Question bank

4,310 interview questions, answered.

Reference answers, what the interviewer is really testing, how it is graded, and the follow-ups that come next.

Easy 4Medium 31Hard 15
Filters · onDifficulty, topic, company
Clear all

50 questions

  1. 1Define hallucinations in LLMs and explain how you would reduce them.Easy· 5 min· OpenAI, Anthropic
  2. 2Design a language model that minimizes harmful outputs while remaining useful and expressive.HardSystem design· 30 min· Anthropic
  3. 3Evaluate LLMs on a toy task and use LLMs to generate additional evaluation data.Hard· 30 min· Anthropic
  4. 4How do you approach generative-AI safety in a consumer product?HardProduct· 20 min· Google, Anthropic
  5. 5What generative-AI skills will be most critical in the next year?Easy· 5 min· Anthropic
  6. 6When is a prompt cache calculated, and what is a model harness?Hard· 20 min· Anthropic
  7. 7Write code rapidly to a specification, then refactor it as new constraints are introduced.HardCoding· 20 min· Anthropic
  8. 8Debug an actual LLM inference step in a 20–30 line Colab skeleton (Anthropic)HardCodingSystem design· Anthropic
  9. 9Design a distributed training pipeline for an LLM with fault toleranceHardCodingSystem design· Anthropic
  10. 10Design a system for safe deployment of AI models in productionHardCodingSystem design· Anthropic
  11. 11Generate outputs for images and pipelines (Anthropic)MediumCodingSystem design· Anthropic
  12. 12Optimize a Python-simulated TPU kernel to beat 1487 cycles (Anthropic performance take-home)HardCodingSystem design· Anthropic
  13. 13Define hallucinations in LLMs.MediumAI MLTechnical· Anthropic, Mastercard
  14. 14Design a web crawler.MediumSystem design· Anthropic, Google
  15. 15Design a key-value store.HardSystem design· Accenture, Anthropic
  16. 16How do you think about AI safety, ethics, guardrails, and governance?MediumAI MLTechnical· Anthropic
  17. 17How would you handle hallucinations in a generative AI model deployed to users?MediumAI MLTechnical· Anthropic, OpenAI
  18. 18Tell me about a time the business problem wasn’t clearly defined. How did you handle it?MediumBehavioral· Anthropic, Uber
  19. 19Design an inference batching system for a single GPU that can handle up to 100 inputs per batch while users wait synchronously, maximizing utilization under compute constraints.HardAI MLSystem design· Anthropic
  20. 20Implement LRU Cache.HardCoding· Accenture, Adobe
  21. 21Describe your process for developing a marketing strategy for new products, especially in highly competitive industries like consumer electronics.HardTechnical· Anthropic
  22. 22Tell me about a time when you received negative feedback and how you handled it.MediumBehavioral· Amazon, American Express
  23. 23How do you stay up to date with advancements in machine learning?MediumBehavioralAI ML· Anthropic, Capital One
  24. 24Identify success metrics for a marketing campaign to get new users, then design an experiment to determine if the campaign should continue.MediumData· Anthropic