AI and ML Interview

Questions
1.What is intelligence?
Intelligence is generally defined as abilities such as learning, inferring, and
reasoning

2. What is AI, ML, DL and Generative AI?

Artificial Intelligence (AI)
Artificial intelligence is a broad field that involves using a computer to simulate
something that would match or exceed human intelligence.
Machine Learning (ML)
Machine Learning is a technology that became popularized around the 2010s. As
its name implies, in ML, the machine is learning.
Instead of being explicitly programmed, the machine is given lots of information,
and it observes things.
Deep Learning (DL)
Deep Learning is considered the next layer in the progression following machine
learning, reaching popularity around the 2010s.
Mechanism: Deep learning utilizes neural networks, which are computer
simulations designed to mimic how the human brain works, to the extent that this
mechanism is understood.
Generative AI (Gen AI)
Generative AI encompasses the most recent advancements in artificial
intelligence. This area has caused an "absolute explosion" in attention and
technology adoption.
Definition: These technologies are classified as generative because they are
generating new content.

3. Is AI boon or curse?
AI's duality makes it a fascinating paradox: a boon for humanity and a potential
curse. It serves as a blessing, driving remarkable advancements in healthcare
and scientific research, and boosting our daily efficiency. However, it also
presents a looming threat, raising concerns about job displacement, the spread
of misinformation, and the perpetuation of societal biases. Ultimately, AI is not
good or evil; its impact is determined by the ethical frameworks we use to guide
its development and application. (Standard answer)

4.What are the different types of AI?

AI can also be categorized into seven types by examining two encompassing
categories: AI capabilities (three types) and AI functionalities (four types).
A. AI Capabilities
These categories distinguish between AI that currently exists (realized) and AI
that is theoretical.
1. Artificial Narrow AI (Weak AI): This is the only type of AI that exists today
(realized AI).
◦ Narrow AI is trained to perform a narrow task and cannot perform outside
of that defined task.
◦ The task it performs might be something a human could not do as well.
2. Artificial General Intelligence (AGI or Strong AI): A favorite of science
fiction, AGI is currently a theoretical concept.
◦ The idea is that AGI could use previous learnings and skills to accomplish
new tasks in a different context without the need for human training. If it wants
to learn a new task, it figures it out itself.
3. Artificial Super AI (Super AI): Also, theoretical.
◦ If realized, Super AI would possess cognitive abilities (thinking, reasoning,
learning, making judgments) that surpass those of human beings.
◦ It would evolve beyond catering to human sentiments, being able to feel
emotions, and possess its own needs, beliefs, and desires.
B. AI Functionalities
These four functionalities are built on the three capabilities, covering both
realized and theoretical AI.
Realized AI (Narrow AI Functionalities)
4. Reactive Machine AI: Systems designed to perform a very specific
specialized task.
◦ It stems from statistical math and analyzes vast amounts of data to produce a
seemingly intelligent output.
◦ Example: IBM's Deep Blue supercomputer from the late 1990s, which beat
a chess grandmaster by analyzing the board and predicting probable outcomes of
each move.
5. Limited Memory AI: This form of AI can recall past events and outcomes
and monitor specific objects or situations over time.
◦ It uses past and present data to decide on a course of action and improves its
performance as it is trained on more data.
◦ Example: Your favorite generative AI chatbot relies on limited memory AI
capabilities to predict the next word, phrase, or visual element within the context
it is generating.
Theoretical AI Functionalities
6. Theory of Mind AI (Related to AGI): This AI would understand the thoughts
and emotions of other entities (specifically humans).
◦ It could infer human motives and reasoning and personalize interactions
based on unique emotional needs and intentions.
◦ Emotion AI is currently in development, with researchers hoping it can
analyze voices, images, and other data to understand and respond to human
feelings.
7. Self-Aware AI (Related to Super AI): This is considered the "scariest AI of
all".
◦ It would have the ability to understand its own internal conditions and
traits, leading to its own set of emotions, needs, and beliefs.

5.Differences between AI and ML?

The key difference between Artificial Intelligence (AI) and Machine Learning
(ML) is that AI is the broad field of endeavor, while ML is a specific set of
techniques and algorithms used within that field. Machine Learning is,
therefore, a subset of Artificial Intelligence.
Here is a detailed breakdown of the differences and relationship between AI and
ML based on the sources:
Artificial Intelligence (AI)
AI is the broad field of study and practice.
• Goal: The fundamental goal of AI is to simulate with a computer something
that would match or exceed human intelligence.
Machine Learning (ML)
Machine Learning is a specific technology and algorithm technique that is
housed within the broader domain of AI. It became more popularized around the
2010s.
• Goal: As the name implies, in ML, the machine is learning.
• Mechanism: ML differs from earlier AI methods because the machine is not
explicitly programmed. Instead, it is given lots of information, and it observes
things.

6.Will AI replace humans?

AI will never replace humans; it enhances and supercharges humanity by
automating repetitive tasks. AI fundamentally relies on humans for guidance, data
annotation, and teaching individual values and intentions. Computers lack
judgment and intelligence, meaning humans must always handle the "gray areas"
and "unknown unknowns" of real-world problems. AI developers focus on
creating tools to free up human time and energy for unique, creative problem-
solving. Therefore, the path forward is human-led AI, allowing people to
embrace the generative aspects of human nature.
https://www.youtube.com/watch?v=iXCmoQDEoe4&pp=ygUVd2lsbCBhaSBy
ZXBsYWNlIGh1bWFu

7.Who is the CEO of Open AI?

The current CEO of OpenAI is Sam Altman. Sam Altman co-founded OpenAI
in 2015 along with Elon Musk and others.

8.What is the full form of GPT?

The full form of GPT is Generative Pre-trained Transformer.
 Generative → It can create (generate) text.
 Pre-trained → It is trained on a large dataset before being fine-tuned.
 Transformer → The deep learning architecture it is based on.

9. What is DALL·E?
DALL·E is an Artificial Intelligence model developed by OpenAI that can
generate images from text descriptions. For example, if you type “a cat wearing
sunglasses and riding a skateboard”, DALL·E can create a realistic or artistic
image of it. It combines natural language understanding with image generation to
turn words into pictures.

10.What is the meaning of “Perplexity” in the company Perplexity

AI?
The word “Perplexity” comes from natural language processing (NLP). It is a
measure of how well a language model predicts a sample of text. Lower
perplexity means the model is better at predicting the next word.The company
Perplexity AI chose this name because they build an AI-powered search and
answer engine. The idea is that their system helps reduce perplexity (confusion)
for users by giving clear, direct answers instead of just links.

11. Can you explain what a CNN is and where it might be used?
A Convolutional Neural Network (CNN) is a type of neural network
specifically designed for processing grid-like data, most notably images. Its core
function is to automatically learn features from the data through a series of
specialized layers. The convolutional layer uses a sliding filter to detect features
like edges and textures. This is followed by pooling layers that downsample the
data, making the network more efficient and robust. Finally, fully connected
layers use these learned features for classification or other tasks. CNNs are
widely used in computer vision, powering everything from facial recognition to
medical image analysis. They are the backbone of systems that "see" and interpret
the world.

12.How many layers are in a neural network?

A simple neural network has at least three layers: an input layer, at least one
hidden layer, and an output layer. However, the number of hidden layers can
be more than one, which is why a neural network with multiple hidden layers is
called a deep neural network or deep learning model. The term "deep" in deep
learning refers to the depth created by stacking these hidden layers.
13.What is the difference between NLP and NLU?
NLP is a broad, overarching field of AI that deals with the interaction between
computers and human language. Its goal is to enable computers to process,
analyze, and generate human language, whether it's written or spoken. NLU is a
subfield of NLP that specifically focuses on "understanding" the meaning and
intent behind human language. It goes beyond the basic structure to interpret the
nuances, context, and ambiguities that are inherent in human communication.

14.Give some examples of Weak AI and Strong AI?

Weak AI (Narrow AI)
Weak AI, also known as narrow AI, is artificial intelligence that is designed and
trained for a specific, single task. It can be highly effective and even outperform
humans in its area, but its intelligence is limited to that domain. It does not have
general cognitive abilities, consciousness, or self-awareness. It's essentially a tool
designed to solve a very specific problem.
Examples of Weak AI:
 Virtual Assistants: Siri, Alexa, and Google Assistant can answer
questions, set timers, and perform specific commands, but they lack the
ability to reason beyond their programming.
 Recommendation Engines: Netflix and Spotify algorithms that suggest
movies or songs based on your past behavior.
 Self-Driving Cars: The AI is highly specialized in navigating, recognizing
objects, and making driving decisions, but it cannot perform tasks outside
of driving.
 Chess-Playing Programs: AI like Deep Blue, which famously beat a
world chess champion, is a perfect example. It is a master of chess but
cannot do anything else.
Strong AI (Artificial General Intelligence - AGI)
Strong AI, or Artificial General Intelligence (AGI), is a theoretical form of AI
that would possess the ability to understand, learn, and apply intelligence to solve
any intellectual task that a human can. It would have a full range of cognitive
abilities, including reasoning, problem-solving, planning, learning, and even
consciousness.
Examples of Strong AI:
 There are no real-world examples of Strong AI. It remains a hypothetical
concept, primarily seen in science fiction.
 Fictional examples include:
o HAL 9000 from 2001: A Space Odyssey
o TARS from Interstellar
o Data from Star Trek: The Next Generation

15. What is the full form of LSTM?

LSTMs are a specialized type of recurrent neural network (RNN) that are
particularly good at processing sequential data, like text or time series. The "long
short-term memory" name comes from its ability to remember information for
long periods of time, which solves a major problem in traditional RNNs called
the vanishing gradient problem. This is achieved through a unique architecture
that includes internal mechanisms called "gates" that regulate the flow of
information. These gates allow the network to selectively remember or forget
information as it processes a sequence of data, making it highly effective for tasks
like speech recognition and language translation.

16. What are the different components of GAN?

A Generative Adversarial Network (GAN) is composed of two main neural
networks that are in an adversarial, or competing, relationship: a Generator and
a Discriminator. Think of it as a game where the Generator is a counterfeiter
trying to create fake money, and the Discriminator is a detective trying to spot the
fakes.
1. The Generator
The Generator's role is to create new, synthetic data that looks as realistic as
possible. It starts with a random noise vector as its input and transforms this noise
into a data sample, such as an image, audio, or text. Initially, its output is poor
and easy to distinguish from real data, but it continuously improves its skills by
receiving feedback from the Discriminator.
2. The Discriminator
The Discriminator is a binary classifier that acts as a judge. It takes two types of
input: real data from a training dataset and fake data from the Generator. Its job
is to determine whether each input is "real" or "fake." The Discriminator is trained
to get better at telling the difference between real and generated data, which in
turn forces the Generator to become more sophisticated.

17. Difference between Bias and variance

Bias
Bias is the error that comes from a model making overly simplistic assumptions
about the data. A model with high bias is too simple to capture the underlying
patterns and relationships in the data. This leads to underfitting, where the model
performs poorly on both the training data and new, unseen data. Think of it
as a model that's too rigid; it can't learn from the nuances of the data. An example
is using a simple straight-line linear regression to model a highly curved, non-
linear dataset.
Variance
Variance is the error that comes from a model being too sensitive to small
fluctuations and noise in the training data. A model with high variance is overly
complex and learns the training data's specific details too well, including its
random noise. This leads to overfitting, where the model performs
exceptionally well on the training data but poorly on new data. Think of it as
a model that's memorized the training data rather than learned to generalize from
it. A very deep, complex neural network with too many parameters can suffer
from high variance.

18. What is bias-variance tradeoff?

Bias is a model being too simple, causing it to miss important patterns.
Variance is a model being too complex, causing it to get confused by noise and
specific details.
The tradeoff is the struggle to find the right balance, making the model complex
enough to learn but simple enough to generalize to new data.

19. What is agentic AI?

Agentic AI systems are proactive systems that pursue goals through a series of
actions, unlike reactive generative AI systems.
They operate through a lifecycle of perceiving, deciding, executing an action, and
learning from the output with minimal human intervention.
Large Language Models (LLMs) provide the necessary reasoning engine, using
chain of thought reasoning to "think" through problems.
This allows the agent to break down complex tasks into logical steps, making
them effective for scenarios requiring ongoing management and multi-step
processes.

20.What is the purpose of Boltzmann machine?

Boltzmann machines are a type of energy-based model which learn a probability
distribution by simulating a system of diverging and converging nodes. These
nodes act like neurons in a neural network, and can be used to build deep learning
models.