Margaret Warren’s Post

Yes, the fate of the planet depends on taxonomy! I humbly submit that the funding required could be provided by a ... tax tax. :) H/T Bob DuCharme for flagging this article, which I've provided to y'all as a gift link. https://lnkd.in/g-D7kMh8

I love deep learning. I wrote my first model in TensorFlow 1 (Who remembers it?) When I first started learning TensorFlow, it seemed so complicated to me that I almost got discouraged. But I was so fascinated by the power of deep learning that I felt, I cannot just let go. With support from my more experienced colleague (these little acts of help from others can really change our lives!), I was able to move forward. I believe that one of the most valuable lessons deep learning taught us is the power of representation learning. It's undeniable. But, there is also a darker side to what I call the "Deep Learning Culture". The Deep Learning Culture is based on a belief that by creating complex enough architectures and feeding them with enough data, we can answer any possible question. It teaches us implicitly, that the data generating process does not matter. The Deep Learning Culture loves extrapolating computational trends, Moore's law and other similar devices and uses them to argue that if we only had a bit more compute, a bit more data, we would undeniably achieve the mythical "AGI". Yet, there's a trap in this thinking. We know that ignoring the data generating process (or the structural properties of the data) makes it in general impossible to learn how our actions will impact the world, how to systematically reason, and more. When the tide goes out and we all discover who's been swimming naked, the entire community may pay the price in the hard currency of lost public trust. #causality #deeplearning #machinelearning #causalAI

What Could Go Wrong When We Start Using LLMs to Organize Knowledge? 7 Pain Points of GraphRAG Alright, tech enthusiasts and AI aficionados. We need to discuss GraphRAG, which Microsoft released yesterday. It's a solution to make LLMs to be more trustworthy by automating the extraction of a rich knowledge graph from any collection of text documents. Seems a good idea on paper, but it comes with a lot of challenges I tried to summarize below and make a small graph illustration about it for fun: 1) Data Quality: Inconsistent data, outdated information, and biased datasets are the triple threat to the foundation of GraphRAG + LLM systems. Garbage In, Garbage Out. We need proper data engineering and subject matter expertise upstream to ensure quality. 2) Retrieval Process: Irrelevant information retrieval, missing context, and information overload restrict the effective use of stored knowledge. 3) Graph Construction: Missing relationships, over-complex graph structure, and incorrect entity linking can compromise the integrity of the knowledge representation. 4) LLM Integration: Interfacing structured knowledge and LLMs don't prevent hallucinations, misinterpretation, and inconsistent reasoning if you don't transform the language problem into a code problem. 5) Knowledge Gaps: How do you handle novel entities? A lack of common sense from the LLM limits the system's ability to deal with new or nuanced information. 6) Scalability & Performance: Slow processing and high computational needs will create significant hurdles for practical, large-scale implementation if there is no strong foundation for building a holistic business ontology. 7) Ethical & Privacy Risks: Information exposure and bias amplification pose serious concerns for user privacy and fair application of the technology. How can you be sure that this system is grounded in your higher-level business processes? Look, I'm not saying we shouldn't use GraphRAG and LLMs. It's cool tech with amazing potential. But let's not kid ourselves: we need data engineers, subject matter experts, and people who understand ontologies to build the Knowledge Graphs that will feed the AI Assistants we are all heading to use in the future of work. So don't get distracted. Question everything, verify relentlessly, and don't use this new shiny shortcut to plan your business future.

This summer marks the ten year anniversary of The Long Now Foundation’s bar, cafe, and event space — The Interval. To commemorate The Interval’s Decennial, we’re reimagining The Interval as not just an award-winning cocktail bar, but as an exhibition on long-term thinking. Over the past decade, The Interval has become a hub for long-term thinking, a home for Long Now Talks, and a gathering place for great conversation and connection. But The Interval could be so much more. That’s why we’re excited to announce that we’re re-opening The Interval this September with a brand new exhibition featuring artifacts from Long Now’s history and a vision for our future. The exhibition will tell the story of long-term thinking in a way that connects deeply with both long-time Long Now members and with first-time visitors to The Interval for whom these concepts might be entirely new. At the core of the exhibition is a painting by Casey Cripe illustrating the long now: the past 10,000 years and the 10,000 years to come. The Interval will close on Wednesday, August 21 to install the new exhibition and reopen on Saturday, September 7. We also hope you’ll join us on September 14 and 15 for The Interval Decennial Member Weekend. Become a Long Now member to learn more. Whether you’re a regular at The Interval or still looking forward to your first visit, we’re excited to welcome you to our reimagined home for long-term thinking. https://hubs.ly/Q02Fj_Xn0

We're halfway through the year. Take a moment, small or large, to consider the Long Now — time not just on the scale of the year or even the decade, but on the scale of millennia and tens of thousands of years.

Panel Discussion: The Functionality of Fact-Based Systems Join us for an engaging panel discussion on "The Functionality of Fact-Based Systems," where we will dive deep into the core components and importance of content, information retrieval, and the systems that underpin our digital interactions. Featuring industry speakers from around the world including: - Beatrice Gamba, WordLift - Margaret Warren, Metadata Authoring Systems - Stratos Kontopoulos, Foodpairing AI - Thalia Barry, IBM Discussion Highlights: - Understanding Content and Its Importance: What is content and why is it crucial for an organization? What roles do information retrieval and the World Wide Web play? Why are search engines and search engine optimization vital? How do we make search engines accessible and consider other aspects of accessibility? - Exploring Fact-Based Systems: How can we describe a fact-based system? What is the importance of trustworthiness and explainability in these systems? What factors must we consider to ensure a reliable fact-based system? - The Devil is in the Details: What is metadata and why is metadata management essential for digital assets like images? What is knowledge representation and reasoning? What is the significance of semantic metadata and how does it enable human and machine interactions? What are the practical applications of semantic technology? - Measuring the Impact: How do you measure ROI through the positive impacts of adding value to an organization? How can we assess the prevention of negative impacts in society, penalties, and liabilities? Join us to uncover the intricate details and essential considerations for building reliable, trustworthy, and explainable fact-based systems. This discussion is a must for anyone interested in the intersection of technology, information management, and organizational strategy. Don't miss this opportunity to engage with experts and gain valuable insights into the functionality of fact-based systems.