Questions with options and answers of TCS Generative AI E1 (Course 6618):
1. How can Generative AI be used in virtual reality applications?
- A) To improve network speed
- B) To simulate a realistic environment
- C) To enhance security features
- D) To create new hardware components
Answer: B) To simulate a realistic environment
2. What is DALLE?
- A) A text generation tool
- B) A virtual assistant
- C) A generative model for images
- D) An AI for analyzing data
Answer: C) A generative model for images
3. How does ChatGPT generate responses?
- A) By retrieving prewritten answers
- B) By analyzing user inputs and generating relevant responses
- C) By using a random number generator
- D) By copying responses from the internet
Answer: B) By analyzing user inputs and generating relevant responses
4. Which of the following are Generative AI tools?
- A) ChatGPT
- B) DALLE
- C) All are correct
- D) None are correct
Answer: C) All are correct
5. Who created ChatGPT?
- A) Google
- B) Microsoft
- C) OpenAI
- D) IBM
Answer: C) OpenAI
6. What is a prompt in the context of Generative AI?
- A) An output image
- B) An input text
- C) A processing unit
- D) A hardware component
Answer: B) An input text
7. What is the use of Generative AI in the Telecom domain?
- A) To simulate network traffic
- B) To analyze call patterns
- C) All are correct
- D) To generate realistic voice responses
Answer: C) All are correct
Question 1: Github CoPilot is useful for
- Generate code
- Generate test cases
- Generate documentation for code
- All are correct
Answer: All are correct
Question 2: Rachel is a data scientist working on a data analysis project in R. She needs to create functions to preprocess the data before analysis. However, she is not familiar with the language. Which tool can Rachel use to assist her in writing R code efficiently?
- Refer to R programming books
- Take an online course on data analysis
- Consult with a data science expert
- Utilize GitHub Copilot to generate R code snippets
Answer: Utilize GitHub Copilot to generate R code snippets
Question 3: What is the purpose of the General Data Protection Regulation (GDPR)?
- To ensure the responsible use of AI technologies.
- To ensure the development of AI technologies.
- To ensure the use of AI technologies.
- To ensure personal data protection and privacy.
Answer: To ensure personal data protection and privacy.
Question 4: Types of LLM applications
- LLM RAG applications
- Fine Tuning LLMs
- Building LLMs from scratch
- All are correct
Answer: All are correct
Question 5: What is the role of Generative Model in RAG pipeline?
- Reranks retrieved passages
- Generates final output
- Pulls relevant passages or documents from a knowledge source or database
- None of the above
Answer: Generates final output
Question 6: What would be an appropriate task to perform after retrieving passages using the Retriever in a RAG model?
- Fine-tune the retriever model on a different dataset
- Pass the retrieved passages to the generator for response generation.
- Tokenize the retrieved passages using the Rag Tokenizer.
- Generate responses directly using the RagRetriever.
Answer: Pass the retrieved passages to the generator for response generation.
Question 7: A company is developing a chatbot using generative AI to assist customers with technical support queries. They want the chatbot to provide accurate and helpful responses. Which technique is suitable for ensuring the chatbot generates relevant and informative answers?
- Crafting prompts with diverse variations
- Using pre-trained prompts from a similar domain
- Incorporating user feedback during model training
- Adjusting prompt length and complexity based on query type
Answer: Incorporating user feedback during model training
Question 8: Which of the following prompts might be an example for "Role prompting"?
- Provide some pointers for a speech on International Yoga Day.
- Generate some good marketing ideas that could be used by a Marketing Executive to drive a campaign on Blood donation.
- Explain the concept of Stock Market risks to a youngster who is around 15-18 years of age
- All are correct
Answer: Only B
Question 9: In developing a Gen AI-driven financial advisory tool, how can the principle of Data Diversity be practically applied?
- By exclusively using data from a single demographic for accuracy.
- By gathering data from various socioeconomic backgrounds and geographic regions
- By prioritizing data from a specific age group
- By ignoring diverse data sources to simplify the analysis
Answer: By gathering data from various socioeconomic backgrounds and geographic regions
Question 10: Mark is developing a web application using JavaScript. He encounters a problem while implementing a complex sorting algorithm and needs assistance to optimize the code. Which tool can provide code suggestions and improvements?
- Use GitHub Copilot to generate optimized code snippets with blocking public code
- Use GitHub Copilot to generate optimized code snippets without blocking public code
- Seek guidance from a mentor
- All are correct
Answer: Use GitHub Copilot to generate optimized code snippets without blocking public code
Question 11: A social media platform is implementing generative AI to generate personalized recommendations for users based on their interests. To ensure the recommendations are accurate and relevant, which aspect of prompt engineering should they focus on ?
- Customizing prompts based on user profiles and preferences
- Using generative prompts with high linguistic diversity
- Incorporating sentiment analysis to gauge user reactions
- Adjusting the model architecture to handle large-scale data processing
Answer: Customizing prompts based on user profiles and preferences
Question 12: What is the role of vector databases in AI applications?
- To store metadata
- To store scalar embeddings for fast retrieval and similarity search
- To store vector embeddings for fast retrieval and similarity search
- To store vector embeddings for fast retrieval and dissimilarity search
Answer: To store vector embeddings for fast retrieval and similarity search
Question 13: In a scenario where you want to generate a paragraph discussing a specific topic, which component of a RAG model would you primarily rely on?
- Generator
- Retriever
- Discriminator
- Encoder
Answer: Generator
Question 14: Consider the following Python code snippet:
```python
from transformers import RagTokenizer, RagRetriever, RagTokenForGeneration
tokenizer = RagTokenizer.from_pretrained("facebook/rag-token-base")
retriever = RagRetriever.from_pretrained("facebook/rag-token-base")
generator = RagTokenForGeneration.from_pretrained("facebook/rag-token-base")
Query = "What is the capital of France?"
context = "France is a country located in Western Europe. Its capital is Paris."
input_dict = tokenizer.prepare_seq2seq_batch(query, return_tensors="pt")
generated = generator.generate(input_ids=input_dict['input_ids'])
print(tokenizer.decode(generated[0], skip_special_tokens=True))
```
What is the purpose of the `tokenizer.prepare_seq2seq_batch()` method in this code?
- To prepare the input query for the sequence-to-sequence generation model by tokenizing the query and converting it into the appropriate tensor format.
- To preprocess the retrieved passages for the RAGRetriever
Answer: To prepare the input query for the sequence-to-sequence generation model by tokenizing the query and converting it into the appropriate tensor format.
Question 15: What is AI governance?
- AI governance is a range of policies, frameworks and practices that organizations and governments implement to ensure the responsible use of AI technologies.
- AI governance is a range of policies, frameworks and practices that organizations and governments implement to ensure the rapid development of AI technologies.
- AI governance is a range of policies, frameworks and practices that organizations and governments implement to ensure the use of AI technologies.
- AI governance is a range of policies, frameworks and practices that organizations and governments implement to ensure the efficient use of AI technologies.
Answer: AI governance is a range of policies, frameworks and practices that organizations and governments implement to ensure the responsible use of AI technologies.
Question 16: What is the purpose of Re-ranking step in RAG pipeline?
- Filters and re-orders the k retrieved passages to pick the most relevant ones for the query
- Concatenates the top re-ranked m passages with the original query
- Generates final output
- None of the above
Answer: Filters and re-orders the k retrieved passages to pick the most relevant ones for the query
Question 17: How can finetuning the embedding model benefit retrieval performance?
- By reducing the size of the model
- By allowing for more meaningful embedding representations over a training distribution of data
- By translating data into multiple languages
- By encrypting the data for security purposes
Answer: By allowing for more meaningful embedding representations over a training distribution of data
Question 18: In the following Python code snippet, what is the purpose of the `skip_special_tokens=True` argument?
```python
from transformers import RagTokenizer, RagTokenForGeneration
tokenizer = RagTokenizer.from_pretrained("facebook/rag-token-base")
generator = RagTokenForGeneration.from_pretrained("facebook/rag-token-base")
input_ids = tokenizer.encode("What is the capital of France?", return_tensors="pt")
generated = generator.generate(input_ids=input_ids)
print(tokenizer.decode(generated[0], skip_special_tokens=True))
```
- Skips tokenization of special characters.
- Skips the generation of special tokens
- Skips the decoding of special tokens.
Answer: Skips the decoding of special tokens.
Question 19: Which of the following is missing in the prompt given below?
Prompt: You are an assistant who can speak only in a language called 'Spices'. This is a new language which always begins as 'Chilli...'. Use this language to respond to any question asked.
- Context is missing
- Examples for the language is missing
- Prompt is very much clear and concise
- All are correct
Answer: Examples for the language is missing
Question 20: What is the role of Guardrails in governing the usage of AI systems?
- Guardrails serve as protective barriers to mitigate risks and threats associated with using GenAI
- Guardrails do not aim to prevent malicious or harmful exploits of GenAI models
- Guardrails can enhance GenAI model weights
- Guardrails are used to safeguard data privacy while training GenAI models
Answer: Guardrails serve as protective barriers to mitigate risks and threats associated with using GenAI.
Question 21: An e-commerce platform is deploying generative AI to generate product descriptions. They want the descriptions to be informative and persuasive. Which prompt engineering technique would be most effective in achieving this objective?
- Incorporating customer reviews and feedback into prompt generation
- Utilizing sentiment analysis to gauge the emotional impact of descriptions
- Fine-tuning the model with product-specific prompts and attributes
- Increasing the model's vocabulary size to improve descriptive capabilities
Answer: Fine-tuning the model with product-specific prompts and attributes
Question 22: You are a new joinee in TCS and you are given access to the TCS AI Experience Zone for Azure. From the given options below choose the correct option?
- Do not share your credentials with any other individual for accessing the TCS AI Experience Zone for Azure VDI as well as for subsequent access to Enterprise Playground App and Enterprise Playground API (Hands-on lab).
- Share your credentials with your friends and colleagues for accessing the TCS AI Experience Zone for Azure
- Share the TCS device from which you are logged into VDI or TCS AI Experience Zone for Azure apps to another individual.
- Engage in screen share and provide access to other individuals via screen share
Answer: Do not share your credentials with any other individual for accessing the TCS AI Experience Zone for Azure VDI as well as for subsequent access to Enterprise Playground App and Enterprise Playground API (Hands-on lab).
Question 23: If you want more non-deterministic response from any text generating tools, which value of the temperature parameter would you choose?
- 0
- 1
- 0.5
- None of the above
Answer: 1
Question 24: What is "data drift" in the context of GenAI accountability?
- Monitoring the performance of AI models
- Ensuring data quality and integrity
- Checking if the model's performance is declining over time
- All are correct
Answer: Checking if the model's performance is declining over time
Question 25: Can code generated by Generative AI tools owned by others lead to any legal issues for the organization?
- Yes, it can lead to legal issues hence it is crucial to read and understand the product terms
- No, it will not lead to any legal issues.
- It depends upon the situation
- None of the above
Answer: Yes, it can lead to legal issues hence it is crucial to read and understand the product terms.
Question 26: Sarah is working on a project in Python and needs to implement a function to calculate the Fibonacci sequence. However, she is not sure about the syntax and the most efficient algorithm to use. Which tool can Sarah leverage to assist her in this situation?
- Use GitHub Copilot to generate optimized code snippets with blocking public code.
- Use Google Duet to generate optimized code snippets
- Use AWS CodeWhisperer to generate optimized code snippets.
- All are correct
Answer: All are correct
Question 27: How does RAG ensure the generated responses are contextually relevant to the query and retrieved passages?
- By using a reinforcement learning-based reward system
- By fine-tuning the model on a specific domain dataset
- By integrating the retrieved passages as input during generation
- By applying post-processing techniques to filter irrelevant responses
Answer: By integrating the retrieved passages as input during generation
Question 28: What benefits do Guardrails offer?
- Mitigating Bias and Fairness
- Preventing Misuse and Harm
- Ensuring Data Privacy and Security
- All of the above
Answer: All of the above
Question 29: Which of the following is missing in the prompt given below?
Prompt: You are an assistant who can speak only in a language called 'Spices'. This is a new language which always begins as 'Chilli...'. Use this language to respond to any question asked.
- Context is missing
- Examples for the language is missing
- Prompt is very much clear and concise
- All are correct
Answer: Examples for the language is missing
Question 30: A research team is training a generative AI model to produce text summaries of news articles. They want to ensure the generated summaries are accurate and do not propagate misinformation. What approach should they take?
- Train the model on a diverse dataset without fact-checking the generated summaries
- Incorporate fact-checking mechanisms to validate the accuracy of the generated summaries
- Increase the model's complexity to minimize the chances of generating misinformation.
- Rely on post-publication correction mechanisms to address any misinformation generated by the model.
Answer: Incorporate fact-checking mechanisms to validate the accuracy of the generated summaries
Q1: What happens if the token limit is exceeded?
- The model refuses the request
- The model processes the request partially
- The model ignores the token limit
- The model crashes
Answer: The model refuses the request
Q2: What is a one shot prompt?
- A prompt without any examples
- A prompt with multiple examples
- A prompt with only one example
- A prompt with negative examples
Answer: A prompt with only one example
Q3: How can you influence GitHub Copilot's suggestions?
- By providing extensive comments
- By using different function class names
- By adding types to your code
- All of the above
Answer: All of the above
Q4: What is chunking?
- Breaking up a large piece of text into smaller chunks
- Combining multiple texts into a single chunk
- Removing unnecessary tokens from a text
- Rearranging the tokens in a text
Answer: Breaking up a large piece of text into smaller chunks
Q5: How can token limits be avoided?
- By restructuring the initial prompt
- By using chunking techniques
- By summarizing parts of the text
- All of the above
Answer: All of the above
Q6: What is the current credit amount given for api keys?
- 3
- 18
- 20
- 5
Answer: 20
Q7: What is the purpose of retry logic in prompt engineering?
- To prevent hallucinations in the Al's response
- To ensure consistent formatting of the response
- To handle cases where the Al fails to provide the expected format
- To improve the reliability of the Al's responses
Answer: To handle cases where the Al fails to provide the expected format
Q8: How can you access GitHub Copilot?
- Through a web browser
- Through Visual Studio Code
- Through GitHub's website
- Through a mobile app
Answer: Through Visual Studio Code
Q9: What is the role of human evaluation in prompt engineering?
- To determine the reliability of the prompt
- To fine-tune the model for specific use cases
- To eliminate the need for prompt engineering
- To prevent prompt injection attacks
Answer: To determine the reliability of the prompt
Q10: How does prompt engineering contribute to the reliability of Al models?
- It decreases the reliability of Al output
- It has no impact on the reliability of Al output
- It increases the reliability of Al output
- It depends on the complexity of the prompt
Answer: It increases the reliability of Al output
Q11: What is the importance of context length?
- It affects the cost of the request
- It determines the model's output
- It affects the token limit
- It determines the model's accuracy
Answer: It affects the token limit
Q12: What can be specified to change the style of an image generated by the AI?
- The type of style needed
- The format of the response
- The number of retry attempts
- The prompt engineering technique
Answer: The type of style needed
Q13: How can you toggle through, multiple suggestions in GitHub Copilot?
- By pressing the Tab key
- By pressing the Enter key
- By pressing the Esc key
- By pressing the Shift key
Answer: By pressing the Tab key
Q14: What does copilot do when you define a complete function?
- It generates the documentation for the function
- It copies some of the original code from the if statement
- It understands and learns your coding patterns
- It scans your entire project directory
Answer: It understands and learns your coding patterns
Q15: When should prompt engineering be used?
- When creating a product name for personal use
- When rigorously testing a prompt for development and subsequently production
- When evaluating image models
- When using toxic words in a prompt
Answer: When rigorously testing a prompt for development and subsequently production
Q16: What is the recommended approach when providing examples in prompt engineering?
- Give two similar examples to constrain the creative space
- Give multiple examples for better results
- Give negative examples to limit Al output
- Give no examples for better results
Answer: Give two similar examples to constrain the creative space
Q17: How can hallucinations be avoided when using language models?
- By providing factual answers
- By adding extra prompts
- By training the models on complete information
- By avoiding biases in the training data
Answer: By adding extra prompts
Q18: What is the correct order of steps to follow when making a request to the Chatgpt API?
- Select post method, paste URL, select body, select raw, select JSON, provide model and messages
- Select get method, paste URL, select body, select raw, select JSON, provide model and messages
- Select post method, paste URL, select body, select raw, select text, provide model and messages
- Select get method, paste URL, select body, select raw, select text, provide model and messages
Answer: Select post method, paste URL, select body, select raw, select JSON, provide model and messages
Q19: What can you adjust in chat mode to increase variability?
- Maximum length
- Model parameters
- Temperature
- Instructions
Answer: Temperature
Q20: What are neural networks?
- Algorithms that analyze big data
- Algorithms that simulate like the human brain
- Algorithms that create new datasets
- Algorithms that solve complex equations
Answer: Algorithms that simulate like the human brain