Andrea De Mauro, PhD

@ademauro.com

788 Followers 582 Following

Data Analytics and AI Exec Advisor and Founder | Book author, published by Pearson, FT, Packt, Apogeo | Professor of AI, Business and Marketing Analytics | Former Exec at Vodafone and P&G | ademauro.com | linkedin.com/in/andread

Diagram illustrating a taxonomy of Generative AI applications in marketing. At the top is “GenAI in Marketing,” branching into two domains: Consumer-facing and Business-facing. Consumer-facing applications include “Improve Shopping Fundamentals” (with Hyper-personalized Communication and Customized Recommendations) and “Improve Consumption Experience” (with Customer Service Chatbots and Content Personalization). Business-facing applications include “Improve Decision Making” (with Market Research and Creativity Boosting) and “Improve Financial Outcomes” (with Dynamic Pricing and Media Optimization). Icons visually represent each of the eight application areas.
Video generated by Sora with prompt: "A Labrador dog with a red tie giving a lecture on marketing at LUISS university in Rome” ". You can see pinus pinaster, the "maritime pine", typical of Rome area outside of the window. How many tails has the Labrador got?
Andrea De Mauro, teaching his students at LUISS University in Rome.
The image presents a framework titled "The Four Levels of RAG" (Retrieval Augmented Generation), which outlines the progressive stages of LLM (Large Language Model) capabilities in integrating external knowledge repositories for enhanced reasoning. The framework spans from "Basic RAG" to "Advanced RAG," showcasing increasing value and complexity over four levels. Level 1: Explicit Fact Retrieval Basic RAG involves locating specific facts within a repository to provide answers. External facts, represented as "chunks," are processed by an LLM. This level is nearly out-of-the-box and requires minimal customization. It has some positive impact. Level 2: Implicit Fact Identification The second level integrates facts dispersed across multiple chunks within a repository. LLMs piece together connected information (e.g., facts "F1, F2, F3") derived from content "C." This level moves beyond simple retrieval by establishing contextual relationships between facts. Level 3: Explicit Rationale Application Advanced RAG begins here. The LLM applies provided rationale frameworks to integrate dispersed facts and generate answers. External rationales are represented through flowcharts or structured logic diagrams, enabling a deeper layer of reasoning. Level 4: Hidden Rationale Identification The most advanced stage, where LLMs analyze facts in a repository and infer hidden rationales to generate answers. An example involves solving a "24-point game," where basic arithmetic operations (e.g., addition, multiplication) are combined strategically to reach the solution. This level requires highly curated knowledge and customized implementations. The chart visually represents value (vertical axis) versus complexity/time to production (horizontal axis). A gradient indicates increasing impact and customization needs across the levels. An upward arrow shows growing potential value. The framework is adapted from Zhao et al., "Retrieval Augmented Generation (RAG) and Beyond," 2024.
Andrea De Mauro and Michele Pacifico, showing their new book on Data and AI, published by Pearson: The Financial Times Guide to Data-Driven Transformation