Tata ELXSI

Dr. Suraj Kamal

03/02/24
 Prologue
The grand challenges

Fundamental Sciences Engineering

• Grand Unification • Nuclear Fusion

• Dark Matter/Energy • Quantum Computing
• Supersymmetry • Quantum Communication
• Quantum Gravity • Hydrogen Synthesis
• Singularities • Advanced Energy Storage
• Origin of Life • Protein Folding
• String Theory • Room Temperature Superconductivity
• Measurement Problem • Carbon Capture and Storage
• The Riemann Hypothesis • Advanced Materials
• The P vs. NP problem • Regenerative Medicine
• The Brain and Consciousness • Universal Vaccines

Artificial Intelligence - AI

Solve Intelligence and solve everything else with it – DeepMind

 The Hierarchy of Natural Intelligence
• Best known case study
• Human Brain
 Science is Justified True Belief
Intelligence?

Epistemological – Study of Knowledge Ontological – Study of Existence

• What constitutes intelligence? • Nature of Intelligence?
• How Intelligence arises? • Existential Basis?
• Processes of Acquisition? • Relationship with Consciousness?
• Limits and Scope? • Universal vs. Individual Intelligence?
• Relativity of Knowledge? • Evolutionary Origins?
• Ethical Implications? • Panpsychism?
• How to measure Intelligence?
Can a machine act intelligently? Can it solve any problem that a person would solve by thinking?
Are human intelligence and machine intelligence the same? Is the human brain essentially a computer?
Can a machine have a mind, mental states, and consciousness in the same sense that a human being can?
Can it feel how things are? What does an intelligent system look like?
Does an AI need—and can it have—emotions, consciousness, empathy, love?
Can we ever achieve AI, even in principle?
How will we know if we’ve done it?
If we can do it, should we?
The Instrumentalist Paradox
 The Era of the Artificials
“Heavier-than-air flying
machines are impossible”
Lord Kelvin 1895

• Mechanical Power • Does a submarine swim?

• Industrial Revolution • Does an aeroplane fly?
• Electricity
• Electronics Revolution
• Mechanical Flight
• Aerospace Revolution • Does a computer think?

Mechanization of intelligence?

Anthroposcene
Moravec’s paradox

Certain things that are very difficult for humans, such as chess or difficult math
problems, are quite easy for computers

But things that are very simple for us humans, such as perceiving objects or using
motor skills to do the washing up, turn out to be very difficult for computers

“It is comparatively easy to make computers exhibit adult level performance on

intelligence tests or playing chess, and difficult or impossible to give them the skills of a
one-year-old when it comes to perception and mobility.”
 Artificial Intelligence?
Human-Based Ideal Rationality
Reasoning-Based: Systems that think like humans Systems that think rationally
Behaviour-Based: Systems that act like humans Systems that act rationally

McCarthy 1955, IBM

“the science and engineering of making intelligent machines” “anything that makes machines act more intelligently”

Definitions
“The imitation of all human intellectual abilities by computers”
Marvin Minsky 1968
“the science of making machines do things that would require “The imitation of various complex human skills by machines”
intelligence if done by men”
“Systems that display intelligent behavior by analyzing their
Russell and Norvig environment and taking actions – with some degree of autonomy –
“the study of intelligent agents that receive precepts from the to achieve specific goals.”
environment and take action”
Wikipedia
Demis Hassabis, CEO and founder of DeepMind 2017 “intelligence exhibited by machines, rather than humans or
“the science of making machines smart” other animals (natural intelligence, NI)”

• Turing's "polite convention": If a machine behaves as intelligently as a human being, then it is as intelligent as a human being
• The Dartmouth proposal: "Every aspect of learning or any other feature of intelligence can in principle be so precisely described that a machine can be
made to simulate it."
• Allen Newell and Herbert A. Simon's physical symbol system hypothesis: "A physical symbol system has the necessary and sufficient means of general
intelligent action."
• John Searle's strong AI hypothesis: "The appropriately programmed computer with the right inputs and outputs would thereby have a mind in exactly
the same sense human beings have minds."
• Hobbes' mechanism: "For 'reason' ... is nothing but 'reckoning,' that is adding and subtracting, of the consequences of general names agreed upon for
the 'marking' and 'signifying' of our thoughts..."
• Intelligence is intangible
• Reasoning
• Learning
• Problem Solving
• Perception/Control
• Linguistic Intelligence
• Emotional Intelligence
Creativity
?
•
• Curiosity
• Wonder

Reasoning − It is the set of processes that enables us to Learning − It is the activity of gaining knowledge Problem Solving − the process in which tries to
provide basis for judgement, making decisions, and or skill by studying, practicing, being taught, or arrive at a desired solution from a present
prediction. experiencing something situation by taking some path.
Inductive Reasoning Deductive Reasoning Auditory Learning – Learn by hearing It includes planning, decision making, which is
It starts with a general Episodic Learning − events one has witnessed the process of selecting the best suitable
It conducts specific
statement and examines Motor Learning − precise control of muscles alternative out of multiple alternatives to
observations to makes
the possibilities to reach a Observational Learning − imitating others reach the desired goal are available.
broad general statements.
specific, logical conclusion. Perceptual Learning − recognize stimuli
Relational Learning − differentiate among
Example − "All women of
Example − "Nita is a stimuli by relational properties
age above 60 years are
teacher. Nita is studious. Spatial Learning − through cognitive maps
grandmothers. Shalini is 65
Therefore, All teachers are Stimulus-Response Learning − perform a
years. Therefore, Shalini is a
studious." particular behavior on certain stimulus
grandmother."
Reasoning Learning Problem Solving Perception/Control Linguistic Intelligence Emotional Intelligence
Symbolic Logic/predicate calculus Machine Learning Planning, Decision Theory Pattern Recognition Natural Language Processing ?

Expert System Bayesian Logic Machine Learning Machine Learning/Symbolic Logic

Software 1.0 Software 2.0 Dynamic Programming Statistical Learning

Generative AI Generative AI Generative AI Generative AI Generative AI Generative AI?

Computer Reinforcement
Science
Learning
Supervised Semi Supervised
Learning Learning
01 Unsupervised
Artificial Self Supervised
Learning

07 02 Intelligence Learning

Machine
Artificial Learning
Intelligence
06 03 Deep
Learning

05 04 Generative
AI

Machine Learning
 Artificial Intelligence – Epistemological categorization

• Reactive machines
• most basic type of artificial intelligence. Machines built in this way don’t possess any knowledge of previous events but instead only
“react” to what is before them in a given moment. As a result, they can only perform certain advanced tasks within a very narrow
scope, such as playing chess, and are incapable of performing tasks outside of their limited context.

• Limited memory machines

• Machines with limited memory possess a limited understanding of past events. They can interact more with the world around them than
reactive machines can. For example, self-driving cars use a form of limited memory to make turns, observe approaching vehicles, and
adjust their speed. However, machines with only limited memory cannot form a complete understanding of the world because their
recall of past events is limited and only used in a narrow band of time

• Theory of mind machines

• Machines that possess a “theory of mind” represent an early form of artificial general intelligence. In addition to being able to create
representations of the world, machines of this type would also have an understanding of other entities that exist within the world. As of
this moment, this reality has still not materialized!

• Self-aware machines
• Machines with self-awareness are the theoretically most advanced type of AI and would possess an understanding of the world, others,
and itself. This is what most people mean when they talk about achieving AGI. Currently, this is a far-off reality.

Professor Arend Hintze of the University of Michigan

 Artificial Intelligence - Terminologies

Strong AI - Artificial General Intelligence

• AI that is capable of human-
level, general intelligence. In
other words, it’s just another way
to say “artificial general
intelligence.”
Artificial Super Intelligence

Weak AI - Narrow AI
• narrow use of widely available AI technology,
such as playing chess, Artificial Narrow
Intelligence (ANI)
The History
in a Nutshell
• The history can be divided roughly into three phases
• early mythical representations of artificial forms of life
and intelligence
• speculations about thinking machines during the
Enlightenment
• the establishment of the theoretical foundation for
the computer
 AI Evolution - Snapshot

Myths and Fantasies Gestation Wave 2 Wave 4

A
Pure Imagination, C
Proposal of foundational E
The Machine Learning Wave E
Generative AI
depicted in mythology concepts and systems, Data-Driven Decision Making Foundation Models
and literature became mainstream
science

Early Attempts Wave 1 Wave 3

B Mechanical Systems to D The Symbolic AI , F Deep Learning
do mental work symbolic reasoning and Deep Neural Networks
rule-based systems
Feature Learning
AI Evolution - Snapshot
 The Source of knowledge

• Empiricism (Francis Bacon 1561-1626)

• John Locke (1632-1704): “Nothing is in the understanding which was not in the senses”
• David Hume (1711-1776): Principle of induction: General rules from repeated associations between their
elements

• Logical positivism: Bertrand Russell (1872-1970)

• All knowledge can be characterized by logical theories connected, ultimately, to observed sentences that
correspond to sensory inputs

• Behaviorism
• a theory of learning based on the idea that all behaviors are acquired through conditioning, and conditioning
occurs through interaction with the environment. Behaviorists believe that our actions are shaped by
environmental stimuli

• Reductionism
• Reductionism is a philosophical position that interprets a complex system as the sum of its parts. Reductionists
analyze a larger system by breaking it down into pieces and determining the connections between the parts
assuming that the isolated components and their structure have sufficient explanatory power to provide an
understanding of the whole system

• Emergentism and Complexity Theory

• The view that there are certain real-world entities necessarily generated from and constituted by other entities
but not fully reducible to them ontologically and/or epistemically
• Complexity theory views complex systems as open systems that interact with their environments. It emphasizes
interactions and feedback loops that constantly change systems
 Early Attempts
(Mostly Hypothetical)

How to put external knowledge/skills into artificial entities

• Antikythera mechanism
• Ancient (2nd century BC) Greek hand-powered mechanism described as the oldest known computer
• Used to predict astronomical positions and eclipses

• Analytical engine/Differential Engine

• Digital mechanical general-purpose computer/mechanical calculator
• Charles Babbage - 1837
 The Gestation

• Church-Turing thesis - 1936

• any computable function is computable via a Turing machine

• McCulloch-Pitts
• Proposed first artificial Neuron - 1942

• Norbert Wiener (November 26, 1894 – March 18, 1964)

• Cybernetics - field of systems science studies circular causal systems – 1948

• Allen .M. Turing

• Computing Machinery and Intelligence – The Imitation Game 1951
 The Birth and the First Wave
• 1951 - Marvin Minsky and Dean Edmonds
• First artificial neural network (ANN) - SNARC
• 3,000 vacuum tubes, network of 40 neurons

• 1952 - Arthur Samuel

• Self-learning Checkers-Playing Program

• 1956 - John McCarthy, Marvin Minsky, Nathaniel Rochester and Claude Shannon
• Coined the term artificial intelligence in a proposal for a workshop
• Recognized as a founding event in the AI field

• 1958 - Frank Rosenblatt

• Developed Perceptron
• Became the foundation for modern neural networks

• 1958 -John McCarthy

• Created LISP (List Processing), the first programming language for AI research, which is still in popular use

• 1959 - Arthur Samuel

• Coined the term machine learning
• Seminal paper - computer could be programmed to outplay its programmer
Let there be AI

We think that significant advance can be made in

one or more of these problems if a carefully selected
a group of scientists work on it together for a summer.

“The Navy revealed the embryo of an electronic

computer today that it expects will be able to walk,
talk, see, write, reproduce itself and be conscious of its
Simulate every aspect of human learning and intelligence existence” - July 8, 1958, the New York Times
Programming omputers to se a Language
Self Improvement (Learn)
Randomness and reativity

The Future of Artificial Intelligence, Santa Fe Institute, Melanie Mitchell

From Perception to Cognition: Towards Human-Understanding and Human-Centricity in AI Kenneth Kwok, PhD
Rosenblatt Perceptron
 The First Wave
• 1964 - Daniel Bobrow
• STUDENT, an early natural language processing (NLP) program

• 1965 - Edward Feigenbaum, Bruce G. Buchanan, Joshua Lederberg and Carl Djerassi
• the first expert system, Dendral, which assisted organic chemists in identifying unknown organic molecules

• 1966 -Joseph Weizenbaum

• Eliza, one of the more celebrated computer programs of all time, capable of engaging in conversations

• 1968 - Terry Winograd

• SHRDLU, the first multimodal AI that could manipulate and reason out a world of blocks according to instructions

• 1969 - Arthur Bryson, Yu-Chi Ho

• proposed backpropagation algorithm to train multilayer ANNs a foundation for deep learning

• 1969 - Marvin Minsky, Seymour Papert

• published the book Perceptrons, which described the limitations of simple neural networks and caused neural network
research to decline and symbolic AI research to thrive

• 1973 -James Lighthill

• released the report "Artificial Intelligence: A General Survey," caused the British govt. to significantly reduce support for AI research
ELIZA [Weizenbaum 1966] Symbolic Incarnation of ChatGPT

H : Men are all alike

E: IN WHAT WAY
H: They're always bugging us about something or other
E: CAN YOU THINK OF A SPECIFIC EXAMPLE
H: Well, my boyfriend made me come here
E: YOUR BOYFRIEND MADE YOU COME HERE
H: He says I'm depressed much of the time
E: I AM SORRY TO HEAR THAT YOU ARE DEPRESSED
H: It's true that I am unhappy
E: DO YOU THINK COMING HERE WILL HELP YOU NOT BE UNHAPPY
H: I need some help, that much seems certain
E: WHAT WOULD IT MEAN TO YOU IF YOU GOT SOME HELP
H: Perhaps I could learn to get along with my mother
E: TELL ME MORE ABOUT YOUR FAMILY
H: My mother takes care of me
E: WHO ELSE IN YOUR FAMILY TAKES CARE OF YOU
The Emergence of the Epistemic Tribes

Symbolists Expert Systems

Analogizers Support Vectors

Connectionists Neural Networks

Evolutionaries Genetic Algorithms

Bayesians Bayes Theorem

Subsumption Theorists: Embodiment Problem

The Second Wave
Machine Learning
 The Second Wave
• 1981 - Danny Hillis
• designed parallel computers for AI and other computational tasks, an architecture similar to modern GPUs

• 1984 - Marvin Minsky, Roger Schank

• coined the term AI winter warning the business community that AI hype would lead to disappointment and the collapse
of the industry, which happened three years later

• 1985 - Judea Pearl

• introduced Bayesian networks causal analysis provides statistical techniques for representing uncertainty in computers

• 1986 - Rumelhart, David E, Hinton, Geoffrey E, Williams, Ronald J.

• "Learning representations by back-propagating errors"

• 1988 - Peter Brown

• "A Statistical Approach to Language Translation," paving the way for a widely studied machine translation method

• 1989 - Yann LeCun, Yoshua Bengio, Patrick Haffner

• demonstrated convolutional neural networks (CNNs) to recognize handwritten characters

• 1997 - Sepp Hochreiter, Jürgen Schmidhuber

• proposed the Long Short-Term Memory recurrent neural network, for sequential data such as speech or video

• 1997 - IBM's Deep Blue

• defeated Garry Kasparov in a historic chess rematch, the first defeat of a reigning world chess champion by a computer under
tournament conditions
IBM's Deep Blue
The Third Wave
Deep Learning Revolution
The resurgence of neural networks
 The Third Wave
• 2006 - Fei-Fei Li
• ImageNet visual database, which became a catalyst for the AI boom and the basis of an annual competition for image
recognition algorithms

• 2009 - Rajat Raina, Anand Madhavan, Andrew Ng

• "Large-Scale Deep Unsupervised Learning Using Graphics Processors," presenting the idea of using GPUs to train large
neural networks

• 2011 - IBM Watson

• In the iconic quiz show Jeopardy, IBM Watson computer system defeated the show's all-time (human) champion, Ken Jennings

• 2012 - Geoffrey Hinton, Ilya Sutskever, Alex Krizhevsky

• introduced a deep CNN architecture that won the ImageNet challenge and triggered the explosion of deep learning research
and implementation

• 2013 - DeepMind
• introduced deep reinforcement learning, a CNN that learned based on rewards and learned to play games through repetition,
surpassing human expert levels

• 2013 - Tomas Mikolov Google research

• introduced Word2vec to automatically identify semantic relationships between words

• 2013 – European Union Human Brain Project (€607 million)

• European Future and Emerging Technologies (FET) Flagship project that ran from 2013 to 2023, pioneered a new paradigm in brain
research, at the interface of computing and technology
 The Third Wave
• 2014 - Ian Goodfellow
• invented generative adversarial networks, a class of machine learning frameworks used to generate photos, transform images
and create deepfakes

• 2014 - Diederik Kingma, Max Welling

• introduced variational autoencoders to generate images, videos and text

• 2014 - Facebook AI
• developed the deep learning facial recognition system DeepFace, which identifies human
faces in digital images with near-human accuracy

• 2016 - DeepMind's AlphaGo

• defeated top Go player Lee Sedol in Seoul, South Korea

• 2017 - Stanford University

• "Deep Unsupervised Learning Using Nonequilibrium Thermodynamics." The technique provides a way to reverse-
engineer the process of adding noise to a final image

• 2017 - Google research

• Introduced transformers in the seminal paper "Attention Is All You Need," inspiring subsequent research into tools that could
automatically parse unlabeled text into large language models (LLMs)

• 2018 - OpenAI GPT

• Generative Pre-trained Transformer, paving the way for subsequent LLMs

• 2020 - DeepMind AlphaFold

• AlphaFold-predicted models of protein structures of nearly the full UniProt proteome of humans and 20 model organisms,
amounting to over 365,000 proteins
Level 0 (No Driving Automation)
Level 1 (Driver Assistance)
Level 2 (Partial Driving Automation)
Level 3 (Conditional Driving Automation)
Level 4 (High Driving Automation)
Level 5 (Full Driving Automation)
 2018 Turing Award - ACM Awards

Yann LeCun Geoffrey Hinton Yoshua Bengio

AlphaFold - the first serious scientific problem solved by AI
AI in Industry
• Manufacturing
• AI can improve efficiency, product quality, and employee safety. It can also be used for
predictive maintenance, quality control, and robotics. AI can also help reduce defects
and ensure consistent standards through data analysis, anomaly detection, and
predictive maintenance.
• Healthcare
• AI can improve preventive care, quality of life, and patient outcomes. It can also help
diagnose and treat patients more accurately. AI can also predict and track the spread
of infectious diseases.
• Finance
• AI can help lenders assess borrowers' risk levels and make lending decisions. Venture
capital firms can use AI to generate customized insights and financial risk management
decisions. AI can also help detect fraudulent transactions and predict the rise and fall of
stock values.
• Information technology
• Machine learning (ML) is a subset of AI that can play a central role in IT companies. ML
enables applications to learn from historical data and accurately predict outcomes
without explicit programming.
The Fourth Wave
The Rise of Generative AI
 Generative AI
• Text
• OpenAI GPT
• Google Bard
• Meta LLMA
• Image
• [Link] Stable Diffusion
• MidJourney
• OpenAI DallE
• Model underlying causal structure that generates new data • Music
• LLM models trained on massive amounts of text data • Meta MusicGen
• Social Media, Web Pages, Books, Code Repositories, • Google MusicLM
Confluence Pages, Test Cases • Speech
• Capable of understanding complex queries and answer • OpenAI Whisper
accordingly
• Based on trained knowledge or custom knowledgebase
Emotional
Reasoning Learning Problem Solving Perception/Control Linguistic Intelligence
Intelligence
Symbolic Logic/predicate Machine Learning Planning, Decision Theory Pattern Recognition Natural Language Processing ?
calculus
Expert System Bayesian Logic Machine Learning Machine Learning/Symbolic
Logic
Software 1.0 Software 2.0 Dynamic Programming Statistical Learning

Generative AI Generative AI Generative AI Generative AI Generative AI Generative AI?

 The Fourth Wave

The Rise of Generative AI

• 2018 - OpenAI
• introduced the GPT -1 (Generative Pretraining Transformer)

• 2018 - Google
• BERT (Bidirectional Encoder Representations from Transformers) was introduced

• 2019 - OpenAI
• released GPT - 2

• 2019 – Google
• Fine tuned Language Model FLAN T5

• 2020 - OpenAI
• released GPT - 3
 The Fourth Wave

The Rise of Generative AI

• 2021 - OpenAI
• introduced the Dall-E multimodal AI system that can generate images from text prompts

• 2022 - OpenAI
• released ChatGPT in November to provide a chat-based interface to its GPT-3.5 LLM

• 2023 - Google
• released Bard chat-based interface to its PaLM2 LLM

• 2023 – Meta AI
• released LLaMA-2 (Large Language Model Meta AI)

• 2023 - OpenAI
• announced the GPT-4 Multi-modal LLM that receives both text and image prompts

• 2023 - Google
• released Bard chat-based interface to its Gemini Pro Multi-modal LLM
Foundation Models – A Paradigm Shift in AI
Bommasani, Rishi, et al. "On the opportunities and risks of foundation models." arXiv preprint arXiv:2108.07258 (2021).

• AI is undergoing a paradigm shift with

the rise of foundation models
• Trained on broad data at scale
• Adaptable to a wide range of
downstream tasks
• Language, vision, robotics,
reasoning, human interaction
 Language is an abstract symbolic system
Neural Networks are Sub-symbolic

Connectionists Bayesians

Predicting the probability of the next word

in the context of current and the previous

Picard F, Friston K. Predictions, perception, and a sense of self. Neurology. 2014 Sep 16;83(12)
Emergent Abilities of Large Language Models
Jason Wei, Yi Tay, Rishi Bommasani, Colin Raffel, Barret Zoph, Sebastian Borgeaud, Dani Yogatama,
Maarten Bosma, Denny Zhou, Donald Metzler, Ed H. Chi, Tatsunori Hashimoto, Oriol Vinyals, Percy Liang,
Jeff Dean, William Fedus
 LLM Training Stages
Andrej Karpathy : State of GPT | BRK216HFS

9/3/20XX Presentation Title 46

The hidden monster

• The pretrained model is an untamed monster

• trained on indiscriminate data scraped from the Internet
• clickbait, misinformation, propaganda, conspiracy theories,
or attacks against certain demographics

• This monster was then finetuned on higher quality data

• Stack Overflow, Quora, GitHub or human annotations
• makes it somewhat socially acceptable

• The finetuned model was further polished using RLHF

• make it aligned with human values
• giving it a smiley face

[Link]
The Brain-scan of GPT
• Transformer based language models are treating words
as particles, which move around under the influence of
each other, generating intriguing patterns

• What determines this structure? Ultimately, it’s presumably

some “neural net encoding” of features of human
language. But as of now, what those features might be is
quite unknown

Prompt Engineering

• Prompting dynamically triggers various parts of the

topography

• In order to trigger proper pathways in the large network,

the input pattern should be able to invoke the right
connections

[Link]
Context Length

[Link]
LLM Leaderboard (15/01/2024)

Rank Model Arena Elo Multimodal Votes Organization License

1 GPT-4-Turbo 1249 ✓ 23069 OpenAI Proprietary

2 GPT-4-0314 1190 ✓ 16237 OpenAI Proprietary

3 GPT-4-0613 1160 ✓ 20884 OpenAI Proprietary

4 Mistral Medium 1150 × 6586 Mistral Proprietary

5 Claude-1 1149 × 16956 Anthropic Proprietary

6 Claude-2.0 1131 ? 11204 Anthropic Proprietary

7 Mixtral-8x7b-Instruct-v0.1 1123 × 12469 Mistral Apache 2.0

8 Gemini Pro (Dev) 1120 ✓ 1898 Google Proprietary

9 Claude-2.1 1119 ? 20883 Anthropic Proprietary

10 GPT-3.5-Turbo-0613 1116 × 26583 OpenAI Proprietary

Collected over 200,000 human preference votes to rank LLMs with the Elo ranking system

[Link]
 Hans Moravec
Max Tegmark: Life 3.0
(Aug/2017)
Where are we Now?
• Perception
• Computer Vision – Good
• Speech Understanding - Good

• Language Understanding
• Generative AI – LLMS - Good

• Reasoning
• Generative AI – LLMS - Evolving

• Planning
• Generative AI – LLMS - Evolving

• Problem Solving
• Generative AI – LLMS - Evolving

[Link]
The Technological Singularity
 LLMs to LMM
Large Multimodal Models (LMMs)

Brain is a joint embedding space of multiple sensory modalities (~20)

Self-Supervised Learning from Images with a Joint-Embedding Predictive Architecture Mahmoud Assran1,2,3* Quentin Duval1
Ishan Misra1 Piotr Bojanowski1 Pascal Vincent1 Michael Rabbat1,3 Yann LeCun1,4 Nicolas Ballas1 1Meta AI (FAIR) 2McGill
University 3 Mila, Quebec AI Institute 4New York University

IMAGEBIND: One Embedding Space To Bind Them All Rohit Girdhar∗ Alaaeldin El-Nouby∗ Zhuang Liu Mannat Singh Kalyan Vasudev
Alwala Armand Joulin Ishan Misra∗ FAIR, Meta AI

The Embodiment Problem

GPT-4 Technical Report OpenAI – Image, Audio, Text

Google Deepmind Gemini OpenAI – Image, Audio, Video and Text

Multimodality and Large Multimodal Models (LMMs) Oct 10, 2023, Chip Huyen
Computers are Growing Again
The Trillion-Parameter Instrument of AI
17,468 vacuum tubes
150 kW of power

NVIDIA DGX GH200

256 NVIDIA Grace Hopper Superchips
144 terabytes (TB) of shared memory
16 kilowatts x 16 Racks

9/3/20XX 62
References

Russell & Norvig, Artificial Intelligence A Modern Approach 2020

Haroon Sheikh, Corien Prin,s Erik Schrijvers, Mission AI The New System Technology

[Link]

Pattern Recognition and Deep Learning Technologies, Enablers of Industry 4.0, and Their Role in
Engineering Research

Epistemological Problems Of Artificial Intelligence John Mccarthy

Stanford University - The Ai Index Report Measuring trends in Artificial Intelligence

 Thank You

Are we seeing the sunlight or

the shadows still