Hey people,
I'm Aishwarya, a Generative AI Educator and a Talent Acquisition Partner.
I'm an experienced Technical Recruiter with a deep-rooted passion for Generative AI, and I excel at the intersection of talent acquisition and cutting-edge technology.
I'm constantly sharpening my skills in identifying, attracting, and engaging top-tier talent, specializing in roles that push the boundaries of innovation and technology.
Let's Connect! Whether you're looking to explore new opportunities, seeking talent for your team, or simply wish to chat about the latest in AI and tech, I'm just a message away.
Anyway, welcome to my presentation on Generative AI, where we'll dive into the world of machine creativity. In this session, we'll explore how Generative AI is reshaping industries and driving innovation.
Discover how machines are learning to think creatively and produce original content in various forms, from art and music to writing and design. We'll discuss the underlying principles of Generative AI in simple terms, along with its real-world applications and implications.
Whether you're new to AI or an industry professional, this presentation will provide valuable insights into the exciting possibilities of Generative AI. Join us as we uncover the potential of machine creativity and its impact on the future of technology.
Feedback is welcome. If you would like customized content and reading material for this topic, please feel free to reach out to me on email.
findingash@outlook.com
2. Introduction
A quick briefing on what Generative AI
means and why it is the future.
Gen AI
Definition
01
02
03
How it’s
different from
other AI
Examples of
Gen AI
Subset of AI that’s
responsible for creation
It focuses on creating
things rather than just
analyzing and
understanding data.
From image generation to
drug discovery. there’s a
countless examples.
AISHWARYA RAMESH
3. Neural Networks
Neural Network is a subset of
Machine Learning. It is a
omputational model inspired by the
human brain, composed of
interconnected artificial neurons
that learn to process information
and make predictions from data.
It consists of 3 layers - input layer,
hidden layer, and the output layer.
Machine Learning
Machine learning is like teaching a
computer to learn from examples.
Instead of giving it explicit instructions,
you let the computer learn on its own.
t's like training a dog – you reward it
when it gets things right, and over
time, it becomes better at tasks like
recognizing patterns, making
predictions, or making decisions.
Mechanics of Generative AI AISHWARYA RAMESH
4. The #1 Key
Model in
Generative AI
The biggest and most common AI
model - used for text generation.
Uses the Attention Mechanism.
Example: NLP, Sentiment Analysis,
Translation
AISHWARYA RAMESH
Transformers
5. Generative Adversarial Network
More
Key Models in
Generative AI
Variational AutoEncoders
Has two networks -
Generator and Discriminator.
Generator generates fake
data from random noise.
Discriminator evaluates if it is
real from the dataset or fake.
They continuously compete
and improve.
Example: Deepfakes
Smart data compressors and
decompressors.
Compression: They compress
complex data into meaningful
code
Decompression: They recreate
the original data from this code
Example: Photos to Art and the
Art back to Photos
AISHWARYA RAMESH
7. AISHWARYA RAMESH
Powerful CPUs Power Consumption Financial Costs
Resource Intensity
These models require
advanced GPUs
(Graphics Processing
Units) for processing.
The need for such
hardware accelerates
the operational costs
significantly.
The energy required to
power these GPUs,
especially during training
is immense. Training a
single model can
consume as much
electricity as a small
town uses in a month.
Medical Image Synthesis
Assisting with Drug
Discovery. Yes.
Personalized medication
based on patient data
Radiology Report
Generation
Healthcare Simulation
8. AISHWARYA RAMESH
Entry Barriers Global Inequality Ongoing Costs
Scalability Challenges
The computational
demands of AI models
pose significant scalability
challenges: Small
organizations and
startups may find the
cost of entry prohibitively
high, limiting innovation
and competition.
Developing countries,
where access to
advanced computational
resources and
sustainable energy
sources can be limited,
might fall further behind
in the AI race, enhancing
the digital divide.
Even after a model is
trained, deploying it for
real-time applications
requires substantial
computational
resources, affecting
long-term scalability and
cost-efficiency.
9. AISHWARYA RAMESH
Model Distillation AI Pruning Efficient Architecture
Efficient Innovations
This involves training a
smaller, more efficient
model to replicate the
performance of a larger,
pre-trained model ,
reducing resource
requirements without
compromising on
performance.
This process removes all
the less important
connections within the
available neural
networks. This in turn
reduces the model's size
and computational
needs while maintaining
accuracy.
Architectures like
EfficientNet and
developments in sparse
transformers aim to
decrease computational
demands by optimizing how
models process and learn
from data, enabling more
sustainable AI development.
10. Generative AI Process
Input Data: Model receives a dataset
(learning examples).
Model Selection: Common Models
include GANs and VAEs.
Training: Continue and improve
Evaluation: Use metrics to evaluate
ability to generate realistic outputs.
Output Data: Use this to generate new
instances
Application: Apply the generated
outputs to the intended use case
OUTPUT DATA
UPDATE MODEL
TRAINING
INPUT DATA AI MODEL ANALYZE
AISHWARYA RAMESH
11. AISHWARYA RAMESH
Creative Arts Business Healthcare
Applications of Generative AI
Artistic Image Generation
Video Game Content
Creation
Music Compoition
(Musenet)
Film Scriptwriting
Virtual Set Deign
Interactive Storytelling
Social Media Marketing
Content Generation
Chatbots for Customer
Service
Revenue Forecasting
Legal Documentations
Product Design and
Prototyping
Medical Image Synthesis
Assisting with Drug
Discovery. Yes.
Personalized medication
based on patient data
Radiology Report
Generation
Healthcare Simulation
12. Image
Generation
with DALL-E
AI model developed by OpenAI.
DALL-E extends the GPT
architecture's capabilities from
purely textual to visual
domains, allowing it to
understand and generate
images (even if it doesn’t exist).
AISHWARYA RAMESH
13. Compose original music pieces, mimic
styles, and collaborate with human
musicians.
Create music in multiple genres, from
classical to pop to jazz.
Assist in songwriting and composition
Examples: OpenAI's Jukebox, Google's
Magenta, or IBM's Watson Beat.
Music Creation
With Gen-AI
AISHWARYA RAMESH
14. Gaming
With Gen-AI
AI can do it all -
from creating game environments to
generating NPC (non-playable
character) dialogues
(Eg - GTA)
It can create textures, landscapes, and
character models, greatly reducing the
time and resources needed.
AISHWARYA RAMESH
15. Healthcare
With Gen-AI
Used to generate medical images
(e.g., X-rays, MRIs). This aids in early
disease detection.
Can predict how different molecules
interact and identify promising
compounds for diseases quickly and
cost-effectively. Eg: AlphaFold
Suggest personalized medication
AISHWARYA RAMESH
16. AISHWARYA RAMESH
Bias in Generative AI
What Bias Is
Te inclination or prejudice towards certain ideas, groups, or individuals in a way that is
considered unfair
Sources of Bias
Biases often stem from the training data used by AI models, reflecting historical, societal, or
representational biases present in the data
Impact of Biased Content
Reinforcement of stereotypes, unfair treatment of certain groups, and misinformation
Strategies to Mitigate Bias
Diverse dataset curation, bias detection algorithms, and ethical AI development practices
17. AISHWARYA RAMESH
Ethical Challenges in Generative AI
Privacy
AI's ability to collect, analyze, and store vast amounts of personal data raises concerns about
individuals' right to privacy.
Surveillance
The use of AI in surveillance technologies, especially by governments and corporations, can
lead to over-monitoring, affecting citizens' freedom and rights.
Autonomy
From personalized recommendations to automated decision-making systems, AI-driven
decisions can influence human choices, affecting individual autonomy.
The Digital Divide
The uneven access to AI technologies across different regions, communities, and socio-
economic groups contributes more to existing inequalities.
18. Consent
Navigating
Ethical AI Challenges
Transparency
Ethical AI deployment requires
obtaining explicit consent
from individuals whose data is
collected and used. This
involves clear communication
about how data is gathered,
used, and stored.
AI systems should be
transparent in their operations,
allowing users to understand
how decisions are made. This
includes disclosing the logic or
rationale behind AI-driven
decisions and ensuring
systems are explainable to
non-expert users.
AISHWARYA RAMESH
19. Clear Guidelines
Auditability
There should be robust
frameworks governing AI
development and
deployment, clarifying
responsibilities and
accountability.
AI systems must be
designed to allow for
auditing and scrutiny,
enabling the tracing of
decisions back to their
source to ensure
accountability.
Navigating
Ethical AI Challenges
AISHWARYA RAMESH
20. AISHWARYA RAMESH
AI Design Trust Level AI Tenets
Ethical AI Frameworks
IEEE's Ethically Aligned
Design provides a
comprehensive set of
principles for prioritizing
human well-being in AI
systems.
EU's Ethics Guidelines for
Trustworthy AI outlines
seven key requirements
for AI systems, including
transparency, fairness,
and accountability.
Partnership on AI's Tenets:
Advocates for responsible
AI development and
usage, focusing on
fairness, transparency,
and collaborative
research.
21. Stay Curious, Stay
Prepared.
Contact Me
aishwarya.ramesh@bytexl.in
Aishwarya R
What's Next?
Let's dive deeper into the art and
science of communicating with AI. Join
me for the next training session on AI
prompts, where we'll unlock the full
potential of Generative AI models
together.
AISHWARYA RAMESH