VendEx Solutions’ Post

Really exciting news from VendEx Solutions! This is defining the next phase of digitizing the data business not only for financial markets, but for all data.

Deciding which LLM to use is hard. Trying to factor in the legal, ethical, or risk considerations of each model is even harder. A lot of this information is hidden deep in 50+ page technical reports, and often the key information necessary to determine whether a model is appropriate to use a task isn't disclosed. We're looking to help solve that problem by releasing our first ever Model Transparency Ratings. We analyzed the public documentation of the top 21 LLMs against the requirements of General Purpose AI Models from the EU AI Act and created a scoring criteria to identify which models may be riskier to use for AI products in the EU. ------- A few insights from our first set of ratings: 1️⃣ Models have been getting LESS transparent over time. Many providers like Meta, Cohere, and Mistral are less transparent about their newer models than their older ones. 2️⃣ Very few LLM providers disclose their data sources. I'd argue that understanding the data sources is MORE important to understanding the risks of an AI system than the model architecture, but we see decent model transparency, and poor data transparency. 3️⃣ "Open Source LLMs" do generally better, but we had to separate the idea of 'open weights' and 'open data' to better distinguish the risks and expectations. ------- Check out our ratings at: aimodelratings.com To read more about Trustible's methodology, check out our blog post here: https://lnkd.in/gShTuTGx We'll be expanding our ratings to include both new models, and include additional risk criteria. Share what you'd want to see in future versions of our ratings in the comments!

"Retrieval Augmented Generation: Everything You Need to Know" Retrieval Augmented Generation (RAG) is an emerging methodology for building Generative AI applications, particularly useful for enterprises looking to utilize private or custom datasets. RAG enhances the capabilities of large language models (LLMs) by integrating them with additional factual data from specific sources, addressing the limitations of LLMs that may struggle with data outside their training sets, leading to inaccuracies or "hallucinations" in responses. The RAG process involves data ingestion, chunking and embedding, query processing, response generation, and optional validation. Key advantages of RAG include reduction of hallucinations, cost-effectiveness, explainability, and enterprise readiness. Vectara offers RAG as a managed service, simplifying the development and deployment of GenAI applications while handling complexities and ensuring enterprise-grade security and performance. RAG is becoming the standard framework for implementing enterprise applications powered by LLMs, providing a robust solution for leveraging custom data effectively while minimizing risks associated with traditional LLM usage. For more details, refer to the full article here:https://lnkd.in/gHKkFeWE

A good synopsis. Lots of value can be delivered through smaller models focused on specific problems and domains.

AI adoption needs more standards and de-facto standards

Coming up for air after a series of *very* interesting discussions with F500 customers on how their companies are investing in GenAI. Here are 3 counterintuitive (maybe) insights I've noted: 1. F500s are realizing just how underutilized their investments in Classical AI have been. This raises questions around how to make the most of what they have vs chasing the shiny object. GenAI and LLMs, perhaps, are not the end-all-be-all for now - more are examining Hybrid AI frameworks from bleeding edge tools like Puller AI. 2. F500s are cleaning house (their data), spending $ millions to harmonize, centralize, unify, prepare their most valuable asset for AI-readiness -- BUT trading in invaluable calendar time to do so. Many are becoming wary of vendor lock-in and time-to-value as the AI arms race blazes forward (side read: take a look at this engaging post by Ashu Garg (link below). This raises the question if there might be a dual-speed approach to AI transformations. (The answer is yes). 3. F500s are acknowledging the breakneck speed of progress in AI, and rightfully concerned about whether security, verifiability, and governance can keep up. Huge opportunity here as it's not immediately clear how wide the gap is between GenAI and compliance at the Enterprise - and at what rate that gap is closing or widening. #EnterpriseAI #GenAI #AIforBI #LLM #BigData

Power up your #RAG with agents and tools👇 Agentic RAG represents an evolution of traditional RAG by integrating AI agent architectures that enhance decision-making capabilities. I was going through this article written by Chris on Agentic RAG and I really enjoyed reading it. The article emphasizes that while traditional RAG focuses on retrieving information, Agentic RAG empowers AI agents to autonomously determine problem-solving steps through a defined processing loop. This involves retrieving data from multiple sources such as community support forums, product documentation, and internal knowledge bases. A critical aspect of Agentic RAG is the use of retrieval functions, which optimize how the AI interacts with its data sources by breaking down complex queries into simpler, more manageable functions. The author (Chris) highlights the importance of prompt engineering and structured responses to ensure accurate and reliable outputs from large language models (LLMs), which often struggle with reasoning. Furthermore, the article discusses the challenges of maintaining up-to-date information and the necessity for ongoing quality assurance in AI agents. Looking ahead, it suggests potential developments in multi-agent systems and autonomous agents that could further enhance the capabilities of Agentic RAG in various applications, including content moderation and quality assurance. Overall, Agentic RAG aims to create more intelligent and responsive AI systems capable of effectively assisting users. Read the complete article: https://lnkd.in/dwe7rQ4x

Power up your #RAG with agents and tools👇

Thanks for sharing this amazing article, Chris! 🙏🙌 The concept of #AgenticRAG is a fascinating evolution of traditional #RAG systems. 🤖🚀 Your explanation on how Agentic RAG empowers #AI agents to autonomously decide problem-solving steps through a structured processing loop is brilliant. 🔄💡 Unlike conventional RAG, which focuses primarily on information retrieval, Agentic RAG introduces a whole new dimension by incorporating #DecisionMaking and #ProblemSolving capabilities. 🧠✨ The emphasis on leveraging retrieval functions to break down complex queries into smaller, manageable parts is a smart way to optimize how AI systems interact with diverse data sources like community forums, internal knowledge bases, and product documentation. 🔍📚 This ensures AI agents are not just retrieving data but understanding it contextually for more accurate and relevant outputs! 🎯 I also appreciated your insights on #PromptEngineering and structured responses to improve reasoning capabilities in large language models (#LLMs). This is crucial for minimizing hallucinations and ensuring reliable outputs. 🔧🤖 The article perfectly captures the potential of these systems to go beyond simple retrieval and become truly autonomous, intelligent assistants. 🚀💪 Looking forward to seeing how #AgenticRAG evolves into #MultiAgentSystems and other innovative applications like content moderation and real-time quality assurance! 🔥 #AIInnovation #MachineLearning #TechTrends #KnowledgeManagement #AutonomousAgents