Nicolas M. Chaillan’s Post

🚀 Exciting News! 🚀 We're thrilled to announce a new plan that makes secure #GenerativeAI more accessible! 🌍 🔥 Introducing our $15/mo/user plan with 8M GPT4o tokens (200,000 Ask Sage tokens)! 🔥 This game-changing offer is perfect for teams on a tight budget who want to unlock the full potential of Ask Sage. With this plan, you'll get access to ALL our features and products, including In a Box (CMMC, ATO, FedRAMP, Acquisition...) and our FedRAMP high capability. Here's what you'll enjoy: ✅ Multi-modal capabilities (text, audio, image) ✅ Access to 20+ LLMs, including GPT 4o, GPT 4-32K, Claude 3, LLMA3 ✅ Dozens of file ingestors and connectors ✅ Access to plugins/agents, custom personas, and prompt template library ✅ Live+ feature, and so much more! But hurry, we're not sure how long this offer will be available or whether it will make long-term financial sense for us. Don't miss out, subscribe now and take advantage of this incredible opportunity! And that's not all! We have some exciting news for our Government customers too: 🔹 GPT 3.5 Gov now supports 16K tokens (input) (40 pages), 4096 output. 🔹 GPT 4 Gov now supports 128K tokens (300 pages!), 4096 pages. These enhancements are a game changer for task automation, including our In a Box iterative generative AI document generation! Ready to experience the power of Ask Sage? Register now at https://chat.asksage.ai and unlock a world of possibilities! #AskSage #AI #ArtificialIntelligence #MachineLearning #Innovation #Technology #SubscriptionPlan #GameChanger

TCO and running costs 💵 are paramount to running a big Ai 🤖 shop. Mixture of Agents (MoA), a novel approach to harness the collective strengths of multiple LLMs. First heard in an article from Snowflake ❄️ and now it seems to be ubiquitous 👍 MoA adopts a layered architecture where each layer comprises several LLM agents. These agents take the outputs from the previous layer as auxiliary information to generate refined responses. This approach allows MoA to effectively integrate diverse capabilities and insights 📈from various models, resulting in a more robust and versatile combined model. 🤖 To effectively leverage the collaboration of multiple LLMs, we categorize their roles based on their strengths in different aspects of collaboration: ✅Proposers: These models generate initial reference responses. While a proposer might produce a high-quality response on its own, its main value lies in offering nuanced and diverse perspectives that serve as valuable references for the aggregator. ✅Aggregators: These models synthesize the different responses from the proposers into a single, high-quality response. A great article and work from Together Ai What appeals me the most is that MOA gives more power ⚡️and still less cost 💲 . The benchmarks domes against GPT4O , current gold shandar proves it ! #LLM #MoA https://lnkd.in/g-MEy9VV

🚀 New Case Study: Revolutionizing Documentation Search with AI! 🚀 We’re excited to share our latest collaboration with WorkFusion, the leading provider of AI digital workforce solutions for Fortune 500 enterprises, banks, insurance, and financial services companies. 🔍 The Challenge: With over 10 versions of each product and extensive documentation, finding necessary information during training was a daunting task. Products often intersected, making it challenging to understand the structural connections within the documentation. 💡 Our Solution: We’ve seamlessly integrated an AI assistant into Moodle’s search, providing users with comprehensive support. Key features include: AI-Powered Server Enhanced Search Results User-Friendly Navigation Popular Queries Tracking This AI-driven approach has significantly enhanced the efficiency and ease of accessing critical information during training, promoting a more user-friendly experience. The client received a unique AI integration in their Moodle platform, enabling users to obtain information easily and quickly. Our solution also supports the continuous improvement of both the products and their documentation. Discover how we’re revolutionizing documentation search with AI in our latest case study: https://lnkd.in/d3dHtBn7 #AI #Innovation #Documentation #CaseStudy #TechInnovation #SmartApp #WorkFusion

For the LLM curious: Encoder - Decoder - Tokenization, Embeddings… What are those? How are they different? How do they connect? Here are some quick definitions that may have been confusing you when it comes to understanding how LLMs work under the hood. Hooray for internet search! 🔸Encoder is a compressed representation of input data that a machine can easily process. The encoder models the context of an input text, often responsible for extracting relevant information from the input. (Source: I’m paraphrasing multiple sources here) 🔸Decoder simply auto-regressively (I know, big word) reconstructs the output in human-readable shape based on encoded context from the encoder. (Source: I’m paraphrasing multiple ones here) 🔸Tokenization is the process of splitting the input and output texts into smaller units that can be processed by the LLM AI models. (Source: Microsoft) 🔸 Embeddings are vectors or arrays of numbers that represent the meaning and the context of the tokens that the model processes and generates. (Source: Microsoft) So basically, encoding is a form of embedding technique to transform the data in a representational state that machines love to consume! (My interpretation) The way data travels through LLMs is like this, as far as I understand: 🔹Gets tokenized 🔹Gets encoded 🔹Gets decoded Well, there you have it. This little exercise is in fact for my own sanity. What have I missed? What should I change? Image source: TeeTracker article. Check it out below. #genai #technology #machinelearning #artificialintelligence Click #linkedangle and follow for more content Set the 🔔 notification on my page, don’t miss a post!

Last month I finally broke through and earned more than the double the subscription cost for Medium! The cutting edge is hard. It's where things usually have a vocabulary, but the way things work are still unclear to most, including the ones that makes things tick, like me. Today I had the privilege of a customer onboarding session for the DFRNT Platform with a true expert in the field! Someone that knows more than most in his area of knowledge, which is connected to what I do We talked about many things, including representations of hypergraphs in the platform, and the use of JSON-LD for their representation. The expert shared how polished and complete the experience in the platform felt and it made my day. I always see the things that are missing, and my roadmap items. Not what is already in there and working great. He did 🙏 So I decided to write another deep dive in my series on hypergraphs and knowledge graphs based on some newfound insights. And to another person that helped me recently, I want to thank him for the spark to continue writing. Thank you Andrew Padilla for the helpful feedback that helped me get back on track with my writing and knowledge sharing! Hypergraphs part 2 (friend link) is here: https://lnkd.in/dENWcnJH Your support in sharing and commenting my articles helps me more than you think. Thank you dear friends! ____ Want to learn more about High Quality Structured Information? Follow me at Philippe Höij + 🔔 to get my every post. Digital Twins of Information expert, co-founder of Twinfox.AI and DFRNT, platforms for model-based validated information using explainable AI!

🤖 Missed our AI-powered Test Data Management demo? No problem!    Check out some highlights from our live demo to see how K2view is revolutionizing software testing: ✅ AI-powered PII discovery and masking in action ✅ Instant test data on demand – see how easy it is to speed up releases ✅ Compliant, AI-generated synthetic data that looks and acts real Watch the full demo now to discover how K2view AI can transform your test data processes 👇  https://lnkd.in/eQJGwG_c   #TestDataManagement, #AI, #SoftwareTesting, #DataCompliance

🚀 Unlock the potential of Enterprise AI! Join our upcoming webinar to discover how to customize Large Language Models (LLMs) & unlock valuable business insights. 🧠🌐 In 60 minute's we'll cover: ✅ Integrating data into LLMs for transformative insights. ✅ Speeding up LLM deployments with data pipeline automation. ✅ Retrieval Augmented Generation (RAG): Enhancing LLMs for richer, context-aware answers. ✅ Real-World Use Cases to leverage LLMs for unmatched business outcomes. Register now to attend live or receive the webinar replay. 👉 https://hubs.li/Q028xVWY0 #EnterpriseAI #DataPipelines #LLM #mlops #llmops #DataPipelineAutomation #dataengineering #dataleaders

An absolutely critical element of being able to leverage data with AI in organizations is how quickly you can make data available to AI in a way that's safe, and that means making your systems as interoperable as possible. Folks will say that if you want to integrate with a legacy system, you need to standardize your connections by adding an API. While doing a bunch of refactoring to add APIs adds billable hours and software licensing revenue to contracts, it simply isn’t necessary, or even feasible in most existing organizations. Standardization doesn't make you go faster, it actually slows you down unless you're starting with a clean slate. You don't want standards, you want modularity - reusable, recomposable building blocks in your system, like Legos. Standardization and modularity are not the same thing. Standards aren’t components, they’re rules, and they don’t work to increase interoperability and process efficiency because: a. They have to be enforced from the top down which is incompatible with the way the IT industry works (unless it benefits the providers of technology over the buyers or the competition) b. They become obsolete over time. As new standards emerge, you end up with systems that don’t talk to each other anyways. If someone tells you standards are the solution to getting systems to talk to each other, they’re simply wrong. The hidden wrong assumption is that you can update all your legacy systems to comply with the new interface standard. Not gonna happen. #AI #API #Datafabric #DevOpsForData

An absolutely critical element of being able to leverage data with AI in organizations is how quickly you can make data available to AI in a way that's safe, and that means making your systems as interoperable as possible. Folks will say that if you want to integrate with a legacy system, you need to standardize your connections by adding an API. While doing a bunch of refactoring to add APIs adds billable hours and software licensing revenue to contracts, it simply isn’t necessary, or even feasible in most existing organizations. Standardization doesn't make you go faster, it actually slows you down unless you're starting with a clean slate. You don't want standards, you want modularity - reusable, recomposable building blocks in your system, like Legos. Standardization and modularity are not the same thing. Standards aren’t components, they’re rules, and they don’t work to increase interoperability and process efficiency because: a. They have to be enforced from the top down which is incompatible with the way the IT industry works (unless it benefits the providers of technology over the buyers or the competition) b. They become obsolete over time. As new standards emerge, you end up with systems that don’t talk to each other anyways. If someone tells you standards are the solution to getting systems to talk to each other, they’re simply wrong. The hidden wrong assumption is that you can update all your legacy systems to comply with the new interface standard. Not gonna happen. #AI #API #Datafabric #DevOpsForData

🚀 Take your Large Language Model (LLM) applications from prototype to production with Proco! As a trusted development platform, Proco offers a suite of tools for prompt engineering, semantic search, version control, quantitative testing, and performance monitoring. 💡 With Proco, you can quickly develop an MVP by experimenting with different prompts, parameters, and even LLM providers to find the best configuration for your use-case. 🔧 Proco acts as a low-latency, highly reliable proxy to LLM providers, allowing you to make version-controlled changes to your prompts – no code changes needed. 📊 Proco collects model inputs, outputs, and user feedback, building up valuable testing datasets that can be used to validate future changes before they go live. 🔍 With Proco, you can see exactly what you're sending to models and what they're giving back, view metrics like quality, latency, and cost over time, and use the best provider and model for the job, swapping when needed. 🌐 Join the Proco community today and focus on your customers, not complex AI tooling. Visit https://buff.ly/3DFTZSm to learn more! #AI #LLM #Proco