Author Archives: Ben Lorica

As AI co-pilots, virtual assistants, and agents become integral to our daily routines, I’ve been reflecting on how these tools handle our most sensitive queries and requests. Whether it’s confidential medical consultations, private legal advice, or personalized mental health support, how can we trust that our interactions remain truly confidential in an age where digitalContinue reading “Building Trust: Enhancing AI with Private Information Retrieval”

Current vector databases often treat embeddings as standalone entities, detached from their original source data. This separation complicates the management of the relationship between embeddings and the data they represent. It requires additional bookkeeping and synchronization efforts to keep embeddings updated with changes in the source data. This approach weakens context and diminishes the effectivenessContinue reading “From Standalone to Integrated: Evolving Vector Embedding Storage”

Text-to-speech (TTS) technology has become a cornerstone of modern content consumption. In our fast-paced society, audio content has exploded, driven by remarkable advancements in TTS systems that produce lifelike, emotive, and clear voices. This shift has transformed how we interact with information, enabling multitasking and reducing screen fatigue. From articles and audiobooks to virtual assistants,Continue reading “F5-TTS: Where Text-to-Speech Meets Zero-Shot Voice Cloning”

By Ben Lorica and Dean Wampler. Imagine a startup developing AI-powered customer service chatbots, agentic workflows, or retrieval-augmented generation (RAG) applications. As their user base grows, they suddenly face skyrocketing costs and deteriorating response times. This scenario is playing out across industries as companies grapple with the challenges of scaling AI workloads efficiently. At theContinue reading “Overcoming AI Scaling Challenges with Ray Compiled Graph”

OpenAI’s Swarm is an experimental open-source framework designed to orchestrate collections of AI agents working collaboratively toward common goals. The idea is compelling: imagine a swarm of lightweight, specialized agents, each handling a piece of a complex task and seamlessly passing information among themselves through natural language. It’s a vision of distributed intelligence that couldContinue reading “Swarming Toward the Future: Navigating the Promise and Pitfalls of Agentic AI”

In his recent book, The Anxious Generation, Jonathan Haidt presents a compelling argument that smartphones and social media have significantly disrupted childhood development, leading to a rise in mental health issues among adolescents. This stance has prompted debate among researchers, with some questioning the way Haidt interprets the supporting evidence. A recent discussion between HaidtContinue reading “Beyond Screens: Factors Influencing Adolescent Mental Health”

Today’s machine learning landscape is defined by large language and foundation models. Unlike traditional ML models of the past, where models were built and trained for specific tasks from scratch, large language models (LLMs) and foundation models now serve as the starting point for most applications. While pre-trained on massive datasets, foundation models often requireContinue reading “Why Your Company Must Invest in Post-Training”

I’ve previously written about GraphRAG, a form of Retrieval-Augmented Generation (RAG) that combines the power of vector embeddings with knowledge graphs (KG). While traditional RAG relies solely on vector-based similarity searches to retrieve relevant information, GraphRAG introduces knowledge graphs to represent and traverse relationships between entities, concepts, and facts. This combination allows GraphRAG to enhanceContinue reading “Enhancing AI Retrieval Systems with Structured Data”