AI LLM Basics MCQ Questions with Answers – Page 2 (Latest 2026)

Q51. Which statement about RLHF is most accurate?

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

Select an answer to check.

Answer: Reinforcement learning from human feedback to align models.

Here, Reinforcement learning from human feedback to align models. is the right choice. Improves helpfulness and safety. It aligns directly with what the question asks about which statement about rlhf is most accurate. Competing choices sound plausible, but they miss the key condition.

Q52. How is RLHF best characterized?

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

Select an answer to check.

Answer: Reinforcement learning from human feedback to align models.

In this case, Reinforcement learning from human feedback to align models. is correct. Improves helpfulness and safety. It aligns directly with what the question asks about how is rlhf best characterized. Competing choices sound plausible, but they miss the key condition.

Q53. Which option best describes supervised fine-tuning?

A. A retrieval index build. B. A pure RL stage. C. Continued training on labeled examples. D. Random reinitialization.

Select an answer to check.

Answer: Continued training on labeled examples.

The best option here is Continued training on labeled examples.. Often precedes RLHF. It aligns directly with what the question asks about which option best describes supervised fine-tuning. Competing choices sound plausible, but they miss the key condition.

Q54. What is the primary purpose of supervised fine-tuning?

A. A retrieval index build. B. A pure RL stage. C. Random reinitialization. D. Continued training on labeled examples.

Select an answer to check.

Answer: Continued training on labeled examples.

For this question, Continued training on labeled examples. is correct. Often precedes RLHF. It aligns directly with what the question asks about what is the primary purpose of supervised fine-tuning. Competing choices sound plausible, but they miss the key condition.

Q55. Which statement about supervised fine-tuning is most accurate?

A. Continued training on labeled examples. B. Random reinitialization. C. A retrieval index build. D. A pure RL stage.

Select an answer to check.

Answer: Continued training on labeled examples.

Continued training on labeled examples. is the correct answer here. Often precedes RLHF. It aligns directly with what the question asks about which statement about supervised fine-tuning is most accurate. Competing choices sound plausible, but they miss the key condition.

Q56. How is supervised fine-tuning best characterized?

A. Continued training on labeled examples. B. A pure RL stage. C. A retrieval index build. D. Random reinitialization.

Select an answer to check.

Answer: Continued training on labeled examples.

Here, Continued training on labeled examples. is the right choice. Often precedes RLHF. This matches the core idea being tested around how is supervised fine-tuning best characterized. Competing choices sound plausible, but they miss the key condition.

Q57. Which option best describes a chat template?

A. A loss function. B. Format combining system/user/assistant turns into model input. C. A regularizer. D. A vector index.

Select an answer to check.

Answer: Format combining system/user/assistant turns into model input.

In this case, Format combining system/user/assistant turns into model input. is correct. Required for chat-tuned models. This matches the core idea being tested around which option best describes a chat template. Competing choices sound plausible, but they miss the key condition.

Q58. What is the primary purpose of a chat template?

A. Format combining system/user/assistant turns into model input. B. A regularizer. C. A vector index. D. A loss function.

Select an answer to check.

Answer: Format combining system/user/assistant turns into model input.

The best option here is Format combining system/user/assistant turns into model input.. Required for chat-tuned models. This matches the core idea being tested around what is the primary purpose of a chat. Competing choices sound plausible, but they miss the key condition.

Q59. Which statement about a chat template is most accurate?

A. A loss function. B. Format combining system/user/assistant turns into model input. C. A regularizer. D. A vector index.

Select an answer to check.

Answer: Format combining system/user/assistant turns into model input.

For this question, Format combining system/user/assistant turns into model input. is correct. Required for chat-tuned models. This matches the core idea being tested around which statement about a chat template is most. Competing choices sound plausible, but they miss the key condition.

Q60. How is a chat template best characterized?

A. A vector index. B. A loss function. C. A regularizer. D. Format combining system/user/assistant turns into model input.

Select an answer to check.

Answer: Format combining system/user/assistant turns into model input.

Format combining system/user/assistant turns into model input. is the correct answer here. Required for chat-tuned models. This matches the core idea being tested around how is a chat template best characterized. Competing choices sound plausible, but they miss the key condition.

Q61. Which option best describes a system prompt?

A. The user's last message. B. A retrieval result. C. Instructions setting persistent role/constraints. D. A tool output only.

Select an answer to check.

Answer: Instructions setting persistent role/constraints.

Here, Instructions setting persistent role/constraints. is the right choice. Shapes assistant behavior. That is exactly the concept behind which option best describes a system prompt in this context. Competing choices sound plausible, but they miss the key condition.

Q62. What is the primary purpose of a system prompt?

A. Instructions setting persistent role/constraints. B. A tool output only. C. A retrieval result. D. The user's last message.

Select an answer to check.

Answer: Instructions setting persistent role/constraints.

In this case, Instructions setting persistent role/constraints. is correct. Shapes assistant behavior. That is exactly the concept behind what is the primary purpose of a system in this context. Competing choices sound plausible, but they miss the key condition.

Q63. Which statement about a system prompt is most accurate?

A. A tool output only. B. Instructions setting persistent role/constraints. C. The user's last message. D. A retrieval result.

Select an answer to check.

Answer: Instructions setting persistent role/constraints.

The best option here is Instructions setting persistent role/constraints.. Shapes assistant behavior. That is exactly the concept behind which statement about a system prompt is most in this context. Competing choices sound plausible, but they miss the key condition.

Q64. How is a system prompt best characterized?

A. Instructions setting persistent role/constraints. B. A tool output only. C. The user's last message. D. A retrieval result.

Select an answer to check.

Answer: Instructions setting persistent role/constraints.

For this question, Instructions setting persistent role/constraints. is correct. Shapes assistant behavior. That is exactly the concept behind how is a system prompt best characterized in this context. Competing choices sound plausible, but they miss the key condition.

Q65. Which option best describes function calling?

A. A model fine-tune. B. Structured tool invocation via JSON schemas. C. A pure retrieval pipeline. D. Free-form text commands the user parses.

Select an answer to check.

Answer: Structured tool invocation via JSON schemas.

Structured tool invocation via JSON schemas. is the correct answer here. Reliable structured outputs. That is exactly the concept behind which option best describes function calling in this context. Competing choices sound plausible, but they miss the key condition.

Q66. What is the primary purpose of function calling?

A. A model fine-tune. B. Structured tool invocation via JSON schemas. C. Free-form text commands the user parses. D. A pure retrieval pipeline.

Select an answer to check.

Answer: Structured tool invocation via JSON schemas.

Here, Structured tool invocation via JSON schemas. is the right choice. Reliable structured outputs. It fits the requirement in the prompt about what is the primary purpose of function calling. Competing choices sound plausible, but they miss the key condition.

Q67. Which statement about function calling is most accurate?

A. A pure retrieval pipeline. B. Free-form text commands the user parses. C. A model fine-tune. D. Structured tool invocation via JSON schemas.

Select an answer to check.

Answer: Structured tool invocation via JSON schemas.

In this case, Structured tool invocation via JSON schemas. is correct. Reliable structured outputs. It fits the requirement in the prompt about which statement about function calling is most accurate. Competing choices sound plausible, but they miss the key condition.

Q68. How is function calling best characterized?

A. A model fine-tune. B. Free-form text commands the user parses. C. A pure retrieval pipeline. D. Structured tool invocation via JSON schemas.

Select an answer to check.

Answer: Structured tool invocation via JSON schemas.

The best option here is Structured tool invocation via JSON schemas.. Reliable structured outputs. It fits the requirement in the prompt about how is function calling best characterized. Competing choices sound plausible, but they miss the key condition.

Q69. Which option best describes hallucination?

A. Output backed by citations. B. Plausible-sounding but unsupported output. C. A schema validation pass. D. A network timeout.

Select an answer to check.

Answer: Plausible-sounding but unsupported output.

For this question, Plausible-sounding but unsupported output. is correct. Mitigated via grounding. It fits the requirement in the prompt about which option best describes hallucination. Competing choices sound plausible, but they miss the key condition.

Q70. What is the primary purpose of hallucination?

A. A network timeout. B. Plausible-sounding but unsupported output. C. A schema validation pass. D. Output backed by citations.

Select an answer to check.

Answer: Plausible-sounding but unsupported output.

Plausible-sounding but unsupported output. is the correct answer here. Mitigated via grounding. It fits the requirement in the prompt about what is the primary purpose of hallucination. Competing choices sound plausible, but they miss the key condition.

Q71. Which statement about hallucination is most accurate?

A. Plausible-sounding but unsupported output. B. Output backed by citations. C. A network timeout. D. A schema validation pass.

Select an answer to check.

Answer: Plausible-sounding but unsupported output.

Here, Plausible-sounding but unsupported output. is the right choice. Mitigated via grounding. This is the most accurate statement for which statement about hallucination is most accurate. Competing choices sound plausible, but they miss the key condition.

Q72. How is hallucination best characterized?

A. Plausible-sounding but unsupported output. B. A schema validation pass. C. Output backed by citations. D. A network timeout.

Select an answer to check.

Answer: Plausible-sounding but unsupported output.

In this case, Plausible-sounding but unsupported output. is correct. Mitigated via grounding. This is the most accurate statement for how is hallucination best characterized. Competing choices sound plausible, but they miss the key condition.

Q73. Which option best describes RAG?

A. A loss function. B. Retrieval-augmented generation; injecting retrieved context. C. A vector store sharding scheme. D. Reinforcement-aided gradient.

Select an answer to check.

Answer: Retrieval-augmented generation; injecting retrieved context.

The best option here is Retrieval-augmented generation; injecting retrieved context.. Grounds answers in evidence. This is the most accurate statement for which option best describes rag. Competing choices sound plausible, but they miss the key condition.

Q74. What is the primary purpose of RAG?

A. A loss function. B. Reinforcement-aided gradient. C. A vector store sharding scheme. D. Retrieval-augmented generation; injecting retrieved context.

Select an answer to check.

Answer: Retrieval-augmented generation; injecting retrieved context.

For this question, Retrieval-augmented generation; injecting retrieved context. is correct. Grounds answers in evidence. This is the most accurate statement for what is the primary purpose of rag. Competing choices sound plausible, but they miss the key condition.

Q75. Which statement about RAG is most accurate?

A. Retrieval-augmented generation; injecting retrieved context. B. A vector store sharding scheme. C. A loss function. D. Reinforcement-aided gradient.

Select an answer to check.

Answer: Retrieval-augmented generation; injecting retrieved context.

Retrieval-augmented generation; injecting retrieved context. is the correct answer here. Grounds answers in evidence. This is the most accurate statement for which statement about rag is most accurate. Competing choices sound plausible, but they miss the key condition.

Q76. How is RAG best characterized?

A. Reinforcement-aided gradient. B. A vector store sharding scheme. C. Retrieval-augmented generation; injecting retrieved context. D. A loss function.

Select an answer to check.

Answer: Retrieval-augmented generation; injecting retrieved context.

Here, Retrieval-augmented generation; injecting retrieved context. is the right choice. Grounds answers in evidence. It aligns directly with what the question asks about how is rag best characterized. The remaining choices fail because they don’t satisfy the full definition.

Q77. Which option best describes the tokenizer?

A. Maps text to/from token IDs. B. A model checkpoint. C. An optimizer. D. A regularizer.

Select an answer to check.

Answer: Maps text to/from token IDs.

In this case, Maps text to/from token IDs. is correct. Determines token boundaries. It aligns directly with what the question asks about which option best describes the tokenizer. The remaining choices fail because they don’t satisfy the full definition.

Q78. What is the primary purpose of the tokenizer?

A. A model checkpoint. B. Maps text to/from token IDs. C. A regularizer. D. An optimizer.

Select an answer to check.

Answer: Maps text to/from token IDs.

The best option here is Maps text to/from token IDs.. Determines token boundaries. It aligns directly with what the question asks about what is the primary purpose of the tokenizer. The remaining choices fail because they don’t satisfy the full definition.

Q79. Which statement about the tokenizer is most accurate?

A. An optimizer. B. A model checkpoint. C. Maps text to/from token IDs. D. A regularizer.

Select an answer to check.

Answer: Maps text to/from token IDs.

For this question, Maps text to/from token IDs. is correct. Determines token boundaries. It aligns directly with what the question asks about which statement about the tokenizer is most accurate. The remaining choices fail because they don’t satisfy the full definition.

Q80. How is the tokenizer best characterized?

A. Maps text to/from token IDs. B. A model checkpoint. C. An optimizer. D. A regularizer.

Select an answer to check.

Answer: Maps text to/from token IDs.

Maps text to/from token IDs. is the correct answer here. Determines token boundaries. It aligns directly with what the question asks about how is the tokenizer best characterized. The remaining choices fail because they don’t satisfy the full definition.

Q81. Which option best describes the embedding matrix?

A. A retrieval index. B. A loss function. C. Maps token IDs to dense vectors. D. A logging shim.

Select an answer to check.

Answer: Maps token IDs to dense vectors.

Here, Maps token IDs to dense vectors. is the right choice. First layer of most LLMs. This matches the core idea being tested around which option best describes the embedding matrix. The remaining choices fail because they don’t satisfy the full definition.

Q82. What is the primary purpose of the embedding matrix?

A. Maps token IDs to dense vectors. B. A logging shim. C. A loss function. D. A retrieval index.

Select an answer to check.

Answer: Maps token IDs to dense vectors.

In this case, Maps token IDs to dense vectors. is correct. First layer of most LLMs. This matches the core idea being tested around what is the primary purpose of the embedding. The remaining choices fail because they don’t satisfy the full definition.

Q83. Which statement about the embedding matrix is most accurate?

A. Maps token IDs to dense vectors. B. A loss function. C. A logging shim. D. A retrieval index.

Select an answer to check.

Answer: Maps token IDs to dense vectors.

The best option here is Maps token IDs to dense vectors.. First layer of most LLMs. This matches the core idea being tested around which statement about the embedding matrix is most. The remaining choices fail because they don’t satisfy the full definition.

Q84. How is the embedding matrix best characterized?

A. A loss function. B. A logging shim. C. A retrieval index. D. Maps token IDs to dense vectors.

Select an answer to check.

Answer: Maps token IDs to dense vectors.

For this question, Maps token IDs to dense vectors. is correct. First layer of most LLMs. This matches the core idea being tested around how is the embedding matrix best characterized. The remaining choices fail because they don’t satisfy the full definition.

Q85. Which option best describes the residual stream?

A. Cumulative hidden state across Transformer layers. B. A retrieval index. C. A vocabulary table. D. A loss function.

Select an answer to check.

Answer: Cumulative hidden state across Transformer layers.

Cumulative hidden state across Transformer layers. is the correct answer here. Residual connections preserve info. This matches the core idea being tested around which option best describes the residual stream. The remaining choices fail because they don’t satisfy the full definition.

Q86. What is the primary purpose of the residual stream?

A. A loss function. B. A vocabulary table. C. Cumulative hidden state across Transformer layers. D. A retrieval index.

Select an answer to check.

Answer: Cumulative hidden state across Transformer layers.

Here, Cumulative hidden state across Transformer layers. is the right choice. Residual connections preserve info. That is exactly the concept behind what is the primary purpose of the residual in this context. The remaining choices fail because they don’t satisfy the full definition.

Q87. Which statement about the residual stream is most accurate?

A. Cumulative hidden state across Transformer layers. B. A retrieval index. C. A loss function. D. A vocabulary table.

Select an answer to check.

Answer: Cumulative hidden state across Transformer layers.

In this case, Cumulative hidden state across Transformer layers. is correct. Residual connections preserve info. That is exactly the concept behind which statement about the residual stream is most in this context. The remaining choices fail because they don’t satisfy the full definition.

Q88. How is the residual stream best characterized?

A. A vocabulary table. B. Cumulative hidden state across Transformer layers. C. A loss function. D. A retrieval index.

Select an answer to check.

Answer: Cumulative hidden state across Transformer layers.

The best option here is Cumulative hidden state across Transformer layers.. Residual connections preserve info. That is exactly the concept behind how is the residual stream best characterized in this context. The remaining choices fail because they don’t satisfy the full definition.

Q89. Which option best describes attention?

A. Mechanism where tokens influence each other. B. A normalization step. C. A pure pooling layer. D. A 1D convolution.

Select an answer to check.

Answer: Mechanism where tokens influence each other.

For this question, Mechanism where tokens influence each other. is correct. Core to Transformers. That is exactly the concept behind which option best describes attention in this context. The remaining choices fail because they don’t satisfy the full definition.

Q90. What is the primary purpose of attention?

A. Mechanism where tokens influence each other. B. A normalization step. C. A pure pooling layer. D. A 1D convolution.

Select an answer to check.

Answer: Mechanism where tokens influence each other.

Mechanism where tokens influence each other. is the correct answer here. Core to Transformers. That is exactly the concept behind what is the primary purpose of attention in this context. The remaining choices fail because they don’t satisfy the full definition.

Q91. Which statement about attention is most accurate?

A. A normalization step. B. A pure pooling layer. C. Mechanism where tokens influence each other. D. A 1D convolution.

Select an answer to check.

Answer: Mechanism where tokens influence each other.

Here, Mechanism where tokens influence each other. is the right choice. Core to Transformers. It fits the requirement in the prompt about which statement about attention is most accurate. The remaining choices fail because they don’t satisfy the full definition.

Q92. How is attention best characterized?

A. Mechanism where tokens influence each other. B. A pure pooling layer. C. A 1D convolution. D. A normalization step.

Select an answer to check.

Answer: Mechanism where tokens influence each other.

In this case, Mechanism where tokens influence each other. is correct. Core to Transformers. It fits the requirement in the prompt about how is attention best characterized. The remaining choices fail because they don’t satisfy the full definition.

Q93. Which option best describes sequence length?

A. Number of layers. B. The model's parameter count. C. Vocabulary size. D. Number of tokens in a single sequence.

Select an answer to check.

Answer: Number of tokens in a single sequence.

The best option here is Number of tokens in a single sequence.. Bounded by context window. It fits the requirement in the prompt about which option best describes sequence length. The remaining choices fail because they don’t satisfy the full definition.

Q94. What is the primary purpose of sequence length?

A. The model's parameter count. B. Number of tokens in a single sequence. C. Number of layers. D. Vocabulary size.

Select an answer to check.

Answer: Number of tokens in a single sequence.

For this question, Number of tokens in a single sequence. is correct. Bounded by context window. It fits the requirement in the prompt about what is the primary purpose of sequence length. The remaining choices fail because they don’t satisfy the full definition.

Q95. Which statement about sequence length is most accurate?

A. The model's parameter count. B. Number of layers. C. Vocabulary size. D. Number of tokens in a single sequence.

Select an answer to check.

Answer: Number of tokens in a single sequence.

Number of tokens in a single sequence. is the correct answer here. Bounded by context window. It fits the requirement in the prompt about which statement about sequence length is most accurate. The remaining choices fail because they don’t satisfy the full definition.

Q96. How is sequence length best characterized?

A. Vocabulary size. B. The model's parameter count. C. Number of tokens in a single sequence. D. Number of layers.

Select an answer to check.

Answer: Number of tokens in a single sequence.

Here, Number of tokens in a single sequence. is the right choice. Bounded by context window. This is the most accurate statement for how is sequence length best characterized. The remaining choices fail because they don’t satisfy the full definition.

Q97. Which option best describes an LLM provider?

A. A monitoring tool. B. A web hosting platform. C. A vector database vendor. D. A service hosting LLM inference (OpenAI, Anthropic, etc.).

Select an answer to check.

Answer: A service hosting LLM inference (OpenAI, Anthropic, etc.).

In this case, A service hosting LLM inference (OpenAI, Anthropic, etc.). is correct. Provides API access to models. This is the most accurate statement for which option best describes an llm provider. The remaining choices fail because they don’t satisfy the full definition.

Q98. What is the primary purpose of an LLM provider?

A. A service hosting LLM inference (OpenAI, Anthropic, etc.). B. A vector database vendor. C. A monitoring tool. D. A web hosting platform.

Select an answer to check.

Answer: A service hosting LLM inference (OpenAI, Anthropic, etc.).

The best option here is A service hosting LLM inference (OpenAI, Anthropic, etc.).. Provides API access to models. This is the most accurate statement for what is the primary purpose of an llm. The remaining choices fail because they don’t satisfy the full definition.

Q99. Which statement about an LLM provider is most accurate?

A. A web hosting platform. B. A monitoring tool. C. A vector database vendor. D. A service hosting LLM inference (OpenAI, Anthropic, etc.).

Select an answer to check.

Answer: A service hosting LLM inference (OpenAI, Anthropic, etc.).

For this question, A service hosting LLM inference (OpenAI, Anthropic, etc.). is correct. Provides API access to models. This is the most accurate statement for which statement about an llm provider is most. The remaining choices fail because they don’t satisfy the full definition.

Q100. How is an LLM provider best characterized?

A. A service hosting LLM inference (OpenAI, Anthropic, etc.). B. A vector database vendor. C. A monitoring tool. D. A web hosting platform.

Select an answer to check.

Answer: A service hosting LLM inference (OpenAI, Anthropic, etc.).

A service hosting LLM inference (OpenAI, Anthropic, etc.). is the correct answer here. Provides API access to models. This is the most accurate statement for how is an llm provider best characterized. The remaining choices fail because they don’t satisfy the full definition.