GEN AI- Prompt Engineering – Interview Questions

What is LLM (Large Language Model)?

Answer:
A Large Language Model (LLM) is a deep learning-based model that is trained on vast amounts
of text data to understand and generate human-like text. The model can generate coherent,
contextually accurate text based on input prompts. LLMs are typically built using the
transformer architecture and have billions or even trillions of parameters, enabling them to
capture complex language patterns and relationships.

Scenario:
Think of an LLM as a sophisticated virtual assistant. If you ask it a question like, "What's the
weather today?" the model will understand the context of the query, generate a relevant
response, and even adjust to different conversational styles based on prior interactions.

What are the key parameters of LLM?

Answer:
The key parameters of an LLM include:

• Number of Parameters: Refers to the weights in the model that are learned during
training. Larger models with more parameters can capture more complex relationships
in data, improving their performance.

• Layers: LLMs consist of multiple layers of neural networks. More layers generally result
in a better ability to process and understand complex language.

• Attention Mechanism: This allows the model to focus on different parts of the input
text. The transformer architecture is built around this concept.

• Training Data: The quality and diversity of the data used to train the model impact its
ability to generalize across different domains.

Scenario:
When training GPT-3, the model was trained on a massive amount of internet text, which
helped it learn language, facts, and even common knowledge in various fields.
What are the metrics of LLM?

Answer:
Some common metrics used to evaluate LLM performance include:

• Perplexity: Measures how well the model predicts the next word. A lower perplexity
indicates better performance. It is often used in language modeling tasks.

• BLEU Score: Commonly used for machine translation tasks, BLEU evaluates the similarity
between generated text and a reference translation.

• ROUGE Score: Often used for summarization tasks, ROUGE compares the overlap of n-
grams between the generated text and reference summaries.

• F1 Score: Measures the balance between precision and recall. It’s commonly used in
classification tasks, such as sentiment analysis.

Scenario:
For instance, when evaluating a chatbot, you might use perplexity to assess how accurately the
model is predicting the next word in the conversation. A higher BLEU score would indicate that
the translations provided by the model are closer to human translations.

What is content filtering?

Answer:
Content filtering refers to the process of screening and removing harmful or undesirable
content generated by AI models. In the context of LLMs, this includes filtering out offensive
language, hate speech, explicit content, or other types of toxic behavior.

Content filtering is often implemented using pre-defined rules or by training models to identify
and block inappropriate content.

Scenario:
In a customer service chatbot, content filtering would ensure that the model doesn't generate
offensive or inappropriate responses, ensuring a safe and professional interaction for users.
What is the RAG model and LangChain?

Answer:

• RAG Model (Retrieval-Augmented Generation):

The RAG model combines retrieval-based and generation-based techniques. It first
retrieves relevant information from an external knowledge base and then generates text
by combining the retrieved information with its understanding of the task. This helps the
model provide more accurate and contextually relevant answers.

• LangChain:
LangChain is a framework designed for building applications with LLMs. It allows
developers to integrate LLMs with external APIs, databases, or documents, enabling
more dynamic and context-aware applications. LangChain helps in creating workflows
that involve multiple LLM calls, enhancing the model's functionality.

Scenario:
In a Q&A system, RAG can retrieve relevant documents from a knowledge base, ensuring that
the generated answer is accurate and grounded in factual information. LangChain can be used
to manage a series of API calls or data manipulations in a pipeline that uses an LLM for
answering queries.

What is NLP? Explain the NLP techniques you’ve worked on.

Answer:
Natural Language Processing (NLP) is a field of AI that focuses on enabling machines to
understand and interpret human language. It involves various techniques to process, analyze,
and generate human-readable text.

Some common NLP techniques include:

• Tokenization: Splitting text into smaller units (tokens), such as words or sentences.

• Named Entity Recognition (NER): Identifying entities (e.g., people, locations,

organizations) in text.

• Part-of-Speech Tagging: Classifying words into parts of speech, like nouns, verbs,
adjectives, etc.

• Text Classification: Categorizing text into predefined classes (e.g., spam detection).

• Sentiment Analysis: Analyzing the sentiment of a piece of text (positive, negative,

neutral).
 • Dependency Parsing: Understanding the grammatical structure of a sentence.

Scenario:
In a sentiment analysis project, I used tokenization and part-of-speech tagging to preprocess the
data before applying machine learning algorithms to predict the sentiment of product reviews.

What is deep learning?

Answer:
Deep Learning is a subset of machine learning that involves neural networks with many layers
(hence “deep” networks). These networks can automatically learn features from data, making
them highly effective in tasks such as image recognition, speech recognition, and natural
language processing.

Deep learning models require a large amount of labeled data and computational power to train.
They have shown remarkable performance in complex tasks like generating human-like text,
translating languages, and even driving autonomous vehicles.

Scenario:
In a project involving image classification, deep learning models like Convolutional Neural
Networks (CNNs) were used to classify images of cats and dogs. The model learned features like
edges, textures, and shapes to make accurate predictions.

What is GPT?

Answer:
GPT (Generative Pretrained Transformer) is a type of LLM that uses the transformer architecture
to generate text. It works by predicting the next word in a sequence, based on the context of
the preceding words. GPT is pretrained on large amounts of text data and can be fine-tuned for
specific tasks such as text generation, summarization, or question-answering.

Scenario:
When you interact with a chatbot built on GPT, the model generates responses based on the
context you provide. It can answer questions, generate creative text, or even hold a
conversation, making it highly versatile.
How would you create a chatbot using the GPT API?

Answer:
To create a chatbot using the GPT API, follow these steps:

1. Set up an OpenAI account and API key.

2. Make API calls to OpenAI’s GPT endpoint.

Send a user’s message as a prompt to the GPT API. For instance, you can pass the
prompt "What's the weather like today?" and receive a response.

3. Process and handle user inputs:

Filter and format the user input, then send it to the API. Process the model’s response
and return it to the user.

4. Enhance the chatbot with state management:

To make the chatbot conversational, store the context of the conversation so that the
model can generate more contextually relevant responses.

Scenario:
In a customer service chatbot, you would send the user’s query (e.g., "I need help with my
order") to the GPT model and format the response into a friendly, helpful message.

How do ChatGPT, Claude, META, and Gemini differ?

Answer:

• ChatGPT: Developed by OpenAI, it's a highly versatile LLM for general-purpose text
generation and conversation.

• Claude: Developed by Anthropic, Claude focuses on safe and ethical AI interactions,

ensuring that responses are aligned with safety guidelines.

• META: Meta’s LLaMA models are optimized for performance in resource-constrained

environments, excelling in research and academic use cases.

• Gemini: A model designed for specific tasks with high performance, often used in
specialized applications that require task-specific optimizations.

Scenario:
If you're building a chatbot for casual use, you might choose ChatGPT. But if you're building a
chatbot for a highly sensitive or regulated environment (e.g., healthcare), Claude might be a
better choice.
Which AI GPT is best for specific use cases?

Answer:

• GPT-3/4: Ideal for general-purpose text generation, writing assistance, chatbots, and
creative content generation.

• Claude: Best for ethical applications, ensuring safe and respectful conversations.

• META (LLaMA): Best for research applications and low-resource environments.

• Gemini: Best for task-specific applications that require specialized performance.

Scenario:
For a marketing campaign chatbot, GPT-3 would be great for generating creative and engaging
responses. For a healthcare assistant, Claude would be preferred for safety reasons.

Given a paragraph, find how many times the word "LLM" is repeated.

Answer:
You can use Python to solve this problem with a simple approach:

def count_llm(paragraph):

words = paragraph.lower().split()

return words.count('llm')

paragraph = "LLM is a powerful tool. LLM can be used for many tasks like text generation. An
LLM is based on deep learning."

print(count_llm(paragraph))

This function will count how many times "LLM" is repeated in the provided paragraph.