Curated · Flashcard
Top 100 Flashcard Interview Questions
Here are the top 100 AI interview questions in Flashcard format, sorted by community quality. Each entry links to a detailed answer with explanations, hints, and source-grade follow-ups. See all in this format →
Sorted by community quality · refreshed as new questions are published.
- 01What is the KV cache in transformer inference?FlashcardEasy·Qual 4.7
- 02PPO vs DPO, what's the practical difference?FlashcardMedium·Qual 4.6AnthropicOpenAI
- 03HNSW vs IVF, when do you pick each for a production vector index?FlashcardMedium·Qual 4.5PineconeQdrant
- 04How does MCP differ from OpenAI-style function calling?FlashcardMedium·Qual 4.5
- 05Estimate the per call cost of a typical RAG chatbot using GPT-4o-mini.FlashcardEasy·Qual 4.4OpenAI
- 06Why is cosine similarity preferred over Euclidean distance for text embeddings?FlashcardEasy·Qual 4.4CohereOpenAI
- 07Why does BPE tokenization use subwords instead of words or characters?FlashcardEasy·Qual 4.4
- 08What is semantic chunking and when does it beat fixed size chunking?FlashcardMedium·Qual 4.3
- 09What causes context pollution in a long-running agent?FlashcardEasy·Qual 4.0
- 10Why run agent-generated code inside an isolated sandbox?FlashcardEasy·Qual 4.0
- 11What does one agent trace capture step by step?FlashcardEasy·Qual 4.0
- 12When should a production agent pause for human approval?FlashcardEasy·Qual 4.0
- 13How is long-term memory typically implemented in an agent?FlashcardEasy·Qual 4.0
- 14Where does an agent keep short-term state during a single task?FlashcardEasy·Qual 4.0
- 15Why cap each agent run with a cost-per-task budget?FlashcardEasy·Qual 4.0
- 16What does observability mean for production LLM agents, and how is it different from app monitoring?FlashcardEasy·Qual 4.0
- 17Define GQA in transformer attentionFlashcardEasy·Qual 4.0
- 18BPE learned 50,000 merge rules during training. What is a merge rule, and where does it live at inference?FlashcardEasy·Qual 4.0
- 19You hear about 'eval harnesses' like Promptfoo and DeepEval. Explain what an eval harness does that a Jupyter notebook cannot.FlashcardEasy·Qual 4.0
- 20What is the attention mechanism, in one sentence, and what makes it structurally different from convolution and recurrence?FlashcardEasy·Qual 4.0
- 21RoPE, spell out the acronym and name exactly which vectors it rotatesFlashcardEasy·Qual 4.0
- 22Flashcard: what is a token in LLM prompting and why does it matter for cost and context?FlashcardEasy·Qual 4.0
- 23Flashcard: what is few shot prompting and what does the 'few' actually refer to?FlashcardEasy·Qual 4.0
- 24GQA, decode the acronym and describe its KV sharing patternFlashcardEasy·Qual 4.0
- 25Why does MCP have a 'prompt' primitive separate from tools and resources?FlashcardEasy·Qual 4.0
- 26Across which axes is the KV cache stored, per layer, per head, per batch?FlashcardEasy·Qual 4.0
- 27Identify PagedAttention and the project it ships withFlashcardEasy·Qual 4.0
- 28Flashcard: what is the ReAct prompting pattern?FlashcardEasy·Qual 4.0
- 29What does a padding mask zero out, and why is it needed when batching variable-length sequences?FlashcardEasy·Qual 4.0
- 30Define LLM-as-judge and explain the one problem it was invented to solve.FlashcardEasy·Qual 4.0
- 31You are building a golden dataset with three annotators. Someone warns you about inter-annotator agreement. What is it and what goes wrong if you skip measuring it?FlashcardEasy·Qual 4.0
- 32How does the temperature parameter reshape the sampling distribution?FlashcardEasy·Qual 4.0
- 33Describe what 'autoregressive' means for LLM generation.FlashcardEasy·Qual 4.0
- 34Describe the residual stream as a read/write surface that sublayers act onFlashcardEasy·Qual 4.0Anthropic
- 35Walk through how top_p (nucleus) sampling truncates the distributionFlashcardEasy·Qual 4.0
- 36Why do hosted LLM APIs charge separate per-million rates for input and output tokens?FlashcardEasy·Qual 4.0
- 37You want to build an MCP server in TypeScript. What package do you use and how does it handle schemas?FlashcardEasy·Qual 4.0
- 38Define LLM inference and explain how it differs from training.FlashcardEasy·Qual 4.0
- 39Where does the MCP client live, and why is it one to one with a server?FlashcardEasy·Qual 4.0
- 40Identify the NVIDIA H100: what generation, what memory, what tensor-core formats?FlashcardEasy·Qual 4.0
- 41TTFT: name the metric it captures and what dominates itFlashcardEasy·Qual 4.0
- 42What is shadow deployment for LLMs and how does it enable safe evaluation?FlashcardEasy·Qual 4.0
- 43In MCP, what exactly is a 'tool' and who decides when to call it?FlashcardEasy·Qual 4.0
- 44Compare hidden_size, d_model, and embedding_dim, three names for the same thing or three different things?FlashcardEasy·Qual 4.0
- 45Your LLM judge gives everything a 4 out of 5. Your colleague says the judge is not calibrated. What does calibration mean here?FlashcardEasy·Qual 4.0
- 46Token 1234 means 'the' in one model and garbage in another. Why?FlashcardEasy·Qual 4.0
- 47Flashcard: is a 0.85 cosine similarity score from your retriever 'high confidence'?FlashcardEasy·Qual 4.0
- 48Rank r in LoRA: what does the number actually count?FlashcardEasy·Qual 4.0
- 49Name the scalar that QK^T is divided by inside scaled dot-product attention.FlashcardEasy·Qual 4.0
- 50BPE grows the vocabulary by adding merges. Unigram does the opposite. Explain the difference.FlashcardEasy·Qual 4.0
- 51Name TGI's maintainer and its niche among serving stacksFlashcardEasy·Qual 4.0Hugging Face
- 52Flashcard: why use Tree of Thoughts search on hard agent planning problems?FlashcardEasy·Qual 4.0
- 53Gradient norm clipping, name the failure mode it exists to preventFlashcardEasy·Qual 4.0
- 54ALiBi, decode the acronym and pinpoint where in the pipeline it injects positionFlashcardEasy·Qual 4.0
- 55Flashcard: what is an 'agent runtime' and what does it do?FlashcardEasy·Qual 4.0
- 56MCP has tools, resources, and prompts. What makes a resource different from a tool?FlashcardEasy·Qual 4.0
- 57Flashcard: why must every agent loop declare when to stop?FlashcardEasy·Qual 4.0
- 58Flashcard: what is similarity search in RAG, and how does it differ from keyword search?FlashcardEasy·Qual 4.0
- 59Bytes, characters, tokens: which one gets billed and how do they relate?FlashcardEasy·Qual 4.0
- 60Why did MCP pick JSON-RPC 2.0 as its wire format instead of REST or gRPC?FlashcardEasy·Qual 4.0
- 61How does Cursor discover and load MCP servers?FlashcardEasy·Qual 4.0Cursor
- 62Why ramp the learning rate during warmup instead of starting at the peak value?FlashcardEasy·Qual 4.0
- 63Flashcard: what role does the LLM's context window play in a RAG pipeline?FlashcardEasy·Qual 4.0
- 64TruthfulQA is designed to catch a specific model failure. Describe what it tests and why larger models sometimes do worse.FlashcardEasy·Qual 4.0
- 65Flashcard: what does it mean when an LLM 'makes a tool call'?FlashcardEasy·Qual 4.0
- 66What happens during the MCP initialize handshake, and why does it matter?FlashcardEasy·Qual 4.0
- 67Your teammate mentions BPE at a code review. What are they talking about?FlashcardEasy·Qual 4.0
- 68Weight quantization in plain terms: what changes and what stays the same?FlashcardEasy·Qual 4.0
- 69Expand MQA and state its KV-cache cost tradeoffFlashcardEasy·Qual 4.0
- 70How does softmax turn an attention score into an attention weight?FlashcardEasy·Qual 4.0
- 71You want to build an MCP server in Python. What do you install and what does the API look like?FlashcardEasy·Qual 4.0
- 72What is the context window of an LLM?FlashcardEasy·Qual 4.0
- 73Explain what gets quantized the second time in QLoRA's double-quantization stepFlashcardEasy·Qual 4.0
- 74Why does dropping the EOS token from SFT labels produce a model that never stops generating?FlashcardEasy·Qual 4.0
- 75Before running LLM-as-judge, someone insists on writing a rubric first. What is a rubric in this context?FlashcardEasy·Qual 4.0
- 76Define supervised fine-tuning (SFT) in one breath.FlashcardEasy·Qual 4.0
- 77Flashcard: when would you pick Plan-and-Execute over ReAct?FlashcardEasy·Qual 4.0
- 78GitHub ships an official MCP server. What can it do, and how does it authenticate?FlashcardEasy·Qual 4.0
- 79Describe the W_O projection in multi-head attention, its shape and what it mixes.FlashcardEasy·Qual 4.0
- 80Decode the acronym MHA in the transformer attention context.FlashcardEasy·Qual 4.0
- 81When someone says 'eval prompt template,' what are they referring to and why does the wording matter so much?FlashcardEasy·Qual 4.0
- 82Server-Sent Events streaming: define it and explain why chat APIs default to itFlashcardEasy·Qual 4.0
- 83You change one word in the system prompt and prompt caching breaks. Why?FlashcardEasy·Qual 4.0
- 84Older tokenizers had an OOV problem. Why did byte-level BPE kill it?FlashcardEasy·Qual 4.0
- 85Explain the difference between pointwise and pairwise evaluation in one breath.FlashcardEasy·Qual 4.0
- 86What integration problem does MCP solve?FlashcardEasy·Qual 4.0
- 87HumanEval is the go-to code generation benchmark. Describe the task it gives the model and how it decides if the answer is correct.FlashcardEasy·Qual 4.0
- 88BLEU score comes up in a meeting. What does it measure and when does it still make sense?FlashcardEasy·Qual 4.0
- 89How is an MCP host different from an MCP client?FlashcardEasy·Qual 4.0
- 90Flashcard: what is an embedding in a RAG pipeline, and what role does it play?FlashcardEasy·Qual 4.0
- 91Define prompt engineering and identify what it actually touches.FlashcardEasy·Qual 4.0
- 92Flashcard: what is a tool in an LLM agent?FlashcardEasy·Qual 4.0
- 93Flashcard: how is agentic RAG different from single-shot RAG?FlashcardEasy·Qual 4.0
- 94Define a single attention head, what does one head actually own and produce?FlashcardEasy·Qual 4.0
- 95You need to add an MCP server to Claude Desktop. Which file do you edit and what goes in it?FlashcardEasy·Qual 4.0Cursor
- 96You hear 'fertility is 2.5 for Hindi on cl100k_base'. What does that number mean and is it good?FlashcardEasy·Qual 4.0
- 97Flashcard: why connect agents through MCP?FlashcardEasy·Qual 4.0
- 98What does Cohen's kappa measure in human LLM evaluation?FlashcardEasy·Qual 4.0
- 99Flashcard: what is an AI agent, in one sentence?FlashcardEasy·Qual 4.0
- 100Chatbot Arena ranks models by Elo rating. Explain the Elo system to someone who has never played competitive chess.FlashcardEasy·Qual 4.0
Related top picks
By experience level
Other formats
By topic