How could OpenAI, a small organization of just 200 employees, managed to shake the foundations of large companies like Google and Meta? Everyone dreams about being a unicorn – having razor sharp focus, high talent-density , rapid speed of innovation but in reality, even startups end up becoming slow organizations very quickly. Why does this happen?
5. Unleashing AI from Labs to Production
AI + Digital
HyperDigital
==
We love building awesome and intuitive applications for our clients
6. Real AI, a deepkapha.ai company, is an Enterprise AI
start-up that develops production-grade AI products,
solutions and services for global enterprises.
OPERATION
CONTINUOUS
EXECUTION
AI
RESEARCH
ENGINEERING
Where we play?
ONE
STOP
SHOP
BUSINESS
STRATEGY
10. Artificial Intelligence
Creating The Assembly Line For Knowledge Workers
Sources: ARK Investment Management LLC, 2023.
Generative AI doubles knowledge worker
productivity with AI coding assistants.
AI training costs dropping 70% annually, expected
to continue.
AI could boost global labor productivity atleast 4x
by ~$200 trillion by 2030.
11. Generative AI – 2022 and beyond
Sources: ARK Investment Management LLC, 2023.
DALL-E 2: "An astronaut riding a
horse” Publicly available September
2022
Meta Make-A-Video
Announced September
2022
Open-Source Stable Diffusion
2.0: Released November 2022
Prompted by a short text, generative AI models can produce images, code, text, audio, and video. In less than one year, dozens of generative AI projects createdmodels that
progressed fromgrainy images to high-quality 3D models and videos.
12. Increasingly Dropping Cost
AI Is Increasing The Productivity Of Knowledge Workers
*Based on data from GitHub. **Generative AI models translated “ a picture of an astronaut on Mars” into multiple images in just a few seconds. Sources: ARK Investment Management LLC, 2023. Kalliamvakou, E. 2022. Forecasts are
inherently limited and cannot be relied upon. For informational purposes only and should not be considered investment advice or a recommendation to buy, sell, or hold any particular security. Past performance is not indicative of
future results.
Time to Complete (Coding) Tasks:
2022* Human
Cost Time
Generative
AI Cost
Time
$150
5 Hours
$0.08
< 1 Minute
180
160
140
120
100
80
60
40
20
0
Human Human + AI
Minutes
-55%
Software engineers completed a coding task in less than half the time with AI
coding assistant GitHub Copilot.
According to our research, AI cancreate agraphic design for just $0.08** in
minutes —a di minimis cost compared to $150 for human labor.
Coding Assistants Generative Image Models
13. The Great Decoupling in coming Next, just like
what happened with Electricity
AI, like electricity, shifts
value from "lower cost of
prediction" to enabling
"vastly more productive
systems" by decoupling
prediction from decision-
making aspects.
The Modern Productivity Paradox in a Not-Too Distant Mirror, 1989
https://warwick.ac.uk/fac/soc/economics/research/workingpapers/1989-1994/twerp339.pdf
14. And Neural Networks Are The Foundations
* Sources: ARK Investment Management LLC, 2023.
Represents an order of magnitude increase in the commercial potential of another
technology.
Adaptive Robotics
3D Printing
Precision Therapies
Programmable Biology
Reusable Rockets
Smart contracts
Next Gen Cloud
Intelligent Devices
Cryptocurrencies
Multiomic Technologies
Autonomous Mobility
Advanced Battery Systems
Digital Wallets
Neural Networks
Relative Importance as a Catalyst
15. AI Should Increase Knowledge Worker Productivity Dramatically
AI adoption could quadruple productivity, rivaling 2030's global GDP projections
Estimated Human Knowledge Work Wages Relative To AI Spend In 2030 AI Total Addressable Market (TAM) Forecast In 2030
$80
$70
$60
$50
$40
$30
$20
$10
$-
Bear Case Base Case
Estimated Knowledge Work Wages AI Spend
Bull Case
2.5
4.5
6.5
1%
$2T
$1T
$0.8T
10%
$8T
$21T
$14T
20%
Software Vendor Value Capture Of Productivity Gain
$41T
$16T
$29T
Trillion
s
Productivity
Uplift
* Sources: ARK Investment Management LLC, 2023.
Manufacturing and AI
17. DISRUPTION IS BEING BE FELT ALREADY
Education companies’ shares fall sharply after warning over ChatGPT | Financial Times
18. DISRUPTION IS BEING BE FELT ALREADY
Industry Vertical Use Case(s) Companies / Professions
Impacted
Education Sector Personalized Learning (Studay plans,
Homework help, continuous
uninterrupted guidance), Automated
Grading & Feedback, Virtual Tutor
Coursera.org, Khan Academy,
Chegg*/ Pearsons Learning*,
Media & Press
Journalists
Automated Content Creation,
Personalized News Delivery, Fact-
Checking and Misinformation Detection
BuzzFeed*, CNN, FoxNews,
Guardian, VICE,
Creatives, IT ,
Marketing
Simplify abstracts, Ad Copy and
Marketing Campaigns
Automated Design Creation,
Personalized User Experiences, Code
Generation and Bug Fixing, Prototyping
and Mockups, Content Personalization
in Design, Automated Testing
Copywriters, Creatives, Designers,
Software Engineers
Beware of the risk
of cannibalization
here
19. PRODUCTIVITY : HARD THINGS AND FAST THINGS
Focus Things Hard Things Fast Things Typical Actors
Things that matter Skin-in-the-game Always looking for
means to speed up all
things. Bound to use
Gen AI for sure.
Entrepreneurs, Startup Founders,
Innovators, Military Generals,
Activists, Passionate Teachers,
Social Workers
Things that matter (to
them alone)
No skin-in-the-game Engage in “fast things”
to create illusion of
speed and productivity
Bureaucrats, Award Collectors,
Professors, Politicians,
Management Consultants
22. HUMAN IN CONTROL
EVOLUTION OF PRODUCTIVITY
< 1850
EXAMPLE
Belt Maker
Herman Paumgartdener is
using a hacksaw and anvil to
punch holes in the belts. This
was another type of
specialized clothing maker.
23. EVOLUTION OF PRODUCTIVITY
1900 – 2000
HUMAN IN CONTROL OF MACHINE EXAMPLE
Fred Johnson is a
worker in a Ford factory
working in assembly
line with these four key
tasks.
Parts
Installation
Quality
Control
Equipment
Operation
Safety
Compliance
24. EVOLUTION OF PRODUCTIVITY
2000 – 2020
HUMAN IN CONTROL WITH MACHINE EXAMPLE
Product
Development
System
Architecture
Design
Security
Optimization Team
Leadership
DevOps
Testing
Soft Skills
Paypal Maffia
A typical "10x engineer"
working for a high-
paced start-up
environment.
33. GREENSCAN – OUR ALT ENERGY AI STARTUP
Scanning the Earth. One Pixel at a time.
21st February 2022
Supported by (Study) Funded by
GreenScan.io
Powered by
We are the first generation to feel the effect of climate change
and the last generation who can do something about it.
- Barack Obama, Former US President
Domain Centric Models
We evaluated several approaches to predict
the wind speed of the storms. From more
traditional naive models for estimating wind
speed using classification model repurposed
for regression task to advanced approach
using computer vision (CNN with LSTM to catch
temporal aspect of the hurricane dataset)
Product Demo - GS Wind
Advanced Deep Learning
Innovation
We trained high dimensional satellite data of
very large dataset to train this complex model to
be able to predict wind speed of emerging
storms as they evolve.
How it works
MLOPs Powered SaaS Platform
Our AI SaaS platform is powered by end-to-
end integrated MLOPs framework which allows
customers to seamlessly onboard the platform
and service their requests. We intend to further
build out this platform to allow enterprise users
build their own Energy Asset & Risk
Management Workflows.
Product Demo - GS Solar
Advanced Deep Learning
Innovation
We trained some 60,000 infrared images of solar
PVs with cutting-edge deep learning computer
vision models to learn about 12 classes to identify
anomalies such as dust, vegetation etc.
Domain Centric Models
For this multi-class problem, we experimented
with various pretrained imagenet models and
used the best performing model on our solar
anomaly dataset.
MLOPs Powered SaaS Platform
Our AI SaaS platform is powered by end-to-
end integrated MLOPs framework which allows
customers to seamlessly onboard the platform
and service their requests. We intend to further
build out this platform to allow enterprise users
build their own Energy Asset & Risk
Management Workflows.
How it works
35. Glimpse AI is our
latest project Feb
2023, start-up project
at the intersection of
manufacturing
domain and
generative AI to help
manufacturers
reduce cost of
operation while
serving personalized
products to their
clients.
Generative AI – Manufacturing – The Hard thing about
customization
36. Catalog Sync
The seller has synced their catalog in the
Glimpse app to ensure a smooth ordering
process. The seller has also enabled
custom product mockups, overlaying a
standard embroidery mockup over their
own product photography.
Shop Sync
Combining integrations with eCommerce
platforms like Shopify and an
understanding of the seller's FAQ's, Glimpse
can automate most customer support
queries and even tasks like adding items to
the cart.
Non-Enterprise Example
Product Design
The seller has customized the default
image generation settings in the Glimpse
app to ensure all designs are created with
with low detail, thick borders, and hex codes
that match their thread colors.
How it works
37. Enterprise Example
How it works
Advanced Custom Preview
Enterprise clients will receive fully
customized solutions to provide accurate
3D previews to their customers.
Manufacturing Line Sync
Enterprise customers benefit from
generative product design that is fully
synced with the production line, ensuring a
direct flow from design to manufacturing.
Advanced Product Design
Generating a preview like this requires
Glimpse to design a 3D model out of
existing Lego pieces. Enterprise clients
require complex upfront development but
allow for fully automated manufacturing.
40. • MOAT 1: Find a really really hard problem
to solve
• MOAT 2: Small Data Will Be King
• MOAT 3: Designing an AI System XAI
grounds up will eventually win (after the
hype wave has claimed all its victims)
Concluding thoughts
41. Questions?
Sources: ARK Investment Management LLC, 2023.
Tarry Singh
Chairman, CEO and Investor/VC
Real AI, deepkapha AI Lab + Startups
e-mail: tarry.singh@deepkapha.com
We’re hiring!!!