Curated · Short Answer
Top 100 Short Answer Interview Questions
Here are the top 100 AI interview questions in Short Answer 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.
- 01Explain scaled dot-product attention.Short AnswerMedium·Qual 4.8
- 02Design a RAG evaluation harness, what do you measure and how?Short AnswerHard·Qual 4.7
- 03When should you fine-tune instead of using RAG?Short AnswerMedium·Qual 4.6
- 04Explain why transformers replaced RNNs for language modeling.Short AnswerEasy·Qual 4.6
- 05Describe the typical agent execution loop.Short AnswerMedium·Qual 4.4
- 06Why does streaming (SSE) improve perceived latency without changing total decode time?Short AnswerEasy·Qual 4.0
- 07Name two situations where Flat brute force search is still the correct production choice in a 2026 vector pipeline.Short AnswerEasy·Qual 4.0Pinecone
- 08Pretraining vs fine-tuning: two key differences in data and objective.Short AnswerEasy·Qual 4.0
- 09Name a 2026 task where an encoder-decoder transformer still beats a decoder-only LLMShort AnswerEasy·Qual 4.0
- 10Walk through why pre-norm trains stably past 100 layers while post-norm needs warmup tricksShort AnswerMedium·Qual 4.0
- 11Why is chat-template integrity the most common 'silent quality killer' in FT?Short AnswerMedium·Qual 4.0
- 12When IS fine-tuning the right call?Short AnswerMedium·Qual 4.0
- 13Describe how continuous batching interacts with the KV cacheShort AnswerMedium·Qual 4.0
- 14Explain effective context for Mistral's 4096 sliding window over a 16k inputShort AnswerMedium·Qual 4.0
- 15Why is idempotency a prerequisite for safe agent retries, and what breaks when a non-idempotent tool is retried?Short AnswerMedium·Qual 4.0
- 16Describe why structured-output mode quietly raises the per-call token billShort AnswerMedium·Qual 4.0OpenAI
- 17Compare the three MCP transports and identify the correct choice for each deployment scenarioShort AnswerMedium·Qual 4.0
- 18Greedy decoding and temperature sampling: does one cost more per token than the other?Short AnswerMedium·Qual 4.0
- 19How does prefix KV-cache sharing reduce serving cost across requests?Short AnswerMedium·Qual 4.0
- 20Writing the LLM-as-judge prompt for an eval pipeline: name three biases to design against and one mitigation each.Short AnswerMedium·Qual 4.0
- 21Why did continuous batching replace static batching for LLM serving?Short AnswerMedium·Qual 4.0
- 22In a few-shot prompt that extracts structured fields from invoices, why is a distinctive stop sequence (like a custom END_INVOICE marker) better than a paragraph break or the natural end of response?Short AnswerMedium·Qual 4.0
- 23Describe each step of the agent loop and explain which component, LLM or runtime, is responsible for itShort AnswerMedium·Qual 4.0
- 24What events must be logged in every agent execution trace for debugging to be tractable?Short AnswerMedium·Qual 4.0Databricks
- 25Same block, two regimes: training parallelism versus inference autoregressionShort AnswerMedium·Qual 4.0
- 26How does a coding agent's scope differ from inline code completion?Short AnswerMedium·Qual 4.0Cursor
- 27LoRA vs full FT on 7B + 10k examples: order of magnitude cost comparisonShort AnswerMedium·Qual 4.0
- 28Define TTFT, TPOT and throughput. Which should a chatbot vs a batch summarizer optimize?Short AnswerMedium·Qual 4.0
- 29Walk through how a prefix-cache hit changes the work that chunked prefill has to doShort AnswerMedium·Qual 4.0
- 30Defend the 4x FFN expansion ratio, then explain why SwiGLU breaks the simple ruleShort AnswerMedium·Qual 4.0
- 31OpenAI's hosted FT for tool-calling models: what does a JSONL row need beyond text SFT?Short AnswerMedium·Qual 4.0OpenAI
- 32Name two recovery moves a serving stack can pull when KV memory runs short mid-request.Short AnswerMedium·Qual 4.0
- 33How is a RAG vector store used as an external long term memory layer inside an agent?Short AnswerMedium·Qual 4.0
- 34Explain the LLM vs runtime split in an agent and why the distinction matters for debuggingShort AnswerMedium·Qual 4.0
- 35A 1200-word system prompt, 200-word query, 150-word reply on GPT-5.5. Estimate the token cost, then name what doubles it.Short AnswerMedium·Qual 4.0
- 36Adam stores two fp32 buffers per trainable parameter: what does that imply for LoRA vs full FT on a 7B?Short AnswerMedium·Qual 4.0
- 37A production model has no ground-truth labels on live traffic. Design a monitoring plan that catches degradation before users complain.Short AnswerMedium·Qual 4.0
- 38Prefix-cache hit lands on a request: does speculative decoding still help that request?Short AnswerMedium·Qual 4.0
- 39Describe the JSON shape of a Claude Opus 4.7 assistant message that calls a toolShort AnswerMedium·Qual 4.0Anthropic
- 40What does the MCP OAuth 2.1 authorization layer protect and why is it marked optional?Short AnswerMedium·Qual 4.0
- 41Where inside the attention sub-layer is dropout typically applied during training, and what does dropping there teach the model?Short AnswerMedium·Qual 4.0
- 42How do online (interactive) and offline (batch) inference workloads differ in optimization target?Short AnswerMedium·Qual 4.0
- 43What does the scale of the MCP server ecosystem in 2026 indicate about protocol adoption?Short AnswerMedium·Qual 4.0
- 44Describe a concrete, production runnable mechanism to detect hallucinations in a RAG answer (claims unsupported by the retrieved chunks), as the answer is generated or just after.Short AnswerMedium·Qual 4.0Perplexity
- 45What is GGUF and how do llama.cpp / mlc-llm enable on-device inference where vLLM cannot run?Short AnswerMedium·Qual 4.0
- 46A team indexed embeddings under cosine but queried with Euclidean, and recall collapsed. Diagnose the cause.Short AnswerMedium·Qual 4.0Pinecone
- 47How do you detect catastrophic forgetting after a domain fine-tune?Short AnswerMedium·Qual 4.0
- 48Why are MCP Tools and Resources distinct primitives rather than one read or write abstraction?Short AnswerMedium·Qual 4.0
- 49Name two production KV-cache eviction policies and what each prioritizesShort AnswerMedium·Qual 4.0
- 50What UX patterns help users make informed MCP approval decisions without approval fatigue?Short AnswerMedium·Qual 4.0Cursor
- 51Explain LIMA's Superficial Alignment Hypothesis and what it implies for your data budgetShort AnswerMedium·Qual 4.0
- 52Closed-API fine-tuning in 2025-2026: options and trade-offs vs open-weight FTShort AnswerMedium·Qual 4.0
- 53Sketch a Llama 3.1 70B self-host versus hosted-API break-even on the back of an envelopeShort AnswerMedium·Qual 4.0
- 54Explain the structural reason BLEU fails for open-ended LLM evaluationShort AnswerMedium·Qual 4.0
- 55Llama-4 fine-tune shipping to 800M MAU: which license clause must the team re-read?Short AnswerMedium·Qual 4.0
- 56For a 64-Q-head model with GQA group size G=8, how many K/V heads exist and what is the cache reduction vs MHA?Short AnswerMedium·Qual 4.0
- 57Why are output tokens typically 3-5× more expensive than input tokens?Short AnswerMedium·Qual 4.0AnthropicOpenAI
- 58Why does prefill saturate compute while decode is bottlenecked on memory bandwidth?Short AnswerMedium·Qual 4.0NVIDIA
- 59How does a host handle tool name collisions when connected to multiple MCP servers?Short AnswerMedium·Qual 4.0Cursor
- 60Picking a cheap but predictive eval to track during training: what do you actually log?Short AnswerMedium·Qual 4.0
- 61Tradeoffs between pairwise LLM-judge comparisons and scalar scoring rubricsShort AnswerMedium·Qual 4.0
- 62How does a Vision Transformer turn an image into a sequence, and what mask runs over it?Short AnswerMedium·Qual 4.0
- 63Compare OpenAI function calling, Anthropic tool use, and ReAct free-form text as agent tool-invocation mechanismsShort AnswerMedium·Qual 4.0OpenAI
- 64Describe two automated hallucination detection techniques and their tradeoffsShort AnswerMedium·Qual 4.0
- 65Design a fine-tune that teaches a model to refuse when the retrieved context lacks the answer.Short AnswerMedium·Qual 4.0
- 66Why is 'reduce FLOPs to speed up decode' the most common beginner misconception in LLM serving?Short AnswerMedium·Qual 4.0NVIDIA
- 67Explain attention as content-addressable memory. Where does the analogy hold and where does it break?Short AnswerMedium·Qual 4.0
- 68Why should agent tools be atomic rather than composite, and what goes wrong when a tool tries to do too much?Short AnswerMedium·Qual 4.0Anthropic
- 69Explain the M×N integration problem that MCP solves and why the framing mattersShort AnswerMedium·Qual 4.0
- 70Name the architectural shifts vLLM v1 made relative to v0Short AnswerMedium·Qual 4.0
- 71How does a chain-of-thought judge differ from a direct-score judge, and when is each better?Short AnswerMedium·Qual 4.0
- 72A distilled model hits 92% on a public benchmark but 31% on an internal hold-out: diagnose.Short AnswerMedium·Qual 4.0
- 73When would you need the offset_mapping feature that HuggingFace fast tokenizers provide?Short AnswerMedium·Qual 4.0Hugging Face
- 74Why is batching the single biggest throughput lever for LLM decode?Short AnswerMedium·Qual 4.0
- 75Memory creeps up for hours then OOMs at 3am: diagnose the KV-cache leak patternShort AnswerMedium·Qual 4.0
- 76What would break if you tokenized a Llama prompt using tiktoken's cl100k_base instead of the Llama tokenizer?Short AnswerMedium·Qual 4.0
- 77Compare offline and online LLM evaluation: when is each the right tool?Short AnswerMedium·Qual 4.0
- 78When does inflating the system prompt actually lower cost per call?Short AnswerMedium·Qual 4.0Anthropic
- 79A consumer-facing LLM product is about to launch. Describe the safety evaluation that should run first.Short AnswerMedium·Qual 4.0Anthropic
- 80You suspect your custom eval set leaked into a model's training data. Describe two practical tests you would run to detect contamination.Short AnswerMedium·Qual 4.0
- 81What does perplexity measure and why is it an insufficient LLM quality signal?Short AnswerMedium·Qual 4.0
- 82What does the first forward pass set up that the second one doesn't pay for?Short AnswerMedium·Qual 4.0
- 83Why might a LoRA that passes every eval check then garble one in five prod requests?Short AnswerMedium·Qual 4.0
- 84Decoder-only LLMs ship without a cross-attention sub-layer, so where does the context go?Short AnswerMedium·Qual 4.0
- 85Contrast SFT and continued pretraining when adapting a base model to medical textShort AnswerMedium·Qual 4.0
- 86The PM asks how to tell if the chatbot got better after the last deploy. Walk through setting up an A/B test.Short AnswerMedium·Qual 4.0
- 87How do you evaluate a customer service chatbot that handles multi-turn conversations? What dimensions matter?Short AnswerMedium·Qual 4.0
- 88Three concrete situations where fine-tuning is the wrong toolShort AnswerMedium·Qual 4.0
- 89Sliding-window attention can't see beyond the window. Describe the trick that recovers older context.Short AnswerMedium·Qual 4.0
- 90Explain how MCP differs from RAG and when to choose each approachShort AnswerMedium·Qual 4.0
- 91When should you choose stdio transport for an MCP server, and when is it the wrong choice?Short AnswerMedium·Qual 4.0
- 92You are about to ship a feature that calls GPT-5.5 per request. Walk through how you estimate the token cost before launch.Short AnswerMedium·Qual 4.0OpenAI
- 93Explain when an architecture uses cross-attention vs self-attention. Give concrete examples.Short AnswerMedium·Qual 4.0
- 94Define recall@10 for a vector index, and spell out exactly what 'ground truth' means in that definition.Short AnswerMedium·Qual 4.0Pinecone
- 95How do you navigate the latency, cost, quality triangle when choosing an eval approach?Short AnswerMedium·Qual 4.0
- 96How do URI templates work in MCP Resources, and how do subscriptions differ from a one-shot read?Short AnswerMedium·Qual 4.0
- 97Why does a deep transformer stack stop training if you remove the residual connections?Short AnswerMedium·Qual 4.0
- 98Describe three patterns for inserting human approval into an agent workflow and their tradeoffsShort AnswerMedium·Qual 4.0Anthropic
- 99What problem does A2A solve that MCP does not, and how do they work together?Short AnswerMedium·Qual 4.0
- 100How does assigning a persona to each agent in a multi-agent system shape its behavior?Short AnswerMedium·Qual 4.0
Related top picks
By experience level
Other formats
By topic