Unit 6: Artificial Intelligence Ecosystem and Future Trends - Practice Quiz

1 What does an Artificial Intelligence development ecosystem primarily include?

2 What is the main role of an AI framework in the development ecosystem?

3 Why is Python widely used in Artificial Intelligence development?

4 Which Python library is commonly used for numerical array operations?

5 What are TensorFlow and PyTorch mainly used for?

6 Which organization originally developed TensorFlow?

7 Which organization originally developed PyTorch?

8 What is Hugging Face best known for in the AI ecosystem?

9 What does the Hugging Face Transformers library provide?

10 What is Ollama commonly used for?

11 What is one benefit of running an AI model locally with Ollama?

12 What is LangChain mainly designed to help developers build?

13 In LangChain, what is a chain?

14 What does a vector database primarily store and search?

15 What is similarity search in a vector database used to find?

16 Which activity usually occurs before training an AI model?

17 Why is an AI model evaluated after training?

18 Which development is a major future trend in Artificial Intelligence?

19 What does Artificial General Intelligence refer to?

20 What is a main goal of Sustainable AI?

21 A team wants to replace its image-classification model without changing the web application that uses it. Which ecosystem design best supports this requirement?

22 An AI service performs well during testing but slows down when many users access it simultaneously. Which ecosystem component should the team examine first?

23 Two Python AI projects require incompatible versions of the same library. What is the most appropriate way to prevent dependency conflicts?

24 A data scientist wants teammates to reproduce the Python environment used to train a model. Which artifact is most directly useful?

25 A researcher frequently changes a neural network during experiments and wants Python-like debugging with immediate execution. Which choice is most suitable?

26 A trained TensorFlow model must run locally on a mobile device with limited memory. Which deployment option is designed for this use case?

27 A developer needs to build a sentiment classifier quickly using an existing transformer model. Which Hugging Face feature offers the most direct starting point?

28 Before using a model downloaded from the Hugging Face Hub in a commercial product, what should a development team verify first?

29 A hospital wants employees to summarize sensitive notes without sending them to an external AI provider. When suitable hardware and governance are available, why might Ollama be useful?

30 A large model runs too slowly on a developer's laptop through Ollama. Which change is most likely to improve local inference feasibility while preserving the same general task?

31 An assistant must answer a question by retrieving company policies, inserting relevant passages into a prompt, and calling a language model. What role can LangChain play?

32 A LangChain agent can call tools that send emails and modify customer records. Which design most effectively reduces the risk of harmful actions?

33 A search system stores document embeddings and receives a query embedding. What does the vector database primarily use to find relevant documents?

34 An enterprise search application must return semantically similar documents, but only from the user's department. Which query strategy is most appropriate?

35 A fraud-detection dataset contains only 1% fraudulent transactions. Why can accuracy alone be a misleading evaluation metric?

36 A recommendation model performs well initially, but its quality declines after customer preferences change. Which workflow practice best addresses this problem?

37 An AI assistant must interpret a chart, read its caption, and answer a spoken question about both. Which future AI direction is most relevant?

38 A manufacturer wants an AI system to inspect products with low latency even when internet connectivity is unreliable. Which trend best supports this goal?

39 Which capability would provide stronger evidence of Artificial General Intelligence than of a narrow AI system?

40 A company wants to reduce the environmental impact of an AI service without removing the service. Which action is most directly aligned with Sustainable AI?

41 A team can reproduce a model's Python packages from a lockfile, but inference outputs differ across machines using the same input and checkpoint. One machine uses a different GPU architecture and CUDA/cuDNN stack. Which action most directly addresses this reproducibility gap?

42 A CPU-bound Python preprocessing function executes pure Python loops and cannot keep a GPU training job supplied with data. Profiling confirms that storage is not the bottleneck. Which change is most likely to improve throughput?

43 A PyTorch model contains dropout and batch-normalization layers. An engineer wraps inference in torch.no_grad(), yet repeated predictions remain inconsistent and batch-normalization statistics continue to change. What is missing?

44 A custom training loop must compute per-example gradients for a loss and then differentiate a regularizer built from those gradients. Which capability is fundamentally required?

45 A production service loads a model and tokenizer from the same Hugging Face repository name. After the repository owner updates the default branch, token IDs change while the service still uses cached model weights. What is the strongest preventive measure?

46 A decoder-only Hugging Face model is batch-generating continuations from prompts of unequal length. With right padding, the system selects logits from the final padded position for shorter prompts and produces degraded output. Which correction best preserves batched generation semantics?

47 An Ollama deployment uses the same quantized model weights in two custom model definitions, but one definition consistently follows chat instructions better. Hardware and generation parameters are identical. Which difference most plausibly explains the result?

48 A LangChain retrieval pipeline must stream model tokens, preserve request-specific metadata, and avoid blocking the event loop while calling an asynchronous vector store. Which design is most appropriate?

49 An agent can call a tool that executes database queries. Retrieved documents may contain malicious text telling the agent to ignore its policy and invoke that tool. Which control most directly limits the resulting security risk?

50 A vector database uses cosine similarity. All stored vectors and the query vector are normalized to unit length. Which ranking statement is correct?

51 A company replaces its embedding model with one having the same output dimension, then continues querying an index containing vectors from the old model. Retrieval quality collapses. Why does matching dimension not prevent this?

52 A fraud model is evaluated using random train-test splitting over transactions. The same customers appear in both sets, and each customer's behavior is highly stable. Offline performance is excellent but performance on newly acquired customers is poor. Which evaluation redesign best estimates the deployment scenario?

53 A model's online accuracy declines, but the marginal distribution of each monitored input feature appears unchanged. Which investigation can still reveal a meaningful form of drift?

54 A benchmark reports that a new autonomous agent completes 80% of tasks, but the benchmark allows unlimited retries and excludes API charges, latency, and human recovery effort. Which conclusion is best supported?

55 An edge device must run a multimodal assistant with intermittent connectivity, strict privacy requirements, and limited memory. Which architecture best reflects likely near-term AI deployment trends?

56 A system exceeds human performance on thousands of benchmark tasks but requires task-specific fine-tuning and cannot transfer knowledge to unfamiliar domains without new labeled data. Which claim is most defensible?

57 Which evaluation design most strongly tests whether an alleged AGI can generalize rather than reproduce memorized benchmark patterns?

58 Two data centers run the same training job with identical energy consumption. Center A reports lower operational carbon emissions than Center B. Which factor can validly explain the difference?

59 A team compares two language models. Model X uses less inference energy per request, but its lower accuracy causes users to retry more often. What is the most appropriate sustainability metric?

60 An AI service pulls an unpinned model artifact containing custom loading code from a public registry during every deployment. Which combination most directly reduces software supply-chain risk?