample from smallest set of tokens with cumulative prob ≥ p.

Here, Sample from smallest set of tokens with cumulative prob ≥ p. is the right choice. Controls quality/diversity. This matches the core idea being tested around which statement about top-p sampling is most accurate. The other options are either incomplete or contextually incorrect.

Q32. How is top-p sampling best characterized?

A. Random sampling with no truncation. B. Sample from smallest set of tokens with cumulative prob ≥ p. C. Always greedy decoding. D. A pure beam search.

Select an answer to check.

Answer: Sample from smallest set of tokens with cumulative prob ≥ p.

In this case, Sample from smallest set of tokens with cumulative prob ≥ p. is correct. Controls quality/diversity. This matches the core idea being tested around how is top-p sampling best characterized. The other options are either incomplete or contextually incorrect.

Q33. Which option best describes top-k sampling?

A. Pure greedy decoding. B. Always uniform sampling. C. Sample from the k highest-probability tokens. D. Sampling from full distribution always.

Select an answer to check.

Answer: Sample from the k highest-probability tokens.

The best option here is Sample from the k highest-probability tokens.. Limits low-probability tail. This matches the core idea being tested around which option best describes top-k sampling. The other options are either incomplete or contextually incorrect.

Q34. What is the primary purpose of top-k sampling?

A. Sampling from full distribution always. B. Sample from the k highest-probability tokens. C. Always uniform sampling. D. Pure greedy decoding.

Select an answer to check.

Answer: Sample from the k highest-probability tokens.

For this question, Sample from the k highest-probability tokens. is correct. Limits low-probability tail. This matches the core idea being tested around what is the primary purpose of top-k sampling. The other options are either incomplete or contextually incorrect.

Q35. Which statement about top-k sampling is most accurate?

A. Always uniform sampling. B. Pure greedy decoding. C. Sampling from full distribution always. D. Sample from the k highest-probability tokens.

Select an answer to check.

Answer: Sample from the k highest-probability tokens.

Sample from the k highest-probability tokens. is the correct answer here. Limits low-probability tail. This matches the core idea being tested around which statement about top-k sampling is most accurate. The other options are either incomplete or contextually incorrect.

Q36. How is top-k sampling best characterized?

A. Always uniform sampling. B. Sample from the k highest-probability tokens. C. Pure greedy decoding. D. Sampling from full distribution always.

Select an answer to check.

Answer: Sample from the k highest-probability tokens.

Here, Sample from the k highest-probability tokens. is the right choice. Limits low-probability tail. That is exactly the concept behind how is top-k sampling best characterized in this context. The other options are either incomplete or contextually incorrect.

Q37. Which option best describes greedy decoding?

A. Always picking the most likely next token. B. Top-p sampling. C. Random token selection. D. Beam search.

Select an answer to check.

Answer: Always picking the most likely next token.

In this case, Always picking the most likely next token. is correct. Deterministic but can be repetitive. That is exactly the concept behind which option best describes greedy decoding in this context. The other options are either incomplete or contextually incorrect.

Q38. What is the primary purpose of greedy decoding?

A. Top-p sampling. B. Random token selection. C. Always picking the most likely next token. D. Beam search.

Select an answer to check.

Answer: Always picking the most likely next token.

The best option here is Always picking the most likely next token.. Deterministic but can be repetitive. That is exactly the concept behind what is the primary purpose of greedy decoding in this context. The other options are either incomplete or contextually incorrect.

Q39. Which statement about greedy decoding is most accurate?

A. Top-p sampling. B. Random token selection. C. Beam search. D. Always picking the most likely next token.

Select an answer to check.

Answer: Always picking the most likely next token.

For this question, Always picking the most likely next token. is correct. Deterministic but can be repetitive. That is exactly the concept behind which statement about greedy decoding is most accurate in this context. The other options are either incomplete or contextually incorrect.

Q40. How is greedy decoding best characterized?

A. Random token selection. B. Beam search. C. Top-p sampling. D. Always picking the most likely next token.

Select an answer to check.

Answer: Always picking the most likely next token.

Always picking the most likely next token. is the correct answer here. Deterministic but can be repetitive. That is exactly the concept behind how is greedy decoding best characterized in this context. The other options are either incomplete or contextually incorrect.

Q41. Which option best describes beam search?

A. Maintain top-k partial sequences and expand each. B. Pure random sampling. C. Greedy decoding. D. Constrained decoding only.

Select an answer to check.

Answer: Maintain top-k partial sequences and expand each.

Here, Maintain top-k partial sequences and expand each. is the right choice. Better for structured tasks. It fits the requirement in the prompt about which option best describes beam search. The other options are either incomplete or contextually incorrect.

Q42. What is the primary purpose of beam search?

A. Pure random sampling. B. Greedy decoding. C. Maintain top-k partial sequences and expand each. D. Constrained decoding only.

Select an answer to check.

Answer: Maintain top-k partial sequences and expand each.

In this case, Maintain top-k partial sequences and expand each. is correct. Better for structured tasks. It fits the requirement in the prompt about what is the primary purpose of beam search. The other options are either incomplete or contextually incorrect.

Q43. Which statement about beam search is most accurate?

A. Pure random sampling. B. Greedy decoding. C. Constrained decoding only. D. Maintain top-k partial sequences and expand each.

Select an answer to check.

Answer: Maintain top-k partial sequences and expand each.

The best option here is Maintain top-k partial sequences and expand each.. Better for structured tasks. It fits the requirement in the prompt about which statement about beam search is most accurate. The other options are either incomplete or contextually incorrect.

Q44. How is beam search best characterized?

A. Maintain top-k partial sequences and expand each. B. Greedy decoding. C. Pure random sampling. D. Constrained decoding only.

Select an answer to check.

Answer: Maintain top-k partial sequences and expand each.

For this question, Maintain top-k partial sequences and expand each. is correct. Better for structured tasks. It fits the requirement in the prompt about how is beam search best characterized. The other options are either incomplete or contextually incorrect.

Q45. Which option best describes an instruction-tuned model?

A. An LLM further trained to follow instructions. B. A retrieval-only system. C. A purely pretrained base model. D. An untrained checkpoint.

Select an answer to check.

Answer: An LLM further trained to follow instructions.

An LLM further trained to follow instructions. is the correct answer here. Examples: chat-tuned models. It fits the requirement in the prompt about which option best describes an instruction-tuned model. The other options are either incomplete or contextually incorrect.

Q46. What is the primary purpose of an instruction-tuned model?

A. A retrieval-only system. B. An untrained checkpoint. C. A purely pretrained base model. D. An LLM further trained to follow instructions.

Select an answer to check.

Answer: An LLM further trained to follow instructions.

Here, An LLM further trained to follow instructions. is the right choice. Examples: chat-tuned models. This is the most accurate statement for what is the primary purpose of an instruction-tuned. The other options are either incomplete or contextually incorrect.

Q47. Which statement about an instruction-tuned model is most accurate?

A. A retrieval-only system. B. An LLM further trained to follow instructions. C. An untrained checkpoint. D. A purely pretrained base model.

Select an answer to check.

Answer: An LLM further trained to follow instructions.

In this case, An LLM further trained to follow instructions. is correct. Examples: chat-tuned models. This is the most accurate statement for which statement about an instruction-tuned model is most. The other options are either incomplete or contextually incorrect.

Q48. How is an instruction-tuned model best characterized?

A. A purely pretrained base model. B. An LLM further trained to follow instructions. C. An untrained checkpoint. D. A retrieval-only system.

Select an answer to check.

Answer: An LLM further trained to follow instructions.

The best option here is An LLM further trained to follow instructions.. Examples: chat-tuned models. This is the most accurate statement for how is an instruction-tuned model best characterized. The other options are either incomplete or contextually incorrect.

Q49. Which option best describes RLHF?

A. Static rule mapping. B. Reinforcement learning from human feedback to align models. C. Random parameter updates. D. Pure unsupervised pretraining.

Select an answer to check.

Answer: Reinforcement learning from human feedback to align models.

For this question, Reinforcement learning from human feedback to align models. is correct. Improves helpfulness and safety. This is the most accurate statement for which option best describes rlhf. The other options are either incomplete or contextually incorrect.

Q50. What is the primary purpose of RLHF?

A. Reinforcement learning from human feedback to align models. B. Static rule mapping. C. Random parameter updates. D. Pure unsupervised pretraining.

Select an answer to check.

Answer: Reinforcement learning from human feedback to align models.

Reinforcement learning from human feedback to align models. is the correct answer here. Improves helpfulness and safety. This is the most accurate statement for what is the primary purpose of rlhf. The other options are either incomplete or contextually incorrect.