Nicely Done 25 Page NVIDIA Free Intro E Book on Large Language Models

E-Book  by NVIDIA

Nicely done 25-age EBook on an Enterprise Guide to Large Language Models

Everything an enterprise needs to know about LLMs.

What’s Included In This eBook?

A comprehensive background on what LLMs are, how they work, and how to evaluate them, paired with use case examples and real-world case studies on the impact LLMs have had for the enterprise.

What Are Large Language Models and How Do They Work?  

Learn about the evolution of LLMs, the role of foundation models, and how the underlying technologies have come together to unlock the power of LLMs for the enterprise.

What Are Large Language Model Examples and Case Studies?

Dive into the LLM applications that are driving the most transformation for enterprises. Examine real-world case studies of companies that adopted LLM-based applications and analyze the impact it had on their business.

How to Build and Evaluate Large Language Models?

Learn the steps to take when building LLMs and how to evaluate whether an LLM is well-suited for your intended use cases.  ... ' 

Exploring Definition of Digital Twins in GPT

 Been exploring how well GPT defines technical terms that I know by experience.   Here is an example: Digital Twins.   In particular its relationship to digital simulation approaches, which I have done in industry for a long time.    Here is an example using ChatGPT, at the first level of definition.  How good is this for the beginner in distinguishing it from simple simulation?   Also taking this further including 'Tell me more' and 'simplify' style prompts.  Reasonable?

 Digital Twin: (Initial Definition)

    Digital twins are virtual representations of physical systems or processes that are designed to simulate and analyze their behavior in real-time. These virtual models are created by combining data from various sources, including sensors, IoT devices, and other sources, to create a real-time representation of the physical system.

Digital twins can be used in a wide range of applications, including manufacturing, healthcare, and transportation. For example, in manufacturing, digital twins can be used to simulate production processes and optimize the manufacturing process, reducing costs and improving quality. In healthcare, digital twins can be used to simulate and analyze patient data to improve diagnosis and treatment, while in transportation, digital twins can be used to optimize traffic flow and reduce congestion.

Digital twins are often used in conjunction with machine learning and artificial intelligence (AI) algorithms to provide predictive insights and optimize performance. By using real-time data from the physical system, digital twins can predict and respond to changes in the environment or the system itself, enabling proactive maintenance, reducing downtime, and improving performance.

Overall, digital twins represent an exciting new technology that has the potential to transform a wide range of industries by enabling real-time simulation, analysis, and optimization of complex physical systems.  STOP    

IFTF's Post Financial Futures Forecasts using GPT-3

Worked for years with IFTF (Institute for the Future) in the enterprise.

The Institute for the Future passes along this new GPT-3 Forecast Model, a nice use example.   I will pass along additional findings.

From IFTF: 

We also conducted a groundbreaking experiment using our forecast provocation and OpenAI's GPT-3 to generate over 1,400 live, unique scenarios of post-financialization futures. If you missed it or would like to try out a scenario again the link is here.  

This is an experimental prototype exploring the use of GPT-3 to help augment and customize IFTF’s human forecasting process to broader audiences and more interactive contexts. 

IFTF has pursued this experiment in the spirit of building our literacy of the capabilities and challenges of large language models, and this experiment should not be construed as an endorsement of OpenAI or GPT-3.

IFTF is not responsible for content produced by the large language model GPT-3. The content provided by GPT-3 does not reflect the opinions or views of IFTF.  (Nor of Franz Dill) 

Be sure to mark your calendars for the next IFTF Ten-Year Forecast series event coming up on June 1. We'll be announcing details on expert speakers and special interactive futures immersions very soon.  

Thank you for being a part of IFTF’s TYF community as we look toward shaping a better future!

Our mailing address is:  Institute for the Future,  201 Hamilton Avenue,   Palo Alto, CA 94301  Iftf.org

On Event Modeling

 Its all about events and how we handle them. 

Adam Dymitruk on Event Modeling    By Software Engineering Radio    November 28, 2022

Credit: Adam Dymitruk/Software Engineering Radio

Adam Dymitruk is CEO and founder of Adaptech Group.

In an interview, Dymitruk explores the event-modeling approach to discovering requirements and designing software systems. Dymitruk explains how the structured approach eliminates the specifics of implementation details and technology decisions, enabling clearer communication for all stakeholders while keeping conversations focused on the business opportunity.

Using concrete examples of event modeling in practice, the interview examines event modeling in the context of other related approaches and methodologies, including event sourcing, event storming, CQRS, and domain-driven design.

From Software Engineering Radio

View Full Article     

New Brain Models

 Alternative, competitive models are always good.

A New Brain Model Could Pave the Way for Conscious AI

A new model of the human brain.

A new study presents a new neurocomputational model of the human brain, which might shed light on how the brain develops complex cognitive skills and advance neural artificial intelligence research. An international team of scientists from the Institut Pasteur and Sorbonne University in Paris, the CHU Sainte-Justine, Mila – Quebec Artificial Intelligence Institute, and the University of Montreal conducted the study.

Guillaume Dumas

The model, which was featured on the cover of the journal Proceedings of the National Academy of Sciences of the United States of America (PNAS), describes neural development over three hierarchical levels of information processing:

the first sensorimotor level explores how the brain’s inner activity learns patterns from perception and associates them with action;

the cognitive level examines how the brain contextually combines those patterns;

lastly, the conscious level considers how the brain dissociates from the outside world and manipulates learned patterns (via memory) no longer accessible to perception.

The model’s emphasis on the interaction between two fundamental types of learning—Hebbian learning, associated with statistical regularity (i.e., repetition), or as neuropsychologist Donald Hebb has put it, “neurons that fire together, wire together”—and reinforcement learning, associated with reward and the dopamine neurotransmitter, provides insights into the fundamental mechanisms underlying cognition.

The model solves three tasks of increasing complexity across those levels, from visual recognition to cognitive manipulation of conscious percepts. Each time, the team introduced a new core mechanism to enable it to progress.  ..'  

AI Model Transferability in Healthcare

 Key issues of how AI models make predictions in healthcare domains. 

ACM OPINION

AI Model Transferability in Healthcare: A Sociotechnical Perspective

By Nature Machine Intelligence, October 24, 2022

Doctor using a tablet to examine patient records.

To deliver value in healthcare AI and ML models must be integrated not only into technology platforms but also into local human and organizational ecosystems and workflows.

Predictive model transferability is gaining more attention as healthcare organizations attempt to implement artificial intelligence (AI)-based prediction tools. Although some machine learning (ML)-based models fail when subjected to retrospective validation across institutions and patient populations, technical improvements show promise for addressing this model efficacy problem. To address the engineering challenges, a technical subfield labelled MLOps has emerged.

However, the focus of MLOps on technical transferability may be obscuring a larger set of obstacles to sociotechnical transferability: organizational, social, and individual challenges of deploying models at scale across contexts, whether institutions, teams or individual roles....

To deliver value in healthcare AI and ML models must be integrated not only into technology platforms but also into local human and organizational ecosystems and workflows. ... 

From Nature Machine Intelligence

View Full Article    

AI Model Transferability in Healthcare

Very good look at this issue,  considering potential application. 

AI Model Transferability in Healthcare: A Sociotechnical Perspective

By Nature Machine Intelligence, October 24, 2022  (Was Introduced to Nature Machine Intelligence) 

Predictive model transferability is gaining more attention as healthcare organizations attempt to implement artificial intelligence (AI)-based prediction tools. Although some machine learning (ML)-based models fail when subjected to retrospective validation across institutions and patient populations, technical improvements show promise for addressing this model efficacy problem. To address the engineering challenges, a technical subfield labelled MLOps has emerged.

However, the focus of MLOps on technical transferability may be obscuring a larger set of obstacles to sociotechnical transferability: organizational, social, and individual challenges of deploying models at scale across contexts, whether institutions, teams or individual roles....

To deliver value in healthcare AI and ML models must be integrated not only into technology platforms but also into local human and organizational ecosystems and workflows.

From Nature Machine Intelligence

View Full Article  

NASA Webb Space Telescope Data Could be Misinterpreted

Data May be Misinterpreted?   Is always an issue.  Models that use the data also must be good.  CNN should look to themselves as a glaring example of bad interpretation.. 

ACM TECHNEWS

NASA Webb Space Telescope Data Could Be Misinterpreted   By CNN

Massachusetts Institute of Technology (MIT) scientists warn of a possible disconnect between the power of the U.S. National Aeronautics and Space Administration's James Webb Space Telescope (JWST) and current datasets.

"The data we will be getting from the JWST will be incredible, but ... our insights will be limited if our models don't match it in quality," said MIT's Clara Sousa-Silva.

New research suggests some of the light-decoding tools researchers employ to understand exoplanets cannot fully accommodate the JWST's spectral data, which could undermine the search for extraterrestrial life.

The researchers are calling for improvements to opacity models to enable them to handle the JWST's subtle measurements.

From CNN

View Full Article  

Building in Explainability into ML Models

 Key to making them effectively Useable

Building Explainability into Components of ML Models

MIT News, Adam Zewe, June 30, 2022

Researchers at the Massachusetts Institute of Technology (MIT) and cybersecurity startup Corelight have developed a taxonomy to help developers create components of machine learning (ML) models that incorporate explainability. The researchers defined properties that make features interpretable for five varieties of users, and that provide instructions for engineering features into formats that will be easier for laypersons to understand. Key to the taxonomy is the precept that there is no universal model for interpretability. The researchers define properties that can make components approximately explainable for different decision-makers, and outline which properties are likely most valuable to users. MIT's Alexandra Zytek said, "The taxonomy says, if you are making interpretable features, to what level are they interpretable? You may not need all levels, depending on the type of domain experts you are working with."  ... 

Building Reliable AI Models

 Below quite interesting,  here just the intro, full article linked to. Reviewing.  I note the point about changing data in context, which I always emphasize in models. 

7 Techniques for Building Reliable AI Models  By Beena Ammanath in future.16z.com

This is an edited excerpt from Trustworthy AI: A Business Guide for Navigating Trust and Ethics in AI by Beena Ammanath (Wiley, March 2022). Ammanath is executive director of the Global Deloitte AI Institute and leads Trustworthy & Ethical Technology at Deloitte. She has held leadership positions in artificial intelligence and data science at multiple companies, and is the founder of Humans For AI, an organization dedicated to increasing diversity in AI.

With AI model training, datasets are a proxy for the real world. Models are trained on one dataset and tested against another, and if the results are similar, there is an expectation that the model functions can translate to the operational environment. What works in the lab should work consistently in the real world, but for how long? Perfect operating scenarios are rare in AI, and real-world data is messy and complex. This has led to what leading AI researcher Andrew Ng called a “proof-of-concept-to-production gap,” where models train as desired but fail once they are deployed. It is partly a problem of robustness and reliability.

When outputs are inconsistently accurate and become worse over time, the result is uncertainty. Data scientists are challenged to build provably robust, consistently accurate AI models in the face of changing real-world data. In the information flux, the algorithm can meander away, with small changes in input cascading into large shifts in function.

To be sure, not all tools operate in environments prone to dramatic change, and not all AI models present the same levels of risk and consequence if they become inaccurate or undependable. The task for enterprises as they grow their AI footprint is to weigh robustness and reliability as a component of their AI strategy and align the processes, people, and technologies that can manage and correct for errors in a dynamic environment. .... ' 

Digital Twins for Virtual Cities

An approach I had not seen, but we had used a simplistic SimCity like model for thinking about interaction in groups.

 Cities Using Digital Twins Like SimCity for Policymakers

Bloomberg CityLab, Linda Poon, April 5, 2022

Cities like Orlando, FL, and Singapore are using digital twins to generate virtual models of themselves, in order to simulate the effects of potential new policies or infrastructure projects that can inform real-world decision-making. For example, the Orlando Economic Partnership and gaming company Unity have developed a three-dimensional (3D) model of the region that the city can show to potential investors as it attempts to expand as a technology hub. Meanwhile, the Virtual Singapore model incorporates over 3 million street-level and 160,000 aerial images, plus billions of 3D data points, exceeding 100 terabytes of raw data. Singapore Land Authority's Victor Khoo said the model differentiates between individual elements, making it easier to test its responses to different conditions in various simulations.  ... '

Computer Models for Vaccines

 Much more in Nature Mag, linked to below.

Could Computer Models Be the Key to Better COVID Vaccines?

In Nature, Elie Dolgin, April 5, 2022

Scientists have been developing mathematical and computational models over the past several years to supplement dosage decision-making for vaccine tests, and some advocates suggest their use would have made COVID vaccines more effective. Previous experience and animal testing typically inform dose selection for experimental vaccines like the COVID-19 vaccine, yielding a range of dosages. Vaccine-dose modeling starts with researchers feeding immune-response results from animal experiments into equations to generate a predicted dose-response curve; this is scaled to humans using clinical data from a smaller dose number, often from historical research on similar vaccines. The result is expected "best" dosages for human testing, which can be further refined as more data becomes available.   .... ' 

Digital Twinning in the Metaverse

Digital Twinning in the Metaverse

Digital twins could be crucial to metaverse adoption?

BY  DANIEL LIPSCOMBE  in GMW3

April 13th 1970: The Apollo 13 spacecraft is 220,000 kilometres from Earth when an explosion rocks the crew and tears off one of the two oxygen tanks from the spacecraft. The blast destroyed one side of the transport, not only removing the oxygen supply from the crew but also water and some electrical systems. This disaster echoed across the world as astronaut Jack Swigert radioed to NASA control, “Houston, we’ve had a problem here.”

From that moment, engineers and scientists at NASA rushed to put their heads together to find a solution to Apollo’s problem. The engineers needed to solve the issue using what the Apollo crew would have to hand, but crucially, without physically seeing the damage firsthand. In the end, the fix was simple; NASA instructed the crew to use cardboard, plastic bags and tape to patch up the craft enough to get them home.

There’s not much NASA could have done at the time to foresee the issue; building several spacecraft to stress test every possible outcome would have burned through budgets swiftly. When the Apollo 13 disaster occurred, NASA engineers could no longer rely on their original designs, as the craft had failed due to an unforeseen hostile environment. The crew in Houston needed a model on Earth that directly mirrored the craft in space.

The Digital Twin

In 2002, NASA coined the term ‘digital twin’, though the original concept is a little older. A digital twin can be described as ‘a digital copy of a physical object: mechanism, building or concept based in reality’. For example, a car manufacturer may create a digital twin of their main assembly plant and use it to implement new safety protocols or install new machinery, by first trying it within a digital safe zone.

Healthcare professionals can use digital twinning to simulate rare illnesses and practice care first hand – albeit digitally – and learn the correct techniques. Planning departments in government can replicate dense population areas of cities to see how new infrastructure will impact the city and society. Environmentalists are simulating extremes of climate change on digital twins based on rainforests and oceans.

A digital twin is a living model of something physical, which, to metaverse aspirers will sound familiar. Digital twins are becoming much more popular and with the advent of more immersive technology – Augmented Reality (AR) and Virtual Reality (VR) – the concept of digital twinning is becoming more mainstream. Not only that, but it points to where the metaverse can aid industry and where digital twinning can benefit from creating the metaverse.

The Impact

To fully realise a digital twin of a location or person, sensors can be placed in the physical space to gauge temperature, humidity, footfall traffic, heart rate, etc. This data is then sent to the digital twin to be replicated and be shown in almost real-time within a 3D metaverse-style space.

The opportunities for this technology are vast and far-reaching and while the positives can be seen, there must be a balance in data use. Any sensors and personal data being beamed back and forth to digital twins must be heavily encrypted and safeguarded. Landmarks and buildings would likely contain blueprints and maps on the interiors and any personal data relating to users must be made safe.

The idea of constant monitoring may be off-putting for some, given the decentralised nature of the metaverse. If the metaverse is to be hosted by millions of users across a blockchain network such as Bitcoin or Etherium, it would make the data much harder to hack, given the security of the ledger technology. Whereas a centralised server hosting this information may be more liable to attack.

On New Open Language Models

Bigscience and open Language Models

Inside BigScience, the quest to build a powerful open language model

Kyle Wiggers  @Kyle_L_Wiggers  in VentureBeat.  

January 10, 2022 9:30 AM

Roughly a year ago, Hugging Face, a Brooklyn, New York-based natural language processing startup, launched BigScience, an international project with more than 900 researchers that is designed to better understand and improve the quality of large natural language models. Large language models (LLMs) — algorithms that can recognize, predict, and generate language on the basis of text-based datasets — have captured the attention of entrepreneurs and tech enthusiasts alike. But the costly hardware required to develop LLMs has kept them largely out of reach of researchers without the resources of companies like OpenAI and DeepMind behind them.

Taking inspiration from organizations like the European Organization for Nuclear Research (also known as CERN),  and the Large Hadron Collider, the goal of BigScience, then, is to create LLMs and large text datasets that will eventually be open-sourced to the broader AI community. The models will be trained on the Jean Zay supercomputer located near Paris, France, which ranks among the most powerful machines in the world.

“From Data to Knowledge”. How the Organization of Data Using LC:NC Can Drastically Reduce the Technical Complexity of Deriving Knowledge From Data._

While the implications for the enterprise might not be immediately clear, efforts like BigScience promise to make LLMs more accessible — and transparent — in the future. With the exception of several models created by EleutherAI, an open AI research group, few trained LLMs exist for research or deployment into production. OpenAI has declined to open source its most powerful model, GPT-3, in favor of exclusively licensing the source code to Microsoft. Meanwhile, companies like Nvidia have released the code for capable LLMs, but left the training of those LLMs to users with sufficiently powerful hardware. ... ' 

Making Decision Makers Use and Understand the Value of Models

Many times had to consider how to get key decision makers to use the results of analytical models.  This article touches on that in some ways. Like to consider further how this could be done consistently. 

Making machine learning more useful to high-stakes decision makers

A visual analytics tool helps child welfare specialists understand machine learning predictions that can assist them in screening cases.

Adam Zewe | MIT News Office

The U.S. Centers for Disease Control and Prevention estimates that one in seven children in the United States experienced abuse or neglect in the past year. Child protective services agencies around the nation receive a high number of reports each year (about 4.4 million in 2019) of alleged neglect or abuse. With so many cases, some agencies are implementing machine learning models to help child welfare specialists screen cases and determine which to recommend for further investigation.

But these models don’t do any good if the humans they are intended to help don’t understand or trust their outputs.

Researchers at MIT and elsewhere launched a research project to identify and tackle machine learning usability challenges in child welfare screening. In collaboration with a child welfare department in Colorado, the researchers studied how call screeners assess cases, with and without the help of machine learning predictions. Based on feedback from the call screeners, they designed a visual analytics tool that uses bar graphs to show how specific factors of a case contribute to the predicted risk that a child will be removed from their home within two years.... ' 

Developing Digital Twins

Interesting, often this means not only the twin but also the context of its use.  Metadata and all. Like to see a full example.

Advanced Technique for Developing Digital Twins Makes Tech Universally Applicable  UT News, May 20, 2021

Researchers at the University of Texas at Austin (UT Austin), the Massachusetts Institute of Technology (MIT), and industry partner The Jessara Group have developed what they’re calling a universally applicable digital twin mathematical model. The framework was designed to facilitate predictive digital twins at scale. MIT's Michael Kapteyn said, "Using probabilistic graphical models, we create a mathematical model of the digital twin that applies broadly across application domains." The researchers used this technique to generate a structural digital twin of a custom-built unmanned aerial vehicle equipped with state-of-the-art sensors. Said Jacob Pretorius of the Jessara Group, “The value of integrated sensing solutions has been recognized for some time, but combining them with the digital twin concept takes that to a new level. We are on the cusp of an exciting future for intelligent engineering systems.”

Ghosts Using Data, Children too

Hmm, can tell you that in any business problem you have to have the correct data to derive something useful.    To walk away with some sort of algorithm that later works.  Either to create or validate a model.   So if the data is badly gathered you don't have much clue.    But it seems humans work for a model.  But from data. So why does that work at all?  ..  Some related thoughts:

The ghosts in the data,   Mar 26, 2021

Bernadette Resha, Gathering of Ghosts (2014)

Something I’ve been thinking about recently as I’ve been working at a company that operates entirely remotely and mostly asynchronously during a time when most companies are working in some variation of this model is the idea of implicit versus explicit knowledge.

Explicit knowledge is anything that you can read about, knowledge that’s easy to share and pass on. Implicit knowledge is knowledge that people gain by context that’s very hard to pull out consciously. The best example of this is from this paper on language acquisition,

Children acquire their first language by engaging with their caretakers in natural meaningful communication. From this “evidence” they automatically acquire complex knowledge of the structure of their language. Yet paradoxically they cannot describe this knowledge, the discovery of which forms the object of the disciplines of theoretical linguistics, psycholinguistics, and child language acquisition.

This is a difference between explicit and implicit knowledge—ask a young child how to form a plural and she says she doesn’t know; ask her“ here is a wug, here is another wug, what have you got?”and she is able to reply,“two wugs.” The acquisition of L1 grammar is implicit and is extracted from experience of usage rather than from explicit rules.  .... 

Pandemic Eviction Modeling

 More models accurately done, with the right data, address the results of specific decisions.

Modeling Shows Pandemic Eviction Bans Protect Entire Communities From Covid-19 Spread

Johns Hopkins Medicine Newsroom

April 19, 2021

Researchers at institutions including Johns Hopkins University and the University of Pennsylvania used computer modeling to determine that eviction bans during the Covid-19 pandemic lowered infection rates, shielding entire communities from the virus. The scientists said they used simulations to predict additional virus infections in major U.S. cities if bans were not authorized in fall 2020. The team initially calibrated its model to reproduce the most common epidemic patterns observed in major cities last year, accounting for infection-rate changes due to public health measures. Another iteration factored in the lifting of eviction bans, determining that people who are evicted or who live in a household that hosts evictees are 1.5 to 2.5 times more likely to become infected than with such bans in place.  .... '

Walking Inside Samples

Worked on something similar, positioned for looking at product designs.   Wlaking through thigs can give you ideas about design. 

Software Allows Scientists to 'Walk Inside' Samples

Australian National University  February 3, 2021

The new Drishti software developed by researchers at Australian National University (ANU) lets scientists visualize data in three dimensions and generate lifelike models of objects like fossil samples so they can "zoom in" on smaller details without damaging the original sample. ANU's Yuzhi Hu said, "After we scan the sample, we then have a set of 3D data which can be digitally dissected effectively using our new tool.” The ANU team said the software could be particularly useful for scientific communication and education. ANU's Ajay Limaye said Drishti's current applications include digitally duplicating a mummy sample. The Drishti software is free and available online... '

Cardiovascular Disease: Improving Therapy with Digital Twins

 Digital twins for modeling therapy. 

An integration of a number of modeling methods are brought together with imaging.  

Cardiovascular Diseases: Computer Model Improves Therapy

Graz University of Technology (Austria), Christoph Pelzl, January 22, 2021

Researchers at Austria's University of Graz and the Graz University of Technology (TU Graz) have generated digital twins of human hearts, which doctors can use to pre-model optimal therapies and improve the likelihood of successful treatment for cardiovascular diseases. Imaging algorithms construct a digital twin from diagnostic data, which provides information to help understand the individual clinical situation and consider various therapeutic scenarios. TU Graz's Thomas Pock said computerized heartbeat simulation requires calculating millions of factors, and demands "complex mathematical procedures, special algorithms, and special hardware that can perform billions of computing actions per second." The new technique can routinely produce anatomically accurate digital twins of patient hearts in clinical environments.  ... '