Capgemini Applied Innovation Exchange - San Francisco’s Post

If you missed today's live event on AI Agents... watch a replay here: https://lnkd.in/ggx3nRu3 We shared an updated list of noteworthy AI Agent startups and recently published papers on the topic. You can see the list of startups and papers below!

Many thanks Andreas Sjostrom for the inspired talk about #AI Agent which is the next evolution of using AI in system. And thanks for the screening of the new start-up ecosystem under construction.

Great Government Technology interview with Brendan Babb, Anchorage CIO. He discusses practical insights and real-world experiences with Generative AI in city government. He mentions using it to simplify complex documents containing land use codes and building codes to make them easier for the public to understand. He also discusses the need for collaboration among government entities and industry partners to address challenges such as privacy and security. Babb emphasizes the importance of building AI literacy and in the near future, predicts that many local governments will adopt generative AI for various use cases, such as, object detection, handling unstructured data, summarization, translation and image processing, which will greatly impact city operations and resident experiences. https://lnkd.in/gZDyBRGE #governmenttechnology #innovation #generativeai #residentexperiences

This is an important article related to it's a very important research that I think should be red widely by every artificial intelligence foundation company, and every artificial intelligence lab. It's based on Research from we're going on at CSAIL at MIT. I believe it will bear fruit rapidly for domain specific applications in narrow AI... and might have an implications in creating forms of specic "common sense"-grounding for both robots and other algorithms. Given this research ... this capability may allow a robot to tie his shoes in thousands of different ways with thousands of different everyday objects.. that make sense to tie shoe with. One of the elements of research just answers the question, "What if an algorithm knew the basic elements and the nature of each word and n-gram it was predicting?" So, If I knew what a bowl was, and knew what a hat was, and instantaneously knew that you could use a bowl for a hat uncertain circumstances. Specified-Common-Sense-Grounding, could sort out the problem described by Marovec Paradox, fairly efficiently. If you don't understand this paradox please do some research. This is probably the reason you're confused about the applications of AI, if you are confused or in a state of powerful skepticism. This is an hypothesis. ☝️ If this hypothesis is true, CSG should measurably reduce the error rate of all LLMs, in a similar but different way that RAG does. it's obviously more efficient than a stack of domain data or reddit conversations, used to create special embeddings. Just like with other conversational data sets from Twitter, email, facebook, threads Instagram etc, Common Sense it's generally better articulated in conversation than in static rules or even knowledge graphs. You'll have to think about that one, it's a complexity that is dynamic in nature and not predictable with linear thinking. I do not believe there is one form of Common Sense because it is mutable across time. what was Common Sense 100 years ago is not and can never be common sense today. That does not make sense. Stay tuned for some summaries of the articles published by the scientists that are referred to in this article that I will create. If you want a simpler version of this article was posted on popular mechanics, but it makes errors in the summary. (It was likely summarized by AI and generated from this article, allegedly). This type of work is for architects, and information scientists as well as the new evolving area of science I'm going to call Knowledge Science. Knowledge Science combines, all of Data Science, Information Theory, complexity Theory, Graph Theory, Probability+Game Theory, and is explained nicely with Euclid and geometry and it's overlap, Set Theory. If Knowledge Science is a differentiated science from ontologies and telologies, it should be able to be consistent with and predicted by Turing's work on computing as well as Von Neumann's and Weiner's. This is a Big Hypothesis and a Big Idea.

What is FLUX.1, a new open-source AI image generator to take on Midjourney?: Built by former Stability AI engineers, the photo-realism of pictures created using FLUX.1 has sparked a lot of buzz online.

Excited to share insights on implementing map reduce in Lang graph for large-scale AI tasks! 🤖 Lang graph is a powerful library for building stateful multi-actor applications with large language models, offering unparalleled advantages in the realm of AI development. 🌟 The process of decomposing tasks for parallel execution using the map reduce algorithm, showcase the importance of statefulness in Lang graph. 📈 The end result? A ranked list of solutions for improving public transportation in urban areas - a pivotal application for our growing cities. 🏙️Let's revolutionize the world of AI together! 💡 #AI #LangGraph #MapReduce #PublicTransportation #GitHub #AIDevelopment

How do AI-systems impact everyday information practices, for example in search and recommender systems? And how do they influence people (and the planet) in ways that often are left unrecognized? Glad to start the year with a publication in JASIST (The Journal of the Association for Information Science and Technology, ASIS&T) on ”Artificial intelligence in the information ecosystem: Affordances for everyday information seeking” by the excellent Noora Hirvonen, Ville Jylhä, YUCONG LAO — all at the University of Oulu — and me (Lunds tekniska högskola, Lunds universitet). https://lnkd.in/dMQg99ha In brief, drawing from the fields of library and information science and science and technology studies and research on affordances, the article responds to the need for more nuanced theoretical insights on the impact and implications of AI on information practices and seeks to develop conceptual tools with which to examine the co-evolution of humans and information systems from a systemic perspective. I found it particularly relevant and interesting to look at generative AI through the lens of Davis’ work (e.g. “How artefacts afford”) and to use Star’s (1989) notion of “The Durkheim Test” (as opposed to a Turing test) in order to argue for the need to consider how AI tools better can meet the values, norms, and needs of communities. Thanks for your great work, Noora, Ville and Yucong! This is for a special issue on “Healthier Information Ecosystems”.

The landscape of GenAI-powered applications in production has been revolutionized by the latest advancements in Prompt Engineering and RAG techniques. These innovations have dramatically enhanced how we leverage generative AI to meet diverse and complex needs. For scenarios where RAG is essential, we now have sophisticated techniques to both validate the content and adapt it according to the user's query. This ensures a significant boost in efficiency and precision, delivering responses that are not only accurate but also highly relevant to the user's context. In the series of Posts below, I talk about three recent techniques for improving RAG-based applications. The three techniques presented and discussed are: Self-RAG: https://lnkd.in/dHuP72qA Adaptive-RAG: https://lnkd.in/dQ6JhCyZ Corrective RAG (or CRAG): https://lnkd.in/ddikFM3N Check out the posts and let me know in the comments, which techniques have you been using to boost the performance and reduce costs in your RAG-based applications? #PromptEngineering #RAG #GenAI #ArtificialIntelligence #TechInnovation #MachineLearning #AIinProduction

Thanks to AICamp for hosting a fantastic AI meetup 🥳. Insightful talks by Arun Prasad and Alan Huynh, and a great time of connection and conversation with the attendees. 👉 The first talk was by Arun Prasad, who discussed vector embeddings, semantic search and evaluating embedding models. - The live demo opened my mind to the power of semantic search versus elementary text-search algorithms. During the live demo, Arun showcased how a query on Stack Exchange that received no results, when passed to a semantic search system, returned numerous results, some even being from Stack Exchange itself! - I found Arun's comment on open-source AI to be insightful. He said that open-sourcing a technology is a great way to validate its usefulness, especially when there is no specific engineering marvel about it. 👉 The second talk was by my buddy Alan Huynh, who explored why no machine learning can sometimes be better than using machine learning. - Alan trained models to classify whether a number is part of the Fibonacci sequence or not. Turns out there's a mathematical rule to figure this out: if N is a Fibonacci number, then 5N^2 + 4 or 5N^2 – 4 must be a square number. As expected, the simple math rule got 100% accuracy, while the XGBoost and KNN algorithms received worse performance. - For detecting clouds in an image, turns out that a simple threshold function that takes 10 minutes to write produces similar results to a segmentation model that took 2 months to develop. This highlights the importance of creating baselines and testing against these baselines. Overall very insightful talks and some fantastic conversations with the attendees. Looking forward to the next one. #ai #meetup #melbourne #aicamp

This is one of the best descriptions of the application of AI in Fin Tech and other industries I’ve seen.