GRC Connect Conference Chicago

 Excellent event and a great discussion. I enjoyed discussing Gen AI as a panelist at the GRC Connect Chicago Conference.

Parsing Documents with GenAI

 A good PDF parser that can understand embedded tables and figures is a necessary condition for building good RAG. Most PDF parsers struggle with representing tables, which sends a confusing representation to the LLM, leading to wrong answers.

That’s where LlamaParse comes in: comparing LlamaParse to PyPDF, PyMuPDF, Textract, and PDFMiner.

Chaos in computation

 

The author, Tim Palmer, presents an exciting discussion on chaos theory and the beneficial behaviors of chaotic systems. For people interested in #AI and #LLMs, chapters that explain the utility of lowering the power supplied to chips are particularly intriguing. Generating noise can translate into many applications in Generative AI. For example, we know that signal noise might benefit a neural network's output quality. The author's examples of low-power and noisy supercomputers applied to hard computational problems like weather prediction are fascinating. It is hard to disagree with the author that the human brain is such an example of a noisy, chaotic system that exhibits lots of creativity. This discussion also reminded me of #Protopia's (protopia.ai) approach to cybersecurity in #GenAI pipelines based on turning the original signal into noise. #chaos #chaoticsystems

Blogchain technologies

 Blockchain technology has many uses in the world of finance, insurance, supply chain, or information security. Scientists and IT leaders are continually investigating new ways to apply blockchain technology. One of the critical challenges the authors address is the connection with existing enterprise infrastructure. Blockchain will play a critical role in IoT and cyber-physical system implementations.
The book is an excellent discussion of blockchain tools, consensus algorithms, and platforms that support interoperability. The chapter on interoperability is a unique study of communication mechanisms. Decentralized algorithms pose unique challenges for handling communication across multiple chains with transparency and data integrity. There are also discussions on solutions for the scalability of extensive data set handling during messaging across chains with large numbers of network nodes. The book includes valuable implementation guides for platforms like Hyperledger Fabric or Multichain.
The authors present an in-depth discussion on cryptoasset security and legal challenges around inheritances and wills.
Blockchain Technology and Applications is an excellent asset for readers at all levels interested in this technology. 

 

Generative AI: presentation for ELEVATE

 

New Perspectives in System Design

Insolvent by Christoph Becker is a comprehensive, transdisciplinary critique of current perspectives on sustainability and justice in system design. The author tackles the complexities of technology and societal interactions. Many concepts discussed here are far-reaching into computing, society, culture, and politics. Those can not be decomposed into atomic problems without losing or changing the scope of their meaning. The author illustrates the impacts of the increasingly opaque information technologies and the rapidly rising amount of computation with significant sustainability questions and unknown societal implications. This is a very timely framework given the generative AI revolution currently underway. The author proposes changes in direction in system dynamics and critical system thinking conceptual frameworks that often show nonlinear counterintuitive effects from minor leverage point influences.
The book's first part discusses the societal impacts of computing systems, including sustainability, societal debts, and an exciting discussion on "problemism," a term that the author uses to describe our preoccupation with problem-solving. In the second part, the author presents new system design approaches focusing on sustainability and justice in the context of critical systems thinking. The author discusses value-neutral, rational decision-making concepts and critical requirements engineering from the systemic point of view. Finally, the author proposes a reorientation of professional experience and presents professional ethics codes in computing along with their societal impacts.
The book is broadly addressed to readers within the computing and neighboring fields. It is a rich discussion with many alternative perspectives and intellectual practices currently shaping computing.

#sustainablecomputing #sustainability #sustainablesystemdesign #ethicalcomputing #systemdesign #LLMs #generativeAI #AI #systemdynamics

Reinforcement Learning

 

After several "AI winters" where there was certain pessimism about our ability to improve the performance of intelligent systems. Today, AI is flourishing thanks to machine learning (ML) advances. Distributed reinforcement learning is one of the essential technologies that can support ML innovation. This book discusses distributed aspects of AI agents where multiple systems seek to maximize their reward functions and contribute to making decisions. The authors assembled an impressive collection of materials that take the reader through analyzing reward distribution in the context of incremental and temporal-difference learning.

#AI #distributedcomputing #reinforcementlearning

Generative AI in Education

 

Generative AI in education has enormous potential. For example, Wolfram's GPT is a step towards revolutionizing education in STEM areas. Automating perception at deep semantic levels, adjusting the level of comprehension based on previous interactions, and combining multiple channels of human telemetry are quick examples of what happens when we exchange knowledge with LLMs. We have a unique opportunity to build systems supporting extensive conceptual discussions about relationships between streams of thought from multiple domains. Even the hallucinations can be harnessed as generative hypotheses subjected to a scientific verification process.

#generatieAI #education #AI #ML #LLM

https://www.cryptopolitan.com/empowering-education-with-openai-innovations-transform-learning-with-ai/

Fixing American Cybersecurity

 Cybersecurity is one of the most critical areas of our lives. Without our technical digital infrastructure functioning as designed, all vital parts of the economy are vulnerable. Unfortunately, many systematic cyberattacks have exposed critical vulnerabilities in our infrastructure. This book is an in-depth look at our adversaries and their strategies. The editor, Larry Clinton, edited an excellent set of chapters that discuss ways to handle the growing cybersecurity problem. 

There are no simple solutions. There are several extensive discussions about the complexity of building secure devices and implementing effective standards. After reading this book, it is apparent that any successful cybersecurity initiative on a national level needs to consider technical, economic, political, legal, and ethical dimensions. Moreover, rapid progress in areas like AI or computing platforms creates asymmetries in the cybersecurity field that can be exploited. 

Given the complexities, the authors stress that public-private partnerships are the only way to move forward. Also, increasing educational efforts combined with market and business incentives should be a part of the strategy. 

The content presented in this book is very up-to-date. The content will provide an excellent overview for readers at all levels.

  

Intelligent Systems for Instruction and Tutoring

The iTalk2Learn tutoring system was quite an interdisciplinary project that combined ML, instructional design, NLP, educational psychology, and many others. Chapter 5 in #AI, The Rise of Lightspeed Learners, covers several essential aspects of using AI in human interactions.

It appears that the latest #LLM-based projects in conversational design, instruction, or other forms of interaction might need a similar engagement scope. While the initial applications of LLMs, GPT, and similar technologies shine in many areas of #NLP, conversational applications lack dialogue design, tone, and engagement. The interactions with #metaverse #avatars or bots are often clumsy and do not feel engaging.

There is a large amount of research that focuses on AI-assisted instruction and tutoring. Key issues include proper content sequencing, task management, hint design, and feedback analytics. Instructional design is often based on several principles, including domain-specific scenarios, conversation (#chat) management, content characteristics, skills, and difficulty-level modeling for individual domains.

Another area of research in AI-driven instructional systems is concerned with modeling students’ emotions. The models can be created explicitly from surveys or implicitly from user input sentiment analysis.

#ai #chatgpt #instructionaldesign #tutoring #tutoringsystems #aiassistant #ML #deeplearning #instructionaltechnology

LLM Finetuning with RAG (Retrieval Augmented Generation)

Enterprises look into capitalizing on the unquestionable qualitative change in generative AI. The central problem of manipulating or combining domain-specific knowledge or local data with the foundation #LLM pipelines is significant. The solutions generally rely on "finetuning" procedures. A few leading methods include transfer learning, #BYOM, and several sequential architectures. #RAG (retrieval augmented generation) is a way to improve LLMs. This paper presents excellent results based on standard evaluation procedures.

A pre-trained document retriever (DPR), used by the authors, is a query understanding and retrieval procedure. A pre-trained #seq2seq encoder/decoder processes the context, which is typically containing the original query and the retrieved (context) documents. The encoder/decoder is based on #BART in this paper.

The parametric memory algorithm that uses a pre-trained "sequence to sequence" #transformer is well-documented in the ML literature. A non-parametric memory component explicitly contains vectorized content from large corpora. Sometimes those representations are simply called document indexes. The characteristics of the solution that relies on non-parametric memory are huge. There is no re-training necessary if we change the source of knowledge. On the other hand, BART or other parametric models need to be re-trained if we need to unfreeze the knowledge base. #ai #training #learning #finetuning #LLM #llms #paremetricmemory #ml

https://arxiv.org/pdf/2005.11401.pdf