Josep Ruano Bou

Brace yourselves for another round of hype and hysteria mixed for us by OpenAI that will happen this fall. In another great article, Ignacio de Gregorio sums up what to expect from a new generation of OpenAI models and marketing materials as their Chief Hype Officers truly mastered the art of getting people's attention to promote company products. Everything indicates that we will get a model (models?) trained on a large dataset of reasoning task examples and with reward models - that will imitate thinking like a seasoned illusionist. We will get another iteration of the "intelligent database" that skillfully imitates raw intelligence (which requires active search / active reasoning instead of retrieval). It will impress the users with better capabilities than anything on the market today, and disappoint the practitioners with failing miserably on complex tasks that can't be retrieved from the model's associative memory. Link to the article in the comments 👇

22

23 comentarios

The Allen Institute for AI (Ai2) released OLMo 2, a fully open-source family of large language models (LLMs) that rivals Meta’s Llama models while providing unprecedented transparency by sharing its training data, weights, and code. The model employs a unique training process, using 3.9 trillion tokens from diverse sources and refined datasets to ensure high-quality responses and domain-specific knowledge. By promoting fully open-source models, OLMo 2 sets a new standard for transparency and collaboration in AI, paving the way for improved open-source alternatives to proprietary solutions like ChatGPT or Google Gemini. https://lnkd.in/ezeAqbim What potential advantages do you think fully open-source AI models like OLMo 2 offer over proprietary alternatives in advancing AI research? #OpenSourceAI #LanguageModels #AIInnovation #OLMo2 #ArtificialIntelligence

1

#GenerativeAI is a type of #AI systems that is capable of creating new content such as images, text or music based on patterns learned from existing data. An interesting speech by Professor Miralla Lapata that describes Generative AI and how it works. Also, she talks about other topics such as #ChatGPT, Large Language Models #LLM and how their impacts will be on society. https://lnkd.in/dRN97FqT

12

4 comentarios

Ever wondered how people are really using #GenAI? Microsoft colleagues from Western Europe recently shared their personal experiences of what a day in the life with a #GenerativeAI #Copilot looks like. From our CEO of Spain Alberto Granados (https://lnkd.in/eZeN3Hxt) over Tina Dunne EA of our Ireland MD (https://lnkd.in/enWEbEg9) to Ronan Geraghty ☁ our Ireland COO (https://lnkd.in/e9a-B-MA). We already know from early-adopters of Microsoft #Copilot that 71% see the AI helping them save time on mundane tasks and 68% said it improved their quality of work (https://lnkd.in/e_DbzvWW). Hearing first-hand where the #ArtificialIntelligence enable #FutureOfWork is going, makes it much more tangible though. If you are curious about more examples of people using Generative AI in their daily lives, have a look at these from teams working Bayer, EY, Wipro and more: https://lnkd.in/e6MqEK8u And in case you're want to know more about how to operationalize Generative AI copilots at scale, have a look at Jared Spataro's insights on the "11x11 tipping point" (https://lnkd.in/e7Rnv5aT) as well as the importance of empowering internal AI champions (https://lnkd.in/ewm4xhcG) CC: Richard, Arjun Vir, Theodora, Dr. Martha, Prof. Dr. Ingrid, Dr. Debashis, Alvin, Efi, Sanjeev, Enrico, Aditya Ranjan, Dinis, Zvonimir, Matt, Zain, Alex, Conor, Guido, FRANKI, Noelle, Tarja, Pablo, Medo, Fabio, Kamales, Tyler, Dr. Marcell, Chuck, Ravit, Joy, T. Scott. Ravi, Philippe, Xiaopeng

116

13 comentarios

IBM and the Spanish Government have agreed to a collaborative partnership to further AI strategy while building Spanish language AI models. #DataScience #AI #ArtificialIntelligence https://hubs.li/Q02t15Wq0 If you enjoyed this post, follow me for more insights on AI and data science!

5

1 comentario

"A Primer on the Inner Workings of Transformer-based Language Models" by Javier Ferrando Monsonís, Gabriele Sarti, Arianna Bisazza, Marta Ruiz Costa-jussà "The rapid progress of research aimed at interpreting the inner workings of advanced language models has highlighted a need for contextualizing the insights gained from years of work in this area. This primer provides a concise technical introduction to the current techniques used to interpret the inner workings of Transformer-based language models, focusing on the generative decoder-only architecture. We conclude by presenting a comprehensive overview of the known internal mechanisms implemented by these models, uncovering connections across popular approaches and active research directions in this area." Paper: https://lnkd.in/eKKTTi-z #transformers #languagemodels

34

Thanks to Codeplay Software and the AERO Project, we are able to accelerate Java programs on CPUs via the #TornadoVM framework and the oneAPI Construction Kit (OCK): https://lnkd.in/dHk2PrDs Acceleration of 3.4x over the execution with Java Streams on the same CPU (32 cores) for running Matrix Multiplications of size 1024x1024. I will probably write a detailed blog post about how to use it and some performance numbers. The good thing about the OCK is that it is a potential CPU implementation of OpenCL, not just for x86 CPUs, but also for ARM and RISC-V accelerators, and even custom accelerators. Stay tuned! #java #acceleration #ai #cpus #gpus

44

Start Your Week with an in-depth AI article "Must-Read"! 🚨 Kicking off Monday with my latest Forbes article, where I explore Dario Amodei’s ambitious vision for the future of #AI. As CEO of Anthropic, Amodei argues that AI could compress 50-100 years of scientific progress into just a decade. While AI presents immense opportunities, the risks are just as significant, especially when it comes to governance and preserving human agency. This is a pivotal moment for us to shape a future where AI truly benefits society. I encourage you to dive in, share your thoughts, and join the conversation about the promise—and peril—of AI. Read the full article and start your week with a thought-provoking discussion. 👇 https://lnkd.in/gvhQUpgs If you enjoy articles like this one, please consider subscribing to my column on Forbes or resharing to spread the discussion. #AI #Technology #Innovation #Leadership #Governance #Healthcare #EconomicDevelopment #FutureOfWork #Ethics #HumanAgency

18

1 comentario

SWARM Community has built an extensive library of interviews and webinars with experts on #AI and #data We are thankful that Joaquin Melara is sharing this valuable content in the Teckedin knowledge base ___________________________________- 'Join Santona Tuli, Head of Data at Upsolver, as she talks through the different levels of AI initiatives and how much each level of involvement requires in terms of commitment. Her broad and extensive knowledge and experience working with data is a testament of her expertise in this area, from working on massive event data from particle collisions at CERN, to improving the development experience for data scientists and machine learning operations in Astronomer, to her present work leading data engineering and science operations. Throughout it all, Santona has been passionate about empowering others to build scalable and useful AI systems.' ABSTRACT: Incorporating AI into the business doesn't look the same for every business. This talk will enumerate the different levels of AI initiative and incorporation into business products and practices. We'll dig into the amount of commitment required for each level and what expected returns might look like. You will come away from the talk with a criteria for determining the level of AI investment that's right for your business. _______________ #artificialintelligence #analytics #dataanalysis #automation #datascience Santona Tuli, Ph.D. https://lnkd.in/gpNU7nBd

5

2 comentarios

Entering a new phase of the process: Validating the draft Open Source AI Definition. Which of these models comply with the principles of Open Source? Volunteer to analyze them! https://lnkd.in/dXaYseBn #OpenSource #AI #ML

8

Third and last day of Machine Learning Summer School and we're heavy into Artificial Neural Networks (ANNs). Thanks Manuel A. Palacios for doing most of the work today! #machinelearning #regression #classification #matlab #python #learnfromhome #shortcourse #processanalytics

26

1 comentario

I found the article "Do You Really Need a #ChiefAIOfficer?" (https://lnkd.in/dWY4D584) from MIT Sloan Management Review to be particularly insightful. The discussion on the role of a CAIO in navigating the complex and rapidly evolving AI landscape is timely and crucial. The article highlights several compelling benefits of appointing a CAIO: ✅ reducing AI #fragmentation, ✅ driving transformational #change, ✅ managing AI #risks centrally. ✅achieving scalable solutions across the organization. However, the article also outlines some challenges: ⭕potential cross-functional conflicts ⭕risk of overemphasizing AI as the go-to solution for every business problem. While these are valid concerns, I believe that with clear mandates, strong #communication, and a #strategicfocus on AI's business impact, the benefits far outweigh the drawbacks. Ultimately, the decision to appoint a CAIO should be closely aligned with the organization's AI maturity and strategic goals. 📢What are your thoughts on the need for a CAIO in today's AI-driven world? #AI #Leadership #DigitalTransformation #businessImpact

36

Report: ENHANCING VALUE CREATION THROUGH INTEROPERABLE DATA SPACES https://lnkd.in/dbK8yVqA I have to read it but the topic of interconnection of data spaces is a hot trend and with lots of work to be done to really achieve it in a seamlessly way. Don't you think so?

7

3 comentarios

Led by Rodrigo Laigner, it is available our recent work "An Empirical Study on Challenges of Event Management in Microservice Architectures", which provides a characterization of event management practices and challenges mined from Stack Overflow. Take a look > https://lnkd.in/dG8gsGJc

11

AllenAI have announced OLMoE, which is a fully open (code, data and weights) state-of-the-art #LLM leveraging sparse #MixtureofExperts (MoE), which combines the strengths of ensemble learning and neural networks to handle complex problems with diverse data patterns. OLMoE-1B-7B has 7 billion (B) parameters but uses only 1B per input token. It's pre-trained on 5 trillion tokens and this is further adapted to create OLMoE-1B-7B-Instruct. It models outperform all available models with similar active parameters, even surpassing larger ones such as Llama2-13B-Chat. #HuggingFace link 👉https://lnkd.in/edhZpKhh Smaller smarter models are really having their day. #ai #enterpriseai #smallLLMs Paper: 👉 https://lnkd.in/egrtZgfz

7

CTO & CAIO together but separate, a CAIO is more than a technologist, a profile that understands the technology but also the problems it faces in society, the challenges that AI poses, a business knowledge with a close working relationship with staff and a team capacity to lead, engage and unite different possibilities and capabilities.  https://lnkd.in/d42FnY4i

3

Multi-modality #RAG is getting a traction, really. I enjoyed reading this paper “ColPali: Efficient Document Retrieval with Vision Language Models” https://lnkd.in/dDZJdphE #ColPali is a model based on a novel model architecture and training strategy based on Vision Language Models (#VLMs) to efficiently index documents from their visual features. It is a #PaliGemma-3B extension that generates #ColBERT- style multi-vector representations of text and images. The integration of the #ColPali project and the #ColQwen VLM model in downstream Vision Retrieval-Augmented Generation applications also involves the use of vector stores to manage image embeddings as long as they support multi-vector architecture or ColBERT-style representations, or simply store the indexes into a disk! ColPali utilizes advanced vision language models to convert document page images into rich semantic embeddings, capturing both textual and visual information. These embeddings are stored in vector databases, facilitating efficient retrieval based on visual semantic similarity. When a user query is processed, the system searches the vector database for embeddings that closely match the query, enabling the retrieval of relevant document pages (as images) without relying on traditional text extraction methods. Then a multi-modal model can process the extracted images to curate a text response to the user. This approach enhances AI systems by integrating visual data, leading to more accurate and context-rich outputs. This approach is particularly beneficial for processing complex documents where visual elements like layouts, images, and charts are crucial. By embedding documents as images, Vision RAG retains comprehensive information, reducing the loss that can occur during text-only parsing. Several GitHub repos implement the paper, and you can fork any of them and expand with Ollama to pull any recent multimodal model like #Llama 3.2 (11B or 90B) and others for the “generation” stage of the ColPali flavor of yours. Sample project is here (localGPT-Vision): https://lnkd.in/djUu7Gt6 However, implementing Vision RAG involves some costs, especially during the indexing stage. Processing and embedding visual data require computational resources, often necessitating powerful hardware such as high-performance GPUs or specialized accelerators. This leads to increased initial setup expenses and ongoing operational costs. Additionally, managing large volumes of image embeddings demands considerable storage capacity and efficient retrieval systems, further contributing to the overall investment. Despite these challenges, the enhanced capabilities of Vision RAG in handling multimodal information make it a valuable tool for applications requiring deep understanding of both textual and visual data.

28