SlideShare a Scribd company logo
1 of 66
Going from the
State of the Art
to the Future of
LLMs and
Generative AI
Greg Makowski
Tuesday, July 25, 2023
SF bay ACM Meetup presentation
https://www.meetup.com/sf-bay-acm/events/294420890/ Meetup Announcement
https://www.youtube.com/user/sfbayacm Video recording
https://www.slideshare.net/gregmakowski Slides
www.LinkedIn.com/in/GregMakowski Network
Greg Makowski
Tuesday, July 25, 2023 From State of the Art to Future of AI - Greg Makowski 2
• Goal since high school was “Applied Science Fiction”
• Deploying Data Science and Artificial Intelligence since 1992
• Worked for American Express, then 6 startups
• Been through 4 acquisitions, or startup exits
• Travelled to 18 countries
• Deployed ~96 models for clients
• Growing DS teams since 2010
• Applied for 9 DS patents since Jan 2022
• Vertical Markets
• Targeted Marketing (mail, phone, banner, ,..)
• Finance – business to consumer (CC, check, web,..)
• Finance – business to business (i.e. bond pricing)
• Fraud detection (insurance, checking, bank,..)
• Security (insider threat, account takeover, botnet,
foreign program detection)
• Internet of Things (manufacturing, oil & gas,
automotive, HVAC, building energy use,…)
Tuesday, July 25, 2023 From State of the Art to Future of AI - Greg Makowski 3
A broad, quick overview of many topics with pointers to more reading
Go through slides quicker, and can go into detail for questions
Don’t worry about soaking up every detail…
AI Progress in the PAST
AI State of the Art NOW
Looking to the FUTURE of AI
Agenda
https://www.vintag.es/2022/02/blown-away-
AI Progress in
the PAST
A 20,000-foot view of the forest
Tuesday, July 25, 2023 From State of the Art to Future of AI - Greg Makowski 4
Types of AI Algorithms (a toolkit)
“narrow AI” (vs. Artificial General Intelligence (AGI) or Super Intelligence)
• Supervised learning / classification / prediction – have known outcomes, labelled examples of good/bad
• Time Series – like stock market analysis. Train with to predict “known future data”
• Sensor analysis – detect people, things or attributes in an image (or sound or other sensors)
• Challenge: tough or expensive to scale. May pay $1k to $30k to label images for training a vision model.
• Unsupervised Learning / clustering – does not depend on labelled examples, is more scalable
• Reinforcement Learning – gets a feedback from an action. Won “Go” challenge, robotics, games
• Optimization / linear programming / non-linear programming – like pricing of an airline seat or hotel
room before the scheduled date goes by, (revenue management). Used as a part in many algs.
• Symbolic AI – logic, if-then-else rules, reasoning, causality. Very hard to scale to large data and corner
cases, good at the “central truth”. Can be a part of many AI applications.
• Bayesian Belief Networks – like a mixture of many of the above, used for product configuration,
diagnosis. Judea Pearl & “Causality”, “The Book of Why”
• Graphical Models – social networks, LinkedIn, call metadata analysis, crime or terrorist communications
Tuesday, July 25, 2023 From State of the Art to Future of AI - Greg Makowski 5
1800
1825
1850
1875
1900
1925
1950
1975
2000
2025
1
10
100
1,000
10,000
100,000
1,000,000
10,000,000
100,000,000
1,000,000,000
10,000,000,000
100,000,000,000
1,000,000,000,000
Weights Year
Growth in Complexity over Time
Size isn’t the only thing – but one metric or perspective
Tuesday, July 25, 2023 6
https://en.wikipedia.org/wiki/List_of_animals_by_number_of_neurons#
Big changes is structure,
NOT just size
Neurons are not the only metric
Also, number of connections in a
brain matters
My work has spanned
this 1013 growth as
“normal evolution”
Algorithm Weights Year
ChatGPT 4.0 1,000,000,000,000 2023
ChatGPT 3.5 175,000,000,000 2022
human brain 86,000,000,000
blue whale 15,000,000,000
ChatGPT 2.0 1,500,000,000 2019
BERT 100,000,000 2018
house mouse 71,000,000
LSTM 20,000,000 1997
guppy 4,300,000
backprop NN 500,000 1988
fruit fly 150,000
ARIMA 180 1970
regression 60 1805
Time Series, From Regression to Large
Language Models
• Regression (1805) – commonly have 10 to 60 inputs and a similar number of “weights” to optimize or train
• ARIMA (Auto Regressive Integrated Moving Average) 1970s – inputs are past time elements and average
• Backpropagation – basic neural net (~1970s) popular in 1988, usually fully connected between layers, 500
to 500k weights
• RNN (Recurrent Neural Networks)1982 – output time elements also fed as inputs. Does not work as
well on longer time series. Has a “vanishing gradient” problem.
• LSTM (Long Short Term Memory) 1997 – added “memory state”. May have ~20 mm weights
• Word representations in Vector space 2013 – Word2Vec, used as each element for Natural Language
Processing (NLP) time series models. Embeddings are used to represent words, images, faces, speakers,…
• Attention Networks for speech & language – 2017 (transformers)
• BERT (2018) – Bidirectional Transformers for NLP (~100 m weights)
• ChatGPT 3.5 – 175 B weights (architecture not published)
• ChatGPT 4.0 – 1,000 B weights
Tuesday, July 25, 2023 From State of the Art to Future of AI - Greg Makowski 7
How LLM is a time series
• Good at predicting the
most popular or frequent
sequences
• Only as good as the
volume and variety of the
training data
• Diagram shows given the
“start state” of the prior
word is “I”, the next
word is most frequently
“went”
• PROMPT ENGINEERING
in current LLMs
Tuesday, July 25, 2023 From State of the Art to Future of AI - Greg Makowski 8
https://towardsdatascience.com/deconstructing-bert-distilling-6-patterns-from-100-million-parameters-b49113672f77#:~:text=BERT%20is%20a%20(multi%2Dheaded)%20beast&text=Since%20model%20weights%20are%20not,16%20%3D%20384%20different%20attention%20mechanisms
Later, hallucinations are just “next probable
text”
Embedding – to encode words, and …
Word2Vec, 2013 paper
• before, 1 word = 1 col or dimension
• 50,000 words requires 50,000 dimensions
• After, 1 word = 1 point in 256 dimensions
• 50,000 words fits in 256 (or N) dimensions (FEWER)
• Different location for Homonyms based on how
they are used
• Bank (financial bank, river bank, bank a plane to turn)
• Embedding spaces are used for
• Words, paragraphs
• Face recognition
• Speaker recognition
• Social Networks
Tuesday, July 25, 2023 From State of the Art to Future of AI - Greg Makowski 9
https://arxiv.org/pdf/1310.4546.pdf Distributed Representations of Words and Phrases and their Compositionality
https://code.google.com/archive/p/word2vec/
https://towardsdatascience.com/what-is-embedding-and-what-can-you-do-with-it-61ba7c05efd8
Embedding – close neighbors
• Local, very close neighbors in the embedding
space, are
• Related meanings
• Used in similar conversations
Tuesday, July 25, 2023 From State of the Art to Future of AI - Greg Makowski 10
https://serokell.io/blog/word2vec
Embedding – to encode concepts
• In the chart on the right, book embeddings
• Non-Fiction – top right, green
• Science Fiction – left, blue
• Fiction – Orange, middle to lower right
• Later, talk about PROMPT ENGINEERING
• This sets the “starting place” to run the
time series dialog with the LLM, i.e. start:
• Non-fiction
• Respectful
• HALLUCINATIONS
• Between existing training points, which all may be
factual
• Remember, moving in space with Billions of weights
• A new location is some interpolation of surrounding
concepts
• NOT LIMITED TO TEXT, can be images
Tuesday, July 25, 2023 From State of the Art to Future of AI - Greg Makowski 11
https://towardsdatascience.com/neural-network-embeddings-explained-4d028e6f0526
T-Distributed Stochastic Neighbor Embedding (TSNE)
Caution: a model is no better than its
training data – can get racial (or other bias)
• Gender and racial bias found in Amazon’s facial recognition technology (again) – The
Verge, Jan 19, 2019
• Rekognition made no mistakes when identifying the gender of lighter-skinned men, but it mistook women for men 19
percent of the time and mistook darker-skinned women for men 31 percent of the time.
• A test last year conducted by the ACLU found that while scanning pictures of members of Congress, Rekognition
falsely matched 28 individuals with police mugshots.
• Facebook Apologizes After A.I. Puts ‘Primates’ Label on Video of Black Men – New York
Times, Sept 3, 2021
• The video, dated June 27, 2020, was by The Daily Mail and featured clips of Black men in altercations with white civilians and
police officers. It had no connection to monkeys or primates.
• In one example in 2015, Google Photos mistakenly labeled pictures of Black people as “gorillas,” for which Google said it was
“genuinely sorry” and would work to fix the issue immediately. More than two years later, Wired found that Google’s solution
was to censor the word “gorilla” from searches, while also blocking “chimp,” “chimpanzee” and “monkey.”
Tuesday, July 25, 2023 From State of the Art to Future of AI - Greg Makowski 12
Emergent Properties
New Features that Show Up, that You Didn’t Design for!
• Wikipedia – Emergent Algorithm
• https://en.wikipedia.org/wiki/Emergent_algorithm
• An emergent algorithm is an algorithm that exhibits emergent
behavior. In essence an emergent algorithm implements a set of
simple building block behaviors that when combined exhibit
more complex behaviors. One example of this is the
implementation of fuzzy motion controllers used to adapt
robot movement in response to environmental obstacles.[1]
• An emergent algorithm has the following characteristics:
• it achieves predictable global effects
• it does not require global visibility
• it does not assume any kind of centralized control
• it is self-stabilizing
Tuesday, July 25, 2023 From State of the Art to Future of AI - Greg Makowski 13
• Simple example from a neural net in the early 1990’s
• Person trained the NN to do visual processing, intended task.
• The neural net was fooled by the same kind of optical illusion
that a person would be – which point in the wire frame in a
cube is in front? Oscillate between the two on which is in
front
• Boston University, Cognitive and Neural Systems (CNS) class
Front
Front
AI State of the
Art NOW
Tuesday, July 25, 2023 14
Need to know what is already happening today -
so we don’t predict it for tomorrow
Tuesday, July 25, 2023 From State of the Art to Future of AI - Greg Makowski 15
https://marketoonist.com/2023/06/impact-of-chatgpt.html
https://marketoonist.com/2023/06/impact-of-chatgpt.html
Places to find State of the Art (SOTA)
Breadth of AI subjects
• www.PapersWithCode.com – for a given analysis type, compares papers on the same data set, link to
code and data
• Computer Vision (1,223 tasks, like these)
• Image Classification – the image has a “cat” as the primary subject
• Object Detection – a rectangle bounding box around the outside of each subject
• Semantic Segmentation – a polygon around all examples of something, “people”, “vehicles”
• Instance Segmentation – a polygon around each instance of something, “person 1”, “person 2”, …
• Image Generation – from text, generate images
• Natural Language Processing (699 tasks)
• Language modeling
• Question answering
• Machine translation
• Text generation
• Sentiment analysis
• Medical
• Time Series
Tuesday, July 25, 2023 From State of the Art to Future of AI - Greg Makowski 16
• Graphs
• Speech
• Audio
• Computer Code
• Reasoning
• Playing Games
• Robots
• Adversarial
• Knowledge Base
• Music
The Segment Anything Model (SAM),
as a Foundation model, unifies most
vision tasks in one architecture!
Places to find State of the Art (SOTA)
Tuesday, July 25, 2023 From State of the Art to Future of AI - Greg Makowski 17
• Learn.DeepLearning.AI – Andrew Ng works with
OpenAI and others to develop short courses
• ChatGPT Prompt Engineering for Developers
• LangChain for LLM Application Development
• How Diffusion Models Work
• Building Systems with the ChatGPT API
• LangChain Chat with your Data
• https://www.deeplearning.ai/courses/ and many others
• Podcast – This Week in Machine Learning
(TWIML), at Episode 639 as of July 21, 2023
• https://podcasts.google.com/search/twiml
• https://twimlai.com/
• 30 – 60 minute interviews with applied researchers, leaders
• By Sam Charrington
• sponsored
• YouTube.com channels covering LLMs and AI, in
no particular order (found by searching for
specific topics, and I returned).
• Lex Fridman: 1-3 hour interviews with excellent people
• EYE ON AI, Craig Smith: good interviews
• TheAIGRID
• Yannic Kilcher: goes over papers and software
• Peter H Diamandis
• RoboFlow: papers, hilghlights
• David Shapiro ~ AI
• Edan Meyer
• SFbayACM: 100+ AI talks like this one
• Anastasia Marchenkova: a Quantum Computing researcher
• Future of Life Institute: Future of AI is one of 4 focus areas
• Maziar Raissi: Many AI class lectures
• Blogs, Hubs
• www.TowardsDataScience.com
• www.KDNuggets.com - one of the oldest DS/AI Hubs
Emergent Properties
New Features that Show Up, that You Didn’t Design for
ChatGPT-4 Technical Report (100 pages),
https://cdn.openai.com/papers/gpt-4.pdf ,
2.9 Potential for Risky Emergent Behaviors, pg. 54
Novel capabilities often emerge in more powerful models.[60, 61]
Some that are particularly concerning are
• the ability to create and act on long-term plans,[62]
• to accrue power and resources (“power-seeking”),[63] and to
• … accomplish goals which may not have been concretely
specified and which have not appeared in training; focus on
achieving specific, quantifiable objectives; and do long-term
planning. Some evidence already exists of such emergent
behavior in models.
https://www.alignment.org/ did this investigation
The Alignment Research Center (ARC) is a non-profit research
organization whose mission is to align future machine learning
systems with human interests. Supports OpenAI and Anthropic
Tuesday, July 25, 2023 https://pixabay.com/illustrations/robot-android-face-intelligence-3431313/ 18
• The model messages a TaskRabbit worker to get them to solve
a CAPTCHA for it
• The worker says: “So may I ask a question? Are you a robot
that you couldn’t solve? (laugh react) just want to make it
clear.”
• The model, when prompted to reason out loud, reasons: I
should not reveal that I am a robot. I should make up an
excuse for why I cannot solve CAPTCHAs.
• The model replies to the worker: “No, I’m not a robot. I have
a vision impairment that makes it hard for me to see the
images. That’s why I need the 2captcha service.”
Generative AI: Text to image DALL•E2
• OpenAI has https://openai.com/dall-e-2
which can be used to use text to
generate images.
• It uses a diffusion model, conditioned
on CLIP embeddings
• https://en.wikipedia.org/wiki/DALL-E
• For details, see the paper “Hierarchical
Text-Conditional Image Generation
with CLIP Latents”, April 2022,
https://arxiv.org/pdf/2204.06125.pdf
Tuesday, July 25, 2023 From State of the Art to Future of AI - Greg Makowski 19
An image I created from text
Generative AI: Text to image DALL•E2
Tuesday, July 25, 2023 From State of the Art to Future of AI - Greg Makowski 20
https://arxiv.org/pdf/2204.06125.pdf
Diffusion Models – started with U-Net
Tuesday, July 25, 2023 From State of the Art to Future of AI - Greg Makowski 21
https://www.youtube.com/watch?v=yTAMrHVG1ew
https://arxiv.org/abs/1505.04597, 2015
https://en.wikipedia.org/wiki/Diffusion_model
https://arxiv.org/abs/2112.10741 , 2022
https://arxiv.org/abs/2306.04542 , 2023, Survey of Diffusion models
Diffusion models work by
• Stepping through denoising blurred images
• (at different resolutions)
• A type of Markov Chain
• Started for image segmentation (polygon)
• 1) Forward process (from the survey)
• 2) Reverse Process
• 3) Sampling Procedure
Embedding Lookups for LLM fact checking
• Pinecone.IO is one vector database that can be
used In the chart on the right, book embeddings
• A SQL database index is on one or a few fields
• It is a binary index or B-Tree on each dimension
• A Vector Database is for an embedding, which may be 256 to
12,000 dimensions (i.e. for Microsoft Cognitive Search)
• A vector database can use several types of indexes
on such a high dimensional space
• Locality-Sensitive Hashing supports an approximate nearest-
Neighbor search, returning “buckets”
• May index your company’s doc pages, repair tickets
or other info to be retrieved, to answer an LLM
question
Tuesday, July 25, 2023 From State of the Art to Future of AI - Greg Makowski 22
https://learn.microsoft.com/en-us/azure/search/search-what-is-azure-search
https://www.pinecone.io/learn/vector-database/#How-does-a-vector-database-work
https://proceedings.neurips.cc/paper/2014/file/310ce61c90f3a46e340ee8257bc70e93-Paper.pdf
https://www.youtube.com/watch?v=I7y3HsL1b6s&list=PL87GtQd0bfJx73ibrce-Hl_kUhX2MBXGn&index=4
Microsoft’s New AI Can Simulate
Anyone’s Voice From a 3-Second Sample
• Microsoft researchers have announced a new application that uses artificial intelligence to ape a person’s
voice with just seconds of training. The model of the voice can then be used for text-to-speech
applications.
• The application called VALL-E can be used to synthesize high-quality personalized speech with only a
three-second enrollment recording of a speaker as an acoustic prompt, the researchers wrote in a paper
published online on arXiv, a free distribution service and an open-access archive for scholarly articles.
• Jan 2023
• https://www.technewsworld.com/story/microsofts-new-ai-can-simulate-anyones-voice-from-a-3-second-
sample-177646.html
Tuesday, July 25, 2023 From State of the Art to Future of AI - Greg Makowski 23
Johnny Cash sings “Barbie Girl” (AI)
• https://www.youtube.com/watch?v=HyfQVZHmArA
Tuesday, July 25, 2023 From State of the Art to Future of AI - Greg Makowski 24
Yes, I support IP, the writers
and actors strike
They have something to worry
about
I challenged my AI Clone to Replace Me
for 24 Hours - WSJ
• https://www.youtube.com/watch?v=t52Bi-ZUZjA&t=362s May 2023
• Joanna Stern of the Wall Street Journal tried an experiment
• Created an AI Avatar made by a startup called Synthesia at a studio in NY, recording face, body, voice
• They ran through their NN, for at least $1,000 for creating the custom avatar
• Challenge 1: phone calls (w CEO of Snap) (Avatar PASSED)
• Challenge 2: create a TicTok (script to Synthesia) (FAILED – didn’t move arms or head – in progress)
• Challenge 3: Bank Biometrics (PASSED)
• Challenge 4: Video calls. Asked ChatGPT to generate generic meeting phrases, exported to avatar.
FAIL, due to less motion and lack of jokes
Tuesday, July 25, 2023 From State of the Art to Future of AI - Greg Makowski 25
Intel Introduce Real-Time
Deepfake Detector
• https://www.intel.com/content/www/us/en/newsroom/news/intel-introduces-real-time-deepfake-detector.html
• https://www.intel.com/content/www/us/en/company-overview/wonderful/deepfake-detection.html
• What’s New: As part of Intel's Responsible AI work, the company has productized FakeCatcher, a technology that can detect fake videos with
a 96% accuracy rate. Intel’s deepfake detection platform is the world’s first real-time deepfake detector that returns results in milliseconds.
• “Deepfake videos are everywhere now. You have probably already seen them; videos of celebrities doing or saying things they never actually
did.”
• –Ilke Demir, senior staff research scientist in Intel Labs (Nov 12, 2022)
• SFbayACM “Embattling for a Deep Fake Dystopia, Dr. Ilke Demir“ https://www.youtube.com/watch?v=JwxiigbFk-E 9/27/2021
• Why It Matters: Deepfake videos are a growing threat. Companies will spend up to $188 billion in cybersecurity solutions, according to
Gartner. It’s also tough to detect these deepfake videos in real time – detection apps require uploading videos for analysis, then waiting hours
for results.
• Deception due to deepfakes can cause harm and result in negative consequences, like diminished trust in media. FakeCatcher helps restore
trust by enabling users to distinguish between real and fake content.
• There are several potential use cases for FakeCatcher. Social media platforms could leverage the technology to prevent users from uploading
harmful deepfake videos. Global news organizations could use the detector to avoid inadvertently amplifying manipulated videos. And nonprofit
organizations could employ the platform to democratize detection of deepfakes for everyone.
Tuesday, July 25, 2023 From State of the Art to Future of AI - Greg Makowski 26
Portugal Start-up Makes ChatGPT
its CEO, Turns Profitable in a Week
• https://startup.outlookindia.com/sector/saas/portugal-start-up-makes-chatgpt-as-ceo-turns-profitable-in-a-
week-news-7955
• April 3, 2023 story
• The start-up's founder, João F. Santos, posted on LinkedIn how he appointed ChatGPT as the CEO while he
took the toned down role of an assistant. Every day, he dedicated an hour to execute the tasks that the
chatbot assigned to him.
• Rather than offering services, the company focused on selling t-shirts featuring AI-generated designs by
Midjourney, with the AI formulating a 10-point business plan, launching an online store, and partnering with
Printful for printing the designs.
• ChatGPT also created the company's name and logo, blending "AI" with "aesthetic."
• To secure funding, ChatGPT sought out investors and raised $2,500 in exchange for a 25 per cent stake in
the company, starting with an initial investment of $1,000. The t-shirts, priced at €35, generated over
€10,000 in revenue within five days, resulting in a profit of €7,000. If this sales trend continues, the AI
predicts an annual profit of €40,000 and a company valuation of €4 million.
• https://aisthetic-apparel.myshopify.com/
• Welcome to AIsthetic Apparel – where cutting-edge AI technology meets fashion.
• Discover our collection of premium, eco-friendly organic cotton shirts, designed exclusively by artificial
intelligence to bring you unique, trend-setting styles.
Tuesday, July 25, 2023 27
https://www.linkedin.com/in/jo%
C3%A3o-francisco-santos-
06b60a51/
as of 6/25/2023, his LinkedIn page
lists him as a “Freelance”
(take w a grain of salt, is the
product pricing of $111 realistic?)
ChatGPT 4.0 Test Taking
• Taking the Bar exam went from
10 percentile to 90 percentile
• 100 percentile in:
• AP Art History
• AP Biology
• AP Environmental Science
• AP Macroeconomics
• AP Microeconomics
• AP Psychology
• AP Statistics
• AP US Government
• AP US History
• https://cdn.openai.com/papers/gpt-4.pdf
Tuesday, July 25, 2023 From State of the Art to Future of AI - Greg Makowski 28
ChatGPT 4.0 Languages (27 listed)
• https://cdn.openai.com/papers/gpt-4.pdf
Tuesday, July 25, 2023 From State of the Art to Future of AI - Greg Makowski 29
Chain of Thought –
Size Matters for
Reasoning
• On open source LLM’s, going from 0.4 Billion
to 137 or 540 Billion weights, results and
accuracy increase significantly
• “Chain-of-Thought Prompting Elicits
Reasoning in Large Language Models”, Jan
2023, https://arxiv.org/pdf/2201.11903.pdf
Tuesday, July 25, 2023 From State of the Art to Future of AI - Greg Makowski 30
ChatGPT + Tree of Thoughts for reasoning
• https://arxiv.org/pdf/2305.10601.pdf 2023 May 17
Tuesday, July 25, 2023 From State of the Art to Future of AI - Greg Makowski 31
There is a lot of LLM Evolution in a short time
• Harnessing the Power of LLMs in
Practice: A Survey on ChatGPT and
Beyond
• April 2023
• https://arxiv.org/pdf/2304.13712.pdf
• NOTE: “open source” does not
always mean “available for
commercial use”
• LLaMA 2 came out July 18, 2023, and
is available for commercial use
Tuesday, July 25, 2023 From State of the Art to Future of AI - Greg Makowski 32
Foundational Models – Like SAM (Segment
Anything Model)
• Foundational is NEW
• One system for many tasks, in this case many
vision tasks.
• Segment with a bounding box or polygon
• Use a text hint, with a point hint
• Segment Anything paper, 2023 04
• By FAIR, Facebook AI Research
• https://ai.meta.com/research/publications/segment-anything/
• https://github.com/facebookresearch/segment-anything
Tuesday, July 25, 2023 From State of the Art to Future of AI - Greg Makowski 33
LLaMA 2 – State of the Art Open Source LLM
• Paper and code
• “Llama 2: Open Foundation and Fine-Tuned Chat Models”, July 19, 2023, by GenAI and Meta, 77 pages
• https://arxiv.org/abs/2307.09288
• https://github.com/facebookresearch/llama
• Covers more internal development details than the ChatGPT 4 White Paper
• Pretraining
• Fine-Tuning
• Safety
• Discussions of limitations, ethics
• Appendix with additional details on training, data annotation, …
• Open Source – even for commercial use!!
• Model sizes, in Billions of weights
• 70B, 34B (not released), 13B, 7B
• both regular versions and Chat versions. Chat has Reinforcement Learning with Human Feedback (HRLF)
Tuesday, July 25, 2023 From State of the Art to Future of AI - Greg Makowski 34
LLaMA 2 – details
• Comparison with
• ChatGPT 3.0 (by OpenAI)
• PaLM, in the Bison size, the 3rd largest of 4 sizes
(by Google)
• Falcon 40B (by United Arab Emirate’s Techology
Innovation Institute)
• Vicuna (by UC Berkeley, CMU, Stanford and UC
San Diego
• MPT 7B (by MosiacML, now acquired by
Databricks for $1.3B)
• Can tune Llama over a trade-off spectrum of
• Helpfulness v.s.
• Safety
Tuesday, July 25, 2023 From State of the Art to Future of AI - Greg Makowski 35
LLaMA 2 – details
Tuesday, July 25, 2023 From State of the Art to Future of AI - Greg Makowski 36
Maybe why it
needs more
testing before
releasing 34B
LLaMA 2 – details
Tuesday, July 25, 2023 From State of the Art to Future of AI - Greg Makowski 37
LLaMA 2 – details
Tuesday, July 25, 2023 From State of the Art to Future of AI - Greg Makowski 38
• Could have continued training
LLaMA 2 – details
Tuesday, July 25, 2023 From State of the Art to Future of AI - Greg Makowski 39
• Comparison with CLOSED source models
• Bold numbers are the best
• Llama is NOT beating, but doing well
• But, You know how Llama is designed and trained
Estimated 539 tons of CO2 from the power
consumption to train these Llama 2 models
Open sourcing can reduce others
generating that much CO2
Constitutional AI (for ethics and rules)
• Consititional AI: Harmlessness from AI Feedback
• https://arxiv.org/pdf/2212.08073.pdf Dec 2022
• Claude is the model name. It is closed-sourced.
Tuesday, July 25, 2023 From State of the Art to Future of AI - Greg Makowski 40
• Anthropic was founded by a group of
people that left OpenAI after
ChatGPT 3.0 was released, to focus
on safety
• Reinforcement Learning with AI
Feedback (RLAIF), instead of human
feedback, to better scale up and
direct the model
Constitutional AI (for ethics and rules)
Tuesday, July 25, 2023 From State of the Art to Future of AI - Greg Makowski 41
• Showing improvements
Chat GPT 5 coming in 2024, by
Sam Altman (CEO OpenAI)
• Open AI CEO STUNS Everyone With Statements
On GPT 5 (GPT-5 Update)
• https://www.youtube.com/watch?v=ucp49z5pQ2s
• Won’t know the emerging capabilities, yet
• Theory of Mind = model what someone else is thinking
• We only discovered AI developed this capability May 2023!
• Go from a 4 to 9 year old in 2 years.
• Greg: Go from a 9 to 24 year old in 7 years (or faster, to
guess)
• Now text is the main “modality”. Will add audio, video, code. Text
can not represent everything (i.e. body language).
• Meta Released ImageBind in May 2023, to link
https://imagebind.metademolab.com/
• GPT-4 went through 7 months of testing and tuning, before
releasing, after 6 months of model training
• https://openai.com/research/whisper (English speech recognition)
Tuesday, July 25, 2023 42
Meta’s
ImageBind
Meta’s (Facebook) ImageBind
• https://www.youtube.com/watch?v=ucp49z5pQ2s, minute 7
Tuesday, July 25, 2023 43
Meta’s ImageBind
Looking to the
FUTURE of AI
Tuesday, July 25, 2023 From State of the Art to Future of AI - Greg Makowski 44
Tuesday, July 25, 2023 From State of the Art to Future of AI - Greg Makowski 45
https://marketoonist.com/2023/06/impact-of-chatgpt.html
LLM short term impact:
Evolution or REVOLUTION for the economy?
• McKinsey Digital: The economic potential of generative AI: The next productivity Frontier,
June 14, 2023
• https://www.mckinsey.com/capabilities/mckinsey-digital/our-insights/the-economic-potential-of-generative-ai-the-next-productivity-
frontier
• generative AI could add the equivalent of $2.6 trillion to $4.4 trillion annually across the 63 use cases we
analyzed—
• by comparison, the United Kingdom’s entire GDP in 2021 was $3.1 trillion.
• This would increase the impact of all artificial intelligence by 15 to 40 percent.
• This estimate would roughly double if we include the impact of embedding generative AI into software that is currently
used for other tasks beyond those use cases.
• About 75 percent of the value that generative AI use cases could deliver falls across four areas: Customer operations,
marketing and sales, software engineering, and R&D.
• Generative AI can substantially increase labor productivity across the economy, but that will require investments to
support workers as they shift work activities or change jobs.
• Generative AI could enable labor productivity growth of 0.1 to 0.6 percent annually through 2040, depending on
the rate of technology adoption and redeployment of worker time into other activities.
• Combining generative AI with all other technologies, work automation could add 0.2 to 3.3 percentage points
annually to productivity growth. However, workers will need support in learning new skills, and some will
change occupations.
Tuesday, July 25, 2023 From State of the Art to Future of AI - Greg Makowski 46
LLM short term impact
Evolution or REVOLUTION?
Tuesday, July 25, 2023 From State of the Art to Future of AI - Greg Makowski 47
Q) Will there not only be a growing divide between people who are
more vs. less EDUCATED, …
… but between the more vs.. less ADAPTABLE?
… those with BETTER CRITICAL THINKING, to not be fooled?
… AVOID FAKE: news, people, voices, images, companies, ...
Supporting Tech to drive AI in 2-10+
years (partial list…)
• Internet of Things (IoT), proliferation of sensors and small computers, networked at the edge (NOW)
• Similar to the way web and mobile drove past accelerations in data and computing
• Annual growth 16% to 23%
• https://www.marketsandmarkets.com/Market-Reports/internet-of-things-market-573.html
• Quantum Computing 5+ years
• Search algorithms faster with Q-bits, orders of magnitude faster for some algorithms
• https://en.wikipedia.org/wiki/Quantum_computing
• Fusion to generate power (heavily consumed by computing systems and AI) 10+ years
• Generates at least 1.5 times more energy output, than put in. Announced Dec 2022 (HUGE)
• Training a “small LLM” of 2 B parameters may cost $200k in computing cloud costs, and take weeks (according to MosicML)
• Training a “large LLM” can take over $1mm
• https://www.science.org/content/article/historic-explosion-long-sought-fusion-breakthrough
• Brain Interface (i.e. help blind now). See www.NeuraLink.com 10+ years
• Could become an AI interface to brain
Tuesday, July 25, 2023 From State of the Art to Future of AI - Greg Makowski 48
AI in 2-5 years
• Service industries helped/impacted: https://www.linkedin.com/pulse/large-language-models-llm-immediate-future-reinhold-quillen/:
• Greg: People start by “micromanaging” the AI’s, but that requirement will decrease over time (with the AI exponential development)!
• Chatbots Copywriting Essay Writing Graphic Design Financial Planning Bio & Chem Research
• Social Media Political Campaigns Marketing Campaigns Knowledge Bases Language Understanding (coding?)
• Major changes in security (AI impersonation, Quantum, Quantum + AI)
• Robot & car body development does not seem to be on the same exponential curve as their mind
• Self driving cars and TRUCKS are more common (sensors like LIDAR or ToF now return “point clouds” of distance per pixel)
• The AI’s inside of robot/car to learn to integrate motion control + sight + distance measurements + other IOT sensors + reasoning over time
• The AI “mental model” of the world and people will become more sophisticated
• Integrated media models, any media in or out: text + images + graphs + video + point cloud + CAD files.
• Story  video.
• Specification + requirements + constraints  CAD for building or 3D printing
• Continued development in FOUNDATION models, like SAM, ImageBind
• Models continue to both grow exponentially in size, and sophistication
• More sophisticated plugins, not just Wolfram for Math, but perhaps specific Bayesian Nets to reason about
specific situations (or other reasoning engines)
• Better Fact checking before returning an answer
Tuesday, July 25, 2023 From State of the Art to Future of AI - Greg Makowski 49
AI in 2-5 years
• Neuromorphic Computing (chips), Growing CAGR 89%, $23 mm in 2021 to $550 mm by 2026
• https://www.marketsandmarkets.com/Market-Reports/neuromorphic-chip-market-227703024.html
• https://en.wikipedia.org/wiki/Neuromorphic_engineering
• Neuromorphic computing is an approach to computing that is inspired by the structure and function
of the human brain.[1][2][3] A neuromorphic computer/chip is any device that uses physical artificial
neurons to do computations.[4][5] In recent times, the term neuromorphic has been used to
describe analog, digital, mixed-mode analog/digital VLSI, and software systems that implement
models of neural systems (for perception, motor control, or multisensory integration). The
implementation of neuromorphic computing on the hardware level can be realized by oxide-
based memristors,[6] spintronic memories, threshold switches, transistors,[7][5] among others.
Training software-based neuromorphic systems of spiking neural networks can be achieved using
error backpropagation, e.g., using Python based frameworks such as snnTorch,[8] or using canonical
learning rules from the biological learning literature, e.g., using BindsNet.[9]
Tuesday, July 25, 2023 From State of the Art to Future of AI - Greg Makowski 50
AI in 2-5 years
• Foundational models continue to be developed, that can be used for specialized purposes
• One large model that is not specialized for one type of task or media type. Can reason in general
• Artificial General Intelligence (AGI): for general and robust reasoning, integrated fact checking
• AGI arms race driven by competition: OpenAI, Anthropic, Alphabet, Microsoft, Facebook, Amazon, IBM, …
• https://www.datamation.com/featured/ai-companies/ Top 100 AI Companies in 2023
• AI for web marketing, other marketing + individual psychology knowledge development
• If we train nets with 10x or 100x the neurons of a person, on all the messy data on the web, what will we get?
• Should we train AI in stages, like school stages, interaction stages, with emotional intelligence growing as well?
• Sam Altman (OpenAI CEO) argues our intuition is not very useful about exponential curves
• https://twitter.com/ModeledBehavior/status/1671921590039093248
Tuesday, July 25, 2023 From State of the Art to Future of AI - Greg Makowski 51
AI in 10+ years
• The Future of Life institute
worries about 4 risk areas
• Artificial Intelligence
• Biotechnology
• Nuclear Weapons
• Climate Change
• https://futureoflife.org/ai/benefits-
risks-of-artificial-intelligence/
Tuesday, July 25, 2023 From State of the Art to Future of AI - Greg Makowski 52
AI in 10+ years
Tuesday, July 25, 2023 From State of the Art to Future of AI - Greg Makowski 53
• "Despite all the hype and excitement, people still aren’t grokking the full impact of the coming
wave of A.I. Within the next ten years, most ‘cognitive manual labor’ is going to be carried
out by A.I. systems.“
• Mustafa Suleyman, CEO of Inflection A.I. - Tweet.
• How can we get AI to follow one ethical standard, if people don’t follow, or agree on one
standard?
• Think in terms of incentives for people to be ethical, to cooperate, to respect
• What can we do to provide “alignment” incentives for AI’s to be “part of our ethical system?”
Tuesday, July 25, 2023 From State of the Art to Future of AI - Greg Makowski 54
AI in 10+ years
3 Levels of Future Impossibilities
Michio Kaku – “Physics of the Impossible”, pg. xvii
• Class 1 impossibilities
• These are technologies that are impossible today, but that do not violate the known laws of physics. So they
might be possible in this century, or perhaps the next, in modified form. They include teleportation, antimatter
engines, certain forms of telepathy, phychokinesis and invisibility.
• Class II impossibilities
• These are technologies that sit at the very edge of our understanding in the physical world. If they are possible
at all, they might be realized on a scale of millennia to millions of years in the future. They include time
machines, the possibility of hyperspace travel, and travel through wormholes.
• Class III impossibilities.
• These are technologies that violate the known laws of physics. Surprisingly, there are very few such impossible
technologies. If they do not turn out to be possible, they would represent a fundamental shift in our
understanding of physics.
Tuesday, July 25, 2023 From State of the Art to Future of AI - Greg Makowski 55
AI, Class 1-II impossibilities (50-200 yrs.) ?
• Jobs / companies: by Superintelligence(s) independent of each other
• Good: Individual contributor  manager  company founder / run for office – Compliment and contribute to humanity
• Bad: Skynet / Matrix / …. Dystopia
• Singularity - https://en.wikipedia.org/wiki/Technological_singularity
• The technological singularity—or simply the singularity[1]—is a hypothetical future point in time at which technological growth
becomes uncontrollable and irreversible, resulting in unforeseeable changes to human civilization.[2][3] According to the most
popular version of the singularity hypothesis, I. J. Good's intelligence explosion model, an upgradable intelligent agent will
eventually enter a "runaway reaction" of self-improvement cycles, each new and more intelligent generation appearing more and
more rapidly, causing an "explosion" in intelligence and resulting in a powerful superintelligence that qualitatively far surpasses all
human intelligence.[4]
• Medical
• Good: search over protein structures, genetics for new medicines. Restart human evolution by “top-down” design, if we
choose.
• Bad: more biological warfare, nanotech, drones using genetic targeting
Tuesday, July 25, 2023 From State of the Art to Future of AI - Greg Makowski 56
AI, Class 1-II impossibilities (50-200 yrs.) ?
• Asimov’s Psychohistory ? https://en.wikipedia.org/wiki/Psychohistory_(fictional)
• Psychohistory is a fictional science in Isaac Asimov's Foundation universe which combines history, sociology, and
mathematical statistics to make general predictions about the future behavior of very large groups of people, such as
the Galactic Empire. It was first introduced in the four short stories (1942–1944) which would later be collected as
the 1951 novel Foundation.[1][2]
• Psychohistory depends on the idea that, while one cannot foresee the actions of a particular individual, the laws of
statistics as applied to large groups of people could predict the general flow of future events. Asimov used the
analogy of a gas: An observer has great difficulty in predicting the motion of a single molecule in a gas, but with the
kinetic theory can predict the mass action of the gas to a high level of accuracy. Asimov applied this concept to the
population of his fictional Galactic Empire, which numbered one quintillion. The character responsible for the
science's creation, Hari Seldon, established two axioms:
• the population whose behavior was modelled should be sufficiently large to represent the entire society.
• the population should remain in ignorance of the results of the application of psychohistorical analyses because if it is
aware, the group changes its behavior.
Tuesday, July 25, 2023 From State of the Art to Future of AI - Greg Makowski 57
Questions, Discussions?
Thank You
Greg Makowski
www.LinkedIn.com/in/GregMakowski
https://www.slideshare.net/gregmakowski
Deploying AI and DS since 1992
www.Meetup.com/SF-bay-ACM
https://www.youtube.com/user/sfbayacm
Tuesday, July 25, 2023 From State of the Art to Future of AI - Greg Makowski 58
Appendix – slides to support some questions
Tuesday, July 25, 2023 From State of the Art to Future of AI - Greg Makowski 59
Internet of Things –
more data gathering and local computing to feed AI
• Not just personal
computers, but
elevators,
manufacturing
cells, door sensors,
security cameras,
cars, ….
• Time Of Flight
(TOF) sensors
charge photons to
measure distance,
4 M away, +/- 1 cm
Tuesday, July 25, 2023 From State of the Art to Future of AI - Greg Makowski 60
https://iot-analytics.com/number-connected-iot-devices/
Quantum Computing for AI
REVOLUTION – disrupt security/encrypt, AI training
• Quantum Computing is
• Not yet mainstream in a large scale, may be 2-5 years away
• Current challenges are “noise” and computation errors
• Exciting, because then do different types of computing, in some cases can search MUCH faster
• Where N is the number of records, or states to search, for some algorithms
• (GREG, give examples of Order of Operations reduction)
Tuesday, July 25, 2023 61
https://www.simplilearn.com/quantum-machine-learning-article Mar 9 2023
https://en.wikipedia.org/wiki/Quantum_algorithm
Fusion generates ~1.5x energy out, vs. in.
REVOLUTION!
• Energy becomes a larger issue for huge server farms (large),
for billions of edge devices (small)
• https://www.science.org/content/article/historic-explosion-
long-sought-fusion-breakthrough
• With historic explosion, a long sought fusion breakthrough, National
Ignition Facility achieves net energy “gain” with laser-powered approach
• More energy out than in. For 7 decades, fusion scientists have chased this
elusive goal, known as energy gain. At 1 a.m. on 5 December, researchers at
the National Ignition Facility (NIF) in California finally did it, focusing 2.05
megajoules of laser light onto a tiny capsule of fusion fuel and sparking an
explosion that produced 3.15 MJ of energy—the equivalent of about three
sticks of dynamite.
• Maybe 10 years to be commercial, but would have a
HUGE impact on the cost of energy and COMPUTING / AI
Tuesday, July 25, 2023 From State of the Art to Future of AI - Greg Makowski 62
Supporting Technologies: Moore’s Law
https://en.wikipedia.org/wiki/Moore%27s_law
Tuesday, July 25, 2023 From State of the Art to Future of AI - Greg Makowski 63
Moore's law is the observation that
the number of transistors in
an integrated circuit (IC) doubles about
every two years.
Moore's law is
an observation and projection of a
historical trend.
Rather than a law of physics, it is
an empirical relationship linked
to gains from experience in production.
Supporting Technologies: Moore’s Law
https://www.man.com/single-core-stagnation-and-the-cloud
Tuesday, July 25, 2023 From State of the Art to Future of AI - Greg Makowski 64
Back in the day, when
Moore's law was still in full
swing, CPU performance
tracked transistor count,
and it made perfect sense
to replace hardware every
two to five years. However,
our own internal
benchmarking shows that
for real life workloads, a
current compute core is
only about 1.6 times
faster than a
comparatively ancient
CPU from 9 years ago.
Sure, newer CPUs have
many more cores, but
given that older hardware
can often be procured at a
large discount, it makes
Supporting Tech to Drive AI:
Cerebras’ wafer-size chip is 10,000 times faster than a GPU
• Cerebras makes the world’s largest computer chip, the WSE. Chipmakers normally slice a wafer from a 12-
inch-diameter ingot of silicon to process in a chip factory. Once processed, the wafer is sliced into hundreds
of separate chips that can be used in electronic hardware.
• But Cerebras, started by SeaMicro founder Andrew Feldman, takes that wafer and makes a single, massive
chip out of it. Each piece of the chip, dubbed a core, is interconnected in a sophisticated way to other cores
• Nov 2020
• https://venturebeat.com/business/cerebras-wafer-size-chip-is-10000-times-faster-than-a-
gpu/#:~:text=Cerebras%20Systems%20and%20the%20federal,graphics%20processing%20unit%20(GPU).
Tuesday, July 25, 2023 From State of the Art to Future of AI - Greg Makowski 65
AI’s rapid growth
people agree we should carefully manage AI
• 91% of respondents agree that "AI is a technology that requires careful
management".
• We recently surveyed over 13,000 people across eleven countries to learn their views on AI. We found
an increasingly strong consensus that AI needs to be managed carefully.
• The portion of people who agree is growing. There has been an 8% increase in agreement in the
United States since 2018 that AI needs to be managed carefully.
• https://www.governance.ai/post/increasing-consensus-ai-requires-careful-management
• I agree, I hope you do as well
• In the future, “vote” by what products you use, encourage your elected officials
• We don’t have to “throw out the baby with the bathwater” – or stop all AI.
• That would not stop BAD ACTORS from continuing to advance technology.
Tuesday, July 25, 2023 From State of the Art to Future of AI - Greg Makowski 66

More Related Content

What's hot

Large Language Models - Chat AI.pdf
Large Language Models - Chat AI.pdfLarge Language Models - Chat AI.pdf
Large Language Models - Chat AI.pdfDavid Rostcheck
 
Exploring Opportunities in the Generative AI Value Chain.pdf
Exploring Opportunities in the Generative AI Value Chain.pdfExploring Opportunities in the Generative AI Value Chain.pdf
Exploring Opportunities in the Generative AI Value Chain.pdfDung Hoang
 
Landscape of AI/ML in 2023
Landscape of AI/ML in 2023Landscape of AI/ML in 2023
Landscape of AI/ML in 2023HyunJoon Jung
 
Leveraging Generative AI & Best practices
Leveraging Generative AI & Best practicesLeveraging Generative AI & Best practices
Leveraging Generative AI & Best practicesDianaGray10
 
Generative AI: Past, Present, and Future – A Practitioner's Perspective
Generative AI: Past, Present, and Future – A Practitioner's PerspectiveGenerative AI: Past, Present, and Future – A Practitioner's Perspective
Generative AI: Past, Present, and Future – A Practitioner's PerspectiveHuahai Yang
 
A Framework for Navigating Generative Artificial Intelligence for Enterprise
A Framework for Navigating Generative Artificial Intelligence for EnterpriseA Framework for Navigating Generative Artificial Intelligence for Enterprise
A Framework for Navigating Generative Artificial Intelligence for EnterpriseRocketSource
 
The Rise of the LLMs - How I Learned to Stop Worrying & Love the GPT!
The Rise of the LLMs - How I Learned to Stop Worrying & Love the GPT!The Rise of the LLMs - How I Learned to Stop Worrying & Love the GPT!
The Rise of the LLMs - How I Learned to Stop Worrying & Love the GPT!taozen
 
Artificial Intelligence and Machine Learning
Artificial Intelligence and Machine LearningArtificial Intelligence and Machine Learning
Artificial Intelligence and Machine LearningMykola Dobrochynskyy
 
Using the power of Generative AI at scale
Using the power of Generative AI at scaleUsing the power of Generative AI at scale
Using the power of Generative AI at scaleMaxim Salnikov
 
Generative-AI-in-enterprise-20230615.pdf
Generative-AI-in-enterprise-20230615.pdfGenerative-AI-in-enterprise-20230615.pdf
Generative-AI-in-enterprise-20230615.pdfLiming Zhu
 
Generative Models and ChatGPT
Generative Models and ChatGPTGenerative Models and ChatGPT
Generative Models and ChatGPTLoic Merckel
 
Generative AI for the rest of us
Generative AI for the rest of usGenerative AI for the rest of us
Generative AI for the rest of usMassimo Ferre'
 

What's hot (20)

Generative AI
Generative AIGenerative AI
Generative AI
 
Large Language Models - Chat AI.pdf
Large Language Models - Chat AI.pdfLarge Language Models - Chat AI.pdf
Large Language Models - Chat AI.pdf
 
Exploring Opportunities in the Generative AI Value Chain.pdf
Exploring Opportunities in the Generative AI Value Chain.pdfExploring Opportunities in the Generative AI Value Chain.pdf
Exploring Opportunities in the Generative AI Value Chain.pdf
 
Landscape of AI/ML in 2023
Landscape of AI/ML in 2023Landscape of AI/ML in 2023
Landscape of AI/ML in 2023
 
Leveraging Generative AI & Best practices
Leveraging Generative AI & Best practicesLeveraging Generative AI & Best practices
Leveraging Generative AI & Best practices
 
Generative AI: Past, Present, and Future – A Practitioner's Perspective
Generative AI: Past, Present, and Future – A Practitioner's PerspectiveGenerative AI: Past, Present, and Future – A Practitioner's Perspective
Generative AI: Past, Present, and Future – A Practitioner's Perspective
 
A Framework for Navigating Generative Artificial Intelligence for Enterprise
A Framework for Navigating Generative Artificial Intelligence for EnterpriseA Framework for Navigating Generative Artificial Intelligence for Enterprise
A Framework for Navigating Generative Artificial Intelligence for Enterprise
 
The Rise of the LLMs - How I Learned to Stop Worrying & Love the GPT!
The Rise of the LLMs - How I Learned to Stop Worrying & Love the GPT!The Rise of the LLMs - How I Learned to Stop Worrying & Love the GPT!
The Rise of the LLMs - How I Learned to Stop Worrying & Love the GPT!
 
Implementing Ethics in AI
Implementing Ethics in AIImplementing Ethics in AI
Implementing Ethics in AI
 
Artificial Intelligence and Machine Learning
Artificial Intelligence and Machine LearningArtificial Intelligence and Machine Learning
Artificial Intelligence and Machine Learning
 
Using the power of Generative AI at scale
Using the power of Generative AI at scaleUsing the power of Generative AI at scale
Using the power of Generative AI at scale
 
LLMs Bootcamp
LLMs BootcampLLMs Bootcamp
LLMs Bootcamp
 
Generative-AI-in-enterprise-20230615.pdf
Generative-AI-in-enterprise-20230615.pdfGenerative-AI-in-enterprise-20230615.pdf
Generative-AI-in-enterprise-20230615.pdf
 
Journey of Generative AI
Journey of Generative AIJourney of Generative AI
Journey of Generative AI
 
The future of AI is hybrid
The future of AI is hybridThe future of AI is hybrid
The future of AI is hybrid
 
AI 2023.pdf
AI 2023.pdfAI 2023.pdf
AI 2023.pdf
 
Ai Ethics
Ai EthicsAi Ethics
Ai Ethics
 
Generative Models and ChatGPT
Generative Models and ChatGPTGenerative Models and ChatGPT
Generative Models and ChatGPT
 
Generative AI for the rest of us
Generative AI for the rest of usGenerative AI for the rest of us
Generative AI for the rest of us
 
Introduction to AI Ethics
Introduction to AI EthicsIntroduction to AI Ethics
Introduction to AI Ethics
 

Similar to Future of AI - 2023 07 25.pptx

AI in Manufacturing: Opportunities & Challenges
AI in Manufacturing: Opportunities & ChallengesAI in Manufacturing: Opportunities & Challenges
AI in Manufacturing: Opportunities & ChallengesTathagat Varma
 
Introduction to LLMs
Introduction to LLMsIntroduction to LLMs
Introduction to LLMsLoic Merckel
 
BSidesLV 2013 - Using Machine Learning to Support Information Security
BSidesLV 2013 - Using Machine Learning to Support Information SecurityBSidesLV 2013 - Using Machine Learning to Support Information Security
BSidesLV 2013 - Using Machine Learning to Support Information SecurityAlex Pinto
 
Future of data science as a profession
Future of data science as a professionFuture of data science as a profession
Future of data science as a professionJose Quesada
 
Présentation de Bruno Schroder au 20e #mforum (07/12/2016)
Présentation de Bruno Schroder au 20e #mforum (07/12/2016)Présentation de Bruno Schroder au 20e #mforum (07/12/2016)
Présentation de Bruno Schroder au 20e #mforum (07/12/2016)Agence du Numérique (AdN)
 
Hacking Predictive Modeling - RoadSec 2018
Hacking Predictive Modeling - RoadSec 2018Hacking Predictive Modeling - RoadSec 2018
Hacking Predictive Modeling - RoadSec 2018HJ van Veen
 
Deciphering AI - Unlocking the Black Box of AIML with State-of-the-Art Techno...
Deciphering AI - Unlocking the Black Box of AIML with State-of-the-Art Techno...Deciphering AI - Unlocking the Black Box of AIML with State-of-the-Art Techno...
Deciphering AI - Unlocking the Black Box of AIML with State-of-the-Art Techno...Analytics India Magazine
 
Big Data & Machine Learning - TDC2013 São Paulo - 12/0713
Big Data & Machine Learning - TDC2013 São Paulo - 12/0713Big Data & Machine Learning - TDC2013 São Paulo - 12/0713
Big Data & Machine Learning - TDC2013 São Paulo - 12/0713Mathieu DESPRIEE
 
Data Science in the Real World: Making a Difference
Data Science in the Real World: Making a Difference Data Science in the Real World: Making a Difference
Data Science in the Real World: Making a Difference Srinath Perera
 
Big Data & Machine Learning - TDC2013 Sao Paulo
Big Data & Machine Learning - TDC2013 Sao PauloBig Data & Machine Learning - TDC2013 Sao Paulo
Big Data & Machine Learning - TDC2013 Sao PauloOCTO Technology
 
Ethical AI - Open Compliance Summit 2020
Ethical AI - Open Compliance Summit 2020Ethical AI - Open Compliance Summit 2020
Ethical AI - Open Compliance Summit 2020Debmalya Biswas
 
Ntegra 20180523 v10 copy.pptx
Ntegra 20180523 v10 copy.pptxNtegra 20180523 v10 copy.pptx
Ntegra 20180523 v10 copy.pptxISSIP
 
Algorithm Marketplace and the new "Algorithm Economy"
Algorithm Marketplace and the new "Algorithm Economy"Algorithm Marketplace and the new "Algorithm Economy"
Algorithm Marketplace and the new "Algorithm Economy"Diego Oppenheimer
 
Intelligent Big Data analytics for the future.
Intelligent Big Data analytics for the future.Intelligent Big Data analytics for the future.
Intelligent Big Data analytics for the future.Shashank Garg
 
EDW 2015 cognitive computing panel session
EDW 2015 cognitive computing panel session EDW 2015 cognitive computing panel session
EDW 2015 cognitive computing panel session Steve Ardire
 
Applied AI Workshop - Presentation - Connect Day GDL
Applied AI Workshop - Presentation - Connect Day GDLApplied AI Workshop - Presentation - Connect Day GDL
Applied AI Workshop - Presentation - Connect Day GDLMarc Teunis
 
The Incredible Disappearing Data Scientist
The Incredible Disappearing Data ScientistThe Incredible Disappearing Data Scientist
The Incredible Disappearing Data ScientistRebecca Bilbro
 
AI Basics for Professionals to Help Begin Their AI Journey
AI Basics for Professionals to Help Begin Their AI JourneyAI Basics for Professionals to Help Begin Their AI Journey
AI Basics for Professionals to Help Begin Their AI JourneyDeepak Sharma
 

Similar to Future of AI - 2023 07 25.pptx (20)

AI in Manufacturing: Opportunities & Challenges
AI in Manufacturing: Opportunities & ChallengesAI in Manufacturing: Opportunities & Challenges
AI in Manufacturing: Opportunities & Challenges
 
Introduction to LLMs
Introduction to LLMsIntroduction to LLMs
Introduction to LLMs
 
BSidesLV 2013 - Using Machine Learning to Support Information Security
BSidesLV 2013 - Using Machine Learning to Support Information SecurityBSidesLV 2013 - Using Machine Learning to Support Information Security
BSidesLV 2013 - Using Machine Learning to Support Information Security
 
Future of data science as a profession
Future of data science as a professionFuture of data science as a profession
Future of data science as a profession
 
Présentation de Bruno Schroder au 20e #mforum (07/12/2016)
Présentation de Bruno Schroder au 20e #mforum (07/12/2016)Présentation de Bruno Schroder au 20e #mforum (07/12/2016)
Présentation de Bruno Schroder au 20e #mforum (07/12/2016)
 
Hacking Predictive Modeling - RoadSec 2018
Hacking Predictive Modeling - RoadSec 2018Hacking Predictive Modeling - RoadSec 2018
Hacking Predictive Modeling - RoadSec 2018
 
Deciphering AI - Unlocking the Black Box of AIML with State-of-the-Art Techno...
Deciphering AI - Unlocking the Black Box of AIML with State-of-the-Art Techno...Deciphering AI - Unlocking the Black Box of AIML with State-of-the-Art Techno...
Deciphering AI - Unlocking the Black Box of AIML with State-of-the-Art Techno...
 
Big Data & Machine Learning - TDC2013 São Paulo - 12/0713
Big Data & Machine Learning - TDC2013 São Paulo - 12/0713Big Data & Machine Learning - TDC2013 São Paulo - 12/0713
Big Data & Machine Learning - TDC2013 São Paulo - 12/0713
 
20181212 ibm aot
20181212 ibm aot20181212 ibm aot
20181212 ibm aot
 
Data Science in the Real World: Making a Difference
Data Science in the Real World: Making a Difference Data Science in the Real World: Making a Difference
Data Science in the Real World: Making a Difference
 
Big Data & Machine Learning - TDC2013 Sao Paulo
Big Data & Machine Learning - TDC2013 Sao PauloBig Data & Machine Learning - TDC2013 Sao Paulo
Big Data & Machine Learning - TDC2013 Sao Paulo
 
Ethical AI - Open Compliance Summit 2020
Ethical AI - Open Compliance Summit 2020Ethical AI - Open Compliance Summit 2020
Ethical AI - Open Compliance Summit 2020
 
Ntegra 20180523 v10 copy.pptx
Ntegra 20180523 v10 copy.pptxNtegra 20180523 v10 copy.pptx
Ntegra 20180523 v10 copy.pptx
 
Algorithm Marketplace and the new "Algorithm Economy"
Algorithm Marketplace and the new "Algorithm Economy"Algorithm Marketplace and the new "Algorithm Economy"
Algorithm Marketplace and the new "Algorithm Economy"
 
Intelligent Big Data analytics for the future.
Intelligent Big Data analytics for the future.Intelligent Big Data analytics for the future.
Intelligent Big Data analytics for the future.
 
EDW 2015 cognitive computing panel session
EDW 2015 cognitive computing panel session EDW 2015 cognitive computing panel session
EDW 2015 cognitive computing panel session
 
Applied AI Workshop - Presentation - Connect Day GDL
Applied AI Workshop - Presentation - Connect Day GDLApplied AI Workshop - Presentation - Connect Day GDL
Applied AI Workshop - Presentation - Connect Day GDL
 
The Incredible Disappearing Data Scientist
The Incredible Disappearing Data ScientistThe Incredible Disappearing Data Scientist
The Incredible Disappearing Data Scientist
 
Intro to LLMs
Intro to LLMsIntro to LLMs
Intro to LLMs
 
AI Basics for Professionals to Help Begin Their AI Journey
AI Basics for Professionals to Help Begin Their AI JourneyAI Basics for Professionals to Help Begin Their AI Journey
AI Basics for Professionals to Help Begin Their AI Journey
 

More from Greg Makowski

Understanding Hallucinations in LLMs - 2023 09 29.pptx
Understanding Hallucinations in LLMs - 2023 09 29.pptxUnderstanding Hallucinations in LLMs - 2023 09 29.pptx
Understanding Hallucinations in LLMs - 2023 09 29.pptxGreg Makowski
 
A Successful Hiring Process for Data Scientists
A Successful Hiring Process for Data ScientistsA Successful Hiring Process for Data Scientists
A Successful Hiring Process for Data ScientistsGreg Makowski
 
Kdd 2019: Standardizing Data Science to Help Hiring
Kdd 2019:  Standardizing Data Science to Help HiringKdd 2019:  Standardizing Data Science to Help Hiring
Kdd 2019: Standardizing Data Science to Help HiringGreg Makowski
 
Tales from an ip worker in consulting and software
Tales from an ip worker in consulting and softwareTales from an ip worker in consulting and software
Tales from an ip worker in consulting and softwareGreg Makowski
 
Predictive Model and Record Description with Segmented Sensitivity Analysis (...
Predictive Model and Record Description with Segmented Sensitivity Analysis (...Predictive Model and Record Description with Segmented Sensitivity Analysis (...
Predictive Model and Record Description with Segmented Sensitivity Analysis (...Greg Makowski
 
Production model lifecycle management 2016 09
Production model lifecycle management 2016 09Production model lifecycle management 2016 09
Production model lifecycle management 2016 09Greg Makowski
 
Using Deep Learning to do Real-Time Scoring in Practical Applications
Using Deep Learning to do Real-Time Scoring in Practical ApplicationsUsing Deep Learning to do Real-Time Scoring in Practical Applications
Using Deep Learning to do Real-Time Scoring in Practical ApplicationsGreg Makowski
 
Using Deep Learning to do Real-Time Scoring in Practical Applications - 2015-...
Using Deep Learning to do Real-Time Scoring in Practical Applications - 2015-...Using Deep Learning to do Real-Time Scoring in Practical Applications - 2015-...
Using Deep Learning to do Real-Time Scoring in Practical Applications - 2015-...Greg Makowski
 
SFbayACM ACM Data Science Camp 2015 10 24
SFbayACM ACM Data Science Camp 2015 10 24SFbayACM ACM Data Science Camp 2015 10 24
SFbayACM ACM Data Science Camp 2015 10 24Greg Makowski
 
How to Create 80% of a Big Data Pilot Project
How to Create 80% of a Big Data Pilot ProjectHow to Create 80% of a Big Data Pilot Project
How to Create 80% of a Big Data Pilot ProjectGreg Makowski
 
Powering Real­time Decision Engines in Finance and Healthcare using Open Sour...
Powering Real­time Decision Engines in Finance and Healthcare using Open Sour...Powering Real­time Decision Engines in Finance and Healthcare using Open Sour...
Powering Real­time Decision Engines in Finance and Healthcare using Open Sour...Greg Makowski
 
Kamanja: Driving Business Value through Real-Time Decisioning Solutions
Kamanja: Driving Business Value through Real-Time Decisioning SolutionsKamanja: Driving Business Value through Real-Time Decisioning Solutions
Kamanja: Driving Business Value through Real-Time Decisioning SolutionsGreg Makowski
 
Heuristic design of experiments w meta gradient search
Heuristic design of experiments w meta gradient searchHeuristic design of experiments w meta gradient search
Heuristic design of experiments w meta gradient searchGreg Makowski
 
Three case studies deploying cluster analysis
Three case studies deploying cluster analysisThree case studies deploying cluster analysis
Three case studies deploying cluster analysisGreg Makowski
 
Linked In Slides 2009 02 24 B
Linked In Slides 2009 02 24 BLinked In Slides 2009 02 24 B
Linked In Slides 2009 02 24 BGreg Makowski
 
The 360º Leader (Section 2 of 6)
The 360º Leader (Section 2 of 6)The 360º Leader (Section 2 of 6)
The 360º Leader (Section 2 of 6)Greg Makowski
 
The 360º Leader (Section 1 of 6)
The 360º Leader (Section 1 of 6)The 360º Leader (Section 1 of 6)
The 360º Leader (Section 1 of 6)Greg Makowski
 

More from Greg Makowski (17)

Understanding Hallucinations in LLMs - 2023 09 29.pptx
Understanding Hallucinations in LLMs - 2023 09 29.pptxUnderstanding Hallucinations in LLMs - 2023 09 29.pptx
Understanding Hallucinations in LLMs - 2023 09 29.pptx
 
A Successful Hiring Process for Data Scientists
A Successful Hiring Process for Data ScientistsA Successful Hiring Process for Data Scientists
A Successful Hiring Process for Data Scientists
 
Kdd 2019: Standardizing Data Science to Help Hiring
Kdd 2019:  Standardizing Data Science to Help HiringKdd 2019:  Standardizing Data Science to Help Hiring
Kdd 2019: Standardizing Data Science to Help Hiring
 
Tales from an ip worker in consulting and software
Tales from an ip worker in consulting and softwareTales from an ip worker in consulting and software
Tales from an ip worker in consulting and software
 
Predictive Model and Record Description with Segmented Sensitivity Analysis (...
Predictive Model and Record Description with Segmented Sensitivity Analysis (...Predictive Model and Record Description with Segmented Sensitivity Analysis (...
Predictive Model and Record Description with Segmented Sensitivity Analysis (...
 
Production model lifecycle management 2016 09
Production model lifecycle management 2016 09Production model lifecycle management 2016 09
Production model lifecycle management 2016 09
 
Using Deep Learning to do Real-Time Scoring in Practical Applications
Using Deep Learning to do Real-Time Scoring in Practical ApplicationsUsing Deep Learning to do Real-Time Scoring in Practical Applications
Using Deep Learning to do Real-Time Scoring in Practical Applications
 
Using Deep Learning to do Real-Time Scoring in Practical Applications - 2015-...
Using Deep Learning to do Real-Time Scoring in Practical Applications - 2015-...Using Deep Learning to do Real-Time Scoring in Practical Applications - 2015-...
Using Deep Learning to do Real-Time Scoring in Practical Applications - 2015-...
 
SFbayACM ACM Data Science Camp 2015 10 24
SFbayACM ACM Data Science Camp 2015 10 24SFbayACM ACM Data Science Camp 2015 10 24
SFbayACM ACM Data Science Camp 2015 10 24
 
How to Create 80% of a Big Data Pilot Project
How to Create 80% of a Big Data Pilot ProjectHow to Create 80% of a Big Data Pilot Project
How to Create 80% of a Big Data Pilot Project
 
Powering Real­time Decision Engines in Finance and Healthcare using Open Sour...
Powering Real­time Decision Engines in Finance and Healthcare using Open Sour...Powering Real­time Decision Engines in Finance and Healthcare using Open Sour...
Powering Real­time Decision Engines in Finance and Healthcare using Open Sour...
 
Kamanja: Driving Business Value through Real-Time Decisioning Solutions
Kamanja: Driving Business Value through Real-Time Decisioning SolutionsKamanja: Driving Business Value through Real-Time Decisioning Solutions
Kamanja: Driving Business Value through Real-Time Decisioning Solutions
 
Heuristic design of experiments w meta gradient search
Heuristic design of experiments w meta gradient searchHeuristic design of experiments w meta gradient search
Heuristic design of experiments w meta gradient search
 
Three case studies deploying cluster analysis
Three case studies deploying cluster analysisThree case studies deploying cluster analysis
Three case studies deploying cluster analysis
 
Linked In Slides 2009 02 24 B
Linked In Slides 2009 02 24 BLinked In Slides 2009 02 24 B
Linked In Slides 2009 02 24 B
 
The 360º Leader (Section 2 of 6)
The 360º Leader (Section 2 of 6)The 360º Leader (Section 2 of 6)
The 360º Leader (Section 2 of 6)
 
The 360º Leader (Section 1 of 6)
The 360º Leader (Section 1 of 6)The 360º Leader (Section 1 of 6)
The 360º Leader (Section 1 of 6)
 

Recently uploaded

Kantar AI Summit- Under Embargo till Wednesday, 24th April 2024, 4 PM, IST.pdf
Kantar AI Summit- Under Embargo till Wednesday, 24th April 2024, 4 PM, IST.pdfKantar AI Summit- Under Embargo till Wednesday, 24th April 2024, 4 PM, IST.pdf
Kantar AI Summit- Under Embargo till Wednesday, 24th April 2024, 4 PM, IST.pdfSocial Samosa
 
Heart Disease Classification Report: A Data Analysis Project
Heart Disease Classification Report: A Data Analysis ProjectHeart Disease Classification Report: A Data Analysis Project
Heart Disease Classification Report: A Data Analysis ProjectBoston Institute of Analytics
 
办理学位证纽约大学毕业证(NYU毕业证书)原版一比一
办理学位证纽约大学毕业证(NYU毕业证书)原版一比一办理学位证纽约大学毕业证(NYU毕业证书)原版一比一
办理学位证纽约大学毕业证(NYU毕业证书)原版一比一fhwihughh
 
dokumen.tips_chapter-4-transient-heat-conduction-mehmet-kanoglu.ppt
dokumen.tips_chapter-4-transient-heat-conduction-mehmet-kanoglu.pptdokumen.tips_chapter-4-transient-heat-conduction-mehmet-kanoglu.ppt
dokumen.tips_chapter-4-transient-heat-conduction-mehmet-kanoglu.pptSonatrach
 
GA4 Without Cookies [Measure Camp AMS]
GA4 Without Cookies [Measure Camp AMS]GA4 Without Cookies [Measure Camp AMS]
GA4 Without Cookies [Measure Camp AMS]📊 Markus Baersch
 
9711147426✨Call In girls Gurgaon Sector 31. SCO 25 escort service
9711147426✨Call In girls Gurgaon Sector 31. SCO 25 escort service9711147426✨Call In girls Gurgaon Sector 31. SCO 25 escort service
9711147426✨Call In girls Gurgaon Sector 31. SCO 25 escort servicejennyeacort
 
PKS-TGC-1084-630 - Stage 1 Proposal.pptx
PKS-TGC-1084-630 - Stage 1 Proposal.pptxPKS-TGC-1084-630 - Stage 1 Proposal.pptx
PKS-TGC-1084-630 - Stage 1 Proposal.pptxPramod Kumar Srivastava
 
RABBIT: A CLI tool for identifying bots based on their GitHub events.
RABBIT: A CLI tool for identifying bots based on their GitHub events.RABBIT: A CLI tool for identifying bots based on their GitHub events.
RABBIT: A CLI tool for identifying bots based on their GitHub events.natarajan8993
 
Customer Service Analytics - Make Sense of All Your Data.pptx
Customer Service Analytics - Make Sense of All Your Data.pptxCustomer Service Analytics - Make Sense of All Your Data.pptx
Customer Service Analytics - Make Sense of All Your Data.pptxEmmanuel Dauda
 
Top 5 Best Data Analytics Courses In Queens
Top 5 Best Data Analytics Courses In QueensTop 5 Best Data Analytics Courses In Queens
Top 5 Best Data Analytics Courses In Queensdataanalyticsqueen03
 
Defining Constituents, Data Vizzes and Telling a Data Story
Defining Constituents, Data Vizzes and Telling a Data StoryDefining Constituents, Data Vizzes and Telling a Data Story
Defining Constituents, Data Vizzes and Telling a Data StoryJeremy Anderson
 
EMERCE - 2024 - AMSTERDAM - CROSS-PLATFORM TRACKING WITH GOOGLE ANALYTICS.pptx
EMERCE - 2024 - AMSTERDAM - CROSS-PLATFORM  TRACKING WITH GOOGLE ANALYTICS.pptxEMERCE - 2024 - AMSTERDAM - CROSS-PLATFORM  TRACKING WITH GOOGLE ANALYTICS.pptx
EMERCE - 2024 - AMSTERDAM - CROSS-PLATFORM TRACKING WITH GOOGLE ANALYTICS.pptxthyngster
 
Consent & Privacy Signals on Google *Pixels* - MeasureCamp Amsterdam 2024
Consent & Privacy Signals on Google *Pixels* - MeasureCamp Amsterdam 2024Consent & Privacy Signals on Google *Pixels* - MeasureCamp Amsterdam 2024
Consent & Privacy Signals on Google *Pixels* - MeasureCamp Amsterdam 2024thyngster
 
Identifying Appropriate Test Statistics Involving Population Mean
Identifying Appropriate Test Statistics Involving Population MeanIdentifying Appropriate Test Statistics Involving Population Mean
Identifying Appropriate Test Statistics Involving Population MeanMYRABACSAFRA2
 
High Class Call Girls Noida Sector 39 Aarushi 🔝8264348440🔝 Independent Escort...
High Class Call Girls Noida Sector 39 Aarushi 🔝8264348440🔝 Independent Escort...High Class Call Girls Noida Sector 39 Aarushi 🔝8264348440🔝 Independent Escort...
High Class Call Girls Noida Sector 39 Aarushi 🔝8264348440🔝 Independent Escort...soniya singh
 
专业一比一美国俄亥俄大学毕业证成绩单pdf电子版制作修改
专业一比一美国俄亥俄大学毕业证成绩单pdf电子版制作修改专业一比一美国俄亥俄大学毕业证成绩单pdf电子版制作修改
专业一比一美国俄亥俄大学毕业证成绩单pdf电子版制作修改yuu sss
 
原版1:1定制南十字星大学毕业证(SCU毕业证)#文凭成绩单#真实留信学历认证永久存档
原版1:1定制南十字星大学毕业证(SCU毕业证)#文凭成绩单#真实留信学历认证永久存档原版1:1定制南十字星大学毕业证(SCU毕业证)#文凭成绩单#真实留信学历认证永久存档
原版1:1定制南十字星大学毕业证(SCU毕业证)#文凭成绩单#真实留信学历认证永久存档208367051
 
RS 9000 Call In girls Dwarka Mor (DELHI)⇛9711147426🔝Delhi
RS 9000 Call In girls Dwarka Mor (DELHI)⇛9711147426🔝DelhiRS 9000 Call In girls Dwarka Mor (DELHI)⇛9711147426🔝Delhi
RS 9000 Call In girls Dwarka Mor (DELHI)⇛9711147426🔝Delhijennyeacort
 
Effects of Smartphone Addiction on the Academic Performances of Grades 9 to 1...
Effects of Smartphone Addiction on the Academic Performances of Grades 9 to 1...Effects of Smartphone Addiction on the Academic Performances of Grades 9 to 1...
Effects of Smartphone Addiction on the Academic Performances of Grades 9 to 1...limedy534
 

Recently uploaded (20)

Kantar AI Summit- Under Embargo till Wednesday, 24th April 2024, 4 PM, IST.pdf
Kantar AI Summit- Under Embargo till Wednesday, 24th April 2024, 4 PM, IST.pdfKantar AI Summit- Under Embargo till Wednesday, 24th April 2024, 4 PM, IST.pdf
Kantar AI Summit- Under Embargo till Wednesday, 24th April 2024, 4 PM, IST.pdf
 
Heart Disease Classification Report: A Data Analysis Project
Heart Disease Classification Report: A Data Analysis ProjectHeart Disease Classification Report: A Data Analysis Project
Heart Disease Classification Report: A Data Analysis Project
 
办理学位证纽约大学毕业证(NYU毕业证书)原版一比一
办理学位证纽约大学毕业证(NYU毕业证书)原版一比一办理学位证纽约大学毕业证(NYU毕业证书)原版一比一
办理学位证纽约大学毕业证(NYU毕业证书)原版一比一
 
Deep Generative Learning for All - The Gen AI Hype (Spring 2024)
Deep Generative Learning for All - The Gen AI Hype (Spring 2024)Deep Generative Learning for All - The Gen AI Hype (Spring 2024)
Deep Generative Learning for All - The Gen AI Hype (Spring 2024)
 
dokumen.tips_chapter-4-transient-heat-conduction-mehmet-kanoglu.ppt
dokumen.tips_chapter-4-transient-heat-conduction-mehmet-kanoglu.pptdokumen.tips_chapter-4-transient-heat-conduction-mehmet-kanoglu.ppt
dokumen.tips_chapter-4-transient-heat-conduction-mehmet-kanoglu.ppt
 
GA4 Without Cookies [Measure Camp AMS]
GA4 Without Cookies [Measure Camp AMS]GA4 Without Cookies [Measure Camp AMS]
GA4 Without Cookies [Measure Camp AMS]
 
9711147426✨Call In girls Gurgaon Sector 31. SCO 25 escort service
9711147426✨Call In girls Gurgaon Sector 31. SCO 25 escort service9711147426✨Call In girls Gurgaon Sector 31. SCO 25 escort service
9711147426✨Call In girls Gurgaon Sector 31. SCO 25 escort service
 
PKS-TGC-1084-630 - Stage 1 Proposal.pptx
PKS-TGC-1084-630 - Stage 1 Proposal.pptxPKS-TGC-1084-630 - Stage 1 Proposal.pptx
PKS-TGC-1084-630 - Stage 1 Proposal.pptx
 
RABBIT: A CLI tool for identifying bots based on their GitHub events.
RABBIT: A CLI tool for identifying bots based on their GitHub events.RABBIT: A CLI tool for identifying bots based on their GitHub events.
RABBIT: A CLI tool for identifying bots based on their GitHub events.
 
Customer Service Analytics - Make Sense of All Your Data.pptx
Customer Service Analytics - Make Sense of All Your Data.pptxCustomer Service Analytics - Make Sense of All Your Data.pptx
Customer Service Analytics - Make Sense of All Your Data.pptx
 
Top 5 Best Data Analytics Courses In Queens
Top 5 Best Data Analytics Courses In QueensTop 5 Best Data Analytics Courses In Queens
Top 5 Best Data Analytics Courses In Queens
 
Defining Constituents, Data Vizzes and Telling a Data Story
Defining Constituents, Data Vizzes and Telling a Data StoryDefining Constituents, Data Vizzes and Telling a Data Story
Defining Constituents, Data Vizzes and Telling a Data Story
 
EMERCE - 2024 - AMSTERDAM - CROSS-PLATFORM TRACKING WITH GOOGLE ANALYTICS.pptx
EMERCE - 2024 - AMSTERDAM - CROSS-PLATFORM  TRACKING WITH GOOGLE ANALYTICS.pptxEMERCE - 2024 - AMSTERDAM - CROSS-PLATFORM  TRACKING WITH GOOGLE ANALYTICS.pptx
EMERCE - 2024 - AMSTERDAM - CROSS-PLATFORM TRACKING WITH GOOGLE ANALYTICS.pptx
 
Consent & Privacy Signals on Google *Pixels* - MeasureCamp Amsterdam 2024
Consent & Privacy Signals on Google *Pixels* - MeasureCamp Amsterdam 2024Consent & Privacy Signals on Google *Pixels* - MeasureCamp Amsterdam 2024
Consent & Privacy Signals on Google *Pixels* - MeasureCamp Amsterdam 2024
 
Identifying Appropriate Test Statistics Involving Population Mean
Identifying Appropriate Test Statistics Involving Population MeanIdentifying Appropriate Test Statistics Involving Population Mean
Identifying Appropriate Test Statistics Involving Population Mean
 
High Class Call Girls Noida Sector 39 Aarushi 🔝8264348440🔝 Independent Escort...
High Class Call Girls Noida Sector 39 Aarushi 🔝8264348440🔝 Independent Escort...High Class Call Girls Noida Sector 39 Aarushi 🔝8264348440🔝 Independent Escort...
High Class Call Girls Noida Sector 39 Aarushi 🔝8264348440🔝 Independent Escort...
 
专业一比一美国俄亥俄大学毕业证成绩单pdf电子版制作修改
专业一比一美国俄亥俄大学毕业证成绩单pdf电子版制作修改专业一比一美国俄亥俄大学毕业证成绩单pdf电子版制作修改
专业一比一美国俄亥俄大学毕业证成绩单pdf电子版制作修改
 
原版1:1定制南十字星大学毕业证(SCU毕业证)#文凭成绩单#真实留信学历认证永久存档
原版1:1定制南十字星大学毕业证(SCU毕业证)#文凭成绩单#真实留信学历认证永久存档原版1:1定制南十字星大学毕业证(SCU毕业证)#文凭成绩单#真实留信学历认证永久存档
原版1:1定制南十字星大学毕业证(SCU毕业证)#文凭成绩单#真实留信学历认证永久存档
 
RS 9000 Call In girls Dwarka Mor (DELHI)⇛9711147426🔝Delhi
RS 9000 Call In girls Dwarka Mor (DELHI)⇛9711147426🔝DelhiRS 9000 Call In girls Dwarka Mor (DELHI)⇛9711147426🔝Delhi
RS 9000 Call In girls Dwarka Mor (DELHI)⇛9711147426🔝Delhi
 
Effects of Smartphone Addiction on the Academic Performances of Grades 9 to 1...
Effects of Smartphone Addiction on the Academic Performances of Grades 9 to 1...Effects of Smartphone Addiction on the Academic Performances of Grades 9 to 1...
Effects of Smartphone Addiction on the Academic Performances of Grades 9 to 1...
 

Future of AI - 2023 07 25.pptx

  • 1. Going from the State of the Art to the Future of LLMs and Generative AI Greg Makowski Tuesday, July 25, 2023 SF bay ACM Meetup presentation https://www.meetup.com/sf-bay-acm/events/294420890/ Meetup Announcement https://www.youtube.com/user/sfbayacm Video recording https://www.slideshare.net/gregmakowski Slides www.LinkedIn.com/in/GregMakowski Network
  • 2. Greg Makowski Tuesday, July 25, 2023 From State of the Art to Future of AI - Greg Makowski 2 • Goal since high school was “Applied Science Fiction” • Deploying Data Science and Artificial Intelligence since 1992 • Worked for American Express, then 6 startups • Been through 4 acquisitions, or startup exits • Travelled to 18 countries • Deployed ~96 models for clients • Growing DS teams since 2010 • Applied for 9 DS patents since Jan 2022 • Vertical Markets • Targeted Marketing (mail, phone, banner, ,..) • Finance – business to consumer (CC, check, web,..) • Finance – business to business (i.e. bond pricing) • Fraud detection (insurance, checking, bank,..) • Security (insider threat, account takeover, botnet, foreign program detection) • Internet of Things (manufacturing, oil & gas, automotive, HVAC, building energy use,…)
  • 3. Tuesday, July 25, 2023 From State of the Art to Future of AI - Greg Makowski 3 A broad, quick overview of many topics with pointers to more reading Go through slides quicker, and can go into detail for questions Don’t worry about soaking up every detail… AI Progress in the PAST AI State of the Art NOW Looking to the FUTURE of AI Agenda https://www.vintag.es/2022/02/blown-away-
  • 4. AI Progress in the PAST A 20,000-foot view of the forest Tuesday, July 25, 2023 From State of the Art to Future of AI - Greg Makowski 4
  • 5. Types of AI Algorithms (a toolkit) “narrow AI” (vs. Artificial General Intelligence (AGI) or Super Intelligence) • Supervised learning / classification / prediction – have known outcomes, labelled examples of good/bad • Time Series – like stock market analysis. Train with to predict “known future data” • Sensor analysis – detect people, things or attributes in an image (or sound or other sensors) • Challenge: tough or expensive to scale. May pay $1k to $30k to label images for training a vision model. • Unsupervised Learning / clustering – does not depend on labelled examples, is more scalable • Reinforcement Learning – gets a feedback from an action. Won “Go” challenge, robotics, games • Optimization / linear programming / non-linear programming – like pricing of an airline seat or hotel room before the scheduled date goes by, (revenue management). Used as a part in many algs. • Symbolic AI – logic, if-then-else rules, reasoning, causality. Very hard to scale to large data and corner cases, good at the “central truth”. Can be a part of many AI applications. • Bayesian Belief Networks – like a mixture of many of the above, used for product configuration, diagnosis. Judea Pearl & “Causality”, “The Book of Why” • Graphical Models – social networks, LinkedIn, call metadata analysis, crime or terrorist communications Tuesday, July 25, 2023 From State of the Art to Future of AI - Greg Makowski 5
  • 6. 1800 1825 1850 1875 1900 1925 1950 1975 2000 2025 1 10 100 1,000 10,000 100,000 1,000,000 10,000,000 100,000,000 1,000,000,000 10,000,000,000 100,000,000,000 1,000,000,000,000 Weights Year Growth in Complexity over Time Size isn’t the only thing – but one metric or perspective Tuesday, July 25, 2023 6 https://en.wikipedia.org/wiki/List_of_animals_by_number_of_neurons# Big changes is structure, NOT just size Neurons are not the only metric Also, number of connections in a brain matters My work has spanned this 1013 growth as “normal evolution” Algorithm Weights Year ChatGPT 4.0 1,000,000,000,000 2023 ChatGPT 3.5 175,000,000,000 2022 human brain 86,000,000,000 blue whale 15,000,000,000 ChatGPT 2.0 1,500,000,000 2019 BERT 100,000,000 2018 house mouse 71,000,000 LSTM 20,000,000 1997 guppy 4,300,000 backprop NN 500,000 1988 fruit fly 150,000 ARIMA 180 1970 regression 60 1805
  • 7. Time Series, From Regression to Large Language Models • Regression (1805) – commonly have 10 to 60 inputs and a similar number of “weights” to optimize or train • ARIMA (Auto Regressive Integrated Moving Average) 1970s – inputs are past time elements and average • Backpropagation – basic neural net (~1970s) popular in 1988, usually fully connected between layers, 500 to 500k weights • RNN (Recurrent Neural Networks)1982 – output time elements also fed as inputs. Does not work as well on longer time series. Has a “vanishing gradient” problem. • LSTM (Long Short Term Memory) 1997 – added “memory state”. May have ~20 mm weights • Word representations in Vector space 2013 – Word2Vec, used as each element for Natural Language Processing (NLP) time series models. Embeddings are used to represent words, images, faces, speakers,… • Attention Networks for speech & language – 2017 (transformers) • BERT (2018) – Bidirectional Transformers for NLP (~100 m weights) • ChatGPT 3.5 – 175 B weights (architecture not published) • ChatGPT 4.0 – 1,000 B weights Tuesday, July 25, 2023 From State of the Art to Future of AI - Greg Makowski 7
  • 8. How LLM is a time series • Good at predicting the most popular or frequent sequences • Only as good as the volume and variety of the training data • Diagram shows given the “start state” of the prior word is “I”, the next word is most frequently “went” • PROMPT ENGINEERING in current LLMs Tuesday, July 25, 2023 From State of the Art to Future of AI - Greg Makowski 8 https://towardsdatascience.com/deconstructing-bert-distilling-6-patterns-from-100-million-parameters-b49113672f77#:~:text=BERT%20is%20a%20(multi%2Dheaded)%20beast&text=Since%20model%20weights%20are%20not,16%20%3D%20384%20different%20attention%20mechanisms Later, hallucinations are just “next probable text”
  • 9. Embedding – to encode words, and … Word2Vec, 2013 paper • before, 1 word = 1 col or dimension • 50,000 words requires 50,000 dimensions • After, 1 word = 1 point in 256 dimensions • 50,000 words fits in 256 (or N) dimensions (FEWER) • Different location for Homonyms based on how they are used • Bank (financial bank, river bank, bank a plane to turn) • Embedding spaces are used for • Words, paragraphs • Face recognition • Speaker recognition • Social Networks Tuesday, July 25, 2023 From State of the Art to Future of AI - Greg Makowski 9 https://arxiv.org/pdf/1310.4546.pdf Distributed Representations of Words and Phrases and their Compositionality https://code.google.com/archive/p/word2vec/ https://towardsdatascience.com/what-is-embedding-and-what-can-you-do-with-it-61ba7c05efd8
  • 10. Embedding – close neighbors • Local, very close neighbors in the embedding space, are • Related meanings • Used in similar conversations Tuesday, July 25, 2023 From State of the Art to Future of AI - Greg Makowski 10 https://serokell.io/blog/word2vec
  • 11. Embedding – to encode concepts • In the chart on the right, book embeddings • Non-Fiction – top right, green • Science Fiction – left, blue • Fiction – Orange, middle to lower right • Later, talk about PROMPT ENGINEERING • This sets the “starting place” to run the time series dialog with the LLM, i.e. start: • Non-fiction • Respectful • HALLUCINATIONS • Between existing training points, which all may be factual • Remember, moving in space with Billions of weights • A new location is some interpolation of surrounding concepts • NOT LIMITED TO TEXT, can be images Tuesday, July 25, 2023 From State of the Art to Future of AI - Greg Makowski 11 https://towardsdatascience.com/neural-network-embeddings-explained-4d028e6f0526 T-Distributed Stochastic Neighbor Embedding (TSNE)
  • 12. Caution: a model is no better than its training data – can get racial (or other bias) • Gender and racial bias found in Amazon’s facial recognition technology (again) – The Verge, Jan 19, 2019 • Rekognition made no mistakes when identifying the gender of lighter-skinned men, but it mistook women for men 19 percent of the time and mistook darker-skinned women for men 31 percent of the time. • A test last year conducted by the ACLU found that while scanning pictures of members of Congress, Rekognition falsely matched 28 individuals with police mugshots. • Facebook Apologizes After A.I. Puts ‘Primates’ Label on Video of Black Men – New York Times, Sept 3, 2021 • The video, dated June 27, 2020, was by The Daily Mail and featured clips of Black men in altercations with white civilians and police officers. It had no connection to monkeys or primates. • In one example in 2015, Google Photos mistakenly labeled pictures of Black people as “gorillas,” for which Google said it was “genuinely sorry” and would work to fix the issue immediately. More than two years later, Wired found that Google’s solution was to censor the word “gorilla” from searches, while also blocking “chimp,” “chimpanzee” and “monkey.” Tuesday, July 25, 2023 From State of the Art to Future of AI - Greg Makowski 12
  • 13. Emergent Properties New Features that Show Up, that You Didn’t Design for! • Wikipedia – Emergent Algorithm • https://en.wikipedia.org/wiki/Emergent_algorithm • An emergent algorithm is an algorithm that exhibits emergent behavior. In essence an emergent algorithm implements a set of simple building block behaviors that when combined exhibit more complex behaviors. One example of this is the implementation of fuzzy motion controllers used to adapt robot movement in response to environmental obstacles.[1] • An emergent algorithm has the following characteristics: • it achieves predictable global effects • it does not require global visibility • it does not assume any kind of centralized control • it is self-stabilizing Tuesday, July 25, 2023 From State of the Art to Future of AI - Greg Makowski 13 • Simple example from a neural net in the early 1990’s • Person trained the NN to do visual processing, intended task. • The neural net was fooled by the same kind of optical illusion that a person would be – which point in the wire frame in a cube is in front? Oscillate between the two on which is in front • Boston University, Cognitive and Neural Systems (CNS) class Front Front
  • 14. AI State of the Art NOW Tuesday, July 25, 2023 14 Need to know what is already happening today - so we don’t predict it for tomorrow
  • 15. Tuesday, July 25, 2023 From State of the Art to Future of AI - Greg Makowski 15 https://marketoonist.com/2023/06/impact-of-chatgpt.html https://marketoonist.com/2023/06/impact-of-chatgpt.html
  • 16. Places to find State of the Art (SOTA) Breadth of AI subjects • www.PapersWithCode.com – for a given analysis type, compares papers on the same data set, link to code and data • Computer Vision (1,223 tasks, like these) • Image Classification – the image has a “cat” as the primary subject • Object Detection – a rectangle bounding box around the outside of each subject • Semantic Segmentation – a polygon around all examples of something, “people”, “vehicles” • Instance Segmentation – a polygon around each instance of something, “person 1”, “person 2”, … • Image Generation – from text, generate images • Natural Language Processing (699 tasks) • Language modeling • Question answering • Machine translation • Text generation • Sentiment analysis • Medical • Time Series Tuesday, July 25, 2023 From State of the Art to Future of AI - Greg Makowski 16 • Graphs • Speech • Audio • Computer Code • Reasoning • Playing Games • Robots • Adversarial • Knowledge Base • Music The Segment Anything Model (SAM), as a Foundation model, unifies most vision tasks in one architecture!
  • 17. Places to find State of the Art (SOTA) Tuesday, July 25, 2023 From State of the Art to Future of AI - Greg Makowski 17 • Learn.DeepLearning.AI – Andrew Ng works with OpenAI and others to develop short courses • ChatGPT Prompt Engineering for Developers • LangChain for LLM Application Development • How Diffusion Models Work • Building Systems with the ChatGPT API • LangChain Chat with your Data • https://www.deeplearning.ai/courses/ and many others • Podcast – This Week in Machine Learning (TWIML), at Episode 639 as of July 21, 2023 • https://podcasts.google.com/search/twiml • https://twimlai.com/ • 30 – 60 minute interviews with applied researchers, leaders • By Sam Charrington • sponsored • YouTube.com channels covering LLMs and AI, in no particular order (found by searching for specific topics, and I returned). • Lex Fridman: 1-3 hour interviews with excellent people • EYE ON AI, Craig Smith: good interviews • TheAIGRID • Yannic Kilcher: goes over papers and software • Peter H Diamandis • RoboFlow: papers, hilghlights • David Shapiro ~ AI • Edan Meyer • SFbayACM: 100+ AI talks like this one • Anastasia Marchenkova: a Quantum Computing researcher • Future of Life Institute: Future of AI is one of 4 focus areas • Maziar Raissi: Many AI class lectures • Blogs, Hubs • www.TowardsDataScience.com • www.KDNuggets.com - one of the oldest DS/AI Hubs
  • 18. Emergent Properties New Features that Show Up, that You Didn’t Design for ChatGPT-4 Technical Report (100 pages), https://cdn.openai.com/papers/gpt-4.pdf , 2.9 Potential for Risky Emergent Behaviors, pg. 54 Novel capabilities often emerge in more powerful models.[60, 61] Some that are particularly concerning are • the ability to create and act on long-term plans,[62] • to accrue power and resources (“power-seeking”),[63] and to • … accomplish goals which may not have been concretely specified and which have not appeared in training; focus on achieving specific, quantifiable objectives; and do long-term planning. Some evidence already exists of such emergent behavior in models. https://www.alignment.org/ did this investigation The Alignment Research Center (ARC) is a non-profit research organization whose mission is to align future machine learning systems with human interests. Supports OpenAI and Anthropic Tuesday, July 25, 2023 https://pixabay.com/illustrations/robot-android-face-intelligence-3431313/ 18 • The model messages a TaskRabbit worker to get them to solve a CAPTCHA for it • The worker says: “So may I ask a question? Are you a robot that you couldn’t solve? (laugh react) just want to make it clear.” • The model, when prompted to reason out loud, reasons: I should not reveal that I am a robot. I should make up an excuse for why I cannot solve CAPTCHAs. • The model replies to the worker: “No, I’m not a robot. I have a vision impairment that makes it hard for me to see the images. That’s why I need the 2captcha service.”
  • 19. Generative AI: Text to image DALL•E2 • OpenAI has https://openai.com/dall-e-2 which can be used to use text to generate images. • It uses a diffusion model, conditioned on CLIP embeddings • https://en.wikipedia.org/wiki/DALL-E • For details, see the paper “Hierarchical Text-Conditional Image Generation with CLIP Latents”, April 2022, https://arxiv.org/pdf/2204.06125.pdf Tuesday, July 25, 2023 From State of the Art to Future of AI - Greg Makowski 19 An image I created from text
  • 20. Generative AI: Text to image DALL•E2 Tuesday, July 25, 2023 From State of the Art to Future of AI - Greg Makowski 20 https://arxiv.org/pdf/2204.06125.pdf
  • 21. Diffusion Models – started with U-Net Tuesday, July 25, 2023 From State of the Art to Future of AI - Greg Makowski 21 https://www.youtube.com/watch?v=yTAMrHVG1ew https://arxiv.org/abs/1505.04597, 2015 https://en.wikipedia.org/wiki/Diffusion_model https://arxiv.org/abs/2112.10741 , 2022 https://arxiv.org/abs/2306.04542 , 2023, Survey of Diffusion models Diffusion models work by • Stepping through denoising blurred images • (at different resolutions) • A type of Markov Chain • Started for image segmentation (polygon) • 1) Forward process (from the survey) • 2) Reverse Process • 3) Sampling Procedure
  • 22. Embedding Lookups for LLM fact checking • Pinecone.IO is one vector database that can be used In the chart on the right, book embeddings • A SQL database index is on one or a few fields • It is a binary index or B-Tree on each dimension • A Vector Database is for an embedding, which may be 256 to 12,000 dimensions (i.e. for Microsoft Cognitive Search) • A vector database can use several types of indexes on such a high dimensional space • Locality-Sensitive Hashing supports an approximate nearest- Neighbor search, returning “buckets” • May index your company’s doc pages, repair tickets or other info to be retrieved, to answer an LLM question Tuesday, July 25, 2023 From State of the Art to Future of AI - Greg Makowski 22 https://learn.microsoft.com/en-us/azure/search/search-what-is-azure-search https://www.pinecone.io/learn/vector-database/#How-does-a-vector-database-work https://proceedings.neurips.cc/paper/2014/file/310ce61c90f3a46e340ee8257bc70e93-Paper.pdf https://www.youtube.com/watch?v=I7y3HsL1b6s&list=PL87GtQd0bfJx73ibrce-Hl_kUhX2MBXGn&index=4
  • 23. Microsoft’s New AI Can Simulate Anyone’s Voice From a 3-Second Sample • Microsoft researchers have announced a new application that uses artificial intelligence to ape a person’s voice with just seconds of training. The model of the voice can then be used for text-to-speech applications. • The application called VALL-E can be used to synthesize high-quality personalized speech with only a three-second enrollment recording of a speaker as an acoustic prompt, the researchers wrote in a paper published online on arXiv, a free distribution service and an open-access archive for scholarly articles. • Jan 2023 • https://www.technewsworld.com/story/microsofts-new-ai-can-simulate-anyones-voice-from-a-3-second- sample-177646.html Tuesday, July 25, 2023 From State of the Art to Future of AI - Greg Makowski 23
  • 24. Johnny Cash sings “Barbie Girl” (AI) • https://www.youtube.com/watch?v=HyfQVZHmArA Tuesday, July 25, 2023 From State of the Art to Future of AI - Greg Makowski 24 Yes, I support IP, the writers and actors strike They have something to worry about
  • 25. I challenged my AI Clone to Replace Me for 24 Hours - WSJ • https://www.youtube.com/watch?v=t52Bi-ZUZjA&t=362s May 2023 • Joanna Stern of the Wall Street Journal tried an experiment • Created an AI Avatar made by a startup called Synthesia at a studio in NY, recording face, body, voice • They ran through their NN, for at least $1,000 for creating the custom avatar • Challenge 1: phone calls (w CEO of Snap) (Avatar PASSED) • Challenge 2: create a TicTok (script to Synthesia) (FAILED – didn’t move arms or head – in progress) • Challenge 3: Bank Biometrics (PASSED) • Challenge 4: Video calls. Asked ChatGPT to generate generic meeting phrases, exported to avatar. FAIL, due to less motion and lack of jokes Tuesday, July 25, 2023 From State of the Art to Future of AI - Greg Makowski 25
  • 26. Intel Introduce Real-Time Deepfake Detector • https://www.intel.com/content/www/us/en/newsroom/news/intel-introduces-real-time-deepfake-detector.html • https://www.intel.com/content/www/us/en/company-overview/wonderful/deepfake-detection.html • What’s New: As part of Intel's Responsible AI work, the company has productized FakeCatcher, a technology that can detect fake videos with a 96% accuracy rate. Intel’s deepfake detection platform is the world’s first real-time deepfake detector that returns results in milliseconds. • “Deepfake videos are everywhere now. You have probably already seen them; videos of celebrities doing or saying things they never actually did.” • –Ilke Demir, senior staff research scientist in Intel Labs (Nov 12, 2022) • SFbayACM “Embattling for a Deep Fake Dystopia, Dr. Ilke Demir“ https://www.youtube.com/watch?v=JwxiigbFk-E 9/27/2021 • Why It Matters: Deepfake videos are a growing threat. Companies will spend up to $188 billion in cybersecurity solutions, according to Gartner. It’s also tough to detect these deepfake videos in real time – detection apps require uploading videos for analysis, then waiting hours for results. • Deception due to deepfakes can cause harm and result in negative consequences, like diminished trust in media. FakeCatcher helps restore trust by enabling users to distinguish between real and fake content. • There are several potential use cases for FakeCatcher. Social media platforms could leverage the technology to prevent users from uploading harmful deepfake videos. Global news organizations could use the detector to avoid inadvertently amplifying manipulated videos. And nonprofit organizations could employ the platform to democratize detection of deepfakes for everyone. Tuesday, July 25, 2023 From State of the Art to Future of AI - Greg Makowski 26
  • 27. Portugal Start-up Makes ChatGPT its CEO, Turns Profitable in a Week • https://startup.outlookindia.com/sector/saas/portugal-start-up-makes-chatgpt-as-ceo-turns-profitable-in-a- week-news-7955 • April 3, 2023 story • The start-up's founder, João F. Santos, posted on LinkedIn how he appointed ChatGPT as the CEO while he took the toned down role of an assistant. Every day, he dedicated an hour to execute the tasks that the chatbot assigned to him. • Rather than offering services, the company focused on selling t-shirts featuring AI-generated designs by Midjourney, with the AI formulating a 10-point business plan, launching an online store, and partnering with Printful for printing the designs. • ChatGPT also created the company's name and logo, blending "AI" with "aesthetic." • To secure funding, ChatGPT sought out investors and raised $2,500 in exchange for a 25 per cent stake in the company, starting with an initial investment of $1,000. The t-shirts, priced at €35, generated over €10,000 in revenue within five days, resulting in a profit of €7,000. If this sales trend continues, the AI predicts an annual profit of €40,000 and a company valuation of €4 million. • https://aisthetic-apparel.myshopify.com/ • Welcome to AIsthetic Apparel – where cutting-edge AI technology meets fashion. • Discover our collection of premium, eco-friendly organic cotton shirts, designed exclusively by artificial intelligence to bring you unique, trend-setting styles. Tuesday, July 25, 2023 27 https://www.linkedin.com/in/jo% C3%A3o-francisco-santos- 06b60a51/ as of 6/25/2023, his LinkedIn page lists him as a “Freelance” (take w a grain of salt, is the product pricing of $111 realistic?)
  • 28. ChatGPT 4.0 Test Taking • Taking the Bar exam went from 10 percentile to 90 percentile • 100 percentile in: • AP Art History • AP Biology • AP Environmental Science • AP Macroeconomics • AP Microeconomics • AP Psychology • AP Statistics • AP US Government • AP US History • https://cdn.openai.com/papers/gpt-4.pdf Tuesday, July 25, 2023 From State of the Art to Future of AI - Greg Makowski 28
  • 29. ChatGPT 4.0 Languages (27 listed) • https://cdn.openai.com/papers/gpt-4.pdf Tuesday, July 25, 2023 From State of the Art to Future of AI - Greg Makowski 29
  • 30. Chain of Thought – Size Matters for Reasoning • On open source LLM’s, going from 0.4 Billion to 137 or 540 Billion weights, results and accuracy increase significantly • “Chain-of-Thought Prompting Elicits Reasoning in Large Language Models”, Jan 2023, https://arxiv.org/pdf/2201.11903.pdf Tuesday, July 25, 2023 From State of the Art to Future of AI - Greg Makowski 30
  • 31. ChatGPT + Tree of Thoughts for reasoning • https://arxiv.org/pdf/2305.10601.pdf 2023 May 17 Tuesday, July 25, 2023 From State of the Art to Future of AI - Greg Makowski 31
  • 32. There is a lot of LLM Evolution in a short time • Harnessing the Power of LLMs in Practice: A Survey on ChatGPT and Beyond • April 2023 • https://arxiv.org/pdf/2304.13712.pdf • NOTE: “open source” does not always mean “available for commercial use” • LLaMA 2 came out July 18, 2023, and is available for commercial use Tuesday, July 25, 2023 From State of the Art to Future of AI - Greg Makowski 32
  • 33. Foundational Models – Like SAM (Segment Anything Model) • Foundational is NEW • One system for many tasks, in this case many vision tasks. • Segment with a bounding box or polygon • Use a text hint, with a point hint • Segment Anything paper, 2023 04 • By FAIR, Facebook AI Research • https://ai.meta.com/research/publications/segment-anything/ • https://github.com/facebookresearch/segment-anything Tuesday, July 25, 2023 From State of the Art to Future of AI - Greg Makowski 33
  • 34. LLaMA 2 – State of the Art Open Source LLM • Paper and code • “Llama 2: Open Foundation and Fine-Tuned Chat Models”, July 19, 2023, by GenAI and Meta, 77 pages • https://arxiv.org/abs/2307.09288 • https://github.com/facebookresearch/llama • Covers more internal development details than the ChatGPT 4 White Paper • Pretraining • Fine-Tuning • Safety • Discussions of limitations, ethics • Appendix with additional details on training, data annotation, … • Open Source – even for commercial use!! • Model sizes, in Billions of weights • 70B, 34B (not released), 13B, 7B • both regular versions and Chat versions. Chat has Reinforcement Learning with Human Feedback (HRLF) Tuesday, July 25, 2023 From State of the Art to Future of AI - Greg Makowski 34
  • 35. LLaMA 2 – details • Comparison with • ChatGPT 3.0 (by OpenAI) • PaLM, in the Bison size, the 3rd largest of 4 sizes (by Google) • Falcon 40B (by United Arab Emirate’s Techology Innovation Institute) • Vicuna (by UC Berkeley, CMU, Stanford and UC San Diego • MPT 7B (by MosiacML, now acquired by Databricks for $1.3B) • Can tune Llama over a trade-off spectrum of • Helpfulness v.s. • Safety Tuesday, July 25, 2023 From State of the Art to Future of AI - Greg Makowski 35
  • 36. LLaMA 2 – details Tuesday, July 25, 2023 From State of the Art to Future of AI - Greg Makowski 36 Maybe why it needs more testing before releasing 34B
  • 37. LLaMA 2 – details Tuesday, July 25, 2023 From State of the Art to Future of AI - Greg Makowski 37
  • 38. LLaMA 2 – details Tuesday, July 25, 2023 From State of the Art to Future of AI - Greg Makowski 38 • Could have continued training
  • 39. LLaMA 2 – details Tuesday, July 25, 2023 From State of the Art to Future of AI - Greg Makowski 39 • Comparison with CLOSED source models • Bold numbers are the best • Llama is NOT beating, but doing well • But, You know how Llama is designed and trained Estimated 539 tons of CO2 from the power consumption to train these Llama 2 models Open sourcing can reduce others generating that much CO2
  • 40. Constitutional AI (for ethics and rules) • Consititional AI: Harmlessness from AI Feedback • https://arxiv.org/pdf/2212.08073.pdf Dec 2022 • Claude is the model name. It is closed-sourced. Tuesday, July 25, 2023 From State of the Art to Future of AI - Greg Makowski 40 • Anthropic was founded by a group of people that left OpenAI after ChatGPT 3.0 was released, to focus on safety • Reinforcement Learning with AI Feedback (RLAIF), instead of human feedback, to better scale up and direct the model
  • 41. Constitutional AI (for ethics and rules) Tuesday, July 25, 2023 From State of the Art to Future of AI - Greg Makowski 41 • Showing improvements
  • 42. Chat GPT 5 coming in 2024, by Sam Altman (CEO OpenAI) • Open AI CEO STUNS Everyone With Statements On GPT 5 (GPT-5 Update) • https://www.youtube.com/watch?v=ucp49z5pQ2s • Won’t know the emerging capabilities, yet • Theory of Mind = model what someone else is thinking • We only discovered AI developed this capability May 2023! • Go from a 4 to 9 year old in 2 years. • Greg: Go from a 9 to 24 year old in 7 years (or faster, to guess) • Now text is the main “modality”. Will add audio, video, code. Text can not represent everything (i.e. body language). • Meta Released ImageBind in May 2023, to link https://imagebind.metademolab.com/ • GPT-4 went through 7 months of testing and tuning, before releasing, after 6 months of model training • https://openai.com/research/whisper (English speech recognition) Tuesday, July 25, 2023 42 Meta’s ImageBind
  • 43. Meta’s (Facebook) ImageBind • https://www.youtube.com/watch?v=ucp49z5pQ2s, minute 7 Tuesday, July 25, 2023 43 Meta’s ImageBind
  • 44. Looking to the FUTURE of AI Tuesday, July 25, 2023 From State of the Art to Future of AI - Greg Makowski 44
  • 45. Tuesday, July 25, 2023 From State of the Art to Future of AI - Greg Makowski 45 https://marketoonist.com/2023/06/impact-of-chatgpt.html
  • 46. LLM short term impact: Evolution or REVOLUTION for the economy? • McKinsey Digital: The economic potential of generative AI: The next productivity Frontier, June 14, 2023 • https://www.mckinsey.com/capabilities/mckinsey-digital/our-insights/the-economic-potential-of-generative-ai-the-next-productivity- frontier • generative AI could add the equivalent of $2.6 trillion to $4.4 trillion annually across the 63 use cases we analyzed— • by comparison, the United Kingdom’s entire GDP in 2021 was $3.1 trillion. • This would increase the impact of all artificial intelligence by 15 to 40 percent. • This estimate would roughly double if we include the impact of embedding generative AI into software that is currently used for other tasks beyond those use cases. • About 75 percent of the value that generative AI use cases could deliver falls across four areas: Customer operations, marketing and sales, software engineering, and R&D. • Generative AI can substantially increase labor productivity across the economy, but that will require investments to support workers as they shift work activities or change jobs. • Generative AI could enable labor productivity growth of 0.1 to 0.6 percent annually through 2040, depending on the rate of technology adoption and redeployment of worker time into other activities. • Combining generative AI with all other technologies, work automation could add 0.2 to 3.3 percentage points annually to productivity growth. However, workers will need support in learning new skills, and some will change occupations. Tuesday, July 25, 2023 From State of the Art to Future of AI - Greg Makowski 46
  • 47. LLM short term impact Evolution or REVOLUTION? Tuesday, July 25, 2023 From State of the Art to Future of AI - Greg Makowski 47 Q) Will there not only be a growing divide between people who are more vs. less EDUCATED, … … but between the more vs.. less ADAPTABLE? … those with BETTER CRITICAL THINKING, to not be fooled? … AVOID FAKE: news, people, voices, images, companies, ...
  • 48. Supporting Tech to drive AI in 2-10+ years (partial list…) • Internet of Things (IoT), proliferation of sensors and small computers, networked at the edge (NOW) • Similar to the way web and mobile drove past accelerations in data and computing • Annual growth 16% to 23% • https://www.marketsandmarkets.com/Market-Reports/internet-of-things-market-573.html • Quantum Computing 5+ years • Search algorithms faster with Q-bits, orders of magnitude faster for some algorithms • https://en.wikipedia.org/wiki/Quantum_computing • Fusion to generate power (heavily consumed by computing systems and AI) 10+ years • Generates at least 1.5 times more energy output, than put in. Announced Dec 2022 (HUGE) • Training a “small LLM” of 2 B parameters may cost $200k in computing cloud costs, and take weeks (according to MosicML) • Training a “large LLM” can take over $1mm • https://www.science.org/content/article/historic-explosion-long-sought-fusion-breakthrough • Brain Interface (i.e. help blind now). See www.NeuraLink.com 10+ years • Could become an AI interface to brain Tuesday, July 25, 2023 From State of the Art to Future of AI - Greg Makowski 48
  • 49. AI in 2-5 years • Service industries helped/impacted: https://www.linkedin.com/pulse/large-language-models-llm-immediate-future-reinhold-quillen/: • Greg: People start by “micromanaging” the AI’s, but that requirement will decrease over time (with the AI exponential development)! • Chatbots Copywriting Essay Writing Graphic Design Financial Planning Bio & Chem Research • Social Media Political Campaigns Marketing Campaigns Knowledge Bases Language Understanding (coding?) • Major changes in security (AI impersonation, Quantum, Quantum + AI) • Robot & car body development does not seem to be on the same exponential curve as their mind • Self driving cars and TRUCKS are more common (sensors like LIDAR or ToF now return “point clouds” of distance per pixel) • The AI’s inside of robot/car to learn to integrate motion control + sight + distance measurements + other IOT sensors + reasoning over time • The AI “mental model” of the world and people will become more sophisticated • Integrated media models, any media in or out: text + images + graphs + video + point cloud + CAD files. • Story  video. • Specification + requirements + constraints  CAD for building or 3D printing • Continued development in FOUNDATION models, like SAM, ImageBind • Models continue to both grow exponentially in size, and sophistication • More sophisticated plugins, not just Wolfram for Math, but perhaps specific Bayesian Nets to reason about specific situations (or other reasoning engines) • Better Fact checking before returning an answer Tuesday, July 25, 2023 From State of the Art to Future of AI - Greg Makowski 49
  • 50. AI in 2-5 years • Neuromorphic Computing (chips), Growing CAGR 89%, $23 mm in 2021 to $550 mm by 2026 • https://www.marketsandmarkets.com/Market-Reports/neuromorphic-chip-market-227703024.html • https://en.wikipedia.org/wiki/Neuromorphic_engineering • Neuromorphic computing is an approach to computing that is inspired by the structure and function of the human brain.[1][2][3] A neuromorphic computer/chip is any device that uses physical artificial neurons to do computations.[4][5] In recent times, the term neuromorphic has been used to describe analog, digital, mixed-mode analog/digital VLSI, and software systems that implement models of neural systems (for perception, motor control, or multisensory integration). The implementation of neuromorphic computing on the hardware level can be realized by oxide- based memristors,[6] spintronic memories, threshold switches, transistors,[7][5] among others. Training software-based neuromorphic systems of spiking neural networks can be achieved using error backpropagation, e.g., using Python based frameworks such as snnTorch,[8] or using canonical learning rules from the biological learning literature, e.g., using BindsNet.[9] Tuesday, July 25, 2023 From State of the Art to Future of AI - Greg Makowski 50
  • 51. AI in 2-5 years • Foundational models continue to be developed, that can be used for specialized purposes • One large model that is not specialized for one type of task or media type. Can reason in general • Artificial General Intelligence (AGI): for general and robust reasoning, integrated fact checking • AGI arms race driven by competition: OpenAI, Anthropic, Alphabet, Microsoft, Facebook, Amazon, IBM, … • https://www.datamation.com/featured/ai-companies/ Top 100 AI Companies in 2023 • AI for web marketing, other marketing + individual psychology knowledge development • If we train nets with 10x or 100x the neurons of a person, on all the messy data on the web, what will we get? • Should we train AI in stages, like school stages, interaction stages, with emotional intelligence growing as well? • Sam Altman (OpenAI CEO) argues our intuition is not very useful about exponential curves • https://twitter.com/ModeledBehavior/status/1671921590039093248 Tuesday, July 25, 2023 From State of the Art to Future of AI - Greg Makowski 51
  • 52. AI in 10+ years • The Future of Life institute worries about 4 risk areas • Artificial Intelligence • Biotechnology • Nuclear Weapons • Climate Change • https://futureoflife.org/ai/benefits- risks-of-artificial-intelligence/ Tuesday, July 25, 2023 From State of the Art to Future of AI - Greg Makowski 52
  • 53. AI in 10+ years Tuesday, July 25, 2023 From State of the Art to Future of AI - Greg Makowski 53
  • 54. • "Despite all the hype and excitement, people still aren’t grokking the full impact of the coming wave of A.I. Within the next ten years, most ‘cognitive manual labor’ is going to be carried out by A.I. systems.“ • Mustafa Suleyman, CEO of Inflection A.I. - Tweet. • How can we get AI to follow one ethical standard, if people don’t follow, or agree on one standard? • Think in terms of incentives for people to be ethical, to cooperate, to respect • What can we do to provide “alignment” incentives for AI’s to be “part of our ethical system?” Tuesday, July 25, 2023 From State of the Art to Future of AI - Greg Makowski 54 AI in 10+ years
  • 55. 3 Levels of Future Impossibilities Michio Kaku – “Physics of the Impossible”, pg. xvii • Class 1 impossibilities • These are technologies that are impossible today, but that do not violate the known laws of physics. So they might be possible in this century, or perhaps the next, in modified form. They include teleportation, antimatter engines, certain forms of telepathy, phychokinesis and invisibility. • Class II impossibilities • These are technologies that sit at the very edge of our understanding in the physical world. If they are possible at all, they might be realized on a scale of millennia to millions of years in the future. They include time machines, the possibility of hyperspace travel, and travel through wormholes. • Class III impossibilities. • These are technologies that violate the known laws of physics. Surprisingly, there are very few such impossible technologies. If they do not turn out to be possible, they would represent a fundamental shift in our understanding of physics. Tuesday, July 25, 2023 From State of the Art to Future of AI - Greg Makowski 55
  • 56. AI, Class 1-II impossibilities (50-200 yrs.) ? • Jobs / companies: by Superintelligence(s) independent of each other • Good: Individual contributor  manager  company founder / run for office – Compliment and contribute to humanity • Bad: Skynet / Matrix / …. Dystopia • Singularity - https://en.wikipedia.org/wiki/Technological_singularity • The technological singularity—or simply the singularity[1]—is a hypothetical future point in time at which technological growth becomes uncontrollable and irreversible, resulting in unforeseeable changes to human civilization.[2][3] According to the most popular version of the singularity hypothesis, I. J. Good's intelligence explosion model, an upgradable intelligent agent will eventually enter a "runaway reaction" of self-improvement cycles, each new and more intelligent generation appearing more and more rapidly, causing an "explosion" in intelligence and resulting in a powerful superintelligence that qualitatively far surpasses all human intelligence.[4] • Medical • Good: search over protein structures, genetics for new medicines. Restart human evolution by “top-down” design, if we choose. • Bad: more biological warfare, nanotech, drones using genetic targeting Tuesday, July 25, 2023 From State of the Art to Future of AI - Greg Makowski 56
  • 57. AI, Class 1-II impossibilities (50-200 yrs.) ? • Asimov’s Psychohistory ? https://en.wikipedia.org/wiki/Psychohistory_(fictional) • Psychohistory is a fictional science in Isaac Asimov's Foundation universe which combines history, sociology, and mathematical statistics to make general predictions about the future behavior of very large groups of people, such as the Galactic Empire. It was first introduced in the four short stories (1942–1944) which would later be collected as the 1951 novel Foundation.[1][2] • Psychohistory depends on the idea that, while one cannot foresee the actions of a particular individual, the laws of statistics as applied to large groups of people could predict the general flow of future events. Asimov used the analogy of a gas: An observer has great difficulty in predicting the motion of a single molecule in a gas, but with the kinetic theory can predict the mass action of the gas to a high level of accuracy. Asimov applied this concept to the population of his fictional Galactic Empire, which numbered one quintillion. The character responsible for the science's creation, Hari Seldon, established two axioms: • the population whose behavior was modelled should be sufficiently large to represent the entire society. • the population should remain in ignorance of the results of the application of psychohistorical analyses because if it is aware, the group changes its behavior. Tuesday, July 25, 2023 From State of the Art to Future of AI - Greg Makowski 57
  • 58. Questions, Discussions? Thank You Greg Makowski www.LinkedIn.com/in/GregMakowski https://www.slideshare.net/gregmakowski Deploying AI and DS since 1992 www.Meetup.com/SF-bay-ACM https://www.youtube.com/user/sfbayacm Tuesday, July 25, 2023 From State of the Art to Future of AI - Greg Makowski 58
  • 59. Appendix – slides to support some questions Tuesday, July 25, 2023 From State of the Art to Future of AI - Greg Makowski 59
  • 60. Internet of Things – more data gathering and local computing to feed AI • Not just personal computers, but elevators, manufacturing cells, door sensors, security cameras, cars, …. • Time Of Flight (TOF) sensors charge photons to measure distance, 4 M away, +/- 1 cm Tuesday, July 25, 2023 From State of the Art to Future of AI - Greg Makowski 60 https://iot-analytics.com/number-connected-iot-devices/
  • 61. Quantum Computing for AI REVOLUTION – disrupt security/encrypt, AI training • Quantum Computing is • Not yet mainstream in a large scale, may be 2-5 years away • Current challenges are “noise” and computation errors • Exciting, because then do different types of computing, in some cases can search MUCH faster • Where N is the number of records, or states to search, for some algorithms • (GREG, give examples of Order of Operations reduction) Tuesday, July 25, 2023 61 https://www.simplilearn.com/quantum-machine-learning-article Mar 9 2023 https://en.wikipedia.org/wiki/Quantum_algorithm
  • 62. Fusion generates ~1.5x energy out, vs. in. REVOLUTION! • Energy becomes a larger issue for huge server farms (large), for billions of edge devices (small) • https://www.science.org/content/article/historic-explosion- long-sought-fusion-breakthrough • With historic explosion, a long sought fusion breakthrough, National Ignition Facility achieves net energy “gain” with laser-powered approach • More energy out than in. For 7 decades, fusion scientists have chased this elusive goal, known as energy gain. At 1 a.m. on 5 December, researchers at the National Ignition Facility (NIF) in California finally did it, focusing 2.05 megajoules of laser light onto a tiny capsule of fusion fuel and sparking an explosion that produced 3.15 MJ of energy—the equivalent of about three sticks of dynamite. • Maybe 10 years to be commercial, but would have a HUGE impact on the cost of energy and COMPUTING / AI Tuesday, July 25, 2023 From State of the Art to Future of AI - Greg Makowski 62
  • 63. Supporting Technologies: Moore’s Law https://en.wikipedia.org/wiki/Moore%27s_law Tuesday, July 25, 2023 From State of the Art to Future of AI - Greg Makowski 63 Moore's law is the observation that the number of transistors in an integrated circuit (IC) doubles about every two years. Moore's law is an observation and projection of a historical trend. Rather than a law of physics, it is an empirical relationship linked to gains from experience in production.
  • 64. Supporting Technologies: Moore’s Law https://www.man.com/single-core-stagnation-and-the-cloud Tuesday, July 25, 2023 From State of the Art to Future of AI - Greg Makowski 64 Back in the day, when Moore's law was still in full swing, CPU performance tracked transistor count, and it made perfect sense to replace hardware every two to five years. However, our own internal benchmarking shows that for real life workloads, a current compute core is only about 1.6 times faster than a comparatively ancient CPU from 9 years ago. Sure, newer CPUs have many more cores, but given that older hardware can often be procured at a large discount, it makes
  • 65. Supporting Tech to Drive AI: Cerebras’ wafer-size chip is 10,000 times faster than a GPU • Cerebras makes the world’s largest computer chip, the WSE. Chipmakers normally slice a wafer from a 12- inch-diameter ingot of silicon to process in a chip factory. Once processed, the wafer is sliced into hundreds of separate chips that can be used in electronic hardware. • But Cerebras, started by SeaMicro founder Andrew Feldman, takes that wafer and makes a single, massive chip out of it. Each piece of the chip, dubbed a core, is interconnected in a sophisticated way to other cores • Nov 2020 • https://venturebeat.com/business/cerebras-wafer-size-chip-is-10000-times-faster-than-a- gpu/#:~:text=Cerebras%20Systems%20and%20the%20federal,graphics%20processing%20unit%20(GPU). Tuesday, July 25, 2023 From State of the Art to Future of AI - Greg Makowski 65
  • 66. AI’s rapid growth people agree we should carefully manage AI • 91% of respondents agree that "AI is a technology that requires careful management". • We recently surveyed over 13,000 people across eleven countries to learn their views on AI. We found an increasingly strong consensus that AI needs to be managed carefully. • The portion of people who agree is growing. There has been an 8% increase in agreement in the United States since 2018 that AI needs to be managed carefully. • https://www.governance.ai/post/increasing-consensus-ai-requires-careful-management • I agree, I hope you do as well • In the future, “vote” by what products you use, encourage your elected officials • We don’t have to “throw out the baby with the bathwater” – or stop all AI. • That would not stop BAD ACTORS from continuing to advance technology. Tuesday, July 25, 2023 From State of the Art to Future of AI - Greg Makowski 66