Author Archives: Ben Lorica

GraphRAG represents an emerging set of techniques that merge knowledge graphs with large language models to enhance retrieval-augmented generation. However, the absence of standardization has led to a variety of implementations, each with its own unique strengths and challenges. In a previous post, I explored a system developed by Nvidia and Blackrock that integrates RAG/VectorRAG,Continue reading “GraphRAG Goes Medical: Introducing MedGraphRAG”

Generative AI models are constantly evolving, and so are the strategies we use to improve their performance. Retrieval-Augmented Generation (RAG) and its advanced iterations, such as GraphRAG and Mixture of Memory Experts, represent approaches that incorporate external knowledge to enhance model capabilities. Combining the strengths of different models has proven to be highly effective. ApproachesContinue reading “Unpacking Model Collaboration: Ensembles, Routers, and Merging”

As I’ve noted in previous posts, RAG and GraphRAG have become popular techniques for many AI teams, and for good reason. They enhance large language models (LLMs) by connecting them with external knowledge. This grounding in factual information is key for improving accuracy and reducing hallucinations. Personally, I’ve found RAG and its variants to beContinue reading “Advancing RAG: Best Practices and Evaluation Frameworks”

Building on the discussions from my previous article analyzing the Republican platform’s stance on Crypto and AI, the Democratic National Committee (DNC) 2024 platform presents a distinctly contrasting vision for AI development and regulation.  In terms of scope and depth, the 2024 DNC platform comes in at a comprehensive 41,000+ words, while the 2024 RNCContinue reading “AI and Technology in the 2024 DNC Platform”

A new survey from RAND sheds light on the root causes of failure for AI projects and offers insights into how they can succeed. AI is often touted as the next frontier in technology, promising to revolutionize industries and drive unprecedented efficiencies. Yet, the stark reality is that many AI projects fail.  With the rapidContinue reading “The Reality Check: Why AI Projects Stumble and How to Succeed”

The State of Generative AI Speech-to-Speech Technology Generative AI-powered speech-to-speech technology is forever changing the way we communicate. This groundbreaking innovation enables real-time transformation of one person’s speech into another’s voice or even a different language, opening up a world of possibilities. From enhancing customer service experiences to creating immersive gaming environments, and even aidingContinue reading “Speaking the Future: Generative AI Speech-to-Speech Systems and Their Applications”

Nvidia and Blackrock’s new paper tackles the challenge of extracting meaningful insights from unstructured financial documents, particularly earnings call transcripts. These documents often contain domain-specific language, varied data formats, and complex relationships that can confound traditional language models. It’s worth noting that the HybridRAG system described in this paper appears to be a research prototypeContinue reading “GraphRAG Meets Finance: Enhancing Unstructured Data Analysis in Earnings Calls”

The AI Risk Repository is an interesting new initiative designed to address the fragmented and inconsistent landscape of AI risk frameworks. This comprehensive and accessible living database is designed to serve as a common frame of reference for understanding and addressing the risks posed by AI systems. Unlike fragmented frameworks that have long plagued theContinue reading “Introducing the AI Risk Repository”